159bd9dedSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namespace.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 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 */ 23a07b2000SAl Viro #include <linux/file.h> 24d10577a8SAl Viro #include <linux/uaccess.h> 250bb80f24SDavid Howells #include <linux/proc_ns.h> 2620b4fb48SLinus Torvalds #include <linux/magic.h> 2757c8a661SMike Rapoport #include <linux/memblock.h> 289caccd41SChristian Brauner #include <linux/proc_fs.h> 299ea459e1SAl Viro #include <linux/task_work.h> 309164bb4aSIngo Molnar #include <linux/sched/task.h> 31e262e32dSDavid Howells #include <uapi/linux/mount.h> 329bc61ab1SDavid Howells #include <linux/fs_context.h> 33037f11b4SAl Viro #include <linux/shmem_fs.h> 34bd303368SChristian Brauner #include <linux/mnt_idmapping.h> 359164bb4aSIngo Molnar 3607b20889SRam Pai #include "pnode.h" 37948730b0SAdrian Bunk #include "internal.h" 381da177e4SLinus Torvalds 39d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 40ab171b95SLuis Chamberlain static unsigned int sysctl_mount_max __read_mostly = 100000; 41d2921684SEric W. Biederman 420818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 430818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 440818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 450818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 460818bf27SAl Viro 470818bf27SAl Viro static __initdata unsigned long mhash_entries; 480818bf27SAl Viro static int __init set_mhash_entries(char *str) 490818bf27SAl Viro { 500818bf27SAl Viro if (!str) 510818bf27SAl Viro return 0; 520818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 530818bf27SAl Viro return 1; 540818bf27SAl Viro } 550818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 560818bf27SAl Viro 570818bf27SAl Viro static __initdata unsigned long mphash_entries; 580818bf27SAl Viro static int __init set_mphash_entries(char *str) 590818bf27SAl Viro { 600818bf27SAl Viro if (!str) 610818bf27SAl Viro return 0; 620818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 630818bf27SAl Viro return 1; 640818bf27SAl Viro } 650818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6613f14b4dSEric Dumazet 67c7999c36SAl Viro static u64 event; 6873cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 69719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 705addc5ddSAl Viro 7138129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 720818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 73e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 7459aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 754edbe133SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 764edbe133SAl Viro static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ 771da177e4SLinus Torvalds 782a186721SChristian Brauner struct mount_kattr { 792a186721SChristian Brauner unsigned int attr_set; 802a186721SChristian Brauner unsigned int attr_clr; 812a186721SChristian Brauner unsigned int propagation; 822a186721SChristian Brauner unsigned int lookup_flags; 832a186721SChristian Brauner bool recurse; 849caccd41SChristian Brauner struct user_namespace *mnt_userns; 852a186721SChristian Brauner }; 862a186721SChristian Brauner 87f87fd4c2SMiklos Szeredi /* /sys/fs */ 8800d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 8900d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 90f87fd4c2SMiklos Szeredi 9199b7db7bSNick Piggin /* 9299b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 9399b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 9499b7db7bSNick Piggin * up the tree. 9599b7db7bSNick Piggin * 9699b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 9799b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 9899b7db7bSNick Piggin */ 9948a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 10099b7db7bSNick Piggin 101d033cb67SChristian Brauner static inline void lock_mount_hash(void) 102d033cb67SChristian Brauner { 103d033cb67SChristian Brauner write_seqlock(&mount_lock); 104d033cb67SChristian Brauner } 105d033cb67SChristian Brauner 106d033cb67SChristian Brauner static inline void unlock_mount_hash(void) 107d033cb67SChristian Brauner { 108d033cb67SChristian Brauner write_sequnlock(&mount_lock); 109d033cb67SChristian Brauner } 110d033cb67SChristian Brauner 11138129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 1121da177e4SLinus Torvalds { 1131da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 1141da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 1150818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 1160818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 1170818bf27SAl Viro } 1180818bf27SAl Viro 1190818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 1200818bf27SAl Viro { 1210818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 1220818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 1230818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 1241da177e4SLinus Torvalds } 1251da177e4SLinus Torvalds 126b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 12773cd49ecSMiklos Szeredi { 128169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 12973cd49ecSMiklos Szeredi 130169b480eSMatthew Wilcox if (res < 0) 13173cd49ecSMiklos Szeredi return res; 132169b480eSMatthew Wilcox mnt->mnt_id = res; 133169b480eSMatthew Wilcox return 0; 13473cd49ecSMiklos Szeredi } 13573cd49ecSMiklos Szeredi 136b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 13773cd49ecSMiklos Szeredi { 138169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 13973cd49ecSMiklos Szeredi } 14073cd49ecSMiklos Szeredi 141719f5d7fSMiklos Szeredi /* 142719f5d7fSMiklos Szeredi * Allocate a new peer group ID 143719f5d7fSMiklos Szeredi */ 1444b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 145719f5d7fSMiklos Szeredi { 146169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 147f21f6220SAl Viro 148169b480eSMatthew Wilcox if (res < 0) 149f21f6220SAl Viro return res; 150169b480eSMatthew Wilcox mnt->mnt_group_id = res; 151169b480eSMatthew Wilcox return 0; 152719f5d7fSMiklos Szeredi } 153719f5d7fSMiklos Szeredi 154719f5d7fSMiklos Szeredi /* 155719f5d7fSMiklos Szeredi * Release a peer group ID 156719f5d7fSMiklos Szeredi */ 1574b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 158719f5d7fSMiklos Szeredi { 159169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 16015169fe7SAl Viro mnt->mnt_group_id = 0; 161719f5d7fSMiklos Szeredi } 162719f5d7fSMiklos Szeredi 163b3e19d92SNick Piggin /* 164b3e19d92SNick Piggin * vfsmount lock must be held for read 165b3e19d92SNick Piggin */ 16683adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 167b3e19d92SNick Piggin { 168b3e19d92SNick Piggin #ifdef CONFIG_SMP 16968e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 170b3e19d92SNick Piggin #else 171b3e19d92SNick Piggin preempt_disable(); 17268e8a9feSAl Viro mnt->mnt_count += n; 173b3e19d92SNick Piggin preempt_enable(); 174b3e19d92SNick Piggin #endif 175b3e19d92SNick Piggin } 176b3e19d92SNick Piggin 177b3e19d92SNick Piggin /* 178b3e19d92SNick Piggin * vfsmount lock must be held for write 179b3e19d92SNick Piggin */ 180edf7ddbfSEric Biggers int mnt_get_count(struct mount *mnt) 181b3e19d92SNick Piggin { 182b3e19d92SNick Piggin #ifdef CONFIG_SMP 183edf7ddbfSEric Biggers int count = 0; 184b3e19d92SNick Piggin int cpu; 185b3e19d92SNick Piggin 186b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 18768e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 188b3e19d92SNick Piggin } 189b3e19d92SNick Piggin 190b3e19d92SNick Piggin return count; 191b3e19d92SNick Piggin #else 19268e8a9feSAl Viro return mnt->mnt_count; 193b3e19d92SNick Piggin #endif 194b3e19d92SNick Piggin } 195b3e19d92SNick Piggin 196b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1971da177e4SLinus Torvalds { 198c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 199c63181e6SAl Viro if (mnt) { 20073cd49ecSMiklos Szeredi int err; 20173cd49ecSMiklos Szeredi 202c63181e6SAl Viro err = mnt_alloc_id(mnt); 20388b38782SLi Zefan if (err) 20488b38782SLi Zefan goto out_free_cache; 20588b38782SLi Zefan 20688b38782SLi Zefan if (name) { 20779f6540bSVasily Averin mnt->mnt_devname = kstrdup_const(name, 20879f6540bSVasily Averin GFP_KERNEL_ACCOUNT); 209c63181e6SAl Viro if (!mnt->mnt_devname) 21088b38782SLi Zefan goto out_free_id; 21173cd49ecSMiklos Szeredi } 21273cd49ecSMiklos Szeredi 213b3e19d92SNick Piggin #ifdef CONFIG_SMP 214c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 215c63181e6SAl Viro if (!mnt->mnt_pcp) 216b3e19d92SNick Piggin goto out_free_devname; 217b3e19d92SNick Piggin 218c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 219b3e19d92SNick Piggin #else 220c63181e6SAl Viro mnt->mnt_count = 1; 221c63181e6SAl Viro mnt->mnt_writers = 0; 222b3e19d92SNick Piggin #endif 223b3e19d92SNick Piggin 22438129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 225c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 226c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 227c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 228c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 229c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 230c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 231c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2320a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 23399b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 23456cbb429SAl Viro INIT_HLIST_HEAD(&mnt->mnt_stuck_children); 235a6435940SChristian Brauner mnt->mnt.mnt_userns = &init_user_ns; 2361da177e4SLinus Torvalds } 237c63181e6SAl Viro return mnt; 23888b38782SLi Zefan 239d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 240d3ef3d73Snpiggin@suse.de out_free_devname: 241fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 242d3ef3d73Snpiggin@suse.de #endif 24388b38782SLi Zefan out_free_id: 244c63181e6SAl Viro mnt_free_id(mnt); 24588b38782SLi Zefan out_free_cache: 246c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 24788b38782SLi Zefan return NULL; 2481da177e4SLinus Torvalds } 2491da177e4SLinus Torvalds 2508366025eSDave Hansen /* 2518366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2528366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2538366025eSDave Hansen * We must keep track of when those operations start 2548366025eSDave Hansen * (for permission checks) and when they end, so that 2558366025eSDave Hansen * we can determine when writes are able to occur to 2568366025eSDave Hansen * a filesystem. 2578366025eSDave Hansen */ 2583d733633SDave Hansen /* 2593d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2603d733633SDave Hansen * @mnt: the mount to check for its write status 2613d733633SDave Hansen * 2623d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2633d733633SDave Hansen * It does not guarantee that the filesystem will stay 2643d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2653d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2663d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2673d733633SDave Hansen * r/w. 2683d733633SDave Hansen */ 26943f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 2703d733633SDave Hansen { 27143f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 2723d733633SDave Hansen } 2733d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2743d733633SDave Hansen 27583adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2763d733633SDave Hansen { 277d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 27868e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 279d3ef3d73Snpiggin@suse.de #else 28068e8a9feSAl Viro mnt->mnt_writers++; 281d3ef3d73Snpiggin@suse.de #endif 2823d733633SDave Hansen } 2833d733633SDave Hansen 28483adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2853d733633SDave Hansen { 286d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 28768e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 288d3ef3d73Snpiggin@suse.de #else 28968e8a9feSAl Viro mnt->mnt_writers--; 290d3ef3d73Snpiggin@suse.de #endif 291d3ef3d73Snpiggin@suse.de } 292d3ef3d73Snpiggin@suse.de 29383adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 294d3ef3d73Snpiggin@suse.de { 295d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 296d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2973d733633SDave Hansen int cpu; 2983d733633SDave Hansen 2993d733633SDave Hansen for_each_possible_cpu(cpu) { 30068e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 3013d733633SDave Hansen } 3023d733633SDave Hansen 303d3ef3d73Snpiggin@suse.de return count; 304d3ef3d73Snpiggin@suse.de #else 305d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 306d3ef3d73Snpiggin@suse.de #endif 3073d733633SDave Hansen } 3083d733633SDave Hansen 3094ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3104ed5e82fSMiklos Szeredi { 3114ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3124ed5e82fSMiklos Szeredi return 1; 3134ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3144ed5e82fSMiklos Szeredi smp_rmb(); 3154ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3164ed5e82fSMiklos Szeredi } 3174ed5e82fSMiklos Szeredi 3183d733633SDave Hansen /* 319eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 320eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 321eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 322eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3233d733633SDave Hansen */ 3248366025eSDave Hansen /** 325eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 32683adc753SAl Viro * @m: the mount on which to take a write 3278366025eSDave Hansen * 328eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 329eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 330eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 331eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 332eb04c282SJan Kara * called. This is effectively a refcount. 3338366025eSDave Hansen */ 334eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3358366025eSDave Hansen { 33683adc753SAl Viro struct mount *mnt = real_mount(m); 3373d733633SDave Hansen int ret = 0; 3383d733633SDave Hansen 339d3ef3d73Snpiggin@suse.de preempt_disable(); 340c6653a83SNick Piggin mnt_inc_writers(mnt); 341d3ef3d73Snpiggin@suse.de /* 342c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 343d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 344d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 345d3ef3d73Snpiggin@suse.de */ 346d3ef3d73Snpiggin@suse.de smp_mb(); 3470f8821daSSebastian Andrzej Siewior might_lock(&mount_lock.lock); 3480f8821daSSebastian Andrzej Siewior while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) { 3490f8821daSSebastian Andrzej Siewior if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 350d3ef3d73Snpiggin@suse.de cpu_relax(); 3510f8821daSSebastian Andrzej Siewior } else { 3520f8821daSSebastian Andrzej Siewior /* 3530f8821daSSebastian Andrzej Siewior * This prevents priority inversion, if the task 3540f8821daSSebastian Andrzej Siewior * setting MNT_WRITE_HOLD got preempted on a remote 3550f8821daSSebastian Andrzej Siewior * CPU, and it prevents life lock if the task setting 3560f8821daSSebastian Andrzej Siewior * MNT_WRITE_HOLD has a lower priority and is bound to 3570f8821daSSebastian Andrzej Siewior * the same CPU as the task that is spinning here. 3580f8821daSSebastian Andrzej Siewior */ 3590f8821daSSebastian Andrzej Siewior preempt_enable(); 3600f8821daSSebastian Andrzej Siewior lock_mount_hash(); 3610f8821daSSebastian Andrzej Siewior unlock_mount_hash(); 3620f8821daSSebastian Andrzej Siewior preempt_disable(); 3630f8821daSSebastian Andrzej Siewior } 3640f8821daSSebastian Andrzej Siewior } 365d3ef3d73Snpiggin@suse.de /* 366d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 367d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 368d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 369d3ef3d73Snpiggin@suse.de */ 370d3ef3d73Snpiggin@suse.de smp_rmb(); 3714ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 372c6653a83SNick Piggin mnt_dec_writers(mnt); 3733d733633SDave Hansen ret = -EROFS; 3743d733633SDave Hansen } 375d3ef3d73Snpiggin@suse.de preempt_enable(); 376eb04c282SJan Kara 377eb04c282SJan Kara return ret; 378eb04c282SJan Kara } 379eb04c282SJan Kara 380eb04c282SJan Kara /** 381eb04c282SJan Kara * mnt_want_write - get write access to a mount 382eb04c282SJan Kara * @m: the mount on which to take a write 383eb04c282SJan Kara * 384eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 385eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 386eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 387eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 388eb04c282SJan Kara */ 389eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 390eb04c282SJan Kara { 391eb04c282SJan Kara int ret; 392eb04c282SJan Kara 393eb04c282SJan Kara sb_start_write(m->mnt_sb); 394eb04c282SJan Kara ret = __mnt_want_write(m); 395eb04c282SJan Kara if (ret) 396eb04c282SJan Kara sb_end_write(m->mnt_sb); 3973d733633SDave Hansen return ret; 3988366025eSDave Hansen } 3998366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 4008366025eSDave Hansen 4018366025eSDave Hansen /** 402eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 403eb04c282SJan Kara * @file: the file who's mount on which to take a write 404eb04c282SJan Kara * 40514e43bf4SEric Biggers * This is like __mnt_want_write, but if the file is already open for writing it 40614e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 40714e43bf4SEric Biggers * and instead only does the check for emergency r/o remounts. This must be 40814e43bf4SEric Biggers * paired with __mnt_drop_write_file. 409eb04c282SJan Kara */ 410eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 411eb04c282SJan Kara { 41214e43bf4SEric Biggers if (file->f_mode & FMODE_WRITER) { 41314e43bf4SEric Biggers /* 41414e43bf4SEric Biggers * Superblock may have become readonly while there are still 41514e43bf4SEric Biggers * writable fd's, e.g. due to a fs error with errors=remount-ro 41614e43bf4SEric Biggers */ 41714e43bf4SEric Biggers if (__mnt_is_readonly(file->f_path.mnt)) 41814e43bf4SEric Biggers return -EROFS; 41914e43bf4SEric Biggers return 0; 42014e43bf4SEric Biggers } 421eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 422eb04c282SJan Kara } 423eb04c282SJan Kara 424eb04c282SJan Kara /** 4257c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4267c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4277c6893e3SMiklos Szeredi * 42814e43bf4SEric Biggers * This is like mnt_want_write, but if the file is already open for writing it 42914e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 43014e43bf4SEric Biggers * and instead only does the freeze protection and the check for emergency r/o 43114e43bf4SEric Biggers * remounts. This must be paired with mnt_drop_write_file. 4327c6893e3SMiklos Szeredi */ 4337c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4347c6893e3SMiklos Szeredi { 4357c6893e3SMiklos Szeredi int ret; 4367c6893e3SMiklos Szeredi 4377c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4387c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4397c6893e3SMiklos Szeredi if (ret) 4407c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4417c6893e3SMiklos Szeredi return ret; 4427c6893e3SMiklos Szeredi } 44396029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 44496029c4eSnpiggin@suse.de 44596029c4eSnpiggin@suse.de /** 446eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4478366025eSDave Hansen * @mnt: the mount on which to give up write access 4488366025eSDave Hansen * 4498366025eSDave Hansen * Tells the low-level filesystem that we are done 4508366025eSDave Hansen * performing writes to it. Must be matched with 451eb04c282SJan Kara * __mnt_want_write() call above. 4528366025eSDave Hansen */ 453eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4548366025eSDave Hansen { 455d3ef3d73Snpiggin@suse.de preempt_disable(); 45683adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 457d3ef3d73Snpiggin@suse.de preempt_enable(); 4588366025eSDave Hansen } 459eb04c282SJan Kara 460eb04c282SJan Kara /** 461eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 462eb04c282SJan Kara * @mnt: the mount on which to give up write access 463eb04c282SJan Kara * 464eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 465eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 466eb04c282SJan Kara * mnt_want_write() call above. 467eb04c282SJan Kara */ 468eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 469eb04c282SJan Kara { 470eb04c282SJan Kara __mnt_drop_write(mnt); 471eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 472eb04c282SJan Kara } 4738366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4748366025eSDave Hansen 475eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 476eb04c282SJan Kara { 47714e43bf4SEric Biggers if (!(file->f_mode & FMODE_WRITER)) 478eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 479eb04c282SJan Kara } 480eb04c282SJan Kara 4817c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4827c6893e3SMiklos Szeredi { 483a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4847c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4857c6893e3SMiklos Szeredi } 4862a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4872a79f17eSAl Viro 488538f4f02SChristian Brauner /** 489538f4f02SChristian Brauner * mnt_hold_writers - prevent write access to the given mount 490538f4f02SChristian Brauner * @mnt: mnt to prevent write access to 491538f4f02SChristian Brauner * 492538f4f02SChristian Brauner * Prevents write access to @mnt if there are no active writers for @mnt. 493538f4f02SChristian Brauner * This function needs to be called and return successfully before changing 494538f4f02SChristian Brauner * properties of @mnt that need to remain stable for callers with write access 495538f4f02SChristian Brauner * to @mnt. 496538f4f02SChristian Brauner * 497538f4f02SChristian Brauner * After this functions has been called successfully callers must pair it with 498538f4f02SChristian Brauner * a call to mnt_unhold_writers() in order to stop preventing write access to 499538f4f02SChristian Brauner * @mnt. 500538f4f02SChristian Brauner * 501538f4f02SChristian Brauner * Context: This function expects lock_mount_hash() to be held serializing 502538f4f02SChristian Brauner * setting MNT_WRITE_HOLD. 503538f4f02SChristian Brauner * Return: On success 0 is returned. 504538f4f02SChristian Brauner * On error, -EBUSY is returned. 505538f4f02SChristian Brauner */ 506fbdc2f6cSChristian Brauner static inline int mnt_hold_writers(struct mount *mnt) 5078366025eSDave Hansen { 50883adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 509d3ef3d73Snpiggin@suse.de /* 510d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 511d3ef3d73Snpiggin@suse.de * should be visible before we do. 512d3ef3d73Snpiggin@suse.de */ 513d3ef3d73Snpiggin@suse.de smp_mb(); 514d3ef3d73Snpiggin@suse.de 515d3ef3d73Snpiggin@suse.de /* 516d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 517d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 518d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 519d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 520d3ef3d73Snpiggin@suse.de * 521d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 522d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 523d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 524d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 525d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 526d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 527d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 528d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 529d3ef3d73Snpiggin@suse.de * we're counting up here. 530d3ef3d73Snpiggin@suse.de */ 531c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 532fbdc2f6cSChristian Brauner return -EBUSY; 533fbdc2f6cSChristian Brauner 534fbdc2f6cSChristian Brauner return 0; 535fbdc2f6cSChristian Brauner } 536fbdc2f6cSChristian Brauner 537538f4f02SChristian Brauner /** 538538f4f02SChristian Brauner * mnt_unhold_writers - stop preventing write access to the given mount 539538f4f02SChristian Brauner * @mnt: mnt to stop preventing write access to 540538f4f02SChristian Brauner * 541538f4f02SChristian Brauner * Stop preventing write access to @mnt allowing callers to gain write access 542538f4f02SChristian Brauner * to @mnt again. 543538f4f02SChristian Brauner * 544538f4f02SChristian Brauner * This function can only be called after a successful call to 545538f4f02SChristian Brauner * mnt_hold_writers(). 546538f4f02SChristian Brauner * 547538f4f02SChristian Brauner * Context: This function expects lock_mount_hash() to be held. 548538f4f02SChristian Brauner */ 549fbdc2f6cSChristian Brauner static inline void mnt_unhold_writers(struct mount *mnt) 550fbdc2f6cSChristian Brauner { 551d3ef3d73Snpiggin@suse.de /* 552d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 553d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 554d3ef3d73Snpiggin@suse.de */ 555d3ef3d73Snpiggin@suse.de smp_wmb(); 55683adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 557fbdc2f6cSChristian Brauner } 558fbdc2f6cSChristian Brauner 559fbdc2f6cSChristian Brauner static int mnt_make_readonly(struct mount *mnt) 560fbdc2f6cSChristian Brauner { 561fbdc2f6cSChristian Brauner int ret; 562fbdc2f6cSChristian Brauner 563fbdc2f6cSChristian Brauner ret = mnt_hold_writers(mnt); 564fbdc2f6cSChristian Brauner if (!ret) 565fbdc2f6cSChristian Brauner mnt->mnt.mnt_flags |= MNT_READONLY; 566fbdc2f6cSChristian Brauner mnt_unhold_writers(mnt); 5673d733633SDave Hansen return ret; 5683d733633SDave Hansen } 5698366025eSDave Hansen 5704ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5714ed5e82fSMiklos Szeredi { 5724ed5e82fSMiklos Szeredi struct mount *mnt; 5734ed5e82fSMiklos Szeredi int err = 0; 5744ed5e82fSMiklos Szeredi 5758e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5768e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5778e8b8796SMiklos Szeredi return -EBUSY; 5788e8b8796SMiklos Szeredi 579719ea2fbSAl Viro lock_mount_hash(); 5804ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5814ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 582ad1844a0SChristian Brauner err = mnt_hold_writers(mnt); 583ad1844a0SChristian Brauner if (err) 5844ed5e82fSMiklos Szeredi break; 5854ed5e82fSMiklos Szeredi } 5864ed5e82fSMiklos Szeredi } 5878e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5888e8b8796SMiklos Szeredi err = -EBUSY; 5898e8b8796SMiklos Szeredi 5904ed5e82fSMiklos Szeredi if (!err) { 5914ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5924ed5e82fSMiklos Szeredi smp_wmb(); 5934ed5e82fSMiklos Szeredi } 5944ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5954ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5964ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5974ed5e82fSMiklos Szeredi } 598719ea2fbSAl Viro unlock_mount_hash(); 5994ed5e82fSMiklos Szeredi 6004ed5e82fSMiklos Szeredi return err; 6014ed5e82fSMiklos Szeredi } 6024ed5e82fSMiklos Szeredi 603b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 6041da177e4SLinus Torvalds { 605a6435940SChristian Brauner struct user_namespace *mnt_userns; 606a6435940SChristian Brauner 607a6435940SChristian Brauner mnt_userns = mnt_user_ns(&mnt->mnt); 608bd303368SChristian Brauner if (!initial_idmapping(mnt_userns)) 609a6435940SChristian Brauner put_user_ns(mnt_userns); 610fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 611d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 61268e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 613d3ef3d73Snpiggin@suse.de #endif 614b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 6151da177e4SLinus Torvalds } 6161da177e4SLinus Torvalds 6178ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 6188ffcb32eSDavid Howells { 6198ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 6208ffcb32eSDavid Howells } 6218ffcb32eSDavid Howells 62248a066e7SAl Viro /* call under rcu_read_lock */ 623294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 62448a066e7SAl Viro { 62548a066e7SAl Viro struct mount *mnt; 62648a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 627294d71ffSAl Viro return 1; 62848a066e7SAl Viro if (bastard == NULL) 629294d71ffSAl Viro return 0; 63048a066e7SAl Viro mnt = real_mount(bastard); 63148a066e7SAl Viro mnt_add_count(mnt, 1); 632119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 63348a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 634294d71ffSAl Viro return 0; 63548a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 63648a066e7SAl Viro mnt_add_count(mnt, -1); 637294d71ffSAl Viro return 1; 63848a066e7SAl Viro } 639119e1ef8SAl Viro lock_mount_hash(); 640119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 641119e1ef8SAl Viro mnt_add_count(mnt, -1); 642119e1ef8SAl Viro unlock_mount_hash(); 643119e1ef8SAl Viro return 1; 644119e1ef8SAl Viro } 645119e1ef8SAl Viro unlock_mount_hash(); 646119e1ef8SAl Viro /* caller will mntput() */ 647294d71ffSAl Viro return -1; 648294d71ffSAl Viro } 649294d71ffSAl Viro 650294d71ffSAl Viro /* call under rcu_read_lock */ 6517e4745a0SAl Viro static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 652294d71ffSAl Viro { 653294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 654294d71ffSAl Viro if (likely(!res)) 655294d71ffSAl Viro return true; 656294d71ffSAl Viro if (unlikely(res < 0)) { 65748a066e7SAl Viro rcu_read_unlock(); 65848a066e7SAl Viro mntput(bastard); 65948a066e7SAl Viro rcu_read_lock(); 660294d71ffSAl Viro } 66148a066e7SAl Viro return false; 66248a066e7SAl Viro } 66348a066e7SAl Viro 6641da177e4SLinus Torvalds /* 665474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 66648a066e7SAl Viro * call under rcu_read_lock() 6671da177e4SLinus Torvalds */ 668474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6691da177e4SLinus Torvalds { 67038129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 671474279dcSAl Viro struct mount *p; 6721da177e4SLinus Torvalds 67338129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 674474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 675474279dcSAl Viro return p; 676474279dcSAl Viro return NULL; 6771da177e4SLinus Torvalds } 678474279dcSAl Viro 679474279dcSAl Viro /* 680f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 681f015f126SDavid Howells * 682f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 683f015f126SDavid Howells * following mounts: 684f015f126SDavid Howells * 685f015f126SDavid Howells * mount /dev/sda1 /mnt 686f015f126SDavid Howells * mount /dev/sda2 /mnt 687f015f126SDavid Howells * mount /dev/sda3 /mnt 688f015f126SDavid Howells * 689f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 690f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 691f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 692f015f126SDavid Howells * 693f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 694a05964f3SRam Pai */ 695ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 696a05964f3SRam Pai { 697c7105365SAl Viro struct mount *child_mnt; 69848a066e7SAl Viro struct vfsmount *m; 69948a066e7SAl Viro unsigned seq; 70099b7db7bSNick Piggin 70148a066e7SAl Viro rcu_read_lock(); 70248a066e7SAl Viro do { 70348a066e7SAl Viro seq = read_seqbegin(&mount_lock); 704474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 70548a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 70648a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 70748a066e7SAl Viro rcu_read_unlock(); 70848a066e7SAl Viro return m; 709a05964f3SRam Pai } 710a05964f3SRam Pai 7119f6c61f9SMiklos Szeredi static inline void lock_ns_list(struct mnt_namespace *ns) 7129f6c61f9SMiklos Szeredi { 7139f6c61f9SMiklos Szeredi spin_lock(&ns->ns_lock); 7149f6c61f9SMiklos Szeredi } 7159f6c61f9SMiklos Szeredi 7169f6c61f9SMiklos Szeredi static inline void unlock_ns_list(struct mnt_namespace *ns) 7179f6c61f9SMiklos Szeredi { 7189f6c61f9SMiklos Szeredi spin_unlock(&ns->ns_lock); 7199f6c61f9SMiklos Szeredi } 7209f6c61f9SMiklos Szeredi 7219f6c61f9SMiklos Szeredi static inline bool mnt_is_cursor(struct mount *mnt) 7229f6c61f9SMiklos Szeredi { 7239f6c61f9SMiklos Szeredi return mnt->mnt.mnt_flags & MNT_CURSOR; 7249f6c61f9SMiklos Szeredi } 7259f6c61f9SMiklos Szeredi 7267af1364fSEric W. Biederman /* 7277af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 7287af1364fSEric W. Biederman * current mount namespace. 7297af1364fSEric W. Biederman * 7307af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 7317af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 7327af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 7337af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 7347af1364fSEric W. Biederman * is a mountpoint. 7357af1364fSEric W. Biederman * 7367af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 7377af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 7387af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 7397af1364fSEric W. Biederman * parent mount. 7407af1364fSEric W. Biederman */ 7417af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 7427af1364fSEric W. Biederman { 7437af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7447af1364fSEric W. Biederman struct mount *mnt; 7457af1364fSEric W. Biederman bool is_covered = false; 7467af1364fSEric W. Biederman 7477af1364fSEric W. Biederman down_read(&namespace_sem); 7489f6c61f9SMiklos Szeredi lock_ns_list(ns); 7497af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7509f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 7519f6c61f9SMiklos Szeredi continue; 7527af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7537af1364fSEric W. Biederman if (is_covered) 7547af1364fSEric W. Biederman break; 7557af1364fSEric W. Biederman } 7569f6c61f9SMiklos Szeredi unlock_ns_list(ns); 7577af1364fSEric W. Biederman up_read(&namespace_sem); 7585ad05cc8SNikolay Borisov 7597af1364fSEric W. Biederman return is_covered; 7607af1364fSEric W. Biederman } 7617af1364fSEric W. Biederman 762e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 76384d17192SAl Viro { 7640818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 76584d17192SAl Viro struct mountpoint *mp; 76684d17192SAl Viro 7670818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 76884d17192SAl Viro if (mp->m_dentry == dentry) { 76984d17192SAl Viro mp->m_count++; 77084d17192SAl Viro return mp; 77184d17192SAl Viro } 77284d17192SAl Viro } 773e2dfa935SEric W. Biederman return NULL; 774e2dfa935SEric W. Biederman } 775e2dfa935SEric W. Biederman 7763895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 777e2dfa935SEric W. Biederman { 7783895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 779e2dfa935SEric W. Biederman int ret; 78084d17192SAl Viro 7813895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7821e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7831e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7841e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7853895dbf8SEric W. Biederman mountpoint: 7863895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7873895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7883895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7893895dbf8SEric W. Biederman if (mp) 7903895dbf8SEric W. Biederman goto done; 79184d17192SAl Viro } 792eed81007SMiklos Szeredi 7933895dbf8SEric W. Biederman if (!new) 7943895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7953895dbf8SEric W. Biederman if (!new) 7963895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7973895dbf8SEric W. Biederman 7983895dbf8SEric W. Biederman 7993895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 8003895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 8013895dbf8SEric W. Biederman 8023895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 8033895dbf8SEric W. Biederman if (ret == -EBUSY) 8043895dbf8SEric W. Biederman goto mountpoint; 8053895dbf8SEric W. Biederman 8063895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 8073895dbf8SEric W. Biederman mp = ERR_PTR(ret); 8083895dbf8SEric W. Biederman if (ret) 8093895dbf8SEric W. Biederman goto done; 8103895dbf8SEric W. Biederman 8113895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 8123895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 8134edbe133SAl Viro new->m_dentry = dget(dentry); 8143895dbf8SEric W. Biederman new->m_count = 1; 8153895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 8163895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 8173895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 8183895dbf8SEric W. Biederman 8193895dbf8SEric W. Biederman mp = new; 8203895dbf8SEric W. Biederman new = NULL; 8213895dbf8SEric W. Biederman done: 8223895dbf8SEric W. Biederman kfree(new); 82384d17192SAl Viro return mp; 82484d17192SAl Viro } 82584d17192SAl Viro 8264edbe133SAl Viro /* 8274edbe133SAl Viro * vfsmount lock must be held. Additionally, the caller is responsible 8284edbe133SAl Viro * for serializing calls for given disposal list. 8294edbe133SAl Viro */ 8304edbe133SAl Viro static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) 83184d17192SAl Viro { 83284d17192SAl Viro if (!--mp->m_count) { 83384d17192SAl Viro struct dentry *dentry = mp->m_dentry; 8340a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 83584d17192SAl Viro spin_lock(&dentry->d_lock); 83684d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 83784d17192SAl Viro spin_unlock(&dentry->d_lock); 8384edbe133SAl Viro dput_to_list(dentry, list); 8390818bf27SAl Viro hlist_del(&mp->m_hash); 84084d17192SAl Viro kfree(mp); 84184d17192SAl Viro } 84284d17192SAl Viro } 84384d17192SAl Viro 8444edbe133SAl Viro /* called with namespace_lock and vfsmount lock */ 8454edbe133SAl Viro static void put_mountpoint(struct mountpoint *mp) 8464edbe133SAl Viro { 8474edbe133SAl Viro __put_mountpoint(mp, &ex_mountpoints); 8484edbe133SAl Viro } 8494edbe133SAl Viro 850143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 8511da177e4SLinus Torvalds { 8526b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 8531da177e4SLinus Torvalds } 8541da177e4SLinus Torvalds 85599b7db7bSNick Piggin /* 85699b7db7bSNick Piggin * vfsmount lock must be held for write 85799b7db7bSNick Piggin */ 8586b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 8595addc5ddSAl Viro { 8605addc5ddSAl Viro if (ns) { 8615addc5ddSAl Viro ns->event = ++event; 8625addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8635addc5ddSAl Viro } 8645addc5ddSAl Viro } 8655addc5ddSAl Viro 86699b7db7bSNick Piggin /* 86799b7db7bSNick Piggin * vfsmount lock must be held for write 86899b7db7bSNick Piggin */ 8696b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8705addc5ddSAl Viro { 8715addc5ddSAl Viro if (ns && ns->event != event) { 8725addc5ddSAl Viro ns->event = event; 8735addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8745addc5ddSAl Viro } 8755addc5ddSAl Viro } 8765addc5ddSAl Viro 87799b7db7bSNick Piggin /* 87899b7db7bSNick Piggin * vfsmount lock must be held for write 87999b7db7bSNick Piggin */ 880e4e59906SAl Viro static struct mountpoint *unhash_mnt(struct mount *mnt) 8811da177e4SLinus Torvalds { 882e4e59906SAl Viro struct mountpoint *mp; 8830714a533SAl Viro mnt->mnt_parent = mnt; 884a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8856b41d536SAl Viro list_del_init(&mnt->mnt_child); 88638129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8870a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 888e4e59906SAl Viro mp = mnt->mnt_mp; 88984d17192SAl Viro mnt->mnt_mp = NULL; 890e4e59906SAl Viro return mp; 8917bdb11deSEric W. Biederman } 8927bdb11deSEric W. Biederman 8937bdb11deSEric W. Biederman /* 8947bdb11deSEric W. Biederman * vfsmount lock must be held for write 8957bdb11deSEric W. Biederman */ 8966a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8976a46c573SEric W. Biederman { 898e4e59906SAl Viro put_mountpoint(unhash_mnt(mnt)); 8996a46c573SEric W. Biederman } 9006a46c573SEric W. Biederman 9016a46c573SEric W. Biederman /* 9026a46c573SEric W. Biederman * vfsmount lock must be held for write 9036a46c573SEric W. Biederman */ 90484d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 90584d17192SAl Viro struct mountpoint *mp, 90644d964d6SAl Viro struct mount *child_mnt) 907b90fa9aeSRam Pai { 90884d17192SAl Viro mp->m_count++; 9093a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 9104edbe133SAl Viro child_mnt->mnt_mountpoint = mp->m_dentry; 9113a2393d7SAl Viro child_mnt->mnt_parent = mnt; 91284d17192SAl Viro child_mnt->mnt_mp = mp; 9130a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 914b90fa9aeSRam Pai } 915b90fa9aeSRam Pai 9161064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 9171064f874SEric W. Biederman { 9181064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 9191064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 9201064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 9211064f874SEric W. Biederman } 9221064f874SEric W. Biederman 92399b7db7bSNick Piggin /* 92499b7db7bSNick Piggin * vfsmount lock must be held for write 92599b7db7bSNick Piggin */ 92684d17192SAl Viro static void attach_mnt(struct mount *mnt, 92784d17192SAl Viro struct mount *parent, 92884d17192SAl Viro struct mountpoint *mp) 9291da177e4SLinus Torvalds { 93084d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 9311064f874SEric W. Biederman __attach_mnt(mnt, parent); 932b90fa9aeSRam Pai } 933b90fa9aeSRam Pai 9341064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 93512a5b529SAl Viro { 9361064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 9371064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 9381064f874SEric W. Biederman 9391064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 9401064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 9411064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 9421064f874SEric W. Biederman 9431064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 9441064f874SEric W. Biederman 9451064f874SEric W. Biederman put_mountpoint(old_mp); 9461064f874SEric W. Biederman mnt_add_count(old_parent, -1); 94712a5b529SAl Viro } 94812a5b529SAl Viro 949b90fa9aeSRam Pai /* 95099b7db7bSNick Piggin * vfsmount lock must be held for write 951b90fa9aeSRam Pai */ 9521064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 953b90fa9aeSRam Pai { 9540714a533SAl Viro struct mount *parent = mnt->mnt_parent; 95583adc753SAl Viro struct mount *m; 956b90fa9aeSRam Pai LIST_HEAD(head); 957143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 958b90fa9aeSRam Pai 9590714a533SAl Viro BUG_ON(parent == mnt); 960b90fa9aeSRam Pai 9611a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 962f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 963143c8c91SAl Viro m->mnt_ns = n; 964f03c6599SAl Viro 965b90fa9aeSRam Pai list_splice(&head, n->list.prev); 966b90fa9aeSRam Pai 967d2921684SEric W. Biederman n->mounts += n->pending_mounts; 968d2921684SEric W. Biederman n->pending_mounts = 0; 969d2921684SEric W. Biederman 9701064f874SEric W. Biederman __attach_mnt(mnt, parent); 9716b3286edSKirill Korotaev touch_mnt_namespace(n); 9721da177e4SLinus Torvalds } 9731da177e4SLinus Torvalds 974909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9751da177e4SLinus Torvalds { 9766b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9776b41d536SAl Viro if (next == &p->mnt_mounts) { 9781da177e4SLinus Torvalds while (1) { 979909b0a88SAl Viro if (p == root) 9801da177e4SLinus Torvalds return NULL; 9816b41d536SAl Viro next = p->mnt_child.next; 9826b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9831da177e4SLinus Torvalds break; 9840714a533SAl Viro p = p->mnt_parent; 9851da177e4SLinus Torvalds } 9861da177e4SLinus Torvalds } 9876b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9881da177e4SLinus Torvalds } 9891da177e4SLinus Torvalds 990315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9919676f0c6SRam Pai { 9926b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9936b41d536SAl Viro while (prev != &p->mnt_mounts) { 9946b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9956b41d536SAl Viro prev = p->mnt_mounts.prev; 9969676f0c6SRam Pai } 9979676f0c6SRam Pai return p; 9989676f0c6SRam Pai } 9999676f0c6SRam Pai 10008f291889SAl Viro /** 10018f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 10028f291889SAl Viro * @fc: The configuration context with the superblock attached 10038f291889SAl Viro * 10048f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 10058f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 10068f291889SAl Viro * 10078f291889SAl Viro * Note that this does not attach the mount to anything. 10088f291889SAl Viro */ 10098f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 10109d412a43SAl Viro { 1011b105e270SAl Viro struct mount *mnt; 1012bd303368SChristian Brauner struct user_namespace *fs_userns; 10139d412a43SAl Viro 10148f291889SAl Viro if (!fc->root) 10158f291889SAl Viro return ERR_PTR(-EINVAL); 10169d412a43SAl Viro 10178f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 10189d412a43SAl Viro if (!mnt) 10199d412a43SAl Viro return ERR_PTR(-ENOMEM); 10209d412a43SAl Viro 10218f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 1022b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 10239d412a43SAl Viro 10248f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 10258f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 10268f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 1027a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10280714a533SAl Viro mnt->mnt_parent = mnt; 10298f291889SAl Viro 1030bd303368SChristian Brauner fs_userns = mnt->mnt.mnt_sb->s_user_ns; 1031bd303368SChristian Brauner if (!initial_idmapping(fs_userns)) 1032bd303368SChristian Brauner mnt->mnt.mnt_userns = get_user_ns(fs_userns); 1033bd303368SChristian Brauner 1034719ea2fbSAl Viro lock_mount_hash(); 10358f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 1036719ea2fbSAl Viro unlock_mount_hash(); 1037b105e270SAl Viro return &mnt->mnt; 10389d412a43SAl Viro } 10398f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 10408f291889SAl Viro 10418f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 10428f291889SAl Viro { 10438f291889SAl Viro int err = vfs_get_tree(fc); 10448f291889SAl Viro if (!err) { 10458f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 10468f291889SAl Viro return vfs_create_mount(fc); 10478f291889SAl Viro } 10488f291889SAl Viro return ERR_PTR(err); 10498f291889SAl Viro } 10508f291889SAl Viro EXPORT_SYMBOL(fc_mount); 10518f291889SAl Viro 10529bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 10539bc61ab1SDavid Howells int flags, const char *name, 10549bc61ab1SDavid Howells void *data) 10551da177e4SLinus Torvalds { 10569bc61ab1SDavid Howells struct fs_context *fc; 10578f291889SAl Viro struct vfsmount *mnt; 10589bc61ab1SDavid Howells int ret = 0; 10599d412a43SAl Viro 10609d412a43SAl Viro if (!type) 10613e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 10629d412a43SAl Viro 10639bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 10649bc61ab1SDavid Howells if (IS_ERR(fc)) 10659bc61ab1SDavid Howells return ERR_CAST(fc); 10669bc61ab1SDavid Howells 10673e1aeb00SDavid Howells if (name) 10683e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 10693e1aeb00SDavid Howells name, strlen(name)); 10709bc61ab1SDavid Howells if (!ret) 10719bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 10729bc61ab1SDavid Howells if (!ret) 10738f291889SAl Viro mnt = fc_mount(fc); 10748f291889SAl Viro else 10758f291889SAl Viro mnt = ERR_PTR(ret); 10769d412a43SAl Viro 10779bc61ab1SDavid Howells put_fs_context(fc); 10788f291889SAl Viro return mnt; 10799d412a43SAl Viro } 10809d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10819d412a43SAl Viro 108293faccbbSEric W. Biederman struct vfsmount * 108393faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 108493faccbbSEric W. Biederman const char *name, void *data) 108593faccbbSEric W. Biederman { 108693faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 108793faccbbSEric W. Biederman * through from the parent mount to the submount don't support 108893faccbbSEric W. Biederman * unprivileged mounts with submounts. 108993faccbbSEric W. Biederman */ 109093faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 109193faccbbSEric W. Biederman return ERR_PTR(-EPERM); 109293faccbbSEric W. Biederman 1093e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 109493faccbbSEric W. Biederman } 109593faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 109693faccbbSEric W. Biederman 109787129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 109836341f64SRam Pai int flag) 10991da177e4SLinus Torvalds { 110087129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1101be34d1a3SDavid Howells struct mount *mnt; 1102be34d1a3SDavid Howells int err; 11031da177e4SLinus Torvalds 1104be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1105be34d1a3SDavid Howells if (!mnt) 1106be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1107be34d1a3SDavid Howells 11087a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 110915169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1110719f5d7fSMiklos Szeredi else 111115169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1112719f5d7fSMiklos Szeredi 111315169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1114be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1115719f5d7fSMiklos Szeredi if (err) 1116719f5d7fSMiklos Szeredi goto out_free; 1117719f5d7fSMiklos Szeredi } 1118719f5d7fSMiklos Szeredi 111916a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 112016a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 11215ff9d8a6SEric W. Biederman 11221da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1123a6435940SChristian Brauner mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt); 1124bd303368SChristian Brauner if (!initial_idmapping(mnt->mnt.mnt_userns)) 1125a6435940SChristian Brauner mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns); 1126b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1127b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1128a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 11290714a533SAl Viro mnt->mnt_parent = mnt; 1130719ea2fbSAl Viro lock_mount_hash(); 113139f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1132719ea2fbSAl Viro unlock_mount_hash(); 1133b90fa9aeSRam Pai 11347a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 11357a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 11366776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 113732301920SAl Viro mnt->mnt_master = old; 1138fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 11398aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1140fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 11416776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1142d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 11436776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1144d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 11455235d448SAl Viro } else { 11465235d448SAl Viro CLEAR_MNT_SHARED(mnt); 11475afe0022SRam Pai } 1148b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 11490f0afb1dSAl Viro set_mnt_shared(mnt); 11501da177e4SLinus Torvalds 11511da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 11521da177e4SLinus Torvalds * as the original if that was on one */ 115336341f64SRam Pai if (flag & CL_EXPIRE) { 11546776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 11556776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 11561da177e4SLinus Torvalds } 1157be34d1a3SDavid Howells 1158cb338d06SAl Viro return mnt; 1159719f5d7fSMiklos Szeredi 1160719f5d7fSMiklos Szeredi out_free: 11618ffcb32eSDavid Howells mnt_free_id(mnt); 1162719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1163be34d1a3SDavid Howells return ERR_PTR(err); 11641da177e4SLinus Torvalds } 11651da177e4SLinus Torvalds 11669ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11679ea459e1SAl Viro { 116856cbb429SAl Viro struct hlist_node *p; 116956cbb429SAl Viro struct mount *m; 11709ea459e1SAl Viro /* 117156cbb429SAl Viro * The warning here probably indicates that somebody messed 117256cbb429SAl Viro * up a mnt_want/drop_write() pair. If this happens, the 117356cbb429SAl Viro * filesystem was probably unable to make r/w->r/o transitions. 11749ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11759ea459e1SAl Viro * so mnt_get_writers() below is safe. 11769ea459e1SAl Viro */ 11779ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11789ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11799ea459e1SAl Viro mnt_pin_kill(mnt); 118056cbb429SAl Viro hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { 118156cbb429SAl Viro hlist_del(&m->mnt_umount); 118256cbb429SAl Viro mntput(&m->mnt); 118356cbb429SAl Viro } 11849ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11859ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11869ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11879ea459e1SAl Viro mnt_free_id(mnt); 11889ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11899ea459e1SAl Viro } 11909ea459e1SAl Viro 11919ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11929ea459e1SAl Viro { 11939ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11949ea459e1SAl Viro } 11959ea459e1SAl Viro 11969ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11979ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11989ea459e1SAl Viro { 11999ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 120029785735SByungchul Park struct mount *m, *t; 12019ea459e1SAl Viro 120229785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 120329785735SByungchul Park cleanup_mnt(m); 12049ea459e1SAl Viro } 12059ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 12069ea459e1SAl Viro 1207900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 12087b7b1aceSAl Viro { 12094edbe133SAl Viro LIST_HEAD(list); 1210edf7ddbfSEric Biggers int count; 12114edbe133SAl Viro 121248a066e7SAl Viro rcu_read_lock(); 12139ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 12149ea0a46cSAl Viro /* 12159ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 12169ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 12179ea0a46cSAl Viro * non-NULL, then there's a reference that won't 12189ea0a46cSAl Viro * be dropped until after an RCU delay done after 12199ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 12209ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 12219ea0a46cSAl Viro * we are dropping is not the final one. 12229ea0a46cSAl Viro */ 1223aa9c0e07SAl Viro mnt_add_count(mnt, -1); 122448a066e7SAl Viro rcu_read_unlock(); 122599b7db7bSNick Piggin return; 1226b3e19d92SNick Piggin } 1227719ea2fbSAl Viro lock_mount_hash(); 1228119e1ef8SAl Viro /* 1229119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1230119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1231119e1ef8SAl Viro */ 1232119e1ef8SAl Viro smp_mb(); 12339ea0a46cSAl Viro mnt_add_count(mnt, -1); 1234edf7ddbfSEric Biggers count = mnt_get_count(mnt); 1235edf7ddbfSEric Biggers if (count != 0) { 1236edf7ddbfSEric Biggers WARN_ON(count < 0); 123748a066e7SAl Viro rcu_read_unlock(); 1238719ea2fbSAl Viro unlock_mount_hash(); 123999b7db7bSNick Piggin return; 124099b7db7bSNick Piggin } 124148a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 124248a066e7SAl Viro rcu_read_unlock(); 124348a066e7SAl Viro unlock_mount_hash(); 124448a066e7SAl Viro return; 124548a066e7SAl Viro } 124648a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 124748a066e7SAl Viro rcu_read_unlock(); 1248962830dfSAndi Kleen 124939f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1250ce07d891SEric W. Biederman 1251ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1252ce07d891SEric W. Biederman struct mount *p, *tmp; 1253ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 12544edbe133SAl Viro __put_mountpoint(unhash_mnt(p), &list); 125556cbb429SAl Viro hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); 1256ce07d891SEric W. Biederman } 1257ce07d891SEric W. Biederman } 1258719ea2fbSAl Viro unlock_mount_hash(); 12594edbe133SAl Viro shrink_dentry_list(&list); 1260649a795aSAl Viro 12619ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 12629ea459e1SAl Viro struct task_struct *task = current; 12639ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 12649ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 126591989c70SJens Axboe if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) 12669ea459e1SAl Viro return; 12679ea459e1SAl Viro } 12689ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 12699ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 12709ea459e1SAl Viro return; 12719ea459e1SAl Viro } 12729ea459e1SAl Viro cleanup_mnt(mnt); 1273b3e19d92SNick Piggin } 1274b3e19d92SNick Piggin 1275b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1276b3e19d92SNick Piggin { 1277b3e19d92SNick Piggin if (mnt) { 1278863d684fSAl Viro struct mount *m = real_mount(mnt); 1279b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1280863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1281863d684fSAl Viro m->mnt_expiry_mark = 0; 1282863d684fSAl Viro mntput_no_expire(m); 1283b3e19d92SNick Piggin } 1284b3e19d92SNick Piggin } 1285b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1286b3e19d92SNick Piggin 1287b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1288b3e19d92SNick Piggin { 1289b3e19d92SNick Piggin if (mnt) 129083adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1291b3e19d92SNick Piggin return mnt; 1292b3e19d92SNick Piggin } 1293b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1294b3e19d92SNick Piggin 12951f287bc4SRandy Dunlap /** 12961f287bc4SRandy Dunlap * path_is_mountpoint() - Check if path is a mount in the current namespace. 12971f287bc4SRandy Dunlap * @path: path to check 1298c6609c0aSIan Kent * 1299c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1300c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1301c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1302c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1303c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1304c6609c0aSIan Kent * alone. 1305c6609c0aSIan Kent */ 1306c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1307c6609c0aSIan Kent { 1308c6609c0aSIan Kent unsigned seq; 1309c6609c0aSIan Kent bool res; 1310c6609c0aSIan Kent 1311c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1312c6609c0aSIan Kent return false; 1313c6609c0aSIan Kent 1314c6609c0aSIan Kent rcu_read_lock(); 1315c6609c0aSIan Kent do { 1316c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1317c6609c0aSIan Kent res = __path_is_mountpoint(path); 1318c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1319c6609c0aSIan Kent rcu_read_unlock(); 1320c6609c0aSIan Kent 1321c6609c0aSIan Kent return res; 1322c6609c0aSIan Kent } 1323c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1324c6609c0aSIan Kent 1325ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 13267b7b1aceSAl Viro { 13273064c356SAl Viro struct mount *p; 13283064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 13293064c356SAl Viro if (IS_ERR(p)) 13303064c356SAl Viro return ERR_CAST(p); 13313064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 13323064c356SAl Viro return &p->mnt; 13337b7b1aceSAl Viro } 13341da177e4SLinus Torvalds 1335a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 13369f6c61f9SMiklos Szeredi static struct mount *mnt_list_next(struct mnt_namespace *ns, 13379f6c61f9SMiklos Szeredi struct list_head *p) 13389f6c61f9SMiklos Szeredi { 13399f6c61f9SMiklos Szeredi struct mount *mnt, *ret = NULL; 13409f6c61f9SMiklos Szeredi 13419f6c61f9SMiklos Szeredi lock_ns_list(ns); 13429f6c61f9SMiklos Szeredi list_for_each_continue(p, &ns->list) { 13439f6c61f9SMiklos Szeredi mnt = list_entry(p, typeof(*mnt), mnt_list); 13449f6c61f9SMiklos Szeredi if (!mnt_is_cursor(mnt)) { 13459f6c61f9SMiklos Szeredi ret = mnt; 13469f6c61f9SMiklos Szeredi break; 13479f6c61f9SMiklos Szeredi } 13489f6c61f9SMiklos Szeredi } 13499f6c61f9SMiklos Szeredi unlock_ns_list(ns); 13509f6c61f9SMiklos Szeredi 13519f6c61f9SMiklos Szeredi return ret; 13529f6c61f9SMiklos Szeredi } 13539f6c61f9SMiklos Szeredi 13540226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 13551da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 13561da177e4SLinus Torvalds { 1357ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13589f6c61f9SMiklos Szeredi struct list_head *prev; 13591da177e4SLinus Torvalds 1360390c6843SRam Pai down_read(&namespace_sem); 13619f6c61f9SMiklos Szeredi if (!*pos) { 13629f6c61f9SMiklos Szeredi prev = &p->ns->list; 13639f6c61f9SMiklos Szeredi } else { 13649f6c61f9SMiklos Szeredi prev = &p->cursor.mnt_list; 13659f6c61f9SMiklos Szeredi 13669f6c61f9SMiklos Szeredi /* Read after we'd reached the end? */ 13679f6c61f9SMiklos Szeredi if (list_empty(prev)) 13689f6c61f9SMiklos Szeredi return NULL; 1369c7999c36SAl Viro } 1370c7999c36SAl Viro 13719f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, prev); 13721da177e4SLinus Torvalds } 13731da177e4SLinus Torvalds 13741da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13751da177e4SLinus Torvalds { 1376ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13779f6c61f9SMiklos Szeredi struct mount *mnt = v; 1378b0765fb8SPavel Emelianov 13799f6c61f9SMiklos Szeredi ++*pos; 13809f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, &mnt->mnt_list); 13811da177e4SLinus Torvalds } 13821da177e4SLinus Torvalds 13831da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13841da177e4SLinus Torvalds { 13859f6c61f9SMiklos Szeredi struct proc_mounts *p = m->private; 13869f6c61f9SMiklos Szeredi struct mount *mnt = v; 13879f6c61f9SMiklos Szeredi 13889f6c61f9SMiklos Szeredi lock_ns_list(p->ns); 13899f6c61f9SMiklos Szeredi if (mnt) 13909f6c61f9SMiklos Szeredi list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); 13919f6c61f9SMiklos Szeredi else 13929f6c61f9SMiklos Szeredi list_del_init(&p->cursor.mnt_list); 13939f6c61f9SMiklos Szeredi unlock_ns_list(p->ns); 1394390c6843SRam Pai up_read(&namespace_sem); 13951da177e4SLinus Torvalds } 13961da177e4SLinus Torvalds 13970226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13989f5596afSAl Viro { 1399ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 14009f6c61f9SMiklos Szeredi struct mount *r = v; 14010226f492SAl Viro return p->show(m, &r->mnt); 14021da177e4SLinus Torvalds } 14031da177e4SLinus Torvalds 1404a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 14051da177e4SLinus Torvalds .start = m_start, 14061da177e4SLinus Torvalds .next = m_next, 14071da177e4SLinus Torvalds .stop = m_stop, 14080226f492SAl Viro .show = m_show, 1409b4629fe2SChuck Lever }; 14109f6c61f9SMiklos Szeredi 14119f6c61f9SMiklos Szeredi void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) 14129f6c61f9SMiklos Szeredi { 14139f6c61f9SMiklos Szeredi down_read(&namespace_sem); 14149f6c61f9SMiklos Szeredi lock_ns_list(ns); 14159f6c61f9SMiklos Szeredi list_del(&cursor->mnt_list); 14169f6c61f9SMiklos Szeredi unlock_ns_list(ns); 14179f6c61f9SMiklos Szeredi up_read(&namespace_sem); 14189f6c61f9SMiklos Szeredi } 1419a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1420b4629fe2SChuck Lever 14211da177e4SLinus Torvalds /** 14221da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 14231f287bc4SRandy Dunlap * @m: root of mount tree 14241da177e4SLinus Torvalds * 14251da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 14261da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 14271da177e4SLinus Torvalds * busy. 14281da177e4SLinus Torvalds */ 1429909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 14301da177e4SLinus Torvalds { 1431909b0a88SAl Viro struct mount *mnt = real_mount(m); 143236341f64SRam Pai int actual_refs = 0; 143336341f64SRam Pai int minimum_refs = 0; 1434315fc83eSAl Viro struct mount *p; 1435909b0a88SAl Viro BUG_ON(!m); 14361da177e4SLinus Torvalds 1437b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1438719ea2fbSAl Viro lock_mount_hash(); 1439909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 144083adc753SAl Viro actual_refs += mnt_get_count(p); 14411da177e4SLinus Torvalds minimum_refs += 2; 14421da177e4SLinus Torvalds } 1443719ea2fbSAl Viro unlock_mount_hash(); 14441da177e4SLinus Torvalds 14451da177e4SLinus Torvalds if (actual_refs > minimum_refs) 14461da177e4SLinus Torvalds return 0; 1447e3474a8eSIan Kent 1448e3474a8eSIan Kent return 1; 14491da177e4SLinus Torvalds } 14501da177e4SLinus Torvalds 14511da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 14521da177e4SLinus Torvalds 14531da177e4SLinus Torvalds /** 14541da177e4SLinus Torvalds * may_umount - check if a mount point is busy 14551da177e4SLinus Torvalds * @mnt: root of mount 14561da177e4SLinus Torvalds * 14571da177e4SLinus Torvalds * This is called to check if a mount point has any 14581da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 14591da177e4SLinus Torvalds * mount has sub mounts this will return busy 14601da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 14611da177e4SLinus Torvalds * 14621da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 14631da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 14641da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 14651da177e4SLinus Torvalds */ 14661da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 14671da177e4SLinus Torvalds { 1468e3474a8eSIan Kent int ret = 1; 14698ad08d8aSAl Viro down_read(&namespace_sem); 1470719ea2fbSAl Viro lock_mount_hash(); 14711ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1472e3474a8eSIan Kent ret = 0; 1473719ea2fbSAl Viro unlock_mount_hash(); 14748ad08d8aSAl Viro up_read(&namespace_sem); 1475a05964f3SRam Pai return ret; 14761da177e4SLinus Torvalds } 14771da177e4SLinus Torvalds 14781da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14791da177e4SLinus Torvalds 148097216be0SAl Viro static void namespace_unlock(void) 14811da177e4SLinus Torvalds { 1482a3b3c562SEric W. Biederman struct hlist_head head; 148356cbb429SAl Viro struct hlist_node *p; 148456cbb429SAl Viro struct mount *m; 14854edbe133SAl Viro LIST_HEAD(list); 148697216be0SAl Viro 1487a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 14884edbe133SAl Viro list_splice_init(&ex_mountpoints, &list); 1489a3b3c562SEric W. Biederman 149097216be0SAl Viro up_write(&namespace_sem); 1491a3b3c562SEric W. Biederman 14924edbe133SAl Viro shrink_dentry_list(&list); 14934edbe133SAl Viro 1494a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 149597216be0SAl Viro return; 149697216be0SAl Viro 149722cb7405SNeilBrown synchronize_rcu_expedited(); 149848a066e7SAl Viro 149956cbb429SAl Viro hlist_for_each_entry_safe(m, p, &head, mnt_umount) { 150056cbb429SAl Viro hlist_del(&m->mnt_umount); 150156cbb429SAl Viro mntput(&m->mnt); 150256cbb429SAl Viro } 150370fbcdf4SRam Pai } 150470fbcdf4SRam Pai 150597216be0SAl Viro static inline void namespace_lock(void) 1506e3197d83SAl Viro { 150797216be0SAl Viro down_write(&namespace_sem); 1508e3197d83SAl Viro } 1509e3197d83SAl Viro 1510e819f152SEric W. Biederman enum umount_tree_flags { 1511e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1512e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1513e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1514e819f152SEric W. Biederman }; 1515f2d0a123SEric W. Biederman 1516f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1517f2d0a123SEric W. Biederman { 1518f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1519f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1520f2d0a123SEric W. Biederman return true; 1521f2d0a123SEric W. Biederman 1522f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1523f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1524f2d0a123SEric W. Biederman return true; 1525f2d0a123SEric W. Biederman 1526f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1527f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1528f2d0a123SEric W. Biederman * connected to mounted mounts. 1529f2d0a123SEric W. Biederman */ 1530f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1531f2d0a123SEric W. Biederman return true; 1532f2d0a123SEric W. Biederman 1533f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1534f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1535f2d0a123SEric W. Biederman return false; 1536f2d0a123SEric W. Biederman 1537f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1538f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1539f2d0a123SEric W. Biederman return false; 1540f2d0a123SEric W. Biederman 1541f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1542f2d0a123SEric W. Biederman return true; 1543f2d0a123SEric W. Biederman } 1544f2d0a123SEric W. Biederman 154599b7db7bSNick Piggin /* 154648a066e7SAl Viro * mount_lock must be held 154799b7db7bSNick Piggin * namespace_sem must be held for write 154899b7db7bSNick Piggin */ 1549e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 155070fbcdf4SRam Pai { 1551c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1552315fc83eSAl Viro struct mount *p; 155370fbcdf4SRam Pai 15545d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 15555d88457eSEric W. Biederman propagate_mount_unlock(mnt); 15565d88457eSEric W. Biederman 1557c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1558590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1559590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1560c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1561590ce4bcSEric W. Biederman } 1562c003b26fSEric W. Biederman 1563411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1564c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1565c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 156638129a13SAl Viro } 156770fbcdf4SRam Pai 1568c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1569e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 15707b8a53fdSAl Viro propagate_umount(&tmp_list); 1571a05964f3SRam Pai 1572c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1573d2921684SEric W. Biederman struct mnt_namespace *ns; 1574ce07d891SEric W. Biederman bool disconnect; 1575c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 15766776db3dSAl Viro list_del_init(&p->mnt_expire); 15771a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1578d2921684SEric W. Biederman ns = p->mnt_ns; 1579d2921684SEric W. Biederman if (ns) { 1580d2921684SEric W. Biederman ns->mounts--; 1581d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1582d2921684SEric W. Biederman } 158363d37a84SAl Viro p->mnt_ns = NULL; 1584e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 158548a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 158687b95ce0SAl Viro 1587f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1588676da58dSAl Viro if (mnt_has_parent(p)) { 158981b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1590ce07d891SEric W. Biederman if (!disconnect) { 1591ce07d891SEric W. Biederman /* Don't forget about p */ 1592ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1593ce07d891SEric W. Biederman } else { 15946a46c573SEric W. Biederman umount_mnt(p); 15957c4b93d8SAl Viro } 1596ce07d891SEric W. Biederman } 15970f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 159819a1c409SAl Viro if (disconnect) 159919a1c409SAl Viro hlist_add_head(&p->mnt_umount, &unmounted); 160038129a13SAl Viro } 16011da177e4SLinus Torvalds } 16021da177e4SLinus Torvalds 1603b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1604c35038beSAl Viro 16058d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 16068d0347f6SDavid Howells { 16078d0347f6SDavid Howells int ret = 0; 16088d0347f6SDavid Howells 16098d0347f6SDavid Howells down_write(&sb->s_umount); 16108d0347f6SDavid Howells if (!sb_rdonly(sb)) { 16118d0347f6SDavid Howells struct fs_context *fc; 16128d0347f6SDavid Howells 16138d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 16148d0347f6SDavid Howells SB_RDONLY); 16158d0347f6SDavid Howells if (IS_ERR(fc)) { 16168d0347f6SDavid Howells ret = PTR_ERR(fc); 16178d0347f6SDavid Howells } else { 16188d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 16198d0347f6SDavid Howells if (!ret) 16208d0347f6SDavid Howells ret = reconfigure_super(fc); 16218d0347f6SDavid Howells put_fs_context(fc); 16228d0347f6SDavid Howells } 16238d0347f6SDavid Howells } 16248d0347f6SDavid Howells up_write(&sb->s_umount); 16258d0347f6SDavid Howells return ret; 16268d0347f6SDavid Howells } 16278d0347f6SDavid Howells 16281ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 16291da177e4SLinus Torvalds { 16301ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 16311da177e4SLinus Torvalds int retval; 16321da177e4SLinus Torvalds 16331ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 16341da177e4SLinus Torvalds if (retval) 16351da177e4SLinus Torvalds return retval; 16361da177e4SLinus Torvalds 16371da177e4SLinus Torvalds /* 16381da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 16391da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 16401da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 16411da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 16421da177e4SLinus Torvalds */ 16431da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 16441ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 16451da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 16461da177e4SLinus Torvalds return -EINVAL; 16471da177e4SLinus Torvalds 1648b3e19d92SNick Piggin /* 1649b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1650b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1651b3e19d92SNick Piggin */ 1652719ea2fbSAl Viro lock_mount_hash(); 165383adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1654719ea2fbSAl Viro unlock_mount_hash(); 16551da177e4SLinus Torvalds return -EBUSY; 1656b3e19d92SNick Piggin } 1657719ea2fbSAl Viro unlock_mount_hash(); 16581da177e4SLinus Torvalds 1659863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 16601da177e4SLinus Torvalds return -EAGAIN; 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds /* 16641da177e4SLinus Torvalds * If we may have to abort operations to get out of this 16651da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 16661da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 16671da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 16681da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 16691da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 16701da177e4SLinus Torvalds * about for the moment. 16711da177e4SLinus Torvalds */ 16721da177e4SLinus Torvalds 167342faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 167442faad99SAl Viro sb->s_op->umount_begin(sb); 167542faad99SAl Viro } 16761da177e4SLinus Torvalds 16771da177e4SLinus Torvalds /* 16781da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 16791da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 16801da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 16811da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 16821da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 16831da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 16841da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 16851da177e4SLinus Torvalds */ 16861ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 16871da177e4SLinus Torvalds /* 16881da177e4SLinus Torvalds * Special case for "unmounting" root ... 16891da177e4SLinus Torvalds * we just try to remount it readonly. 16901da177e4SLinus Torvalds */ 1691bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1692a1480dccSAndy Lutomirski return -EPERM; 16938d0347f6SDavid Howells return do_umount_root(sb); 16941da177e4SLinus Torvalds } 16951da177e4SLinus Torvalds 169697216be0SAl Viro namespace_lock(); 1697719ea2fbSAl Viro lock_mount_hash(); 16981da177e4SLinus Torvalds 169925d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 170025d202edSEric W. Biederman retval = -EINVAL; 170125d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 170225d202edSEric W. Biederman goto out; 170325d202edSEric W. Biederman 170425d202edSEric W. Biederman event++; 170548a066e7SAl Viro if (flags & MNT_DETACH) { 170648a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1707e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 170848a066e7SAl Viro retval = 0; 170948a066e7SAl Viro } else { 1710b54b9be7SAl Viro shrink_submounts(mnt); 17111da177e4SLinus Torvalds retval = -EBUSY; 171248a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 17131a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1714e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 17151da177e4SLinus Torvalds retval = 0; 17161da177e4SLinus Torvalds } 171748a066e7SAl Viro } 171825d202edSEric W. Biederman out: 1719719ea2fbSAl Viro unlock_mount_hash(); 1720e3197d83SAl Viro namespace_unlock(); 17211da177e4SLinus Torvalds return retval; 17221da177e4SLinus Torvalds } 17231da177e4SLinus Torvalds 17241da177e4SLinus Torvalds /* 172580b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 172680b5dce8SEric W. Biederman * 172780b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 172880b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 172980b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 173080b5dce8SEric W. Biederman * leaking them. 173180b5dce8SEric W. Biederman * 173280b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 173380b5dce8SEric W. Biederman */ 173480b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 173580b5dce8SEric W. Biederman { 173680b5dce8SEric W. Biederman struct mountpoint *mp; 173780b5dce8SEric W. Biederman struct mount *mnt; 173880b5dce8SEric W. Biederman 173980b5dce8SEric W. Biederman namespace_lock(); 17403895dbf8SEric W. Biederman lock_mount_hash(); 174180b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1742adc9b5c0SAl Viro if (!mp) 174380b5dce8SEric W. Biederman goto out_unlock; 174480b5dce8SEric W. Biederman 1745e06b933eSAndrey Ulanov event++; 174680b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 174780b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1748ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1749fe78fcc8SEric W. Biederman umount_mnt(mnt); 175056cbb429SAl Viro hlist_add_head(&mnt->mnt_umount, &unmounted); 1751ce07d891SEric W. Biederman } 1752e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 175380b5dce8SEric W. Biederman } 175480b5dce8SEric W. Biederman put_mountpoint(mp); 175580b5dce8SEric W. Biederman out_unlock: 17563895dbf8SEric W. Biederman unlock_mount_hash(); 175780b5dce8SEric W. Biederman namespace_unlock(); 175880b5dce8SEric W. Biederman } 175980b5dce8SEric W. Biederman 176080b5dce8SEric W. Biederman /* 17619b40bc90SAl Viro * Is the caller allowed to modify his namespace? 17629b40bc90SAl Viro */ 1763a5f85d78SAl Viro bool may_mount(void) 17649b40bc90SAl Viro { 17659b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 17669b40bc90SAl Viro } 17679b40bc90SAl Viro 1768f7e33bdbSJeff Layton static void warn_mandlock(void) 17699e8925b6SJeff Layton { 1770f7e33bdbSJeff Layton pr_warn_once("=======================================================\n" 1771f7e33bdbSJeff Layton "WARNING: The mand mount option has been deprecated and\n" 1772f7e33bdbSJeff Layton " and is ignored by this kernel. Remove the mand\n" 1773f7e33bdbSJeff Layton " option from the mount to silence this warning.\n" 1774f7e33bdbSJeff Layton "=======================================================\n"); 17759e8925b6SJeff Layton } 17769e8925b6SJeff Layton 177725ccd24fSChristoph Hellwig static int can_umount(const struct path *path, int flags) 17781da177e4SLinus Torvalds { 177925ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 17801da177e4SLinus Torvalds 17819b40bc90SAl Viro if (!may_mount()) 17829b40bc90SAl Viro return -EPERM; 178341525f56SChristoph Hellwig if (path->dentry != path->mnt->mnt_root) 178425ccd24fSChristoph Hellwig return -EINVAL; 1785143c8c91SAl Viro if (!check_mnt(mnt)) 178625ccd24fSChristoph Hellwig return -EINVAL; 178725d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 178825ccd24fSChristoph Hellwig return -EINVAL; 1789b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 179025ccd24fSChristoph Hellwig return -EPERM; 179125ccd24fSChristoph Hellwig return 0; 179225ccd24fSChristoph Hellwig } 17931da177e4SLinus Torvalds 1794a0a6df9aSAl Viro // caller is responsible for flags being sane 179525ccd24fSChristoph Hellwig int path_umount(struct path *path, int flags) 179625ccd24fSChristoph Hellwig { 179725ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 179825ccd24fSChristoph Hellwig int ret; 179925ccd24fSChristoph Hellwig 180025ccd24fSChristoph Hellwig ret = can_umount(path, flags); 180125ccd24fSChristoph Hellwig if (!ret) 180225ccd24fSChristoph Hellwig ret = do_umount(mnt, flags); 180325ccd24fSChristoph Hellwig 1804429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 180541525f56SChristoph Hellwig dput(path->dentry); 1806900148dcSAl Viro mntput_no_expire(mnt); 180725ccd24fSChristoph Hellwig return ret; 18081da177e4SLinus Torvalds } 18091da177e4SLinus Torvalds 181009267defSChristoph Hellwig static int ksys_umount(char __user *name, int flags) 181141525f56SChristoph Hellwig { 181241525f56SChristoph Hellwig int lookup_flags = LOOKUP_MOUNTPOINT; 181341525f56SChristoph Hellwig struct path path; 181441525f56SChristoph Hellwig int ret; 181541525f56SChristoph Hellwig 1816a0a6df9aSAl Viro // basic validity checks done first 1817a0a6df9aSAl Viro if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1818a0a6df9aSAl Viro return -EINVAL; 1819a0a6df9aSAl Viro 182041525f56SChristoph Hellwig if (!(flags & UMOUNT_NOFOLLOW)) 182141525f56SChristoph Hellwig lookup_flags |= LOOKUP_FOLLOW; 182241525f56SChristoph Hellwig ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); 182341525f56SChristoph Hellwig if (ret) 182441525f56SChristoph Hellwig return ret; 182541525f56SChristoph Hellwig return path_umount(&path, flags); 182641525f56SChristoph Hellwig } 182741525f56SChristoph Hellwig 18283a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 18293a18ef5cSDominik Brodowski { 18303a18ef5cSDominik Brodowski return ksys_umount(name, flags); 18313a18ef5cSDominik Brodowski } 18323a18ef5cSDominik Brodowski 18331da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 18341da177e4SLinus Torvalds 18351da177e4SLinus Torvalds /* 18361da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 18371da177e4SLinus Torvalds */ 1838bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 18391da177e4SLinus Torvalds { 18403a18ef5cSDominik Brodowski return ksys_umount(name, 0); 18411da177e4SLinus Torvalds } 18421da177e4SLinus Torvalds 18431da177e4SLinus Torvalds #endif 18441da177e4SLinus Torvalds 18454ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 18468823c079SEric W. Biederman { 18474ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1848e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1849e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 18504ce5d2b1SEric W. Biederman } 18514ce5d2b1SEric W. Biederman 1852213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 185358be2825SAl Viro { 185458be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 185558be2825SAl Viro } 185658be2825SAl Viro 1857303cc571SChristian Brauner struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) 1858303cc571SChristian Brauner { 1859303cc571SChristian Brauner return &mnt->ns; 1860303cc571SChristian Brauner } 1861303cc571SChristian Brauner 18624ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 18634ce5d2b1SEric W. Biederman { 18644ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 18654ce5d2b1SEric W. Biederman * mount namespace loop? 18664ce5d2b1SEric W. Biederman */ 18674ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 18684ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 18694ce5d2b1SEric W. Biederman return false; 18704ce5d2b1SEric W. Biederman 1871f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 18728823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 18738823c079SEric W. Biederman } 18748823c079SEric W. Biederman 187587129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 187636341f64SRam Pai int flag) 18771da177e4SLinus Torvalds { 187884d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 18791da177e4SLinus Torvalds 18804ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 18814ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 18824ce5d2b1SEric W. Biederman 18834ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1884be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 18859676f0c6SRam Pai 188636341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1887be34d1a3SDavid Howells if (IS_ERR(q)) 1888be34d1a3SDavid Howells return q; 1889be34d1a3SDavid Howells 1890a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 18911da177e4SLinus Torvalds 18921da177e4SLinus Torvalds p = mnt; 18936b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1894315fc83eSAl Viro struct mount *s; 18957ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 18961da177e4SLinus Torvalds continue; 18971da177e4SLinus Torvalds 1898909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 18994ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 19004ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1901df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1902df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1903df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1904df7342b2SEric W. Biederman goto out; 1905df7342b2SEric W. Biederman } else { 19064ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 19074ce5d2b1SEric W. Biederman continue; 19084ce5d2b1SEric W. Biederman } 1909df7342b2SEric W. Biederman } 19104ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 19114ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 19129676f0c6SRam Pai s = skip_mnt_tree(s); 19139676f0c6SRam Pai continue; 19149676f0c6SRam Pai } 19150714a533SAl Viro while (p != s->mnt_parent) { 19160714a533SAl Viro p = p->mnt_parent; 19170714a533SAl Viro q = q->mnt_parent; 19181da177e4SLinus Torvalds } 191987129cc0SAl Viro p = s; 192084d17192SAl Viro parent = q; 192187129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1922be34d1a3SDavid Howells if (IS_ERR(q)) 1923be34d1a3SDavid Howells goto out; 1924719ea2fbSAl Viro lock_mount_hash(); 19251a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 19261064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1927719ea2fbSAl Viro unlock_mount_hash(); 19281da177e4SLinus Torvalds } 19291da177e4SLinus Torvalds } 19301da177e4SLinus Torvalds return res; 1931be34d1a3SDavid Howells out: 19321da177e4SLinus Torvalds if (res) { 1933719ea2fbSAl Viro lock_mount_hash(); 1934e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1935719ea2fbSAl Viro unlock_mount_hash(); 19361da177e4SLinus Torvalds } 1937be34d1a3SDavid Howells return q; 19381da177e4SLinus Torvalds } 19391da177e4SLinus Torvalds 1940be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1941be34d1a3SDavid Howells 1942ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 19438aec0809SAl Viro { 1944cb338d06SAl Viro struct mount *tree; 194597216be0SAl Viro namespace_lock(); 1946cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1947cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1948cd4a4017SEric W. Biederman else 194987129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 195087129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1951328e6d90SAl Viro namespace_unlock(); 1952be34d1a3SDavid Howells if (IS_ERR(tree)) 195352e220d3SDan Carpenter return ERR_CAST(tree); 1954be34d1a3SDavid Howells return &tree->mnt; 19558aec0809SAl Viro } 19568aec0809SAl Viro 1957a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1958a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1959a07b2000SAl Viro 1960a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1961a07b2000SAl Viro { 1962a07b2000SAl Viro struct mnt_namespace *ns; 1963a07b2000SAl Viro namespace_lock(); 1964a07b2000SAl Viro lock_mount_hash(); 1965a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 196644dfd84aSDavid Howells if (ns) { 196744dfd84aSDavid Howells if (is_anon_ns(ns)) 1968a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 196944dfd84aSDavid Howells else 197044dfd84aSDavid Howells ns = NULL; 197144dfd84aSDavid Howells } 1972a07b2000SAl Viro unlock_mount_hash(); 1973a07b2000SAl Viro namespace_unlock(); 197444dfd84aSDavid Howells if (ns) 1975a07b2000SAl Viro free_mnt_ns(ns); 1976a07b2000SAl Viro } 1977a07b2000SAl Viro 19788aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 19798aec0809SAl Viro { 198097216be0SAl Viro namespace_lock(); 1981719ea2fbSAl Viro lock_mount_hash(); 19829c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1983719ea2fbSAl Viro unlock_mount_hash(); 19843ab6abeeSAl Viro namespace_unlock(); 19858aec0809SAl Viro } 19868aec0809SAl Viro 1987427215d8SMiklos Szeredi static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 1988427215d8SMiklos Szeredi { 1989427215d8SMiklos Szeredi struct mount *child; 1990427215d8SMiklos Szeredi 1991427215d8SMiklos Szeredi list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 1992427215d8SMiklos Szeredi if (!is_subdir(child->mnt_mountpoint, dentry)) 1993427215d8SMiklos Szeredi continue; 1994427215d8SMiklos Szeredi 1995427215d8SMiklos Szeredi if (child->mnt.mnt_flags & MNT_LOCKED) 1996427215d8SMiklos Szeredi return true; 1997427215d8SMiklos Szeredi } 1998427215d8SMiklos Szeredi return false; 1999427215d8SMiklos Szeredi } 2000427215d8SMiklos Szeredi 2001c771d683SMiklos Szeredi /** 2002c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 20031f287bc4SRandy Dunlap * @path: path to clone 2004c771d683SMiklos Szeredi * 20051f287bc4SRandy Dunlap * This creates a new vfsmount, which will be the clone of @path. The new mount 20061f287bc4SRandy Dunlap * will not be attached anywhere in the namespace and will be private (i.e. 20071f287bc4SRandy Dunlap * changes to the originating mount won't be propagated into this). 2008c771d683SMiklos Szeredi * 2009c771d683SMiklos Szeredi * Release with mntput(). 2010c771d683SMiklos Szeredi */ 2011ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 2012c771d683SMiklos Szeredi { 2013c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 2014c771d683SMiklos Szeredi struct mount *new_mnt; 2015c771d683SMiklos Szeredi 2016427215d8SMiklos Szeredi down_read(&namespace_sem); 2017c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 2018427215d8SMiklos Szeredi goto invalid; 2019427215d8SMiklos Szeredi 2020427215d8SMiklos Szeredi if (!check_mnt(old_mnt)) 2021427215d8SMiklos Szeredi goto invalid; 2022427215d8SMiklos Szeredi 2023427215d8SMiklos Szeredi if (has_locked_children(old_mnt, path->dentry)) 2024427215d8SMiklos Szeredi goto invalid; 2025c771d683SMiklos Szeredi 2026c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 2027427215d8SMiklos Szeredi up_read(&namespace_sem); 2028427215d8SMiklos Szeredi 2029c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 2030c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 2031c771d683SMiklos Szeredi 2032df820f8dSMiklos Szeredi /* Longterm mount to be removed by kern_unmount*() */ 2033df820f8dSMiklos Szeredi new_mnt->mnt_ns = MNT_NS_INTERNAL; 2034df820f8dSMiklos Szeredi 2035c771d683SMiklos Szeredi return &new_mnt->mnt; 2036427215d8SMiklos Szeredi 2037427215d8SMiklos Szeredi invalid: 2038427215d8SMiklos Szeredi up_read(&namespace_sem); 2039427215d8SMiklos Szeredi return ERR_PTR(-EINVAL); 2040c771d683SMiklos Szeredi } 2041c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 2042c771d683SMiklos Szeredi 20431f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 20441f707137SAl Viro struct vfsmount *root) 20451f707137SAl Viro { 20461a4eeaf2SAl Viro struct mount *mnt; 20471f707137SAl Viro int res = f(root, arg); 20481f707137SAl Viro if (res) 20491f707137SAl Viro return res; 20501a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 20511a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 20521f707137SAl Viro if (res) 20531f707137SAl Viro return res; 20541f707137SAl Viro } 20551f707137SAl Viro return 0; 20561f707137SAl Viro } 20571f707137SAl Viro 20583bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 20593bd045ccSAl Viro { 20603bd045ccSAl Viro struct mount *p; 20613bd045ccSAl Viro 20623bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 20633bd045ccSAl Viro int flags = p->mnt.mnt_flags; 20643bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 20653bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 20663bd045ccSAl Viro 20673bd045ccSAl Viro if (flags & MNT_READONLY) 20683bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 20693bd045ccSAl Viro 20703bd045ccSAl Viro if (flags & MNT_NODEV) 20713bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 20723bd045ccSAl Viro 20733bd045ccSAl Viro if (flags & MNT_NOSUID) 20743bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 20753bd045ccSAl Viro 20763bd045ccSAl Viro if (flags & MNT_NOEXEC) 20773bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 20783bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 20793bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 20803bd045ccSAl Viro flags |= MNT_LOCKED; 20813bd045ccSAl Viro p->mnt.mnt_flags = flags; 20823bd045ccSAl Viro } 20833bd045ccSAl Viro } 20843bd045ccSAl Viro 20854b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 2086719f5d7fSMiklos Szeredi { 2087315fc83eSAl Viro struct mount *p; 2088719f5d7fSMiklos Szeredi 2089909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 2090fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 20914b8b21f4SAl Viro mnt_release_group_id(p); 2092719f5d7fSMiklos Szeredi } 2093719f5d7fSMiklos Szeredi } 2094719f5d7fSMiklos Szeredi 20954b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 2096719f5d7fSMiklos Szeredi { 2097315fc83eSAl Viro struct mount *p; 2098719f5d7fSMiklos Szeredi 2099909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 2100fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 21014b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 2102719f5d7fSMiklos Szeredi if (err) { 21034b8b21f4SAl Viro cleanup_group_ids(mnt, p); 2104719f5d7fSMiklos Szeredi return err; 2105719f5d7fSMiklos Szeredi } 2106719f5d7fSMiklos Szeredi } 2107719f5d7fSMiklos Szeredi } 2108719f5d7fSMiklos Szeredi 2109719f5d7fSMiklos Szeredi return 0; 2110719f5d7fSMiklos Szeredi } 2111719f5d7fSMiklos Szeredi 2112d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 2113d2921684SEric W. Biederman { 2114d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 2115124f75f8SAl Viro unsigned int mounts = 0; 2116d2921684SEric W. Biederman struct mount *p; 2117d2921684SEric W. Biederman 2118124f75f8SAl Viro if (ns->mounts >= max) 2119124f75f8SAl Viro return -ENOSPC; 2120124f75f8SAl Viro max -= ns->mounts; 2121124f75f8SAl Viro if (ns->pending_mounts >= max) 2122124f75f8SAl Viro return -ENOSPC; 2123124f75f8SAl Viro max -= ns->pending_mounts; 2124124f75f8SAl Viro 2125d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 2126d2921684SEric W. Biederman mounts++; 2127d2921684SEric W. Biederman 2128124f75f8SAl Viro if (mounts > max) 2129d2921684SEric W. Biederman return -ENOSPC; 2130d2921684SEric W. Biederman 2131124f75f8SAl Viro ns->pending_mounts += mounts; 2132d2921684SEric W. Biederman return 0; 2133d2921684SEric W. Biederman } 2134d2921684SEric W. Biederman 2135b90fa9aeSRam Pai /* 2136b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 2137b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 213821444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 213921444403SRam Pai * store the parent mount and mountpoint dentry. 214021444403SRam Pai * (done when source_mnt is moved) 2141b90fa9aeSRam Pai * 2142b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 2143b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 21449676f0c6SRam Pai * --------------------------------------------------------------------------- 2145b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 21469676f0c6SRam Pai * |************************************************************************** 21479676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21489676f0c6SRam Pai * | dest | | | | | 21499676f0c6SRam Pai * | | | | | | | 21509676f0c6SRam Pai * | v | | | | | 21519676f0c6SRam Pai * |************************************************************************** 21529676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 21535afe0022SRam Pai * | | | | | | 21549676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 21559676f0c6SRam Pai * *************************************************************************** 2156b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 2157b90fa9aeSRam Pai * destination mount. 2158b90fa9aeSRam Pai * 2159b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2160b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2161b90fa9aeSRam Pai * the peer group of the source mount. 2162b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2163b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2164b90fa9aeSRam Pai * mount. 21655afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 21665afe0022SRam Pai * of the destination mount and the cloned mount is made slave 21675afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 21685afe0022SRam Pai * is marked as 'shared and slave'. 21695afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 21705afe0022SRam Pai * source mount. 21715afe0022SRam Pai * 21729676f0c6SRam Pai * --------------------------------------------------------------------------- 217321444403SRam Pai * | MOVE MOUNT OPERATION | 21749676f0c6SRam Pai * |************************************************************************** 21759676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21769676f0c6SRam Pai * | dest | | | | | 21779676f0c6SRam Pai * | | | | | | | 21789676f0c6SRam Pai * | v | | | | | 21799676f0c6SRam Pai * |************************************************************************** 21809676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 21815afe0022SRam Pai * | | | | | | 21829676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 21839676f0c6SRam Pai * *************************************************************************** 21845afe0022SRam Pai * 21855afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 21865afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 218721444403SRam Pai * (+*) the mount is moved to the destination. 21885afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 21895afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 21905afe0022SRam Pai * the mount is marked as 'shared and slave'. 21915afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2192b90fa9aeSRam Pai * 2193b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2194b90fa9aeSRam Pai * applied to each mount in the tree. 2195b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2196b90fa9aeSRam Pai * in allocations. 2197b90fa9aeSRam Pai */ 21980fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 219984d17192SAl Viro struct mount *dest_mnt, 220084d17192SAl Viro struct mountpoint *dest_mp, 22012763d119SAl Viro bool moving) 2202b90fa9aeSRam Pai { 22033bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 220438129a13SAl Viro HLIST_HEAD(tree_list); 2205d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 22061064f874SEric W. Biederman struct mountpoint *smp; 2207315fc83eSAl Viro struct mount *child, *p; 220838129a13SAl Viro struct hlist_node *n; 2209719f5d7fSMiklos Szeredi int err; 2210b90fa9aeSRam Pai 22111064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 22121064f874SEric W. Biederman * to be tucked under other mounts. 22131064f874SEric W. Biederman */ 22141064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 22151064f874SEric W. Biederman if (IS_ERR(smp)) 22161064f874SEric W. Biederman return PTR_ERR(smp); 22171064f874SEric W. Biederman 2218d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 22192763d119SAl Viro if (!moving) { 2220d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2221d2921684SEric W. Biederman if (err) 2222d2921684SEric W. Biederman goto out; 2223d2921684SEric W. Biederman } 2224d2921684SEric W. Biederman 2225fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 22260fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2227719f5d7fSMiklos Szeredi if (err) 2228719f5d7fSMiklos Szeredi goto out; 222984d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2230f2ebb3a9SAl Viro lock_mount_hash(); 2231719f5d7fSMiklos Szeredi if (err) 2232719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2233909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 22340f0afb1dSAl Viro set_mnt_shared(p); 22350b1b901bSAl Viro } else { 22360b1b901bSAl Viro lock_mount_hash(); 2237b90fa9aeSRam Pai } 22382763d119SAl Viro if (moving) { 22392763d119SAl Viro unhash_mnt(source_mnt); 224084d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2241143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 224221444403SRam Pai } else { 224344dfd84aSDavid Howells if (source_mnt->mnt_ns) { 224444dfd84aSDavid Howells /* move from anon - the caller will destroy */ 224544dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 224644dfd84aSDavid Howells } 224784d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 22481064f874SEric W. Biederman commit_tree(source_mnt); 224921444403SRam Pai } 2250b90fa9aeSRam Pai 225138129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 22521d6a32acSAl Viro struct mount *q; 225338129a13SAl Viro hlist_del_init(&child->mnt_hash); 22541064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 22551d6a32acSAl Viro child->mnt_mountpoint); 22561064f874SEric W. Biederman if (q) 22571064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 22583bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 22593bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 22603bd045ccSAl Viro lock_mnt_tree(child); 2261d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 22621064f874SEric W. Biederman commit_tree(child); 2263b90fa9aeSRam Pai } 22641064f874SEric W. Biederman put_mountpoint(smp); 2265719ea2fbSAl Viro unlock_mount_hash(); 226699b7db7bSNick Piggin 2267b90fa9aeSRam Pai return 0; 2268719f5d7fSMiklos Szeredi 2269719f5d7fSMiklos Szeredi out_cleanup_ids: 2270f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2271f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2272d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2273e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2274f2ebb3a9SAl Viro } 2275f2ebb3a9SAl Viro unlock_mount_hash(); 22760fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2277719f5d7fSMiklos Szeredi out: 2278d2921684SEric W. Biederman ns->pending_mounts = 0; 22791064f874SEric W. Biederman 22801064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 22811064f874SEric W. Biederman put_mountpoint(smp); 22821064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 22831064f874SEric W. Biederman 2284719f5d7fSMiklos Szeredi return err; 2285b90fa9aeSRam Pai } 2286b90fa9aeSRam Pai 228784d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2288b12cea91SAl Viro { 2289b12cea91SAl Viro struct vfsmount *mnt; 229084d17192SAl Viro struct dentry *dentry = path->dentry; 2291b12cea91SAl Viro retry: 22925955102cSAl Viro inode_lock(dentry->d_inode); 229384d17192SAl Viro if (unlikely(cant_mount(dentry))) { 22945955102cSAl Viro inode_unlock(dentry->d_inode); 229584d17192SAl Viro return ERR_PTR(-ENOENT); 2296b12cea91SAl Viro } 229797216be0SAl Viro namespace_lock(); 2298b12cea91SAl Viro mnt = lookup_mnt(path); 229984d17192SAl Viro if (likely(!mnt)) { 23003895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 230184d17192SAl Viro if (IS_ERR(mp)) { 230297216be0SAl Viro namespace_unlock(); 23035955102cSAl Viro inode_unlock(dentry->d_inode); 230484d17192SAl Viro return mp; 230584d17192SAl Viro } 230684d17192SAl Viro return mp; 230784d17192SAl Viro } 230897216be0SAl Viro namespace_unlock(); 23095955102cSAl Viro inode_unlock(path->dentry->d_inode); 2310b12cea91SAl Viro path_put(path); 2311b12cea91SAl Viro path->mnt = mnt; 231284d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2313b12cea91SAl Viro goto retry; 2314b12cea91SAl Viro } 2315b12cea91SAl Viro 231684d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2317b12cea91SAl Viro { 231884d17192SAl Viro struct dentry *dentry = where->m_dentry; 23193895dbf8SEric W. Biederman 23203895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 232184d17192SAl Viro put_mountpoint(where); 23223895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 23233895dbf8SEric W. Biederman 2324328e6d90SAl Viro namespace_unlock(); 23255955102cSAl Viro inode_unlock(dentry->d_inode); 2326b12cea91SAl Viro } 2327b12cea91SAl Viro 232884d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 23291da177e4SLinus Torvalds { 2330e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 23311da177e4SLinus Torvalds return -EINVAL; 23321da177e4SLinus Torvalds 2333e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2334e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 23351da177e4SLinus Torvalds return -ENOTDIR; 23361da177e4SLinus Torvalds 23372763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 23381da177e4SLinus Torvalds } 23391da177e4SLinus Torvalds 23401da177e4SLinus Torvalds /* 23417a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 23427a2e8a8fSValerie Aurora */ 23437a2e8a8fSValerie Aurora 2344e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 23457a2e8a8fSValerie Aurora { 2346e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 23477a2e8a8fSValerie Aurora 23487a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 23497a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 23507a2e8a8fSValerie Aurora return 0; 23517a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 23527a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 23537a2e8a8fSValerie Aurora return 0; 23547a2e8a8fSValerie Aurora return type; 23557a2e8a8fSValerie Aurora } 23567a2e8a8fSValerie Aurora 23577a2e8a8fSValerie Aurora /* 235807b20889SRam Pai * recursively change the type of the mountpoint. 235907b20889SRam Pai */ 2360e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 236107b20889SRam Pai { 2362315fc83eSAl Viro struct mount *m; 23634b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2364e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 23657a2e8a8fSValerie Aurora int type; 2366719f5d7fSMiklos Szeredi int err = 0; 236707b20889SRam Pai 23682d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 236907b20889SRam Pai return -EINVAL; 237007b20889SRam Pai 2371e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 23727a2e8a8fSValerie Aurora if (!type) 23737a2e8a8fSValerie Aurora return -EINVAL; 23747a2e8a8fSValerie Aurora 237597216be0SAl Viro namespace_lock(); 2376719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2377719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2378719f5d7fSMiklos Szeredi if (err) 2379719f5d7fSMiklos Szeredi goto out_unlock; 2380719f5d7fSMiklos Szeredi } 2381719f5d7fSMiklos Szeredi 2382719ea2fbSAl Viro lock_mount_hash(); 2383909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 23840f0afb1dSAl Viro change_mnt_propagation(m, type); 2385719ea2fbSAl Viro unlock_mount_hash(); 2386719f5d7fSMiklos Szeredi 2387719f5d7fSMiklos Szeredi out_unlock: 238897216be0SAl Viro namespace_unlock(); 2389719f5d7fSMiklos Szeredi return err; 239007b20889SRam Pai } 239107b20889SRam Pai 2392a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2393a07b2000SAl Viro { 2394a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2395a07b2000SAl Viro 2396a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2397a07b2000SAl Viro return mnt; 2398a07b2000SAl Viro 2399a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2400a07b2000SAl Viro return mnt; 2401a07b2000SAl Viro 2402a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2403a07b2000SAl Viro return mnt; 2404a07b2000SAl Viro 2405a07b2000SAl Viro if (recurse) 2406a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2407a07b2000SAl Viro else 2408a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2409a07b2000SAl Viro 2410a07b2000SAl Viro if (!IS_ERR(mnt)) 2411a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2412a07b2000SAl Viro 2413a07b2000SAl Viro return mnt; 2414a07b2000SAl Viro } 2415a07b2000SAl Viro 241607b20889SRam Pai /* 24171da177e4SLinus Torvalds * do loopback mount. 24181da177e4SLinus Torvalds */ 2419808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 24202dafe1c4SEric Sandeen int recurse) 24211da177e4SLinus Torvalds { 24222d92ab3cSAl Viro struct path old_path; 2423a07b2000SAl Viro struct mount *mnt = NULL, *parent; 242484d17192SAl Viro struct mountpoint *mp; 242557eccb83SAl Viro int err; 24261da177e4SLinus Torvalds if (!old_name || !*old_name) 24271da177e4SLinus Torvalds return -EINVAL; 2428815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 24291da177e4SLinus Torvalds if (err) 24301da177e4SLinus Torvalds return err; 24311da177e4SLinus Torvalds 24328823c079SEric W. Biederman err = -EINVAL; 24334ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 24348823c079SEric W. Biederman goto out; 24358823c079SEric W. Biederman 243684d17192SAl Viro mp = lock_mount(path); 2437a07b2000SAl Viro if (IS_ERR(mp)) { 243884d17192SAl Viro err = PTR_ERR(mp); 24399676f0c6SRam Pai goto out; 2440a07b2000SAl Viro } 24419676f0c6SRam Pai 244284d17192SAl Viro parent = real_mount(path->mnt); 2443e149ed2bSAl Viro if (!check_mnt(parent)) 2444e149ed2bSAl Viro goto out2; 2445e149ed2bSAl Viro 2446a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2447be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2448be34d1a3SDavid Howells err = PTR_ERR(mnt); 2449e9c5d8a5SAndrey Vagin goto out2; 2450be34d1a3SDavid Howells } 2451ccd48bc7SAl Viro 245284d17192SAl Viro err = graft_tree(mnt, parent, mp); 24531da177e4SLinus Torvalds if (err) { 2454719ea2fbSAl Viro lock_mount_hash(); 2455e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2456719ea2fbSAl Viro unlock_mount_hash(); 24575b83d2c5SRam Pai } 2458b12cea91SAl Viro out2: 245984d17192SAl Viro unlock_mount(mp); 2460ccd48bc7SAl Viro out: 24612d92ab3cSAl Viro path_put(&old_path); 24621da177e4SLinus Torvalds return err; 24631da177e4SLinus Torvalds } 24641da177e4SLinus Torvalds 2465a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2466a07b2000SAl Viro { 2467a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2468a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2469a07b2000SAl Viro struct mount *mnt, *p; 2470a07b2000SAl Viro struct file *file; 2471a07b2000SAl Viro 2472a07b2000SAl Viro if (IS_ERR(ns)) 2473a07b2000SAl Viro return ERR_CAST(ns); 2474a07b2000SAl Viro 2475a07b2000SAl Viro namespace_lock(); 2476a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2477a07b2000SAl Viro if (IS_ERR(mnt)) { 2478a07b2000SAl Viro namespace_unlock(); 2479a07b2000SAl Viro free_mnt_ns(ns); 2480a07b2000SAl Viro return ERR_CAST(mnt); 2481a07b2000SAl Viro } 2482a07b2000SAl Viro 2483a07b2000SAl Viro lock_mount_hash(); 2484a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2485a07b2000SAl Viro p->mnt_ns = ns; 2486a07b2000SAl Viro ns->mounts++; 2487a07b2000SAl Viro } 2488a07b2000SAl Viro ns->root = mnt; 2489a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2490a07b2000SAl Viro mntget(&mnt->mnt); 2491a07b2000SAl Viro unlock_mount_hash(); 2492a07b2000SAl Viro namespace_unlock(); 2493a07b2000SAl Viro 2494a07b2000SAl Viro mntput(path->mnt); 2495a07b2000SAl Viro path->mnt = &mnt->mnt; 2496a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2497a07b2000SAl Viro if (IS_ERR(file)) 2498a07b2000SAl Viro dissolve_on_fput(path->mnt); 2499a07b2000SAl Viro else 2500a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2501a07b2000SAl Viro return file; 2502a07b2000SAl Viro } 2503a07b2000SAl Viro 25042658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2505a07b2000SAl Viro { 2506a07b2000SAl Viro struct file *file; 2507a07b2000SAl Viro struct path path; 2508a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2509a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2510a07b2000SAl Viro int error; 2511a07b2000SAl Viro int fd; 2512a07b2000SAl Viro 2513a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2514a07b2000SAl Viro 2515a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2516a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2517a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2518a07b2000SAl Viro return -EINVAL; 2519a07b2000SAl Viro 2520a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2521a07b2000SAl Viro return -EINVAL; 2522a07b2000SAl Viro 2523a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2524a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2525a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2526a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2527a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2528a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2529a07b2000SAl Viro 2530a07b2000SAl Viro if (detached && !may_mount()) 2531a07b2000SAl Viro return -EPERM; 2532a07b2000SAl Viro 2533a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2534a07b2000SAl Viro if (fd < 0) 2535a07b2000SAl Viro return fd; 2536a07b2000SAl Viro 2537a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2538a07b2000SAl Viro if (unlikely(error)) { 2539a07b2000SAl Viro file = ERR_PTR(error); 2540a07b2000SAl Viro } else { 2541a07b2000SAl Viro if (detached) 2542a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2543a07b2000SAl Viro else 2544a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2545a07b2000SAl Viro path_put(&path); 2546a07b2000SAl Viro } 2547a07b2000SAl Viro if (IS_ERR(file)) { 2548a07b2000SAl Viro put_unused_fd(fd); 2549a07b2000SAl Viro return PTR_ERR(file); 2550a07b2000SAl Viro } 2551a07b2000SAl Viro fd_install(fd, file); 2552a07b2000SAl Viro return fd; 2553a07b2000SAl Viro } 2554a07b2000SAl Viro 255543f5e655SDavid Howells /* 255643f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 255743f5e655SDavid Howells * 255843f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 255943f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 256043f5e655SDavid Howells */ 256143f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 25622e4b7fcdSDave Hansen { 256343f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 25642e4b7fcdSDave Hansen 256543f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 256643f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 256743f5e655SDavid Howells return false; 256843f5e655SDavid Howells 256943f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 257043f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 257143f5e655SDavid Howells return false; 257243f5e655SDavid Howells 257343f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 257443f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 257543f5e655SDavid Howells return false; 257643f5e655SDavid Howells 257743f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 257843f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 257943f5e655SDavid Howells return false; 258043f5e655SDavid Howells 258143f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 258243f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 258343f5e655SDavid Howells return false; 258443f5e655SDavid Howells 258543f5e655SDavid Howells return true; 258643f5e655SDavid Howells } 258743f5e655SDavid Howells 258843f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 258943f5e655SDavid Howells { 259043f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 259143f5e655SDavid Howells 259243f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 25932e4b7fcdSDave Hansen return 0; 25942e4b7fcdSDave Hansen 25952e4b7fcdSDave Hansen if (readonly_request) 259643f5e655SDavid Howells return mnt_make_readonly(mnt); 259743f5e655SDavid Howells 259868847c94SChristian Brauner mnt->mnt.mnt_flags &= ~MNT_READONLY; 259968847c94SChristian Brauner return 0; 260043f5e655SDavid Howells } 260143f5e655SDavid Howells 260243f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 260343f5e655SDavid Howells { 260443f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 260543f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 260643f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 260743f5e655SDavid Howells } 260843f5e655SDavid Howells 2609f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2610f8b92ba6SDeepa Dinamani { 2611f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2612f8b92ba6SDeepa Dinamani 2613f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2614a128b054SAnthony Iliopoulos (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) && 2615f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2616f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2617f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2618f8b92ba6SDeepa Dinamani struct tm tm; 2619f8b92ba6SDeepa Dinamani 2620f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2621f8b92ba6SDeepa Dinamani 26220ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 26230ecee669SEric Biggers sb->s_type->name, 26240ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 26250ecee669SEric Biggers mntpath, 2626f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2627f8b92ba6SDeepa Dinamani 2628f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2629a128b054SAnthony Iliopoulos sb->s_iflags |= SB_I_TS_EXPIRY_WARNED; 2630f8b92ba6SDeepa Dinamani } 2631f8b92ba6SDeepa Dinamani } 2632f8b92ba6SDeepa Dinamani 263343f5e655SDavid Howells /* 263443f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 263543f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 263643f5e655SDavid Howells * to mount(2). 263743f5e655SDavid Howells */ 263843f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 263943f5e655SDavid Howells { 264043f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 264143f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 264243f5e655SDavid Howells int ret; 264343f5e655SDavid Howells 264443f5e655SDavid Howells if (!check_mnt(mnt)) 264543f5e655SDavid Howells return -EINVAL; 264643f5e655SDavid Howells 264743f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 264843f5e655SDavid Howells return -EINVAL; 264943f5e655SDavid Howells 265043f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 265143f5e655SDavid Howells return -EPERM; 265243f5e655SDavid Howells 2653e58ace1aSChristian Brauner /* 2654e58ace1aSChristian Brauner * We're only checking whether the superblock is read-only not 2655e58ace1aSChristian Brauner * changing it, so only take down_read(&sb->s_umount). 2656e58ace1aSChristian Brauner */ 2657e58ace1aSChristian Brauner down_read(&sb->s_umount); 265868847c94SChristian Brauner lock_mount_hash(); 265943f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 266043f5e655SDavid Howells if (ret == 0) 266143f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 266268847c94SChristian Brauner unlock_mount_hash(); 2663e58ace1aSChristian Brauner up_read(&sb->s_umount); 2664f8b92ba6SDeepa Dinamani 2665f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2666f8b92ba6SDeepa Dinamani 266743f5e655SDavid Howells return ret; 26682e4b7fcdSDave Hansen } 26692e4b7fcdSDave Hansen 26701da177e4SLinus Torvalds /* 26711da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 26721da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 26731da177e4SLinus Torvalds * on it - tough luck. 26741da177e4SLinus Torvalds */ 2675e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2676e462ec50SDavid Howells int mnt_flags, void *data) 26771da177e4SLinus Torvalds { 26781da177e4SLinus Torvalds int err; 26792d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2680143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 26818d0347f6SDavid Howells struct fs_context *fc; 26821da177e4SLinus Torvalds 2683143c8c91SAl Viro if (!check_mnt(mnt)) 26841da177e4SLinus Torvalds return -EINVAL; 26851da177e4SLinus Torvalds 26862d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 26871da177e4SLinus Torvalds return -EINVAL; 26881da177e4SLinus Torvalds 268943f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 269007b64558SEric W. Biederman return -EPERM; 26919566d674SEric W. Biederman 26928d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 26938d0347f6SDavid Howells if (IS_ERR(fc)) 26948d0347f6SDavid Howells return PTR_ERR(fc); 2695ff36fe2cSEric Paris 2696b330966fSMiklos Szeredi fc->oldapi = true; 26978d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 26988d0347f6SDavid Howells if (!err) { 26991da177e4SLinus Torvalds down_write(&sb->s_umount); 270057eccb83SAl Viro err = -EPERM; 270143f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 27028d0347f6SDavid Howells err = reconfigure_super(fc); 270368847c94SChristian Brauner if (!err) { 270468847c94SChristian Brauner lock_mount_hash(); 270543f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 270668847c94SChristian Brauner unlock_mount_hash(); 270768847c94SChristian Brauner } 27080e55a7ccSDan Williams } 27096339dab8SAl Viro up_write(&sb->s_umount); 27108d0347f6SDavid Howells } 2711f8b92ba6SDeepa Dinamani 2712f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2713f8b92ba6SDeepa Dinamani 27148d0347f6SDavid Howells put_fs_context(fc); 27151da177e4SLinus Torvalds return err; 27161da177e4SLinus Torvalds } 27171da177e4SLinus Torvalds 2718cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 27199676f0c6SRam Pai { 2720315fc83eSAl Viro struct mount *p; 2721909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2722fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 27239676f0c6SRam Pai return 1; 27249676f0c6SRam Pai } 27259676f0c6SRam Pai return 0; 27269676f0c6SRam Pai } 27279676f0c6SRam Pai 272844dfd84aSDavid Howells /* 272944dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 273044dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 273144dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 273244dfd84aSDavid Howells * cycles to be made. 273344dfd84aSDavid Howells */ 273444dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 273544dfd84aSDavid Howells { 273644dfd84aSDavid Howells struct mount *p; 273744dfd84aSDavid Howells bool ret = false; 273844dfd84aSDavid Howells 273944dfd84aSDavid Howells lock_mount_hash(); 274044dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 274144dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 274244dfd84aSDavid Howells goto out; 274344dfd84aSDavid Howells 274444dfd84aSDavid Howells ret = true; 274544dfd84aSDavid Howells out: 274644dfd84aSDavid Howells unlock_mount_hash(); 274744dfd84aSDavid Howells return ret; 274844dfd84aSDavid Howells } 274944dfd84aSDavid Howells 27509ffb14efSPavel Tikhomirov static int do_set_group(struct path *from_path, struct path *to_path) 27519ffb14efSPavel Tikhomirov { 27529ffb14efSPavel Tikhomirov struct mount *from, *to; 27539ffb14efSPavel Tikhomirov int err; 27549ffb14efSPavel Tikhomirov 27559ffb14efSPavel Tikhomirov from = real_mount(from_path->mnt); 27569ffb14efSPavel Tikhomirov to = real_mount(to_path->mnt); 27579ffb14efSPavel Tikhomirov 27589ffb14efSPavel Tikhomirov namespace_lock(); 27599ffb14efSPavel Tikhomirov 27609ffb14efSPavel Tikhomirov err = -EINVAL; 27619ffb14efSPavel Tikhomirov /* To and From must be mounted */ 27629ffb14efSPavel Tikhomirov if (!is_mounted(&from->mnt)) 27639ffb14efSPavel Tikhomirov goto out; 27649ffb14efSPavel Tikhomirov if (!is_mounted(&to->mnt)) 27659ffb14efSPavel Tikhomirov goto out; 27669ffb14efSPavel Tikhomirov 27679ffb14efSPavel Tikhomirov err = -EPERM; 27689ffb14efSPavel Tikhomirov /* We should be allowed to modify mount namespaces of both mounts */ 27699ffb14efSPavel Tikhomirov if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27709ffb14efSPavel Tikhomirov goto out; 27719ffb14efSPavel Tikhomirov if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27729ffb14efSPavel Tikhomirov goto out; 27739ffb14efSPavel Tikhomirov 27749ffb14efSPavel Tikhomirov err = -EINVAL; 27759ffb14efSPavel Tikhomirov /* To and From paths should be mount roots */ 27769ffb14efSPavel Tikhomirov if (from_path->dentry != from_path->mnt->mnt_root) 27779ffb14efSPavel Tikhomirov goto out; 27789ffb14efSPavel Tikhomirov if (to_path->dentry != to_path->mnt->mnt_root) 27799ffb14efSPavel Tikhomirov goto out; 27809ffb14efSPavel Tikhomirov 27819ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed across same superblock */ 27829ffb14efSPavel Tikhomirov if (from->mnt.mnt_sb != to->mnt.mnt_sb) 27839ffb14efSPavel Tikhomirov goto out; 27849ffb14efSPavel Tikhomirov 27859ffb14efSPavel Tikhomirov /* From mount root should be wider than To mount root */ 27869ffb14efSPavel Tikhomirov if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) 27879ffb14efSPavel Tikhomirov goto out; 27889ffb14efSPavel Tikhomirov 27899ffb14efSPavel Tikhomirov /* From mount should not have locked children in place of To's root */ 27909ffb14efSPavel Tikhomirov if (has_locked_children(from, to->mnt.mnt_root)) 27919ffb14efSPavel Tikhomirov goto out; 27929ffb14efSPavel Tikhomirov 27939ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed on private mounts */ 27949ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) 27959ffb14efSPavel Tikhomirov goto out; 27969ffb14efSPavel Tikhomirov 27979ffb14efSPavel Tikhomirov /* From should not be private */ 27989ffb14efSPavel Tikhomirov if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) 27999ffb14efSPavel Tikhomirov goto out; 28009ffb14efSPavel Tikhomirov 28019ffb14efSPavel Tikhomirov if (IS_MNT_SLAVE(from)) { 28029ffb14efSPavel Tikhomirov struct mount *m = from->mnt_master; 28039ffb14efSPavel Tikhomirov 28049ffb14efSPavel Tikhomirov list_add(&to->mnt_slave, &m->mnt_slave_list); 28059ffb14efSPavel Tikhomirov to->mnt_master = m; 28069ffb14efSPavel Tikhomirov } 28079ffb14efSPavel Tikhomirov 28089ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(from)) { 28099ffb14efSPavel Tikhomirov to->mnt_group_id = from->mnt_group_id; 28109ffb14efSPavel Tikhomirov list_add(&to->mnt_share, &from->mnt_share); 28119ffb14efSPavel Tikhomirov lock_mount_hash(); 28129ffb14efSPavel Tikhomirov set_mnt_shared(to); 28139ffb14efSPavel Tikhomirov unlock_mount_hash(); 28149ffb14efSPavel Tikhomirov } 28159ffb14efSPavel Tikhomirov 28169ffb14efSPavel Tikhomirov err = 0; 28179ffb14efSPavel Tikhomirov out: 28189ffb14efSPavel Tikhomirov namespace_unlock(); 28199ffb14efSPavel Tikhomirov return err; 28209ffb14efSPavel Tikhomirov } 28219ffb14efSPavel Tikhomirov 28222db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 28231da177e4SLinus Torvalds { 282444dfd84aSDavid Howells struct mnt_namespace *ns; 2825676da58dSAl Viro struct mount *p; 28260fb54e50SAl Viro struct mount *old; 28272763d119SAl Viro struct mount *parent; 28282763d119SAl Viro struct mountpoint *mp, *old_mp; 282957eccb83SAl Viro int err; 283044dfd84aSDavid Howells bool attached; 28311da177e4SLinus Torvalds 28322db154b3SDavid Howells mp = lock_mount(new_path); 283384d17192SAl Viro if (IS_ERR(mp)) 28342db154b3SDavid Howells return PTR_ERR(mp); 2835cc53ce53SDavid Howells 28362db154b3SDavid Howells old = real_mount(old_path->mnt); 28372db154b3SDavid Howells p = real_mount(new_path->mnt); 28382763d119SAl Viro parent = old->mnt_parent; 283944dfd84aSDavid Howells attached = mnt_has_parent(old); 28402763d119SAl Viro old_mp = old->mnt_mp; 284144dfd84aSDavid Howells ns = old->mnt_ns; 2842143c8c91SAl Viro 28431da177e4SLinus Torvalds err = -EINVAL; 284444dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 284544dfd84aSDavid Howells if (!check_mnt(p)) 28462db154b3SDavid Howells goto out; 28471da177e4SLinus Torvalds 2848570d7a98SEric Biggers /* The thing moved must be mounted... */ 2849570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 285044dfd84aSDavid Howells goto out; 285144dfd84aSDavid Howells 2852570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2853570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 28542db154b3SDavid Howells goto out; 28551da177e4SLinus Torvalds 28562db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 28572db154b3SDavid Howells goto out; 28582db154b3SDavid Howells 28592db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 28602db154b3SDavid Howells goto out; 28612db154b3SDavid Howells 28622db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 28632db154b3SDavid Howells d_is_dir(old_path->dentry)) 28642db154b3SDavid Howells goto out; 286521444403SRam Pai /* 286621444403SRam Pai * Don't move a mount residing in a shared parent. 286721444403SRam Pai */ 28682763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 28692db154b3SDavid Howells goto out; 28709676f0c6SRam Pai /* 28719676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 28729676f0c6SRam Pai * mount which is shared. 28739676f0c6SRam Pai */ 2874fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 28752db154b3SDavid Howells goto out; 28761da177e4SLinus Torvalds err = -ELOOP; 287744dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 287844dfd84aSDavid Howells goto out; 2879fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2880676da58dSAl Viro if (p == old) 28812db154b3SDavid Howells goto out; 28821da177e4SLinus Torvalds 28832db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 28842763d119SAl Viro attached); 28854ac91378SJan Blunck if (err) 28862db154b3SDavid Howells goto out; 28871da177e4SLinus Torvalds 28881da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 28891da177e4SLinus Torvalds * automatically */ 28906776db3dSAl Viro list_del_init(&old->mnt_expire); 28912763d119SAl Viro if (attached) 28922763d119SAl Viro put_mountpoint(old_mp); 28931da177e4SLinus Torvalds out: 28942db154b3SDavid Howells unlock_mount(mp); 289544dfd84aSDavid Howells if (!err) { 28962763d119SAl Viro if (attached) 28972763d119SAl Viro mntput_no_expire(parent); 28982763d119SAl Viro else 289944dfd84aSDavid Howells free_mnt_ns(ns); 290044dfd84aSDavid Howells } 29012db154b3SDavid Howells return err; 29022db154b3SDavid Howells } 29032db154b3SDavid Howells 29042db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 29052db154b3SDavid Howells { 29062db154b3SDavid Howells struct path old_path; 29072db154b3SDavid Howells int err; 29082db154b3SDavid Howells 29092db154b3SDavid Howells if (!old_name || !*old_name) 29102db154b3SDavid Howells return -EINVAL; 29112db154b3SDavid Howells 29122db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 29132db154b3SDavid Howells if (err) 29142db154b3SDavid Howells return err; 29152db154b3SDavid Howells 29162db154b3SDavid Howells err = do_move_mount(&old_path, path); 29172d92ab3cSAl Viro path_put(&old_path); 29181da177e4SLinus Torvalds return err; 29191da177e4SLinus Torvalds } 29201da177e4SLinus Torvalds 29219d412a43SAl Viro /* 29229d412a43SAl Viro * add a mount into a namespace's mount tree 29239d412a43SAl Viro */ 29248f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 29251e2d8464SAl Viro const struct path *path, int mnt_flags) 29269d412a43SAl Viro { 29278f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 29289d412a43SAl Viro 2929f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 29309d412a43SAl Viro 293184d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2932156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2933156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 29348f11538eSAl Viro return -EINVAL; 2935156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 293684d17192SAl Viro if (!parent->mnt_ns) 29378f11538eSAl Viro return -EINVAL; 2938156cacb1SAl Viro } 29399d412a43SAl Viro 29409d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 294195bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 29429d412a43SAl Viro path->mnt->mnt_root == path->dentry) 29438f11538eSAl Viro return -EBUSY; 29449d412a43SAl Viro 2945e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 29468f11538eSAl Viro return -EINVAL; 29479d412a43SAl Viro 294895bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 29498f11538eSAl Viro return graft_tree(newmnt, parent, mp); 29509d412a43SAl Viro } 2951b1e75df4SAl Viro 2952132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2953132e4608SDavid Howells 2954132e4608SDavid Howells /* 2955132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2956132e4608SDavid Howells * be added to the namespace tree. 2957132e4608SDavid Howells */ 2958132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2959132e4608SDavid Howells unsigned int mnt_flags) 2960132e4608SDavid Howells { 2961132e4608SDavid Howells struct vfsmount *mnt; 29628f11538eSAl Viro struct mountpoint *mp; 2963132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2964132e4608SDavid Howells int error; 2965132e4608SDavid Howells 2966c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2967c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2968c9ce29edSAl Viro error = -EPERM; 2969c9ce29edSAl Viro 2970c9ce29edSAl Viro if (unlikely(error)) { 2971c9ce29edSAl Viro fc_drop_locked(fc); 2972c9ce29edSAl Viro return error; 2973132e4608SDavid Howells } 2974132e4608SDavid Howells 2975132e4608SDavid Howells up_write(&sb->s_umount); 2976132e4608SDavid Howells 2977132e4608SDavid Howells mnt = vfs_create_mount(fc); 2978132e4608SDavid Howells if (IS_ERR(mnt)) 2979132e4608SDavid Howells return PTR_ERR(mnt); 2980132e4608SDavid Howells 2981f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 2982f8b92ba6SDeepa Dinamani 29838f11538eSAl Viro mp = lock_mount(mountpoint); 29848f11538eSAl Viro if (IS_ERR(mp)) { 29858f11538eSAl Viro mntput(mnt); 29868f11538eSAl Viro return PTR_ERR(mp); 29878f11538eSAl Viro } 29888f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 29898f11538eSAl Viro unlock_mount(mp); 29900ecee669SEric Biggers if (error < 0) 29910ecee669SEric Biggers mntput(mnt); 2992f8b92ba6SDeepa Dinamani return error; 2993f8b92ba6SDeepa Dinamani } 2994f8b92ba6SDeepa Dinamani 29951da177e4SLinus Torvalds /* 29961da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 29971da177e4SLinus Torvalds * namespace's tree 29981da177e4SLinus Torvalds */ 2999e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 3000808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 30011da177e4SLinus Torvalds { 30020c55cfc4SEric W. Biederman struct file_system_type *type; 3003a0c9a8b8SAl Viro struct fs_context *fc; 3004a0c9a8b8SAl Viro const char *subtype = NULL; 3005a0c9a8b8SAl Viro int err = 0; 30061da177e4SLinus Torvalds 30070c55cfc4SEric W. Biederman if (!fstype) 30081da177e4SLinus Torvalds return -EINVAL; 30091da177e4SLinus Torvalds 30100c55cfc4SEric W. Biederman type = get_fs_type(fstype); 30110c55cfc4SEric W. Biederman if (!type) 30120c55cfc4SEric W. Biederman return -ENODEV; 30130c55cfc4SEric W. Biederman 3014a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 3015a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 3016a0c9a8b8SAl Viro if (subtype) { 3017a0c9a8b8SAl Viro subtype++; 3018a0c9a8b8SAl Viro if (!*subtype) { 30190c55cfc4SEric W. Biederman put_filesystem(type); 3020a0c9a8b8SAl Viro return -EINVAL; 3021a0c9a8b8SAl Viro } 3022a0c9a8b8SAl Viro } 30238654df4eSEric W. Biederman } 30248654df4eSEric W. Biederman 3025a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 3026a0c9a8b8SAl Viro put_filesystem(type); 3027a0c9a8b8SAl Viro if (IS_ERR(fc)) 3028a0c9a8b8SAl Viro return PTR_ERR(fc); 3029a0c9a8b8SAl Viro 30303e1aeb00SDavid Howells if (subtype) 30313e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 30323e1aeb00SDavid Howells subtype, strlen(subtype)); 30333e1aeb00SDavid Howells if (!err && name) 30343e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 3035a0c9a8b8SAl Viro if (!err) 3036a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 3037c3aabf07SAl Viro if (!err && !mount_capable(fc)) 3038c3aabf07SAl Viro err = -EPERM; 3039a0c9a8b8SAl Viro if (!err) 3040a0c9a8b8SAl Viro err = vfs_get_tree(fc); 3041132e4608SDavid Howells if (!err) 3042132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 3043a0c9a8b8SAl Viro 3044a0c9a8b8SAl Viro put_fs_context(fc); 304515f9a3f3SAl Viro return err; 30461da177e4SLinus Torvalds } 30471da177e4SLinus Torvalds 30481e2d8464SAl Viro int finish_automount(struct vfsmount *m, const struct path *path) 304919a167afSAl Viro { 305026df6034SAl Viro struct dentry *dentry = path->dentry; 30518f11538eSAl Viro struct mountpoint *mp; 305225e195aaSAl Viro struct mount *mnt; 305319a167afSAl Viro int err; 305425e195aaSAl Viro 305525e195aaSAl Viro if (!m) 305625e195aaSAl Viro return 0; 305725e195aaSAl Viro if (IS_ERR(m)) 305825e195aaSAl Viro return PTR_ERR(m); 305925e195aaSAl Viro 306025e195aaSAl Viro mnt = real_mount(m); 306119a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 306219a167afSAl Viro * expired before we get a chance to add it 306319a167afSAl Viro */ 30646776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 306519a167afSAl Viro 306619a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 306726df6034SAl Viro m->mnt_root == dentry) { 3068b1e75df4SAl Viro err = -ELOOP; 306926df6034SAl Viro goto discard; 307019a167afSAl Viro } 307119a167afSAl Viro 307226df6034SAl Viro /* 307326df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 307426df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 307526df6034SAl Viro * got", not "try to mount it on top". 307626df6034SAl Viro */ 307726df6034SAl Viro inode_lock(dentry->d_inode); 307826df6034SAl Viro namespace_lock(); 307926df6034SAl Viro if (unlikely(cant_mount(dentry))) { 308026df6034SAl Viro err = -ENOENT; 308126df6034SAl Viro goto discard_locked; 308226df6034SAl Viro } 308326df6034SAl Viro rcu_read_lock(); 308426df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 308526df6034SAl Viro rcu_read_unlock(); 308626df6034SAl Viro err = 0; 308726df6034SAl Viro goto discard_locked; 308826df6034SAl Viro } 308926df6034SAl Viro rcu_read_unlock(); 309026df6034SAl Viro mp = get_mountpoint(dentry); 30918f11538eSAl Viro if (IS_ERR(mp)) { 30928f11538eSAl Viro err = PTR_ERR(mp); 309326df6034SAl Viro goto discard_locked; 30948f11538eSAl Viro } 309526df6034SAl Viro 30968f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 30978f11538eSAl Viro unlock_mount(mp); 309826df6034SAl Viro if (unlikely(err)) 309926df6034SAl Viro goto discard; 310026df6034SAl Viro mntput(m); 3101b1e75df4SAl Viro return 0; 310226df6034SAl Viro 310326df6034SAl Viro discard_locked: 310426df6034SAl Viro namespace_unlock(); 310526df6034SAl Viro inode_unlock(dentry->d_inode); 310626df6034SAl Viro discard: 3107b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 31086776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 310997216be0SAl Viro namespace_lock(); 31106776db3dSAl Viro list_del_init(&mnt->mnt_expire); 311197216be0SAl Viro namespace_unlock(); 311219a167afSAl Viro } 3113b1e75df4SAl Viro mntput(m); 3114b1e75df4SAl Viro mntput(m); 311519a167afSAl Viro return err; 311619a167afSAl Viro } 311719a167afSAl Viro 3118ea5b778aSDavid Howells /** 3119ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 3120ea5b778aSDavid Howells * @mnt: The mount to list. 3121ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 3122ea5b778aSDavid Howells */ 3123ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 3124ea5b778aSDavid Howells { 312597216be0SAl Viro namespace_lock(); 3126ea5b778aSDavid Howells 31276776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 3128ea5b778aSDavid Howells 312997216be0SAl Viro namespace_unlock(); 3130ea5b778aSDavid Howells } 3131ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 3132ea5b778aSDavid Howells 3133ea5b778aSDavid Howells /* 31341da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 31351da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 31361da177e4SLinus Torvalds * here 31371da177e4SLinus Torvalds */ 31381da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 31391da177e4SLinus Torvalds { 3140761d5c38SAl Viro struct mount *mnt, *next; 31411da177e4SLinus Torvalds LIST_HEAD(graveyard); 31421da177e4SLinus Torvalds 31431da177e4SLinus Torvalds if (list_empty(mounts)) 31441da177e4SLinus Torvalds return; 31451da177e4SLinus Torvalds 314697216be0SAl Viro namespace_lock(); 3147719ea2fbSAl Viro lock_mount_hash(); 31481da177e4SLinus Torvalds 31491da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 31501da177e4SLinus Torvalds * following criteria: 31511da177e4SLinus Torvalds * - only referenced by its parent vfsmount 31521da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 31531da177e4SLinus Torvalds * cleared by mntput()) 31541da177e4SLinus Torvalds */ 31556776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 3156863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 31571ab59738SAl Viro propagate_mount_busy(mnt, 1)) 31581da177e4SLinus Torvalds continue; 31596776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 31601da177e4SLinus Torvalds } 3161bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 31626776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 3163143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 3164e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3165bcc5c7d2SAl Viro } 3166719ea2fbSAl Viro unlock_mount_hash(); 31673ab6abeeSAl Viro namespace_unlock(); 31681da177e4SLinus Torvalds } 31691da177e4SLinus Torvalds 31701da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 31711da177e4SLinus Torvalds 31721da177e4SLinus Torvalds /* 31735528f911STrond Myklebust * Ripoff of 'select_parent()' 31745528f911STrond Myklebust * 31755528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 31765528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 31775528f911STrond Myklebust */ 3178692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 31795528f911STrond Myklebust { 3180692afc31SAl Viro struct mount *this_parent = parent; 31815528f911STrond Myklebust struct list_head *next; 31825528f911STrond Myklebust int found = 0; 31835528f911STrond Myklebust 31845528f911STrond Myklebust repeat: 31856b41d536SAl Viro next = this_parent->mnt_mounts.next; 31865528f911STrond Myklebust resume: 31876b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 31885528f911STrond Myklebust struct list_head *tmp = next; 31896b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 31905528f911STrond Myklebust 31915528f911STrond Myklebust next = tmp->next; 3192692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 31935528f911STrond Myklebust continue; 31945528f911STrond Myklebust /* 31955528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 31965528f911STrond Myklebust */ 31976b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 31985528f911STrond Myklebust this_parent = mnt; 31995528f911STrond Myklebust goto repeat; 32005528f911STrond Myklebust } 32015528f911STrond Myklebust 32021ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 32036776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 32045528f911STrond Myklebust found++; 32055528f911STrond Myklebust } 32065528f911STrond Myklebust } 32075528f911STrond Myklebust /* 32085528f911STrond Myklebust * All done at this level ... ascend and resume the search 32095528f911STrond Myklebust */ 32105528f911STrond Myklebust if (this_parent != parent) { 32116b41d536SAl Viro next = this_parent->mnt_child.next; 32120714a533SAl Viro this_parent = this_parent->mnt_parent; 32135528f911STrond Myklebust goto resume; 32145528f911STrond Myklebust } 32155528f911STrond Myklebust return found; 32165528f911STrond Myklebust } 32175528f911STrond Myklebust 32185528f911STrond Myklebust /* 32195528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 32205528f911STrond Myklebust * submounts of a specific parent mountpoint 322199b7db7bSNick Piggin * 322248a066e7SAl Viro * mount_lock must be held for write 32235528f911STrond Myklebust */ 3224b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 32255528f911STrond Myklebust { 32265528f911STrond Myklebust LIST_HEAD(graveyard); 3227761d5c38SAl Viro struct mount *m; 32285528f911STrond Myklebust 32295528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 3230c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3231bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3232761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 32336776db3dSAl Viro mnt_expire); 3234143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3235e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3236bcc5c7d2SAl Viro } 3237bcc5c7d2SAl Viro } 32385528f911STrond Myklebust } 32395528f911STrond Myklebust 3240028abd92SChristoph Hellwig static void *copy_mount_options(const void __user * data) 32411da177e4SLinus Torvalds { 3242b40ef869SAl Viro char *copy; 3243d563d678SCatalin Marinas unsigned left, offset; 32441da177e4SLinus Torvalds 32451da177e4SLinus Torvalds if (!data) 3246b40ef869SAl Viro return NULL; 32471da177e4SLinus Torvalds 3248b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3249b40ef869SAl Viro if (!copy) 3250b40ef869SAl Viro return ERR_PTR(-ENOMEM); 32511da177e4SLinus Torvalds 3252d563d678SCatalin Marinas left = copy_from_user(copy, data, PAGE_SIZE); 32531da177e4SLinus Torvalds 3254d563d678SCatalin Marinas /* 3255d563d678SCatalin Marinas * Not all architectures have an exact copy_from_user(). Resort to 3256d563d678SCatalin Marinas * byte at a time. 3257d563d678SCatalin Marinas */ 3258d563d678SCatalin Marinas offset = PAGE_SIZE - left; 3259d563d678SCatalin Marinas while (left) { 3260d563d678SCatalin Marinas char c; 3261d563d678SCatalin Marinas if (get_user(c, (const char __user *)data + offset)) 3262d563d678SCatalin Marinas break; 3263d563d678SCatalin Marinas copy[offset] = c; 3264d563d678SCatalin Marinas left--; 3265d563d678SCatalin Marinas offset++; 3266d563d678SCatalin Marinas } 3267d563d678SCatalin Marinas 3268d563d678SCatalin Marinas if (left == PAGE_SIZE) { 3269b40ef869SAl Viro kfree(copy); 3270b40ef869SAl Viro return ERR_PTR(-EFAULT); 32711da177e4SLinus Torvalds } 3272d563d678SCatalin Marinas 3273b40ef869SAl Viro return copy; 32741da177e4SLinus Torvalds } 32751da177e4SLinus Torvalds 3276028abd92SChristoph Hellwig static char *copy_mount_string(const void __user *data) 3277eca6f534SVegard Nossum { 3278fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3279eca6f534SVegard Nossum } 3280eca6f534SVegard Nossum 32811da177e4SLinus Torvalds /* 32821da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 32831da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 32841da177e4SLinus Torvalds * 32851da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 32861da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 32871da177e4SLinus Torvalds * information (or be NULL). 32881da177e4SLinus Torvalds * 32891da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 32901da177e4SLinus Torvalds * When the flags word was introduced its top half was required 32911da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 32921da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 32931da177e4SLinus Torvalds * and must be discarded. 32941da177e4SLinus Torvalds */ 3295c60166f0SChristoph Hellwig int path_mount(const char *dev_name, struct path *path, 3296808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 32971da177e4SLinus Torvalds { 3298e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 3299a1e6aaa3SChristoph Hellwig int ret; 33001da177e4SLinus Torvalds 33011da177e4SLinus Torvalds /* Discard magic */ 33021da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 33031da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 33041da177e4SLinus Torvalds 33051da177e4SLinus Torvalds /* Basic sanity checks */ 33061da177e4SLinus Torvalds if (data_page) 33071da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 33081da177e4SLinus Torvalds 3309e462ec50SDavid Howells if (flags & MS_NOUSER) 3310e462ec50SDavid Howells return -EINVAL; 3311e462ec50SDavid Howells 3312a1e6aaa3SChristoph Hellwig ret = security_sb_mount(dev_name, path, type_page, flags, data_page); 3313a1e6aaa3SChristoph Hellwig if (ret) 3314a1e6aaa3SChristoph Hellwig return ret; 3315a1e6aaa3SChristoph Hellwig if (!may_mount()) 3316a1e6aaa3SChristoph Hellwig return -EPERM; 3317f7e33bdbSJeff Layton if (flags & SB_MANDLOCK) 3318f7e33bdbSJeff Layton warn_mandlock(); 3319a27ab9f2STetsuo Handa 3320613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3321613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 33220a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 33230a1c01c9SMatthew Garrett 33241da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 33251da177e4SLinus Torvalds if (flags & MS_NOSUID) 33261da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 33271da177e4SLinus Torvalds if (flags & MS_NODEV) 33281da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 33291da177e4SLinus Torvalds if (flags & MS_NOEXEC) 33301da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3331fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3332fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3333fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3334fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3335d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3336d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3337a9e5b732SDavid Howells if (flags & MS_RDONLY) 33382e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3339dab741e0SMattias Nissler if (flags & MS_NOSYMFOLLOW) 3340dab741e0SMattias Nissler mnt_flags |= MNT_NOSYMFOLLOW; 3341fc33a7bbSChristoph Hellwig 3342ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3343ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3344ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3345ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3346ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3347a1e6aaa3SChristoph Hellwig mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; 3348ffbc6f0eSEric W. Biederman } 3349ffbc6f0eSEric W. Biederman 3350e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3351e462ec50SDavid Howells SB_SYNCHRONOUS | 3352e462ec50SDavid Howells SB_MANDLOCK | 3353e462ec50SDavid Howells SB_DIRSYNC | 3354e462ec50SDavid Howells SB_SILENT | 3355917086ffSMimi Zohar SB_POSIXACL | 3356d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3357917086ffSMimi Zohar SB_I_VERSION); 33581da177e4SLinus Torvalds 335943f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 3360a1e6aaa3SChristoph Hellwig return do_reconfigure_mnt(path, mnt_flags); 3361a1e6aaa3SChristoph Hellwig if (flags & MS_REMOUNT) 3362a1e6aaa3SChristoph Hellwig return do_remount(path, flags, sb_flags, mnt_flags, data_page); 3363a1e6aaa3SChristoph Hellwig if (flags & MS_BIND) 3364a1e6aaa3SChristoph Hellwig return do_loopback(path, dev_name, flags & MS_REC); 3365a1e6aaa3SChristoph Hellwig if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 3366a1e6aaa3SChristoph Hellwig return do_change_type(path, flags); 3367a1e6aaa3SChristoph Hellwig if (flags & MS_MOVE) 3368a1e6aaa3SChristoph Hellwig return do_move_mount_old(path, dev_name); 3369a1e6aaa3SChristoph Hellwig 3370a1e6aaa3SChristoph Hellwig return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, 33711da177e4SLinus Torvalds data_page); 3372a1e6aaa3SChristoph Hellwig } 3373a1e6aaa3SChristoph Hellwig 3374a1e6aaa3SChristoph Hellwig long do_mount(const char *dev_name, const char __user *dir_name, 3375a1e6aaa3SChristoph Hellwig const char *type_page, unsigned long flags, void *data_page) 3376a1e6aaa3SChristoph Hellwig { 3377a1e6aaa3SChristoph Hellwig struct path path; 3378a1e6aaa3SChristoph Hellwig int ret; 3379a1e6aaa3SChristoph Hellwig 3380a1e6aaa3SChristoph Hellwig ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3381a1e6aaa3SChristoph Hellwig if (ret) 3382a1e6aaa3SChristoph Hellwig return ret; 3383a1e6aaa3SChristoph Hellwig ret = path_mount(dev_name, &path, type_page, flags, data_page); 33842d92ab3cSAl Viro path_put(&path); 3385a1e6aaa3SChristoph Hellwig return ret; 33861da177e4SLinus Torvalds } 33871da177e4SLinus Torvalds 3388537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3389537f7ccbSEric W. Biederman { 3390537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3391537f7ccbSEric W. Biederman } 3392537f7ccbSEric W. Biederman 3393537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3394537f7ccbSEric W. Biederman { 3395537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3396537f7ccbSEric W. Biederman } 3397537f7ccbSEric W. Biederman 3398771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3399771b1371SEric W. Biederman { 340074e83122SAl Viro if (!is_anon_ns(ns)) 34016344c433SAl Viro ns_free_inum(&ns->ns); 3402537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3403771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3404771b1371SEric W. Biederman kfree(ns); 3405771b1371SEric W. Biederman } 3406771b1371SEric W. Biederman 34078823c079SEric W. Biederman /* 34088823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 34098823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 34108823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 34118823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 34128823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 34138823c079SEric W. Biederman */ 34148823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 34158823c079SEric W. Biederman 341674e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3417cf8d2c11STrond Myklebust { 3418cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3419537f7ccbSEric W. Biederman struct ucounts *ucounts; 342098f842e6SEric W. Biederman int ret; 3421cf8d2c11STrond Myklebust 3422537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3423537f7ccbSEric W. Biederman if (!ucounts) 3424df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3425537f7ccbSEric W. Biederman 342630acd0bdSVasily Averin new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); 3427537f7ccbSEric W. Biederman if (!new_ns) { 3428537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3429cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3430537f7ccbSEric W. Biederman } 343174e83122SAl Viro if (!anon) { 34326344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 343398f842e6SEric W. Biederman if (ret) { 343498f842e6SEric W. Biederman kfree(new_ns); 3435537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 343698f842e6SEric W. Biederman return ERR_PTR(ret); 343798f842e6SEric W. Biederman } 343874e83122SAl Viro } 343933c42940SAl Viro new_ns->ns.ops = &mntns_operations; 344074e83122SAl Viro if (!anon) 34418823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 34421a7b8969SKirill Tkhai refcount_set(&new_ns->ns.count, 1); 3443cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3444cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 34459f6c61f9SMiklos Szeredi spin_lock_init(&new_ns->ns_lock); 3446771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3447537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3448cf8d2c11STrond Myklebust return new_ns; 3449cf8d2c11STrond Myklebust } 3450cf8d2c11STrond Myklebust 34510766f788SEmese Revfy __latent_entropy 34529559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 34539559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 34541da177e4SLinus Torvalds { 34556b3286edSKirill Korotaev struct mnt_namespace *new_ns; 34567f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3457315fc83eSAl Viro struct mount *p, *q; 34589559f689SAl Viro struct mount *old; 3459cb338d06SAl Viro struct mount *new; 34607a472ef4SEric W. Biederman int copy_flags; 34611da177e4SLinus Torvalds 34629559f689SAl Viro BUG_ON(!ns); 34639559f689SAl Viro 34649559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 34659559f689SAl Viro get_mnt_ns(ns); 34669559f689SAl Viro return ns; 34679559f689SAl Viro } 34689559f689SAl Viro 34699559f689SAl Viro old = ns->root; 34709559f689SAl Viro 347174e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3472cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3473cf8d2c11STrond Myklebust return new_ns; 34741da177e4SLinus Torvalds 347597216be0SAl Viro namespace_lock(); 34761da177e4SLinus Torvalds /* First pass: copy the tree topology */ 34774ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 34789559f689SAl Viro if (user_ns != ns->user_ns) 34793bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 34807a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3481be34d1a3SDavid Howells if (IS_ERR(new)) { 3482328e6d90SAl Viro namespace_unlock(); 3483771b1371SEric W. Biederman free_mnt_ns(new_ns); 3484be34d1a3SDavid Howells return ERR_CAST(new); 34851da177e4SLinus Torvalds } 34863bd045ccSAl Viro if (user_ns != ns->user_ns) { 34873bd045ccSAl Viro lock_mount_hash(); 34883bd045ccSAl Viro lock_mnt_tree(new); 34893bd045ccSAl Viro unlock_mount_hash(); 34903bd045ccSAl Viro } 3491be08d6d2SAl Viro new_ns->root = new; 34921a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 34931da177e4SLinus Torvalds 34941da177e4SLinus Torvalds /* 34951da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 34961da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 34971da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 34981da177e4SLinus Torvalds */ 3499909b0a88SAl Viro p = old; 3500cb338d06SAl Viro q = new; 35011da177e4SLinus Torvalds while (p) { 3502143c8c91SAl Viro q->mnt_ns = new_ns; 3503d2921684SEric W. Biederman new_ns->mounts++; 35049559f689SAl Viro if (new_fs) { 35059559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 35069559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3507315fc83eSAl Viro rootmnt = &p->mnt; 35081da177e4SLinus Torvalds } 35099559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 35109559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3511315fc83eSAl Viro pwdmnt = &p->mnt; 35121da177e4SLinus Torvalds } 35131da177e4SLinus Torvalds } 3514909b0a88SAl Viro p = next_mnt(p, old); 3515909b0a88SAl Viro q = next_mnt(q, new); 35164ce5d2b1SEric W. Biederman if (!q) 35174ce5d2b1SEric W. Biederman break; 35184ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 35194ce5d2b1SEric W. Biederman p = next_mnt(p, old); 35201da177e4SLinus Torvalds } 3521328e6d90SAl Viro namespace_unlock(); 35221da177e4SLinus Torvalds 35231da177e4SLinus Torvalds if (rootmnt) 3524f03c6599SAl Viro mntput(rootmnt); 35251da177e4SLinus Torvalds if (pwdmnt) 3526f03c6599SAl Viro mntput(pwdmnt); 35271da177e4SLinus Torvalds 3528741a2951SJANAK DESAI return new_ns; 3529741a2951SJANAK DESAI } 3530741a2951SJANAK DESAI 353174e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3532cf8d2c11STrond Myklebust { 35331a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3534ea441d11SAl Viro struct mnt_namespace *ns; 3535d31da0f0SAl Viro struct super_block *s; 3536ea441d11SAl Viro struct path path; 3537ea441d11SAl Viro int err; 3538ea441d11SAl Viro 353974e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 354074e83122SAl Viro if (IS_ERR(ns)) { 354174e83122SAl Viro mntput(m); 3542ea441d11SAl Viro return ERR_CAST(ns); 354374e83122SAl Viro } 354474e83122SAl Viro mnt->mnt_ns = ns; 354574e83122SAl Viro ns->root = mnt; 354674e83122SAl Viro ns->mounts++; 354774e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3548ea441d11SAl Viro 354974e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3550ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3551ea441d11SAl Viro 3552ea441d11SAl Viro put_mnt_ns(ns); 3553ea441d11SAl Viro 3554ea441d11SAl Viro if (err) 3555ea441d11SAl Viro return ERR_PTR(err); 3556ea441d11SAl Viro 3557ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3558d31da0f0SAl Viro s = path.mnt->mnt_sb; 3559d31da0f0SAl Viro atomic_inc(&s->s_active); 3560ea441d11SAl Viro mntput(path.mnt); 3561ea441d11SAl Viro /* lock the sucker */ 3562d31da0f0SAl Viro down_write(&s->s_umount); 3563ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3564ea441d11SAl Viro return path.dentry; 3565ea441d11SAl Viro } 3566ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3567ea441d11SAl Viro 3568cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3569cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 35701da177e4SLinus Torvalds { 3571eca6f534SVegard Nossum int ret; 3572eca6f534SVegard Nossum char *kernel_type; 3573eca6f534SVegard Nossum char *kernel_dev; 3574b40ef869SAl Viro void *options; 35751da177e4SLinus Torvalds 3576b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3577b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3578b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3579eca6f534SVegard Nossum goto out_type; 35801da177e4SLinus Torvalds 3581b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3582b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3583b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3584eca6f534SVegard Nossum goto out_dev; 35851da177e4SLinus Torvalds 3586b40ef869SAl Viro options = copy_mount_options(data); 3587b40ef869SAl Viro ret = PTR_ERR(options); 3588b40ef869SAl Viro if (IS_ERR(options)) 3589eca6f534SVegard Nossum goto out_data; 35901da177e4SLinus Torvalds 3591b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3592eca6f534SVegard Nossum 3593b40ef869SAl Viro kfree(options); 3594eca6f534SVegard Nossum out_data: 3595eca6f534SVegard Nossum kfree(kernel_dev); 3596eca6f534SVegard Nossum out_dev: 3597eca6f534SVegard Nossum kfree(kernel_type); 3598eca6f534SVegard Nossum out_type: 3599eca6f534SVegard Nossum return ret; 36001da177e4SLinus Torvalds } 36011da177e4SLinus Torvalds 36025b490500SChristian Brauner #define FSMOUNT_VALID_FLAGS \ 36035b490500SChristian Brauner (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ 3604dd8b477fSChristian Brauner MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ 3605dd8b477fSChristian Brauner MOUNT_ATTR_NOSYMFOLLOW) 36065b490500SChristian Brauner 36079caccd41SChristian Brauner #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) 36082a186721SChristian Brauner 36092a186721SChristian Brauner #define MOUNT_SETATTR_PROPAGATION_FLAGS \ 36102a186721SChristian Brauner (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) 36112a186721SChristian Brauner 36125b490500SChristian Brauner static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) 36135b490500SChristian Brauner { 36145b490500SChristian Brauner unsigned int mnt_flags = 0; 36155b490500SChristian Brauner 36165b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_RDONLY) 36175b490500SChristian Brauner mnt_flags |= MNT_READONLY; 36185b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOSUID) 36195b490500SChristian Brauner mnt_flags |= MNT_NOSUID; 36205b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODEV) 36215b490500SChristian Brauner mnt_flags |= MNT_NODEV; 36225b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOEXEC) 36235b490500SChristian Brauner mnt_flags |= MNT_NOEXEC; 36245b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODIRATIME) 36255b490500SChristian Brauner mnt_flags |= MNT_NODIRATIME; 3626dd8b477fSChristian Brauner if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) 3627dd8b477fSChristian Brauner mnt_flags |= MNT_NOSYMFOLLOW; 36285b490500SChristian Brauner 36295b490500SChristian Brauner return mnt_flags; 36305b490500SChristian Brauner } 36315b490500SChristian Brauner 36321da177e4SLinus Torvalds /* 363393766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 363493766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 363593766fbdSDavid Howells */ 363693766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 363793766fbdSDavid Howells unsigned int, attr_flags) 363893766fbdSDavid Howells { 363993766fbdSDavid Howells struct mnt_namespace *ns; 364093766fbdSDavid Howells struct fs_context *fc; 364193766fbdSDavid Howells struct file *file; 364293766fbdSDavid Howells struct path newmount; 364393766fbdSDavid Howells struct mount *mnt; 364493766fbdSDavid Howells struct fd f; 364593766fbdSDavid Howells unsigned int mnt_flags = 0; 364693766fbdSDavid Howells long ret; 364793766fbdSDavid Howells 364893766fbdSDavid Howells if (!may_mount()) 364993766fbdSDavid Howells return -EPERM; 365093766fbdSDavid Howells 365193766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 365293766fbdSDavid Howells return -EINVAL; 365393766fbdSDavid Howells 36545b490500SChristian Brauner if (attr_flags & ~FSMOUNT_VALID_FLAGS) 365593766fbdSDavid Howells return -EINVAL; 365693766fbdSDavid Howells 36575b490500SChristian Brauner mnt_flags = attr_flags_to_mnt_flags(attr_flags); 365893766fbdSDavid Howells 365993766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 366093766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 366193766fbdSDavid Howells break; 366293766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 366393766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 366493766fbdSDavid Howells break; 366593766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 366693766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 366793766fbdSDavid Howells break; 366893766fbdSDavid Howells default: 366993766fbdSDavid Howells return -EINVAL; 367093766fbdSDavid Howells } 367193766fbdSDavid Howells 367293766fbdSDavid Howells f = fdget(fs_fd); 367393766fbdSDavid Howells if (!f.file) 367493766fbdSDavid Howells return -EBADF; 367593766fbdSDavid Howells 367693766fbdSDavid Howells ret = -EINVAL; 367793766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 367893766fbdSDavid Howells goto err_fsfd; 367993766fbdSDavid Howells 368093766fbdSDavid Howells fc = f.file->private_data; 368193766fbdSDavid Howells 368293766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 368393766fbdSDavid Howells if (ret < 0) 368493766fbdSDavid Howells goto err_fsfd; 368593766fbdSDavid Howells 368693766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 368793766fbdSDavid Howells ret = -EINVAL; 368893766fbdSDavid Howells if (!fc->root) 368993766fbdSDavid Howells goto err_unlock; 369093766fbdSDavid Howells 369193766fbdSDavid Howells ret = -EPERM; 369293766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 369393766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 369493766fbdSDavid Howells goto err_unlock; 369593766fbdSDavid Howells } 369693766fbdSDavid Howells 369793766fbdSDavid Howells ret = -EBUSY; 369893766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 369993766fbdSDavid Howells goto err_unlock; 370093766fbdSDavid Howells 3701f7e33bdbSJeff Layton if (fc->sb_flags & SB_MANDLOCK) 3702f7e33bdbSJeff Layton warn_mandlock(); 370393766fbdSDavid Howells 370493766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 370593766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 370693766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 370793766fbdSDavid Howells goto err_unlock; 370893766fbdSDavid Howells } 370993766fbdSDavid Howells newmount.dentry = dget(fc->root); 371093766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 371193766fbdSDavid Howells 371293766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 371393766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 371493766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 371593766fbdSDavid Howells * don't want to have to handle any errors incurred. 371693766fbdSDavid Howells */ 371793766fbdSDavid Howells vfs_clean_context(fc); 371893766fbdSDavid Howells 371993766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 372093766fbdSDavid Howells if (IS_ERR(ns)) { 372193766fbdSDavid Howells ret = PTR_ERR(ns); 372293766fbdSDavid Howells goto err_path; 372393766fbdSDavid Howells } 372493766fbdSDavid Howells mnt = real_mount(newmount.mnt); 372593766fbdSDavid Howells mnt->mnt_ns = ns; 372693766fbdSDavid Howells ns->root = mnt; 372793766fbdSDavid Howells ns->mounts = 1; 372893766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 37291b0b9cc8SEric Biggers mntget(newmount.mnt); 373093766fbdSDavid Howells 373193766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 373293766fbdSDavid Howells * it, not just simply put it. 373393766fbdSDavid Howells */ 373493766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 373593766fbdSDavid Howells if (IS_ERR(file)) { 373693766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 373793766fbdSDavid Howells ret = PTR_ERR(file); 373893766fbdSDavid Howells goto err_path; 373993766fbdSDavid Howells } 374093766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 374193766fbdSDavid Howells 374293766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 374393766fbdSDavid Howells if (ret >= 0) 374493766fbdSDavid Howells fd_install(ret, file); 374593766fbdSDavid Howells else 374693766fbdSDavid Howells fput(file); 374793766fbdSDavid Howells 374893766fbdSDavid Howells err_path: 374993766fbdSDavid Howells path_put(&newmount); 375093766fbdSDavid Howells err_unlock: 375193766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 375293766fbdSDavid Howells err_fsfd: 375393766fbdSDavid Howells fdput(f); 375493766fbdSDavid Howells return ret; 375593766fbdSDavid Howells } 375693766fbdSDavid Howells 375793766fbdSDavid Howells /* 375893766fbdSDavid Howells * Move a mount from one place to another. In combination with 375993766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 376093766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 376193766fbdSDavid Howells * a mount subtree. 37622db154b3SDavid Howells * 37632db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 37642db154b3SDavid Howells */ 37652db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 37662658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 37672658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 37682db154b3SDavid Howells unsigned int, flags) 37692db154b3SDavid Howells { 37702db154b3SDavid Howells struct path from_path, to_path; 37712db154b3SDavid Howells unsigned int lflags; 37722db154b3SDavid Howells int ret = 0; 37732db154b3SDavid Howells 37742db154b3SDavid Howells if (!may_mount()) 37752db154b3SDavid Howells return -EPERM; 37762db154b3SDavid Howells 37772db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 37782db154b3SDavid Howells return -EINVAL; 37792db154b3SDavid Howells 37802db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 37812db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 37822db154b3SDavid Howells * to clear it afterwards. 37832db154b3SDavid Howells */ 37842db154b3SDavid Howells lflags = 0; 37852db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37862db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37872db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 37882db154b3SDavid Howells 37892db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 37902db154b3SDavid Howells if (ret < 0) 37912db154b3SDavid Howells return ret; 37922db154b3SDavid Howells 37932db154b3SDavid Howells lflags = 0; 37942db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37952db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37962db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 37972db154b3SDavid Howells 37982db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 37992db154b3SDavid Howells if (ret < 0) 38002db154b3SDavid Howells goto out_from; 38012db154b3SDavid Howells 38022db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 38032db154b3SDavid Howells if (ret < 0) 38042db154b3SDavid Howells goto out_to; 38052db154b3SDavid Howells 38069ffb14efSPavel Tikhomirov if (flags & MOVE_MOUNT_SET_GROUP) 38079ffb14efSPavel Tikhomirov ret = do_set_group(&from_path, &to_path); 38089ffb14efSPavel Tikhomirov else 38092db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 38102db154b3SDavid Howells 38112db154b3SDavid Howells out_to: 38122db154b3SDavid Howells path_put(&to_path); 38132db154b3SDavid Howells out_from: 38142db154b3SDavid Howells path_put(&from_path); 38152db154b3SDavid Howells return ret; 38162db154b3SDavid Howells } 38172db154b3SDavid Howells 38182db154b3SDavid Howells /* 3819afac7cbaSAl Viro * Return true if path is reachable from root 3820afac7cbaSAl Viro * 382148a066e7SAl Viro * namespace_sem or mount_lock is held 3822afac7cbaSAl Viro */ 3823643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3824afac7cbaSAl Viro const struct path *root) 3825afac7cbaSAl Viro { 3826643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3827a73324daSAl Viro dentry = mnt->mnt_mountpoint; 38280714a533SAl Viro mnt = mnt->mnt_parent; 3829afac7cbaSAl Viro } 3830643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3831afac7cbaSAl Viro } 3832afac7cbaSAl Viro 3833640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3834afac7cbaSAl Viro { 383525ab4c9bSYaowei Bai bool res; 383648a066e7SAl Viro read_seqlock_excl(&mount_lock); 3837643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 383848a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3839afac7cbaSAl Viro return res; 3840afac7cbaSAl Viro } 3841afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3842afac7cbaSAl Viro 3843afac7cbaSAl Viro /* 38441da177e4SLinus Torvalds * pivot_root Semantics: 38451da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 38461da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 38471da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 38481da177e4SLinus Torvalds * 38491da177e4SLinus Torvalds * Restrictions: 38501da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 38511da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 38521da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 38531da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 38541da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 38551da177e4SLinus Torvalds * 38564a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 38570c1bc6b8SMauro Carvalho Chehab * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives 38584a0d11faSNeil Brown * in this situation. 38594a0d11faSNeil Brown * 38601da177e4SLinus Torvalds * Notes: 38611da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 38621da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 38631da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 38641da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 38651da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 38661da177e4SLinus Torvalds * first. 38671da177e4SLinus Torvalds */ 38683480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 38693480b257SHeiko Carstens const char __user *, put_old) 38701da177e4SLinus Torvalds { 38712763d119SAl Viro struct path new, old, root; 38722763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 387384d17192SAl Viro struct mountpoint *old_mp, *root_mp; 38741da177e4SLinus Torvalds int error; 38751da177e4SLinus Torvalds 38769b40bc90SAl Viro if (!may_mount()) 38771da177e4SLinus Torvalds return -EPERM; 38781da177e4SLinus Torvalds 3879ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3880ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 38811da177e4SLinus Torvalds if (error) 38821da177e4SLinus Torvalds goto out0; 38831da177e4SLinus Torvalds 3884ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 3885ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 38861da177e4SLinus Torvalds if (error) 38871da177e4SLinus Torvalds goto out1; 38881da177e4SLinus Torvalds 38892d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3890b12cea91SAl Viro if (error) 3891b12cea91SAl Viro goto out2; 38921da177e4SLinus Torvalds 3893f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 389484d17192SAl Viro old_mp = lock_mount(&old); 389584d17192SAl Viro error = PTR_ERR(old_mp); 389684d17192SAl Viro if (IS_ERR(old_mp)) 3897b12cea91SAl Viro goto out3; 3898b12cea91SAl Viro 38991da177e4SLinus Torvalds error = -EINVAL; 3900419148daSAl Viro new_mnt = real_mount(new.mnt); 3901419148daSAl Viro root_mnt = real_mount(root.mnt); 390284d17192SAl Viro old_mnt = real_mount(old.mnt); 39032763d119SAl Viro ex_parent = new_mnt->mnt_parent; 39042763d119SAl Viro root_parent = root_mnt->mnt_parent; 390584d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 39062763d119SAl Viro IS_MNT_SHARED(ex_parent) || 39072763d119SAl Viro IS_MNT_SHARED(root_parent)) 3908b12cea91SAl Viro goto out4; 3909143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3910b12cea91SAl Viro goto out4; 39115ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 39125ff9d8a6SEric W. Biederman goto out4; 39131da177e4SLinus Torvalds error = -ENOENT; 3914f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3915b12cea91SAl Viro goto out4; 39161da177e4SLinus Torvalds error = -EBUSY; 391784d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3918b12cea91SAl Viro goto out4; /* loop, on the same file system */ 39191da177e4SLinus Torvalds error = -EINVAL; 39208c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3921b12cea91SAl Viro goto out4; /* not a mountpoint */ 3922676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3923b12cea91SAl Viro goto out4; /* not attached */ 39242d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3925b12cea91SAl Viro goto out4; /* not a mountpoint */ 3926676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3927b12cea91SAl Viro goto out4; /* not attached */ 39284ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 392984d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3930b12cea91SAl Viro goto out4; 39310d082601SEric W. Biederman /* make certain new is below the root */ 39320d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 39330d082601SEric W. Biederman goto out4; 3934719ea2fbSAl Viro lock_mount_hash(); 39352763d119SAl Viro umount_mnt(new_mnt); 39362763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 39375ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 39385ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 39395ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 39405ff9d8a6SEric W. Biederman } 39414ac91378SJan Blunck /* mount old root on put_old */ 394284d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 39434ac91378SJan Blunck /* mount new_root on / */ 39442763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 39452763d119SAl Viro mnt_add_count(root_parent, -1); 39466b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 39474fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 39484fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 39493895dbf8SEric W. Biederman put_mountpoint(root_mp); 3950719ea2fbSAl Viro unlock_mount_hash(); 39512d8f3038SAl Viro chroot_fs_refs(&root, &new); 39521da177e4SLinus Torvalds error = 0; 3953b12cea91SAl Viro out4: 395484d17192SAl Viro unlock_mount(old_mp); 39552763d119SAl Viro if (!error) 39562763d119SAl Viro mntput_no_expire(ex_parent); 3957b12cea91SAl Viro out3: 39588c3ee42eSAl Viro path_put(&root); 3959b12cea91SAl Viro out2: 39602d8f3038SAl Viro path_put(&old); 39611da177e4SLinus Torvalds out1: 39622d8f3038SAl Viro path_put(&new); 39631da177e4SLinus Torvalds out0: 39641da177e4SLinus Torvalds return error; 39651da177e4SLinus Torvalds } 39661da177e4SLinus Torvalds 39672a186721SChristian Brauner static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) 39682a186721SChristian Brauner { 39692a186721SChristian Brauner unsigned int flags = mnt->mnt.mnt_flags; 39702a186721SChristian Brauner 39712a186721SChristian Brauner /* flags to clear */ 39722a186721SChristian Brauner flags &= ~kattr->attr_clr; 39732a186721SChristian Brauner /* flags to raise */ 39742a186721SChristian Brauner flags |= kattr->attr_set; 39752a186721SChristian Brauner 39762a186721SChristian Brauner return flags; 39772a186721SChristian Brauner } 39782a186721SChristian Brauner 39799caccd41SChristian Brauner static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 39809caccd41SChristian Brauner { 39819caccd41SChristian Brauner struct vfsmount *m = &mnt->mnt; 3982bd303368SChristian Brauner struct user_namespace *fs_userns = m->mnt_sb->s_user_ns; 39839caccd41SChristian Brauner 39849caccd41SChristian Brauner if (!kattr->mnt_userns) 39859caccd41SChristian Brauner return 0; 39869caccd41SChristian Brauner 39879caccd41SChristian Brauner /* 3988bd303368SChristian Brauner * Creating an idmapped mount with the filesystem wide idmapping 3989bd303368SChristian Brauner * doesn't make sense so block that. We don't allow mushy semantics. 3990bd303368SChristian Brauner */ 3991bd303368SChristian Brauner if (kattr->mnt_userns == fs_userns) 3992bd303368SChristian Brauner return -EINVAL; 3993bd303368SChristian Brauner 3994bd303368SChristian Brauner /* 39959caccd41SChristian Brauner * Once a mount has been idmapped we don't allow it to change its 39969caccd41SChristian Brauner * mapping. It makes things simpler and callers can just create 39979caccd41SChristian Brauner * another bind-mount they can idmap if they want to. 39989caccd41SChristian Brauner */ 3999bb49e9e7SChristian Brauner if (is_idmapped_mnt(m)) 40009caccd41SChristian Brauner return -EPERM; 40019caccd41SChristian Brauner 40029caccd41SChristian Brauner /* The underlying filesystem doesn't support idmapped mounts yet. */ 40039caccd41SChristian Brauner if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) 40049caccd41SChristian Brauner return -EINVAL; 40059caccd41SChristian Brauner 40069caccd41SChristian Brauner /* We're not controlling the superblock. */ 4007bd303368SChristian Brauner if (!ns_capable(fs_userns, CAP_SYS_ADMIN)) 40089caccd41SChristian Brauner return -EPERM; 40099caccd41SChristian Brauner 40109caccd41SChristian Brauner /* Mount has already been visible in the filesystem hierarchy. */ 40119caccd41SChristian Brauner if (!is_anon_ns(mnt->mnt_ns)) 40129caccd41SChristian Brauner return -EINVAL; 40139caccd41SChristian Brauner 40149caccd41SChristian Brauner return 0; 40159caccd41SChristian Brauner } 40169caccd41SChristian Brauner 4017a26f788bSChristian Brauner /** 4018a26f788bSChristian Brauner * mnt_allow_writers() - check whether the attribute change allows writers 4019a26f788bSChristian Brauner * @kattr: the new mount attributes 4020a26f788bSChristian Brauner * @mnt: the mount to which @kattr will be applied 4021a26f788bSChristian Brauner * 4022a26f788bSChristian Brauner * Check whether thew new mount attributes in @kattr allow concurrent writers. 4023a26f788bSChristian Brauner * 4024a26f788bSChristian Brauner * Return: true if writers need to be held, false if not 4025a26f788bSChristian Brauner */ 4026a26f788bSChristian Brauner static inline bool mnt_allow_writers(const struct mount_kattr *kattr, 4027a26f788bSChristian Brauner const struct mount *mnt) 40282a186721SChristian Brauner { 4029e1bbcd27SChristian Brauner return (!(kattr->attr_set & MNT_READONLY) || 4030e1bbcd27SChristian Brauner (mnt->mnt.mnt_flags & MNT_READONLY)) && 4031e1bbcd27SChristian Brauner !kattr->mnt_userns; 40322a186721SChristian Brauner } 40332a186721SChristian Brauner 403487bb5b60SChristian Brauner static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt) 40352a186721SChristian Brauner { 4036e257039fSAl Viro struct mount *m; 4037e257039fSAl Viro int err; 40382a186721SChristian Brauner 4039e257039fSAl Viro for (m = mnt; m; m = next_mnt(m, mnt)) { 4040e257039fSAl Viro if (!can_change_locked_flags(m, recalc_flags(kattr, m))) { 4041e257039fSAl Viro err = -EPERM; 4042e257039fSAl Viro break; 40432a186721SChristian Brauner } 40442a186721SChristian Brauner 404587bb5b60SChristian Brauner err = can_idmap_mount(kattr, m); 404687bb5b60SChristian Brauner if (err) 4047e257039fSAl Viro break; 40482a186721SChristian Brauner 4049e257039fSAl Viro if (!mnt_allow_writers(kattr, m)) { 405087bb5b60SChristian Brauner err = mnt_hold_writers(m); 405187bb5b60SChristian Brauner if (err) 4052e257039fSAl Viro break; 40532a186721SChristian Brauner } 40542a186721SChristian Brauner 4055e257039fSAl Viro if (!kattr->recurse) 405687bb5b60SChristian Brauner return 0; 40572a186721SChristian Brauner } 40582a186721SChristian Brauner 4059e257039fSAl Viro if (err) { 4060e257039fSAl Viro struct mount *p; 4061e257039fSAl Viro 40620014edaeSChristian Brauner /* 40630014edaeSChristian Brauner * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will 40640014edaeSChristian Brauner * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all 40650014edaeSChristian Brauner * mounts and needs to take care to include the first mount. 40660014edaeSChristian Brauner */ 40670014edaeSChristian Brauner for (p = mnt; p; p = next_mnt(p, mnt)) { 4068e257039fSAl Viro /* If we had to hold writers unblock them. */ 4069e257039fSAl Viro if (p->mnt.mnt_flags & MNT_WRITE_HOLD) 4070e257039fSAl Viro mnt_unhold_writers(p); 40710014edaeSChristian Brauner 40720014edaeSChristian Brauner /* 40730014edaeSChristian Brauner * We're done once the first mount we changed got 40740014edaeSChristian Brauner * MNT_WRITE_HOLD unset. 40750014edaeSChristian Brauner */ 40760014edaeSChristian Brauner if (p == m) 40770014edaeSChristian Brauner break; 4078e257039fSAl Viro } 4079e257039fSAl Viro } 4080e257039fSAl Viro return err; 40812a186721SChristian Brauner } 40822a186721SChristian Brauner 40839caccd41SChristian Brauner static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 40849caccd41SChristian Brauner { 4085bd303368SChristian Brauner struct user_namespace *mnt_userns, *old_mnt_userns; 40869caccd41SChristian Brauner 40879caccd41SChristian Brauner if (!kattr->mnt_userns) 40889caccd41SChristian Brauner return; 40899caccd41SChristian Brauner 4090bd303368SChristian Brauner /* 4091bd303368SChristian Brauner * We're the only ones able to change the mount's idmapping. So 4092bd303368SChristian Brauner * mnt->mnt.mnt_userns is stable and we can retrieve it directly. 4093bd303368SChristian Brauner */ 4094bd303368SChristian Brauner old_mnt_userns = mnt->mnt.mnt_userns; 4095bd303368SChristian Brauner 40969caccd41SChristian Brauner mnt_userns = get_user_ns(kattr->mnt_userns); 40979caccd41SChristian Brauner /* Pairs with smp_load_acquire() in mnt_user_ns(). */ 40989caccd41SChristian Brauner smp_store_release(&mnt->mnt.mnt_userns, mnt_userns); 4099bd303368SChristian Brauner 4100bd303368SChristian Brauner /* 4101bd303368SChristian Brauner * If this is an idmapped filesystem drop the reference we've taken 4102bd303368SChristian Brauner * in vfs_create_mount() before. 4103bd303368SChristian Brauner */ 4104bd303368SChristian Brauner if (!initial_idmapping(old_mnt_userns)) 4105bd303368SChristian Brauner put_user_ns(old_mnt_userns); 41069caccd41SChristian Brauner } 41079caccd41SChristian Brauner 4108e257039fSAl Viro static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt) 41092a186721SChristian Brauner { 4110e257039fSAl Viro struct mount *m; 41112a186721SChristian Brauner 4112e257039fSAl Viro for (m = mnt; m; m = next_mnt(m, mnt)) { 41132a186721SChristian Brauner unsigned int flags; 41142a186721SChristian Brauner 41159caccd41SChristian Brauner do_idmap_mount(kattr, m); 41162a186721SChristian Brauner flags = recalc_flags(kattr, m); 41172a186721SChristian Brauner WRITE_ONCE(m->mnt.mnt_flags, flags); 41182a186721SChristian Brauner 411903b6abeeSChristian Brauner /* If we had to hold writers unblock them. */ 412003b6abeeSChristian Brauner if (m->mnt.mnt_flags & MNT_WRITE_HOLD) 41212a186721SChristian Brauner mnt_unhold_writers(m); 41222a186721SChristian Brauner 4123e257039fSAl Viro if (kattr->propagation) 41242a186721SChristian Brauner change_mnt_propagation(m, kattr->propagation); 4125e257039fSAl Viro if (!kattr->recurse) 41262a186721SChristian Brauner break; 4127e257039fSAl Viro } 41282a186721SChristian Brauner touch_mnt_namespace(mnt->mnt_ns); 41292a186721SChristian Brauner } 41302a186721SChristian Brauner 41312a186721SChristian Brauner static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) 41322a186721SChristian Brauner { 413387bb5b60SChristian Brauner struct mount *mnt = real_mount(path->mnt); 41342a186721SChristian Brauner int err = 0; 41352a186721SChristian Brauner 41362a186721SChristian Brauner if (path->dentry != mnt->mnt.mnt_root) 41372a186721SChristian Brauner return -EINVAL; 41382a186721SChristian Brauner 41392a186721SChristian Brauner if (kattr->propagation) { 41402a186721SChristian Brauner /* 41412a186721SChristian Brauner * Only take namespace_lock() if we're actually changing 41422a186721SChristian Brauner * propagation. 41432a186721SChristian Brauner */ 41442a186721SChristian Brauner namespace_lock(); 41452a186721SChristian Brauner if (kattr->propagation == MS_SHARED) { 41462a186721SChristian Brauner err = invent_group_ids(mnt, kattr->recurse); 41472a186721SChristian Brauner if (err) { 41482a186721SChristian Brauner namespace_unlock(); 41492a186721SChristian Brauner return err; 41502a186721SChristian Brauner } 41512a186721SChristian Brauner } 41522a186721SChristian Brauner } 41532a186721SChristian Brauner 415487bb5b60SChristian Brauner err = -EINVAL; 41552a186721SChristian Brauner lock_mount_hash(); 41562a186721SChristian Brauner 415787bb5b60SChristian Brauner /* Ensure that this isn't anything purely vfs internal. */ 415887bb5b60SChristian Brauner if (!is_mounted(&mnt->mnt)) 415987bb5b60SChristian Brauner goto out; 41602a186721SChristian Brauner 416187bb5b60SChristian Brauner /* 416287bb5b60SChristian Brauner * If this is an attached mount make sure it's located in the callers 416387bb5b60SChristian Brauner * mount namespace. If it's not don't let the caller interact with it. 416487bb5b60SChristian Brauner * If this is a detached mount make sure it has an anonymous mount 416587bb5b60SChristian Brauner * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE. 416687bb5b60SChristian Brauner */ 416787bb5b60SChristian Brauner if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns))) 416887bb5b60SChristian Brauner goto out; 416987bb5b60SChristian Brauner 417087bb5b60SChristian Brauner /* 417187bb5b60SChristian Brauner * First, we get the mount tree in a shape where we can change mount 417287bb5b60SChristian Brauner * properties without failure. If we succeeded to do so we commit all 417387bb5b60SChristian Brauner * changes and if we failed we clean up. 417487bb5b60SChristian Brauner */ 417587bb5b60SChristian Brauner err = mount_setattr_prepare(kattr, mnt); 4176e257039fSAl Viro if (!err) 4177e257039fSAl Viro mount_setattr_commit(kattr, mnt); 417887bb5b60SChristian Brauner 417987bb5b60SChristian Brauner out: 41802a186721SChristian Brauner unlock_mount_hash(); 41812a186721SChristian Brauner 41822a186721SChristian Brauner if (kattr->propagation) { 41832a186721SChristian Brauner namespace_unlock(); 41842a186721SChristian Brauner if (err) 41852a186721SChristian Brauner cleanup_group_ids(mnt, NULL); 41862a186721SChristian Brauner } 41872a186721SChristian Brauner 41882a186721SChristian Brauner return err; 41892a186721SChristian Brauner } 41902a186721SChristian Brauner 41919caccd41SChristian Brauner static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, 41929caccd41SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 41939caccd41SChristian Brauner { 41949caccd41SChristian Brauner int err = 0; 41959caccd41SChristian Brauner struct ns_common *ns; 41969caccd41SChristian Brauner struct user_namespace *mnt_userns; 41979caccd41SChristian Brauner struct file *file; 41989caccd41SChristian Brauner 41999caccd41SChristian Brauner if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) 42009caccd41SChristian Brauner return 0; 42019caccd41SChristian Brauner 42029caccd41SChristian Brauner /* 42039caccd41SChristian Brauner * We currently do not support clearing an idmapped mount. If this ever 42049caccd41SChristian Brauner * is a use-case we can revisit this but for now let's keep it simple 42059caccd41SChristian Brauner * and not allow it. 42069caccd41SChristian Brauner */ 42079caccd41SChristian Brauner if (attr->attr_clr & MOUNT_ATTR_IDMAP) 42089caccd41SChristian Brauner return -EINVAL; 42099caccd41SChristian Brauner 42109caccd41SChristian Brauner if (attr->userns_fd > INT_MAX) 42119caccd41SChristian Brauner return -EINVAL; 42129caccd41SChristian Brauner 42139caccd41SChristian Brauner file = fget(attr->userns_fd); 42149caccd41SChristian Brauner if (!file) 42159caccd41SChristian Brauner return -EBADF; 42169caccd41SChristian Brauner 42179caccd41SChristian Brauner if (!proc_ns_file(file)) { 42189caccd41SChristian Brauner err = -EINVAL; 42199caccd41SChristian Brauner goto out_fput; 42209caccd41SChristian Brauner } 42219caccd41SChristian Brauner 42229caccd41SChristian Brauner ns = get_proc_ns(file_inode(file)); 42239caccd41SChristian Brauner if (ns->ops->type != CLONE_NEWUSER) { 42249caccd41SChristian Brauner err = -EINVAL; 42259caccd41SChristian Brauner goto out_fput; 42269caccd41SChristian Brauner } 42279caccd41SChristian Brauner 42289caccd41SChristian Brauner /* 4229bd303368SChristian Brauner * The initial idmapping cannot be used to create an idmapped 4230bd303368SChristian Brauner * mount. We use the initial idmapping as an indicator of a mount 4231bd303368SChristian Brauner * that is not idmapped. It can simply be passed into helpers that 4232bd303368SChristian Brauner * are aware of idmapped mounts as a convenient shortcut. A user 4233bd303368SChristian Brauner * can just create a dedicated identity mapping to achieve the same 4234bd303368SChristian Brauner * result. 42359caccd41SChristian Brauner */ 42369caccd41SChristian Brauner mnt_userns = container_of(ns, struct user_namespace, ns); 4237bd303368SChristian Brauner if (initial_idmapping(mnt_userns)) { 42389caccd41SChristian Brauner err = -EPERM; 42399caccd41SChristian Brauner goto out_fput; 42409caccd41SChristian Brauner } 4241*bf1ac16eSSeth Forshee 4242*bf1ac16eSSeth Forshee /* We're not controlling the target namespace. */ 4243*bf1ac16eSSeth Forshee if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) { 4244*bf1ac16eSSeth Forshee err = -EPERM; 4245*bf1ac16eSSeth Forshee goto out_fput; 4246*bf1ac16eSSeth Forshee } 4247*bf1ac16eSSeth Forshee 42489caccd41SChristian Brauner kattr->mnt_userns = get_user_ns(mnt_userns); 42499caccd41SChristian Brauner 42509caccd41SChristian Brauner out_fput: 42519caccd41SChristian Brauner fput(file); 42529caccd41SChristian Brauner return err; 42539caccd41SChristian Brauner } 42549caccd41SChristian Brauner 42559caccd41SChristian Brauner static int build_mount_kattr(const struct mount_attr *attr, size_t usize, 42562a186721SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 42572a186721SChristian Brauner { 42582a186721SChristian Brauner unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 42592a186721SChristian Brauner 42602a186721SChristian Brauner if (flags & AT_NO_AUTOMOUNT) 42612a186721SChristian Brauner lookup_flags &= ~LOOKUP_AUTOMOUNT; 42622a186721SChristian Brauner if (flags & AT_SYMLINK_NOFOLLOW) 42632a186721SChristian Brauner lookup_flags &= ~LOOKUP_FOLLOW; 42642a186721SChristian Brauner if (flags & AT_EMPTY_PATH) 42652a186721SChristian Brauner lookup_flags |= LOOKUP_EMPTY; 42662a186721SChristian Brauner 42672a186721SChristian Brauner *kattr = (struct mount_kattr) { 42682a186721SChristian Brauner .lookup_flags = lookup_flags, 42692a186721SChristian Brauner .recurse = !!(flags & AT_RECURSIVE), 42702a186721SChristian Brauner }; 42712a186721SChristian Brauner 42722a186721SChristian Brauner if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) 42732a186721SChristian Brauner return -EINVAL; 42742a186721SChristian Brauner if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) 42752a186721SChristian Brauner return -EINVAL; 42762a186721SChristian Brauner kattr->propagation = attr->propagation; 42772a186721SChristian Brauner 42782a186721SChristian Brauner if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) 42792a186721SChristian Brauner return -EINVAL; 42802a186721SChristian Brauner 42812a186721SChristian Brauner kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); 42822a186721SChristian Brauner kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); 42832a186721SChristian Brauner 42842a186721SChristian Brauner /* 42852a186721SChristian Brauner * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, 42862a186721SChristian Brauner * users wanting to transition to a different atime setting cannot 42872a186721SChristian Brauner * simply specify the atime setting in @attr_set, but must also 42882a186721SChristian Brauner * specify MOUNT_ATTR__ATIME in the @attr_clr field. 42892a186721SChristian Brauner * So ensure that MOUNT_ATTR__ATIME can't be partially set in 42902a186721SChristian Brauner * @attr_clr and that @attr_set can't have any atime bits set if 42912a186721SChristian Brauner * MOUNT_ATTR__ATIME isn't set in @attr_clr. 42922a186721SChristian Brauner */ 42932a186721SChristian Brauner if (attr->attr_clr & MOUNT_ATTR__ATIME) { 42942a186721SChristian Brauner if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) 42952a186721SChristian Brauner return -EINVAL; 42962a186721SChristian Brauner 42972a186721SChristian Brauner /* 42982a186721SChristian Brauner * Clear all previous time settings as they are mutually 42992a186721SChristian Brauner * exclusive. 43002a186721SChristian Brauner */ 43012a186721SChristian Brauner kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; 43022a186721SChristian Brauner switch (attr->attr_set & MOUNT_ATTR__ATIME) { 43032a186721SChristian Brauner case MOUNT_ATTR_RELATIME: 43042a186721SChristian Brauner kattr->attr_set |= MNT_RELATIME; 43052a186721SChristian Brauner break; 43062a186721SChristian Brauner case MOUNT_ATTR_NOATIME: 43072a186721SChristian Brauner kattr->attr_set |= MNT_NOATIME; 43082a186721SChristian Brauner break; 43092a186721SChristian Brauner case MOUNT_ATTR_STRICTATIME: 43102a186721SChristian Brauner break; 43112a186721SChristian Brauner default: 43122a186721SChristian Brauner return -EINVAL; 43132a186721SChristian Brauner } 43142a186721SChristian Brauner } else { 43152a186721SChristian Brauner if (attr->attr_set & MOUNT_ATTR__ATIME) 43162a186721SChristian Brauner return -EINVAL; 43172a186721SChristian Brauner } 43182a186721SChristian Brauner 43199caccd41SChristian Brauner return build_mount_idmapped(attr, usize, kattr, flags); 43209caccd41SChristian Brauner } 43219caccd41SChristian Brauner 43229caccd41SChristian Brauner static void finish_mount_kattr(struct mount_kattr *kattr) 43239caccd41SChristian Brauner { 43249caccd41SChristian Brauner put_user_ns(kattr->mnt_userns); 43259caccd41SChristian Brauner kattr->mnt_userns = NULL; 43262a186721SChristian Brauner } 43272a186721SChristian Brauner 43282a186721SChristian Brauner SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, 43292a186721SChristian Brauner unsigned int, flags, struct mount_attr __user *, uattr, 43302a186721SChristian Brauner size_t, usize) 43312a186721SChristian Brauner { 43322a186721SChristian Brauner int err; 43332a186721SChristian Brauner struct path target; 43342a186721SChristian Brauner struct mount_attr attr; 43352a186721SChristian Brauner struct mount_kattr kattr; 43362a186721SChristian Brauner 43372a186721SChristian Brauner BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); 43382a186721SChristian Brauner 43392a186721SChristian Brauner if (flags & ~(AT_EMPTY_PATH | 43402a186721SChristian Brauner AT_RECURSIVE | 43412a186721SChristian Brauner AT_SYMLINK_NOFOLLOW | 43422a186721SChristian Brauner AT_NO_AUTOMOUNT)) 43432a186721SChristian Brauner return -EINVAL; 43442a186721SChristian Brauner 43452a186721SChristian Brauner if (unlikely(usize > PAGE_SIZE)) 43462a186721SChristian Brauner return -E2BIG; 43472a186721SChristian Brauner if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) 43482a186721SChristian Brauner return -EINVAL; 43492a186721SChristian Brauner 43502a186721SChristian Brauner if (!may_mount()) 43512a186721SChristian Brauner return -EPERM; 43522a186721SChristian Brauner 43532a186721SChristian Brauner err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); 43542a186721SChristian Brauner if (err) 43552a186721SChristian Brauner return err; 43562a186721SChristian Brauner 43572a186721SChristian Brauner /* Don't bother walking through the mounts if this is a nop. */ 43582a186721SChristian Brauner if (attr.attr_set == 0 && 43592a186721SChristian Brauner attr.attr_clr == 0 && 43602a186721SChristian Brauner attr.propagation == 0) 43612a186721SChristian Brauner return 0; 43622a186721SChristian Brauner 43639caccd41SChristian Brauner err = build_mount_kattr(&attr, usize, &kattr, flags); 43642a186721SChristian Brauner if (err) 43652a186721SChristian Brauner return err; 43662a186721SChristian Brauner 43672a186721SChristian Brauner err = user_path_at(dfd, path, kattr.lookup_flags, &target); 4368012e3322SChristian Brauner if (!err) { 43692a186721SChristian Brauner err = do_mount_setattr(&target, &kattr); 43702a186721SChristian Brauner path_put(&target); 4371012e3322SChristian Brauner } 4372012e3322SChristian Brauner finish_mount_kattr(&kattr); 43732a186721SChristian Brauner return err; 43742a186721SChristian Brauner } 43752a186721SChristian Brauner 43761da177e4SLinus Torvalds static void __init init_mount_tree(void) 43771da177e4SLinus Torvalds { 43781da177e4SLinus Torvalds struct vfsmount *mnt; 437974e83122SAl Viro struct mount *m; 43806b3286edSKirill Korotaev struct mnt_namespace *ns; 4381ac748a09SJan Blunck struct path root; 43821da177e4SLinus Torvalds 4383fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 43841da177e4SLinus Torvalds if (IS_ERR(mnt)) 43851da177e4SLinus Torvalds panic("Can't create rootfs"); 4386b3e19d92SNick Piggin 438774e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 43883b22edc5STrond Myklebust if (IS_ERR(ns)) 43891da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 439074e83122SAl Viro m = real_mount(mnt); 439174e83122SAl Viro m->mnt_ns = ns; 439274e83122SAl Viro ns->root = m; 439374e83122SAl Viro ns->mounts = 1; 439474e83122SAl Viro list_add(&m->mnt_list, &ns->list); 43956b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 43966b3286edSKirill Korotaev get_mnt_ns(ns); 43971da177e4SLinus Torvalds 4398be08d6d2SAl Viro root.mnt = mnt; 4399be08d6d2SAl Viro root.dentry = mnt->mnt_root; 4400da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 4401ac748a09SJan Blunck 4402ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 4403ac748a09SJan Blunck set_fs_root(current->fs, &root); 44041da177e4SLinus Torvalds } 44051da177e4SLinus Torvalds 440674bf17cfSDenis Cheng void __init mnt_init(void) 44071da177e4SLinus Torvalds { 440815a67dd8SRandy Dunlap int err; 44091da177e4SLinus Torvalds 44107d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 441179f6540bSVasily Averin 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); 44121da177e4SLinus Torvalds 44130818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 441438129a13SAl Viro sizeof(struct hlist_head), 44150818bf27SAl Viro mhash_entries, 19, 44163d375d78SPavel Tatashin HASH_ZERO, 44170818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 44180818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 44190818bf27SAl Viro sizeof(struct hlist_head), 44200818bf27SAl Viro mphash_entries, 19, 44213d375d78SPavel Tatashin HASH_ZERO, 44220818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 44231da177e4SLinus Torvalds 442484d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 44251da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 44261da177e4SLinus Torvalds 44274b93dc9bSTejun Heo kernfs_init(); 44284b93dc9bSTejun Heo 442915a67dd8SRandy Dunlap err = sysfs_init(); 443015a67dd8SRandy Dunlap if (err) 443115a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 44328e24eea7SHarvey Harrison __func__, err); 443300d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 443400d26666SGreg Kroah-Hartman if (!fs_kobj) 44358e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 4436037f11b4SAl Viro shmem_init(); 44371da177e4SLinus Torvalds init_rootfs(); 44381da177e4SLinus Torvalds init_mount_tree(); 44391da177e4SLinus Torvalds } 44401da177e4SLinus Torvalds 4441616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 44421da177e4SLinus Torvalds { 44431a7b8969SKirill Tkhai if (!refcount_dec_and_test(&ns->ns.count)) 4444616511d0STrond Myklebust return; 44457b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 4446771b1371SEric W. Biederman free_mnt_ns(ns); 44471da177e4SLinus Torvalds } 44489d412a43SAl Viro 4449d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 44509d412a43SAl Viro { 4451423e0ab0STim Chen struct vfsmount *mnt; 4452d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 4453423e0ab0STim Chen if (!IS_ERR(mnt)) { 4454423e0ab0STim Chen /* 4455423e0ab0STim Chen * it is a longterm mount, don't release mnt until 4456423e0ab0STim Chen * we unmount before file sys is unregistered 4457423e0ab0STim Chen */ 4458f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 4459423e0ab0STim Chen } 4460423e0ab0STim Chen return mnt; 44619d412a43SAl Viro } 4462d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 4463423e0ab0STim Chen 4464423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 4465423e0ab0STim Chen { 4466423e0ab0STim Chen /* release long term mount so mount point can be released */ 4467423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 4468f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 446948a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 4470423e0ab0STim Chen mntput(mnt); 4471423e0ab0STim Chen } 4472423e0ab0STim Chen } 4473423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 447402125a82SAl Viro 4475df820f8dSMiklos Szeredi void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) 4476df820f8dSMiklos Szeredi { 4477df820f8dSMiklos Szeredi unsigned int i; 4478df820f8dSMiklos Szeredi 4479df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4480df820f8dSMiklos Szeredi if (mnt[i]) 4481df820f8dSMiklos Szeredi real_mount(mnt[i])->mnt_ns = NULL; 4482df820f8dSMiklos Szeredi synchronize_rcu_expedited(); 4483df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4484df820f8dSMiklos Szeredi mntput(mnt[i]); 4485df820f8dSMiklos Szeredi } 4486df820f8dSMiklos Szeredi EXPORT_SYMBOL(kern_unmount_array); 4487df820f8dSMiklos Szeredi 448802125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 448902125a82SAl Viro { 4490143c8c91SAl Viro return check_mnt(real_mount(mnt)); 449102125a82SAl Viro } 44928823c079SEric W. Biederman 44933151527eSEric W. Biederman bool current_chrooted(void) 44943151527eSEric W. Biederman { 44953151527eSEric W. Biederman /* Does the current process have a non-standard root */ 44963151527eSEric W. Biederman struct path ns_root; 44973151527eSEric W. Biederman struct path fs_root; 44983151527eSEric W. Biederman bool chrooted; 44993151527eSEric W. Biederman 45003151527eSEric W. Biederman /* Find the namespace root */ 45013151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 45023151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 45033151527eSEric W. Biederman path_get(&ns_root); 45043151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 45053151527eSEric W. Biederman ; 45063151527eSEric W. Biederman 45073151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 45083151527eSEric W. Biederman 45093151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 45103151527eSEric W. Biederman 45113151527eSEric W. Biederman path_put(&fs_root); 45123151527eSEric W. Biederman path_put(&ns_root); 45133151527eSEric W. Biederman 45143151527eSEric W. Biederman return chrooted; 45153151527eSEric W. Biederman } 45163151527eSEric W. Biederman 4517132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 4518132e4608SDavid Howells const struct super_block *sb, 45198654df4eSEric W. Biederman int *new_mnt_flags) 452087a8ebd6SEric W. Biederman { 45218c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 452287a8ebd6SEric W. Biederman struct mount *mnt; 4523e51db735SEric W. Biederman bool visible = false; 452487a8ebd6SEric W. Biederman 452544bb4385SAl Viro down_read(&namespace_sem); 45269f6c61f9SMiklos Szeredi lock_ns_list(ns); 452787a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 4528e51db735SEric W. Biederman struct mount *child; 452977b1a97dSEric W. Biederman int mnt_flags; 453077b1a97dSEric W. Biederman 45319f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 45329f6c61f9SMiklos Szeredi continue; 45339f6c61f9SMiklos Szeredi 4534132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 4535e51db735SEric W. Biederman continue; 4536e51db735SEric W. Biederman 45377e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 45387e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 45397e96c1b0SEric W. Biederman */ 45407e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 45417e96c1b0SEric W. Biederman continue; 45427e96c1b0SEric W. Biederman 4543a1935c17SEric W. Biederman /* A local view of the mount flags */ 454477b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 454577b1a97dSEric W. Biederman 4546695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 4547bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 4548695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 4549695e9df0SEric W. Biederman 45508c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 45518c6cf9ccSEric W. Biederman * than the proposed new mount. 45528c6cf9ccSEric W. Biederman */ 455377b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 45548c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 45558c6cf9ccSEric W. Biederman continue; 455677b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 455777b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 45588c6cf9ccSEric W. Biederman continue; 45598c6cf9ccSEric W. Biederman 4560ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 4561ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 4562ceeb0e5dSEric W. Biederman * empty directories. 4563e51db735SEric W. Biederman */ 4564e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 4565e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 4566ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 4567d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 4568ceeb0e5dSEric W. Biederman continue; 45697236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 45707236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 4571e51db735SEric W. Biederman goto next; 457287a8ebd6SEric W. Biederman } 45738c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 457477b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 45758c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 4576e51db735SEric W. Biederman visible = true; 4577e51db735SEric W. Biederman goto found; 4578e51db735SEric W. Biederman next: ; 457987a8ebd6SEric W. Biederman } 4580e51db735SEric W. Biederman found: 45819f6c61f9SMiklos Szeredi unlock_ns_list(ns); 458244bb4385SAl Viro up_read(&namespace_sem); 4583e51db735SEric W. Biederman return visible; 458487a8ebd6SEric W. Biederman } 458587a8ebd6SEric W. Biederman 4586132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 45878654df4eSEric W. Biederman { 4588a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 45898654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 45908654df4eSEric W. Biederman unsigned long s_iflags; 45918654df4eSEric W. Biederman 45928654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 45938654df4eSEric W. Biederman return false; 45948654df4eSEric W. Biederman 45958654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 4596132e4608SDavid Howells s_iflags = sb->s_iflags; 45978654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 45988654df4eSEric W. Biederman return false; 45998654df4eSEric W. Biederman 4600a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 4601a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 4602a1935c17SEric W. Biederman required_iflags); 4603a1935c17SEric W. Biederman return true; 4604a1935c17SEric W. Biederman } 4605a1935c17SEric W. Biederman 4606132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 46078654df4eSEric W. Biederman } 46088654df4eSEric W. Biederman 4609380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 4610380cf5baSAndy Lutomirski { 4611380cf5baSAndy Lutomirski /* 4612380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 4613380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 4614380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 4615380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 4616380cf5baSAndy Lutomirski * in other namespaces. 4617380cf5baSAndy Lutomirski */ 4618380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 4619380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 4620380cf5baSAndy Lutomirski } 4621380cf5baSAndy Lutomirski 462264964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 46238823c079SEric W. Biederman { 462458be2825SAl Viro struct ns_common *ns = NULL; 46258823c079SEric W. Biederman struct nsproxy *nsproxy; 46268823c079SEric W. Biederman 4627728dba3aSEric W. Biederman task_lock(task); 4628728dba3aSEric W. Biederman nsproxy = task->nsproxy; 46298823c079SEric W. Biederman if (nsproxy) { 463058be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 463158be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 46328823c079SEric W. Biederman } 4633728dba3aSEric W. Biederman task_unlock(task); 46348823c079SEric W. Biederman 46358823c079SEric W. Biederman return ns; 46368823c079SEric W. Biederman } 46378823c079SEric W. Biederman 463864964528SAl Viro static void mntns_put(struct ns_common *ns) 46398823c079SEric W. Biederman { 464058be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 46418823c079SEric W. Biederman } 46428823c079SEric W. Biederman 4643f2a8d52eSChristian Brauner static int mntns_install(struct nsset *nsset, struct ns_common *ns) 46448823c079SEric W. Biederman { 4645f2a8d52eSChristian Brauner struct nsproxy *nsproxy = nsset->nsproxy; 4646f2a8d52eSChristian Brauner struct fs_struct *fs = nsset->fs; 46474f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 4648f2a8d52eSChristian Brauner struct user_namespace *user_ns = nsset->cred->user_ns; 46498823c079SEric W. Biederman struct path root; 46504f757f3cSAl Viro int err; 46518823c079SEric W. Biederman 46520c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 4653f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_CHROOT) || 4654f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_ADMIN)) 4655ae11e0f1SZhao Hongjiang return -EPERM; 46568823c079SEric W. Biederman 465774e83122SAl Viro if (is_anon_ns(mnt_ns)) 465874e83122SAl Viro return -EINVAL; 465974e83122SAl Viro 46608823c079SEric W. Biederman if (fs->users != 1) 46618823c079SEric W. Biederman return -EINVAL; 46628823c079SEric W. Biederman 46638823c079SEric W. Biederman get_mnt_ns(mnt_ns); 46644f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 46658823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 46668823c079SEric W. Biederman 46678823c079SEric W. Biederman /* Find the root */ 46684f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 46694f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 46704f757f3cSAl Viro if (err) { 46714f757f3cSAl Viro /* revert to old namespace */ 46724f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 46734f757f3cSAl Viro put_mnt_ns(mnt_ns); 46744f757f3cSAl Viro return err; 46754f757f3cSAl Viro } 46768823c079SEric W. Biederman 46774068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 46784068367cSAndrei Vagin 46798823c079SEric W. Biederman /* Update the pwd and root */ 46808823c079SEric W. Biederman set_fs_pwd(fs, &root); 46818823c079SEric W. Biederman set_fs_root(fs, &root); 46828823c079SEric W. Biederman 46838823c079SEric W. Biederman path_put(&root); 46848823c079SEric W. Biederman return 0; 46858823c079SEric W. Biederman } 46868823c079SEric W. Biederman 4687bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 4688bcac25a5SAndrey Vagin { 4689bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 4690bcac25a5SAndrey Vagin } 4691bcac25a5SAndrey Vagin 46928823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 46938823c079SEric W. Biederman .name = "mnt", 46948823c079SEric W. Biederman .type = CLONE_NEWNS, 46958823c079SEric W. Biederman .get = mntns_get, 46968823c079SEric W. Biederman .put = mntns_put, 46978823c079SEric W. Biederman .install = mntns_install, 4698bcac25a5SAndrey Vagin .owner = mntns_owner, 46998823c079SEric W. Biederman }; 4700ab171b95SLuis Chamberlain 4701ab171b95SLuis Chamberlain #ifdef CONFIG_SYSCTL 4702ab171b95SLuis Chamberlain static struct ctl_table fs_namespace_sysctls[] = { 4703ab171b95SLuis Chamberlain { 4704ab171b95SLuis Chamberlain .procname = "mount-max", 4705ab171b95SLuis Chamberlain .data = &sysctl_mount_max, 4706ab171b95SLuis Chamberlain .maxlen = sizeof(unsigned int), 4707ab171b95SLuis Chamberlain .mode = 0644, 4708ab171b95SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 4709ab171b95SLuis Chamberlain .extra1 = SYSCTL_ONE, 4710ab171b95SLuis Chamberlain }, 4711ab171b95SLuis Chamberlain { } 4712ab171b95SLuis Chamberlain }; 4713ab171b95SLuis Chamberlain 4714ab171b95SLuis Chamberlain static int __init init_fs_namespace_sysctls(void) 4715ab171b95SLuis Chamberlain { 4716ab171b95SLuis Chamberlain register_sysctl_init("fs", fs_namespace_sysctls); 4717ab171b95SLuis Chamberlain return 0; 4718ab171b95SLuis Chamberlain } 4719ab171b95SLuis Chamberlain fs_initcall(init_fs_namespace_sysctls); 4720ab171b95SLuis Chamberlain 4721ab171b95SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 4722