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; 85256c8aedSChristian Brauner struct mnt_idmap *mnt_idmap; 862a186721SChristian Brauner }; 872a186721SChristian Brauner 88f87fd4c2SMiklos Szeredi /* /sys/fs */ 8900d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 9000d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 91f87fd4c2SMiklos Szeredi 9299b7db7bSNick Piggin /* 9399b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 9499b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 9599b7db7bSNick Piggin * up the tree. 9699b7db7bSNick Piggin * 9799b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 9899b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 9999b7db7bSNick Piggin */ 10048a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 10199b7db7bSNick Piggin 102d033cb67SChristian Brauner static inline void lock_mount_hash(void) 103d033cb67SChristian Brauner { 104d033cb67SChristian Brauner write_seqlock(&mount_lock); 105d033cb67SChristian Brauner } 106d033cb67SChristian Brauner 107d033cb67SChristian Brauner static inline void unlock_mount_hash(void) 108d033cb67SChristian Brauner { 109d033cb67SChristian Brauner write_sequnlock(&mount_lock); 110d033cb67SChristian Brauner } 111d033cb67SChristian Brauner 11238129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 1131da177e4SLinus Torvalds { 1141da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 1151da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 1160818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 1170818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 1180818bf27SAl Viro } 1190818bf27SAl Viro 1200818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 1210818bf27SAl Viro { 1220818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 1230818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 1240818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 1251da177e4SLinus Torvalds } 1261da177e4SLinus Torvalds 127b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 12873cd49ecSMiklos Szeredi { 129169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 13073cd49ecSMiklos Szeredi 131169b480eSMatthew Wilcox if (res < 0) 13273cd49ecSMiklos Szeredi return res; 133169b480eSMatthew Wilcox mnt->mnt_id = res; 134169b480eSMatthew Wilcox return 0; 13573cd49ecSMiklos Szeredi } 13673cd49ecSMiklos Szeredi 137b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 13873cd49ecSMiklos Szeredi { 139169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 14073cd49ecSMiklos Szeredi } 14173cd49ecSMiklos Szeredi 142719f5d7fSMiklos Szeredi /* 143719f5d7fSMiklos Szeredi * Allocate a new peer group ID 144719f5d7fSMiklos Szeredi */ 1454b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 146719f5d7fSMiklos Szeredi { 147169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 148f21f6220SAl Viro 149169b480eSMatthew Wilcox if (res < 0) 150f21f6220SAl Viro return res; 151169b480eSMatthew Wilcox mnt->mnt_group_id = res; 152169b480eSMatthew Wilcox return 0; 153719f5d7fSMiklos Szeredi } 154719f5d7fSMiklos Szeredi 155719f5d7fSMiklos Szeredi /* 156719f5d7fSMiklos Szeredi * Release a peer group ID 157719f5d7fSMiklos Szeredi */ 1584b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 159719f5d7fSMiklos Szeredi { 160169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 16115169fe7SAl Viro mnt->mnt_group_id = 0; 162719f5d7fSMiklos Szeredi } 163719f5d7fSMiklos Szeredi 164b3e19d92SNick Piggin /* 165b3e19d92SNick Piggin * vfsmount lock must be held for read 166b3e19d92SNick Piggin */ 16783adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 168b3e19d92SNick Piggin { 169b3e19d92SNick Piggin #ifdef CONFIG_SMP 17068e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 171b3e19d92SNick Piggin #else 172b3e19d92SNick Piggin preempt_disable(); 17368e8a9feSAl Viro mnt->mnt_count += n; 174b3e19d92SNick Piggin preempt_enable(); 175b3e19d92SNick Piggin #endif 176b3e19d92SNick Piggin } 177b3e19d92SNick Piggin 178b3e19d92SNick Piggin /* 179b3e19d92SNick Piggin * vfsmount lock must be held for write 180b3e19d92SNick Piggin */ 181edf7ddbfSEric Biggers int mnt_get_count(struct mount *mnt) 182b3e19d92SNick Piggin { 183b3e19d92SNick Piggin #ifdef CONFIG_SMP 184edf7ddbfSEric Biggers int count = 0; 185b3e19d92SNick Piggin int cpu; 186b3e19d92SNick Piggin 187b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 18868e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 189b3e19d92SNick Piggin } 190b3e19d92SNick Piggin 191b3e19d92SNick Piggin return count; 192b3e19d92SNick Piggin #else 19368e8a9feSAl Viro return mnt->mnt_count; 194b3e19d92SNick Piggin #endif 195b3e19d92SNick Piggin } 196b3e19d92SNick Piggin 197b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1981da177e4SLinus Torvalds { 199c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 200c63181e6SAl Viro if (mnt) { 20173cd49ecSMiklos Szeredi int err; 20273cd49ecSMiklos Szeredi 203c63181e6SAl Viro err = mnt_alloc_id(mnt); 20488b38782SLi Zefan if (err) 20588b38782SLi Zefan goto out_free_cache; 20688b38782SLi Zefan 20788b38782SLi Zefan if (name) { 20879f6540bSVasily Averin mnt->mnt_devname = kstrdup_const(name, 20979f6540bSVasily Averin GFP_KERNEL_ACCOUNT); 210c63181e6SAl Viro if (!mnt->mnt_devname) 21188b38782SLi Zefan goto out_free_id; 21273cd49ecSMiklos Szeredi } 21373cd49ecSMiklos Szeredi 214b3e19d92SNick Piggin #ifdef CONFIG_SMP 215c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 216c63181e6SAl Viro if (!mnt->mnt_pcp) 217b3e19d92SNick Piggin goto out_free_devname; 218b3e19d92SNick Piggin 219c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 220b3e19d92SNick Piggin #else 221c63181e6SAl Viro mnt->mnt_count = 1; 222c63181e6SAl Viro mnt->mnt_writers = 0; 223b3e19d92SNick Piggin #endif 224b3e19d92SNick Piggin 22538129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 226c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 227c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 228c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 229c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 230c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 231c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 232c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2330a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 23499b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 23556cbb429SAl Viro INIT_HLIST_HEAD(&mnt->mnt_stuck_children); 236256c8aedSChristian Brauner mnt->mnt.mnt_idmap = &nop_mnt_idmap; 2371da177e4SLinus Torvalds } 238c63181e6SAl Viro return mnt; 23988b38782SLi Zefan 240d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 241d3ef3d73Snpiggin@suse.de out_free_devname: 242fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 243d3ef3d73Snpiggin@suse.de #endif 24488b38782SLi Zefan out_free_id: 245c63181e6SAl Viro mnt_free_id(mnt); 24688b38782SLi Zefan out_free_cache: 247c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 24888b38782SLi Zefan return NULL; 2491da177e4SLinus Torvalds } 2501da177e4SLinus Torvalds 2518366025eSDave Hansen /* 2528366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2538366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2548366025eSDave Hansen * We must keep track of when those operations start 2558366025eSDave Hansen * (for permission checks) and when they end, so that 2568366025eSDave Hansen * we can determine when writes are able to occur to 2578366025eSDave Hansen * a filesystem. 2588366025eSDave Hansen */ 2593d733633SDave Hansen /* 2603d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2613d733633SDave Hansen * @mnt: the mount to check for its write status 2623d733633SDave Hansen * 2633d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2643d733633SDave Hansen * It does not guarantee that the filesystem will stay 2653d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2663d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2673d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2683d733633SDave Hansen * r/w. 2693d733633SDave Hansen */ 27043f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 2713d733633SDave Hansen { 27243f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 2733d733633SDave Hansen } 2743d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2753d733633SDave Hansen 27683adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2773d733633SDave Hansen { 278d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 27968e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 280d3ef3d73Snpiggin@suse.de #else 28168e8a9feSAl Viro mnt->mnt_writers++; 282d3ef3d73Snpiggin@suse.de #endif 2833d733633SDave Hansen } 2843d733633SDave Hansen 28583adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2863d733633SDave Hansen { 287d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 28868e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 289d3ef3d73Snpiggin@suse.de #else 29068e8a9feSAl Viro mnt->mnt_writers--; 291d3ef3d73Snpiggin@suse.de #endif 292d3ef3d73Snpiggin@suse.de } 293d3ef3d73Snpiggin@suse.de 29483adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 295d3ef3d73Snpiggin@suse.de { 296d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 297d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2983d733633SDave Hansen int cpu; 2993d733633SDave Hansen 3003d733633SDave Hansen for_each_possible_cpu(cpu) { 30168e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 3023d733633SDave Hansen } 3033d733633SDave Hansen 304d3ef3d73Snpiggin@suse.de return count; 305d3ef3d73Snpiggin@suse.de #else 306d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 307d3ef3d73Snpiggin@suse.de #endif 3083d733633SDave Hansen } 3093d733633SDave Hansen 3104ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3114ed5e82fSMiklos Szeredi { 3124ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3134ed5e82fSMiklos Szeredi return 1; 3144ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3154ed5e82fSMiklos Szeredi smp_rmb(); 3164ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3174ed5e82fSMiklos Szeredi } 3184ed5e82fSMiklos Szeredi 3193d733633SDave Hansen /* 320eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 321eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 322eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 323eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3243d733633SDave Hansen */ 3258366025eSDave Hansen /** 326eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 32783adc753SAl Viro * @m: the mount on which to take a write 3288366025eSDave Hansen * 329eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 330eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 331eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 332eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 333eb04c282SJan Kara * called. This is effectively a refcount. 3348366025eSDave Hansen */ 335eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3368366025eSDave Hansen { 33783adc753SAl Viro struct mount *mnt = real_mount(m); 3383d733633SDave Hansen int ret = 0; 3393d733633SDave Hansen 340d3ef3d73Snpiggin@suse.de preempt_disable(); 341c6653a83SNick Piggin mnt_inc_writers(mnt); 342d3ef3d73Snpiggin@suse.de /* 343c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 344d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 345d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 346d3ef3d73Snpiggin@suse.de */ 347d3ef3d73Snpiggin@suse.de smp_mb(); 3480f8821daSSebastian Andrzej Siewior might_lock(&mount_lock.lock); 3490f8821daSSebastian Andrzej Siewior while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) { 3500f8821daSSebastian Andrzej Siewior if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 351d3ef3d73Snpiggin@suse.de cpu_relax(); 3520f8821daSSebastian Andrzej Siewior } else { 3530f8821daSSebastian Andrzej Siewior /* 3540f8821daSSebastian Andrzej Siewior * This prevents priority inversion, if the task 3550f8821daSSebastian Andrzej Siewior * setting MNT_WRITE_HOLD got preempted on a remote 3560f8821daSSebastian Andrzej Siewior * CPU, and it prevents life lock if the task setting 3570f8821daSSebastian Andrzej Siewior * MNT_WRITE_HOLD has a lower priority and is bound to 3580f8821daSSebastian Andrzej Siewior * the same CPU as the task that is spinning here. 3590f8821daSSebastian Andrzej Siewior */ 3600f8821daSSebastian Andrzej Siewior preempt_enable(); 3610f8821daSSebastian Andrzej Siewior lock_mount_hash(); 3620f8821daSSebastian Andrzej Siewior unlock_mount_hash(); 3630f8821daSSebastian Andrzej Siewior preempt_disable(); 3640f8821daSSebastian Andrzej Siewior } 3650f8821daSSebastian Andrzej Siewior } 366d3ef3d73Snpiggin@suse.de /* 367d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 368d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 369d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 370d3ef3d73Snpiggin@suse.de */ 371d3ef3d73Snpiggin@suse.de smp_rmb(); 3724ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 373c6653a83SNick Piggin mnt_dec_writers(mnt); 3743d733633SDave Hansen ret = -EROFS; 3753d733633SDave Hansen } 376d3ef3d73Snpiggin@suse.de preempt_enable(); 377eb04c282SJan Kara 378eb04c282SJan Kara return ret; 379eb04c282SJan Kara } 380eb04c282SJan Kara 381eb04c282SJan Kara /** 382eb04c282SJan Kara * mnt_want_write - get write access to a mount 383eb04c282SJan Kara * @m: the mount on which to take a write 384eb04c282SJan Kara * 385eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 386eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 387eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 388eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 389eb04c282SJan Kara */ 390eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 391eb04c282SJan Kara { 392eb04c282SJan Kara int ret; 393eb04c282SJan Kara 394eb04c282SJan Kara sb_start_write(m->mnt_sb); 395eb04c282SJan Kara ret = __mnt_want_write(m); 396eb04c282SJan Kara if (ret) 397eb04c282SJan Kara sb_end_write(m->mnt_sb); 3983d733633SDave Hansen return ret; 3998366025eSDave Hansen } 4008366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 4018366025eSDave Hansen 4028366025eSDave Hansen /** 403eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 404eb04c282SJan Kara * @file: the file who's mount on which to take a write 405eb04c282SJan Kara * 40614e43bf4SEric Biggers * This is like __mnt_want_write, but if the file is already open for writing it 40714e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 40814e43bf4SEric Biggers * and instead only does the check for emergency r/o remounts. This must be 40914e43bf4SEric Biggers * paired with __mnt_drop_write_file. 410eb04c282SJan Kara */ 411eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 412eb04c282SJan Kara { 41314e43bf4SEric Biggers if (file->f_mode & FMODE_WRITER) { 41414e43bf4SEric Biggers /* 41514e43bf4SEric Biggers * Superblock may have become readonly while there are still 41614e43bf4SEric Biggers * writable fd's, e.g. due to a fs error with errors=remount-ro 41714e43bf4SEric Biggers */ 41814e43bf4SEric Biggers if (__mnt_is_readonly(file->f_path.mnt)) 41914e43bf4SEric Biggers return -EROFS; 42014e43bf4SEric Biggers return 0; 42114e43bf4SEric Biggers } 422eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 423eb04c282SJan Kara } 424eb04c282SJan Kara 425eb04c282SJan Kara /** 4267c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4277c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4287c6893e3SMiklos Szeredi * 42914e43bf4SEric Biggers * This is like mnt_want_write, but if the file is already open for writing it 43014e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 43114e43bf4SEric Biggers * and instead only does the freeze protection and the check for emergency r/o 43214e43bf4SEric Biggers * remounts. This must be paired with mnt_drop_write_file. 4337c6893e3SMiklos Szeredi */ 4347c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4357c6893e3SMiklos Szeredi { 4367c6893e3SMiklos Szeredi int ret; 4377c6893e3SMiklos Szeredi 4387c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4397c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4407c6893e3SMiklos Szeredi if (ret) 4417c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4427c6893e3SMiklos Szeredi return ret; 4437c6893e3SMiklos Szeredi } 44496029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 44596029c4eSnpiggin@suse.de 44696029c4eSnpiggin@suse.de /** 447eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4488366025eSDave Hansen * @mnt: the mount on which to give up write access 4498366025eSDave Hansen * 4508366025eSDave Hansen * Tells the low-level filesystem that we are done 4518366025eSDave Hansen * performing writes to it. Must be matched with 452eb04c282SJan Kara * __mnt_want_write() call above. 4538366025eSDave Hansen */ 454eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4558366025eSDave Hansen { 456d3ef3d73Snpiggin@suse.de preempt_disable(); 45783adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 458d3ef3d73Snpiggin@suse.de preempt_enable(); 4598366025eSDave Hansen } 460eb04c282SJan Kara 461eb04c282SJan Kara /** 462eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 463eb04c282SJan Kara * @mnt: the mount on which to give up write access 464eb04c282SJan Kara * 465eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 466eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 467eb04c282SJan Kara * mnt_want_write() call above. 468eb04c282SJan Kara */ 469eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 470eb04c282SJan Kara { 471eb04c282SJan Kara __mnt_drop_write(mnt); 472eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 473eb04c282SJan Kara } 4748366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4758366025eSDave Hansen 476eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 477eb04c282SJan Kara { 47814e43bf4SEric Biggers if (!(file->f_mode & FMODE_WRITER)) 479eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 480eb04c282SJan Kara } 481eb04c282SJan Kara 4827c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4837c6893e3SMiklos Szeredi { 484a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4857c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4867c6893e3SMiklos Szeredi } 4872a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4882a79f17eSAl Viro 489538f4f02SChristian Brauner /** 490538f4f02SChristian Brauner * mnt_hold_writers - prevent write access to the given mount 491538f4f02SChristian Brauner * @mnt: mnt to prevent write access to 492538f4f02SChristian Brauner * 493538f4f02SChristian Brauner * Prevents write access to @mnt if there are no active writers for @mnt. 494538f4f02SChristian Brauner * This function needs to be called and return successfully before changing 495538f4f02SChristian Brauner * properties of @mnt that need to remain stable for callers with write access 496538f4f02SChristian Brauner * to @mnt. 497538f4f02SChristian Brauner * 498538f4f02SChristian Brauner * After this functions has been called successfully callers must pair it with 499538f4f02SChristian Brauner * a call to mnt_unhold_writers() in order to stop preventing write access to 500538f4f02SChristian Brauner * @mnt. 501538f4f02SChristian Brauner * 502538f4f02SChristian Brauner * Context: This function expects lock_mount_hash() to be held serializing 503538f4f02SChristian Brauner * setting MNT_WRITE_HOLD. 504538f4f02SChristian Brauner * Return: On success 0 is returned. 505538f4f02SChristian Brauner * On error, -EBUSY is returned. 506538f4f02SChristian Brauner */ 507fbdc2f6cSChristian Brauner static inline int mnt_hold_writers(struct mount *mnt) 5088366025eSDave Hansen { 50983adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 510d3ef3d73Snpiggin@suse.de /* 511d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 512d3ef3d73Snpiggin@suse.de * should be visible before we do. 513d3ef3d73Snpiggin@suse.de */ 514d3ef3d73Snpiggin@suse.de smp_mb(); 515d3ef3d73Snpiggin@suse.de 516d3ef3d73Snpiggin@suse.de /* 517d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 518d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 519d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 520d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 521d3ef3d73Snpiggin@suse.de * 522d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 523d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 524d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 525d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 526d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 527d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 528d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 529d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 530d3ef3d73Snpiggin@suse.de * we're counting up here. 531d3ef3d73Snpiggin@suse.de */ 532c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 533fbdc2f6cSChristian Brauner return -EBUSY; 534fbdc2f6cSChristian Brauner 535fbdc2f6cSChristian Brauner return 0; 536fbdc2f6cSChristian Brauner } 537fbdc2f6cSChristian Brauner 538538f4f02SChristian Brauner /** 539538f4f02SChristian Brauner * mnt_unhold_writers - stop preventing write access to the given mount 540538f4f02SChristian Brauner * @mnt: mnt to stop preventing write access to 541538f4f02SChristian Brauner * 542538f4f02SChristian Brauner * Stop preventing write access to @mnt allowing callers to gain write access 543538f4f02SChristian Brauner * to @mnt again. 544538f4f02SChristian Brauner * 545538f4f02SChristian Brauner * This function can only be called after a successful call to 546538f4f02SChristian Brauner * mnt_hold_writers(). 547538f4f02SChristian Brauner * 548538f4f02SChristian Brauner * Context: This function expects lock_mount_hash() to be held. 549538f4f02SChristian Brauner */ 550fbdc2f6cSChristian Brauner static inline void mnt_unhold_writers(struct mount *mnt) 551fbdc2f6cSChristian Brauner { 552d3ef3d73Snpiggin@suse.de /* 553d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 554d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 555d3ef3d73Snpiggin@suse.de */ 556d3ef3d73Snpiggin@suse.de smp_wmb(); 55783adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 558fbdc2f6cSChristian Brauner } 559fbdc2f6cSChristian Brauner 560fbdc2f6cSChristian Brauner static int mnt_make_readonly(struct mount *mnt) 561fbdc2f6cSChristian Brauner { 562fbdc2f6cSChristian Brauner int ret; 563fbdc2f6cSChristian Brauner 564fbdc2f6cSChristian Brauner ret = mnt_hold_writers(mnt); 565fbdc2f6cSChristian Brauner if (!ret) 566fbdc2f6cSChristian Brauner mnt->mnt.mnt_flags |= MNT_READONLY; 567fbdc2f6cSChristian Brauner mnt_unhold_writers(mnt); 5683d733633SDave Hansen return ret; 5693d733633SDave Hansen } 5708366025eSDave Hansen 5714ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5724ed5e82fSMiklos Szeredi { 5734ed5e82fSMiklos Szeredi struct mount *mnt; 5744ed5e82fSMiklos Szeredi int err = 0; 5754ed5e82fSMiklos Szeredi 5768e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5778e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5788e8b8796SMiklos Szeredi return -EBUSY; 5798e8b8796SMiklos Szeredi 580719ea2fbSAl Viro lock_mount_hash(); 5814ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5824ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 583ad1844a0SChristian Brauner err = mnt_hold_writers(mnt); 584ad1844a0SChristian Brauner if (err) 5854ed5e82fSMiklos Szeredi break; 5864ed5e82fSMiklos Szeredi } 5874ed5e82fSMiklos Szeredi } 5888e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5898e8b8796SMiklos Szeredi err = -EBUSY; 5908e8b8796SMiklos Szeredi 5914ed5e82fSMiklos Szeredi if (!err) { 5924ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5934ed5e82fSMiklos Szeredi smp_wmb(); 5944ed5e82fSMiklos Szeredi } 5954ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5964ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5974ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5984ed5e82fSMiklos Szeredi } 599719ea2fbSAl Viro unlock_mount_hash(); 6004ed5e82fSMiklos Szeredi 6014ed5e82fSMiklos Szeredi return err; 6024ed5e82fSMiklos Szeredi } 6034ed5e82fSMiklos Szeredi 604b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 6051da177e4SLinus Torvalds { 606256c8aedSChristian Brauner mnt_idmap_put(mnt_idmap(&mnt->mnt)); 607fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 608d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 60968e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 610d3ef3d73Snpiggin@suse.de #endif 611b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 6121da177e4SLinus Torvalds } 6131da177e4SLinus Torvalds 6148ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 6158ffcb32eSDavid Howells { 6168ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 6178ffcb32eSDavid Howells } 6188ffcb32eSDavid Howells 61948a066e7SAl Viro /* call under rcu_read_lock */ 620294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 62148a066e7SAl Viro { 62248a066e7SAl Viro struct mount *mnt; 62348a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 624294d71ffSAl Viro return 1; 62548a066e7SAl Viro if (bastard == NULL) 626294d71ffSAl Viro return 0; 62748a066e7SAl Viro mnt = real_mount(bastard); 62848a066e7SAl Viro mnt_add_count(mnt, 1); 629119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 63048a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 631294d71ffSAl Viro return 0; 63248a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 63348a066e7SAl Viro mnt_add_count(mnt, -1); 634294d71ffSAl Viro return 1; 63548a066e7SAl Viro } 636119e1ef8SAl Viro lock_mount_hash(); 637119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 638119e1ef8SAl Viro mnt_add_count(mnt, -1); 639119e1ef8SAl Viro unlock_mount_hash(); 640119e1ef8SAl Viro return 1; 641119e1ef8SAl Viro } 642119e1ef8SAl Viro unlock_mount_hash(); 643119e1ef8SAl Viro /* caller will mntput() */ 644294d71ffSAl Viro return -1; 645294d71ffSAl Viro } 646294d71ffSAl Viro 647294d71ffSAl Viro /* call under rcu_read_lock */ 6487e4745a0SAl Viro static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 649294d71ffSAl Viro { 650294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 651294d71ffSAl Viro if (likely(!res)) 652294d71ffSAl Viro return true; 653294d71ffSAl Viro if (unlikely(res < 0)) { 65448a066e7SAl Viro rcu_read_unlock(); 65548a066e7SAl Viro mntput(bastard); 65648a066e7SAl Viro rcu_read_lock(); 657294d71ffSAl Viro } 65848a066e7SAl Viro return false; 65948a066e7SAl Viro } 66048a066e7SAl Viro 6611da177e4SLinus Torvalds /* 662474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 66348a066e7SAl Viro * call under rcu_read_lock() 6641da177e4SLinus Torvalds */ 665474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6661da177e4SLinus Torvalds { 66738129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 668474279dcSAl Viro struct mount *p; 6691da177e4SLinus Torvalds 67038129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 671474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 672474279dcSAl Viro return p; 673474279dcSAl Viro return NULL; 6741da177e4SLinus Torvalds } 675474279dcSAl Viro 676474279dcSAl Viro /* 677f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 678f015f126SDavid Howells * 679f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 680f015f126SDavid Howells * following mounts: 681f015f126SDavid Howells * 682f015f126SDavid Howells * mount /dev/sda1 /mnt 683f015f126SDavid Howells * mount /dev/sda2 /mnt 684f015f126SDavid Howells * mount /dev/sda3 /mnt 685f015f126SDavid Howells * 686f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 687f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 688f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 689f015f126SDavid Howells * 690f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 691a05964f3SRam Pai */ 692ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 693a05964f3SRam Pai { 694c7105365SAl Viro struct mount *child_mnt; 69548a066e7SAl Viro struct vfsmount *m; 69648a066e7SAl Viro unsigned seq; 69799b7db7bSNick Piggin 69848a066e7SAl Viro rcu_read_lock(); 69948a066e7SAl Viro do { 70048a066e7SAl Viro seq = read_seqbegin(&mount_lock); 701474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 70248a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 70348a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 70448a066e7SAl Viro rcu_read_unlock(); 70548a066e7SAl Viro return m; 706a05964f3SRam Pai } 707a05964f3SRam Pai 7089f6c61f9SMiklos Szeredi static inline void lock_ns_list(struct mnt_namespace *ns) 7099f6c61f9SMiklos Szeredi { 7109f6c61f9SMiklos Szeredi spin_lock(&ns->ns_lock); 7119f6c61f9SMiklos Szeredi } 7129f6c61f9SMiklos Szeredi 7139f6c61f9SMiklos Szeredi static inline void unlock_ns_list(struct mnt_namespace *ns) 7149f6c61f9SMiklos Szeredi { 7159f6c61f9SMiklos Szeredi spin_unlock(&ns->ns_lock); 7169f6c61f9SMiklos Szeredi } 7179f6c61f9SMiklos Szeredi 7189f6c61f9SMiklos Szeredi static inline bool mnt_is_cursor(struct mount *mnt) 7199f6c61f9SMiklos Szeredi { 7209f6c61f9SMiklos Szeredi return mnt->mnt.mnt_flags & MNT_CURSOR; 7219f6c61f9SMiklos Szeredi } 7229f6c61f9SMiklos Szeredi 7237af1364fSEric W. Biederman /* 7247af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 7257af1364fSEric W. Biederman * current mount namespace. 7267af1364fSEric W. Biederman * 7277af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 7287af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 7297af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 7307af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 7317af1364fSEric W. Biederman * is a mountpoint. 7327af1364fSEric W. Biederman * 7337af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 7347af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 7357af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 7367af1364fSEric W. Biederman * parent mount. 7377af1364fSEric W. Biederman */ 7387af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 7397af1364fSEric W. Biederman { 7407af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7417af1364fSEric W. Biederman struct mount *mnt; 7427af1364fSEric W. Biederman bool is_covered = false; 7437af1364fSEric W. Biederman 7447af1364fSEric W. Biederman down_read(&namespace_sem); 7459f6c61f9SMiklos Szeredi lock_ns_list(ns); 7467af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7479f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 7489f6c61f9SMiklos Szeredi continue; 7497af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7507af1364fSEric W. Biederman if (is_covered) 7517af1364fSEric W. Biederman break; 7527af1364fSEric W. Biederman } 7539f6c61f9SMiklos Szeredi unlock_ns_list(ns); 7547af1364fSEric W. Biederman up_read(&namespace_sem); 7555ad05cc8SNikolay Borisov 7567af1364fSEric W. Biederman return is_covered; 7577af1364fSEric W. Biederman } 7587af1364fSEric W. Biederman 759e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 76084d17192SAl Viro { 7610818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 76284d17192SAl Viro struct mountpoint *mp; 76384d17192SAl Viro 7640818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 76584d17192SAl Viro if (mp->m_dentry == dentry) { 76684d17192SAl Viro mp->m_count++; 76784d17192SAl Viro return mp; 76884d17192SAl Viro } 76984d17192SAl Viro } 770e2dfa935SEric W. Biederman return NULL; 771e2dfa935SEric W. Biederman } 772e2dfa935SEric W. Biederman 7733895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 774e2dfa935SEric W. Biederman { 7753895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 776e2dfa935SEric W. Biederman int ret; 77784d17192SAl Viro 7783895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7791e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7801e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7811e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7823895dbf8SEric W. Biederman mountpoint: 7833895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7843895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7853895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7863895dbf8SEric W. Biederman if (mp) 7873895dbf8SEric W. Biederman goto done; 78884d17192SAl Viro } 789eed81007SMiklos Szeredi 7903895dbf8SEric W. Biederman if (!new) 7913895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7923895dbf8SEric W. Biederman if (!new) 7933895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7943895dbf8SEric W. Biederman 7953895dbf8SEric W. Biederman 7963895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7973895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7983895dbf8SEric W. Biederman 7993895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 8003895dbf8SEric W. Biederman if (ret == -EBUSY) 8013895dbf8SEric W. Biederman goto mountpoint; 8023895dbf8SEric W. Biederman 8033895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 8043895dbf8SEric W. Biederman mp = ERR_PTR(ret); 8053895dbf8SEric W. Biederman if (ret) 8063895dbf8SEric W. Biederman goto done; 8073895dbf8SEric W. Biederman 8083895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 8093895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 8104edbe133SAl Viro new->m_dentry = dget(dentry); 8113895dbf8SEric W. Biederman new->m_count = 1; 8123895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 8133895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 8143895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 8153895dbf8SEric W. Biederman 8163895dbf8SEric W. Biederman mp = new; 8173895dbf8SEric W. Biederman new = NULL; 8183895dbf8SEric W. Biederman done: 8193895dbf8SEric W. Biederman kfree(new); 82084d17192SAl Viro return mp; 82184d17192SAl Viro } 82284d17192SAl Viro 8234edbe133SAl Viro /* 8244edbe133SAl Viro * vfsmount lock must be held. Additionally, the caller is responsible 8254edbe133SAl Viro * for serializing calls for given disposal list. 8264edbe133SAl Viro */ 8274edbe133SAl Viro static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) 82884d17192SAl Viro { 82984d17192SAl Viro if (!--mp->m_count) { 83084d17192SAl Viro struct dentry *dentry = mp->m_dentry; 8310a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 83284d17192SAl Viro spin_lock(&dentry->d_lock); 83384d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 83484d17192SAl Viro spin_unlock(&dentry->d_lock); 8354edbe133SAl Viro dput_to_list(dentry, list); 8360818bf27SAl Viro hlist_del(&mp->m_hash); 83784d17192SAl Viro kfree(mp); 83884d17192SAl Viro } 83984d17192SAl Viro } 84084d17192SAl Viro 8414edbe133SAl Viro /* called with namespace_lock and vfsmount lock */ 8424edbe133SAl Viro static void put_mountpoint(struct mountpoint *mp) 8434edbe133SAl Viro { 8444edbe133SAl Viro __put_mountpoint(mp, &ex_mountpoints); 8454edbe133SAl Viro } 8464edbe133SAl Viro 847143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 8481da177e4SLinus Torvalds { 8496b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 8501da177e4SLinus Torvalds } 8511da177e4SLinus Torvalds 85299b7db7bSNick Piggin /* 85399b7db7bSNick Piggin * vfsmount lock must be held for write 85499b7db7bSNick Piggin */ 8556b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 8565addc5ddSAl Viro { 8575addc5ddSAl Viro if (ns) { 8585addc5ddSAl Viro ns->event = ++event; 8595addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8605addc5ddSAl Viro } 8615addc5ddSAl Viro } 8625addc5ddSAl Viro 86399b7db7bSNick Piggin /* 86499b7db7bSNick Piggin * vfsmount lock must be held for write 86599b7db7bSNick Piggin */ 8666b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8675addc5ddSAl Viro { 8685addc5ddSAl Viro if (ns && ns->event != event) { 8695addc5ddSAl Viro ns->event = event; 8705addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8715addc5ddSAl Viro } 8725addc5ddSAl Viro } 8735addc5ddSAl Viro 87499b7db7bSNick Piggin /* 87599b7db7bSNick Piggin * vfsmount lock must be held for write 87699b7db7bSNick Piggin */ 877e4e59906SAl Viro static struct mountpoint *unhash_mnt(struct mount *mnt) 8781da177e4SLinus Torvalds { 879e4e59906SAl Viro struct mountpoint *mp; 8800714a533SAl Viro mnt->mnt_parent = mnt; 881a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8826b41d536SAl Viro list_del_init(&mnt->mnt_child); 88338129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8840a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 885e4e59906SAl Viro mp = mnt->mnt_mp; 88684d17192SAl Viro mnt->mnt_mp = NULL; 887e4e59906SAl Viro return mp; 8887bdb11deSEric W. Biederman } 8897bdb11deSEric W. Biederman 8907bdb11deSEric W. Biederman /* 8917bdb11deSEric W. Biederman * vfsmount lock must be held for write 8927bdb11deSEric W. Biederman */ 8936a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8946a46c573SEric W. Biederman { 895e4e59906SAl Viro put_mountpoint(unhash_mnt(mnt)); 8966a46c573SEric W. Biederman } 8976a46c573SEric W. Biederman 8986a46c573SEric W. Biederman /* 8996a46c573SEric W. Biederman * vfsmount lock must be held for write 9006a46c573SEric W. Biederman */ 90184d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 90284d17192SAl Viro struct mountpoint *mp, 90344d964d6SAl Viro struct mount *child_mnt) 904b90fa9aeSRam Pai { 90584d17192SAl Viro mp->m_count++; 9063a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 9074edbe133SAl Viro child_mnt->mnt_mountpoint = mp->m_dentry; 9083a2393d7SAl Viro child_mnt->mnt_parent = mnt; 90984d17192SAl Viro child_mnt->mnt_mp = mp; 9100a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 911b90fa9aeSRam Pai } 912b90fa9aeSRam Pai 9131064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 9141064f874SEric W. Biederman { 9151064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 9161064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 9171064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 9181064f874SEric W. Biederman } 9191064f874SEric W. Biederman 92099b7db7bSNick Piggin /* 92199b7db7bSNick Piggin * vfsmount lock must be held for write 92299b7db7bSNick Piggin */ 92384d17192SAl Viro static void attach_mnt(struct mount *mnt, 92484d17192SAl Viro struct mount *parent, 92584d17192SAl Viro struct mountpoint *mp) 9261da177e4SLinus Torvalds { 92784d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 9281064f874SEric W. Biederman __attach_mnt(mnt, parent); 929b90fa9aeSRam Pai } 930b90fa9aeSRam Pai 9311064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 93212a5b529SAl Viro { 9331064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 9341064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 9351064f874SEric W. Biederman 9361064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 9371064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 9381064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 9391064f874SEric W. Biederman 9401064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 9411064f874SEric W. Biederman 9421064f874SEric W. Biederman put_mountpoint(old_mp); 9431064f874SEric W. Biederman mnt_add_count(old_parent, -1); 94412a5b529SAl Viro } 94512a5b529SAl Viro 946b90fa9aeSRam Pai /* 94799b7db7bSNick Piggin * vfsmount lock must be held for write 948b90fa9aeSRam Pai */ 9491064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 950b90fa9aeSRam Pai { 9510714a533SAl Viro struct mount *parent = mnt->mnt_parent; 95283adc753SAl Viro struct mount *m; 953b90fa9aeSRam Pai LIST_HEAD(head); 954143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 955b90fa9aeSRam Pai 9560714a533SAl Viro BUG_ON(parent == mnt); 957b90fa9aeSRam Pai 9581a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 959f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 960143c8c91SAl Viro m->mnt_ns = n; 961f03c6599SAl Viro 962b90fa9aeSRam Pai list_splice(&head, n->list.prev); 963b90fa9aeSRam Pai 964d2921684SEric W. Biederman n->mounts += n->pending_mounts; 965d2921684SEric W. Biederman n->pending_mounts = 0; 966d2921684SEric W. Biederman 9671064f874SEric W. Biederman __attach_mnt(mnt, parent); 9686b3286edSKirill Korotaev touch_mnt_namespace(n); 9691da177e4SLinus Torvalds } 9701da177e4SLinus Torvalds 971909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9721da177e4SLinus Torvalds { 9736b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9746b41d536SAl Viro if (next == &p->mnt_mounts) { 9751da177e4SLinus Torvalds while (1) { 976909b0a88SAl Viro if (p == root) 9771da177e4SLinus Torvalds return NULL; 9786b41d536SAl Viro next = p->mnt_child.next; 9796b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9801da177e4SLinus Torvalds break; 9810714a533SAl Viro p = p->mnt_parent; 9821da177e4SLinus Torvalds } 9831da177e4SLinus Torvalds } 9846b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9851da177e4SLinus Torvalds } 9861da177e4SLinus Torvalds 987315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9889676f0c6SRam Pai { 9896b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9906b41d536SAl Viro while (prev != &p->mnt_mounts) { 9916b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9926b41d536SAl Viro prev = p->mnt_mounts.prev; 9939676f0c6SRam Pai } 9949676f0c6SRam Pai return p; 9959676f0c6SRam Pai } 9969676f0c6SRam Pai 9978f291889SAl Viro /** 9988f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 9998f291889SAl Viro * @fc: The configuration context with the superblock attached 10008f291889SAl Viro * 10018f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 10028f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 10038f291889SAl Viro * 10048f291889SAl Viro * Note that this does not attach the mount to anything. 10058f291889SAl Viro */ 10068f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 10079d412a43SAl Viro { 1008b105e270SAl Viro struct mount *mnt; 10099d412a43SAl Viro 10108f291889SAl Viro if (!fc->root) 10118f291889SAl Viro return ERR_PTR(-EINVAL); 10129d412a43SAl Viro 10138f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 10149d412a43SAl Viro if (!mnt) 10159d412a43SAl Viro return ERR_PTR(-ENOMEM); 10169d412a43SAl Viro 10178f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 1018b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 10199d412a43SAl Viro 10208f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 10218f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 10228f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 1023a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10240714a533SAl Viro mnt->mnt_parent = mnt; 10258f291889SAl Viro 1026719ea2fbSAl Viro lock_mount_hash(); 10278f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 1028719ea2fbSAl Viro unlock_mount_hash(); 1029b105e270SAl Viro return &mnt->mnt; 10309d412a43SAl Viro } 10318f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 10328f291889SAl Viro 10338f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 10348f291889SAl Viro { 10358f291889SAl Viro int err = vfs_get_tree(fc); 10368f291889SAl Viro if (!err) { 10378f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 10388f291889SAl Viro return vfs_create_mount(fc); 10398f291889SAl Viro } 10408f291889SAl Viro return ERR_PTR(err); 10418f291889SAl Viro } 10428f291889SAl Viro EXPORT_SYMBOL(fc_mount); 10438f291889SAl Viro 10449bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 10459bc61ab1SDavid Howells int flags, const char *name, 10469bc61ab1SDavid Howells void *data) 10471da177e4SLinus Torvalds { 10489bc61ab1SDavid Howells struct fs_context *fc; 10498f291889SAl Viro struct vfsmount *mnt; 10509bc61ab1SDavid Howells int ret = 0; 10519d412a43SAl Viro 10529d412a43SAl Viro if (!type) 10533e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 10549d412a43SAl Viro 10559bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 10569bc61ab1SDavid Howells if (IS_ERR(fc)) 10579bc61ab1SDavid Howells return ERR_CAST(fc); 10589bc61ab1SDavid Howells 10593e1aeb00SDavid Howells if (name) 10603e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 10613e1aeb00SDavid Howells name, strlen(name)); 10629bc61ab1SDavid Howells if (!ret) 10639bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 10649bc61ab1SDavid Howells if (!ret) 10658f291889SAl Viro mnt = fc_mount(fc); 10668f291889SAl Viro else 10678f291889SAl Viro mnt = ERR_PTR(ret); 10689d412a43SAl Viro 10699bc61ab1SDavid Howells put_fs_context(fc); 10708f291889SAl Viro return mnt; 10719d412a43SAl Viro } 10729d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10739d412a43SAl Viro 107493faccbbSEric W. Biederman struct vfsmount * 107593faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 107693faccbbSEric W. Biederman const char *name, void *data) 107793faccbbSEric W. Biederman { 107893faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 107993faccbbSEric W. Biederman * through from the parent mount to the submount don't support 108093faccbbSEric W. Biederman * unprivileged mounts with submounts. 108193faccbbSEric W. Biederman */ 108293faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 108393faccbbSEric W. Biederman return ERR_PTR(-EPERM); 108493faccbbSEric W. Biederman 1085e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 108693faccbbSEric W. Biederman } 108793faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 108893faccbbSEric W. Biederman 108987129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 109036341f64SRam Pai int flag) 10911da177e4SLinus Torvalds { 109287129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1093be34d1a3SDavid Howells struct mount *mnt; 1094be34d1a3SDavid Howells int err; 10951da177e4SLinus Torvalds 1096be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1097be34d1a3SDavid Howells if (!mnt) 1098be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1099be34d1a3SDavid Howells 11007a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 110115169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1102719f5d7fSMiklos Szeredi else 110315169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1104719f5d7fSMiklos Szeredi 110515169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1106be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1107719f5d7fSMiklos Szeredi if (err) 1108719f5d7fSMiklos Szeredi goto out_free; 1109719f5d7fSMiklos Szeredi } 1110719f5d7fSMiklos Szeredi 111116a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 111216a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 11135ff9d8a6SEric W. Biederman 11141da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1115256c8aedSChristian Brauner mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt)); 1116256c8aedSChristian Brauner 1117b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1118b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1119a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 11200714a533SAl Viro mnt->mnt_parent = mnt; 1121719ea2fbSAl Viro lock_mount_hash(); 112239f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1123719ea2fbSAl Viro unlock_mount_hash(); 1124b90fa9aeSRam Pai 11257a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 11267a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 11276776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 112832301920SAl Viro mnt->mnt_master = old; 1129fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 11308aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1131fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 11326776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1133d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 11346776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1135d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 11365235d448SAl Viro } else { 11375235d448SAl Viro CLEAR_MNT_SHARED(mnt); 11385afe0022SRam Pai } 1139b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 11400f0afb1dSAl Viro set_mnt_shared(mnt); 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 11431da177e4SLinus Torvalds * as the original if that was on one */ 114436341f64SRam Pai if (flag & CL_EXPIRE) { 11456776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 11466776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 11471da177e4SLinus Torvalds } 1148be34d1a3SDavid Howells 1149cb338d06SAl Viro return mnt; 1150719f5d7fSMiklos Szeredi 1151719f5d7fSMiklos Szeredi out_free: 11528ffcb32eSDavid Howells mnt_free_id(mnt); 1153719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1154be34d1a3SDavid Howells return ERR_PTR(err); 11551da177e4SLinus Torvalds } 11561da177e4SLinus Torvalds 11579ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11589ea459e1SAl Viro { 115956cbb429SAl Viro struct hlist_node *p; 116056cbb429SAl Viro struct mount *m; 11619ea459e1SAl Viro /* 116256cbb429SAl Viro * The warning here probably indicates that somebody messed 116356cbb429SAl Viro * up a mnt_want/drop_write() pair. If this happens, the 116456cbb429SAl Viro * filesystem was probably unable to make r/w->r/o transitions. 11659ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11669ea459e1SAl Viro * so mnt_get_writers() below is safe. 11679ea459e1SAl Viro */ 11689ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11699ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11709ea459e1SAl Viro mnt_pin_kill(mnt); 117156cbb429SAl Viro hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { 117256cbb429SAl Viro hlist_del(&m->mnt_umount); 117356cbb429SAl Viro mntput(&m->mnt); 117456cbb429SAl Viro } 11759ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11769ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11779ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11789ea459e1SAl Viro mnt_free_id(mnt); 11799ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11809ea459e1SAl Viro } 11819ea459e1SAl Viro 11829ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11839ea459e1SAl Viro { 11849ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11859ea459e1SAl Viro } 11869ea459e1SAl Viro 11879ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11889ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11899ea459e1SAl Viro { 11909ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 119129785735SByungchul Park struct mount *m, *t; 11929ea459e1SAl Viro 119329785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 119429785735SByungchul Park cleanup_mnt(m); 11959ea459e1SAl Viro } 11969ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11979ea459e1SAl Viro 1198900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11997b7b1aceSAl Viro { 12004edbe133SAl Viro LIST_HEAD(list); 1201edf7ddbfSEric Biggers int count; 12024edbe133SAl Viro 120348a066e7SAl Viro rcu_read_lock(); 12049ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 12059ea0a46cSAl Viro /* 12069ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 12079ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 12089ea0a46cSAl Viro * non-NULL, then there's a reference that won't 12099ea0a46cSAl Viro * be dropped until after an RCU delay done after 12109ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 12119ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 12129ea0a46cSAl Viro * we are dropping is not the final one. 12139ea0a46cSAl Viro */ 1214aa9c0e07SAl Viro mnt_add_count(mnt, -1); 121548a066e7SAl Viro rcu_read_unlock(); 121699b7db7bSNick Piggin return; 1217b3e19d92SNick Piggin } 1218719ea2fbSAl Viro lock_mount_hash(); 1219119e1ef8SAl Viro /* 1220119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1221119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1222119e1ef8SAl Viro */ 1223119e1ef8SAl Viro smp_mb(); 12249ea0a46cSAl Viro mnt_add_count(mnt, -1); 1225edf7ddbfSEric Biggers count = mnt_get_count(mnt); 1226edf7ddbfSEric Biggers if (count != 0) { 1227edf7ddbfSEric Biggers WARN_ON(count < 0); 122848a066e7SAl Viro rcu_read_unlock(); 1229719ea2fbSAl Viro unlock_mount_hash(); 123099b7db7bSNick Piggin return; 123199b7db7bSNick Piggin } 123248a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 123348a066e7SAl Viro rcu_read_unlock(); 123448a066e7SAl Viro unlock_mount_hash(); 123548a066e7SAl Viro return; 123648a066e7SAl Viro } 123748a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 123848a066e7SAl Viro rcu_read_unlock(); 1239962830dfSAndi Kleen 124039f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1241ce07d891SEric W. Biederman 1242ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1243ce07d891SEric W. Biederman struct mount *p, *tmp; 1244ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 12454edbe133SAl Viro __put_mountpoint(unhash_mnt(p), &list); 124656cbb429SAl Viro hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); 1247ce07d891SEric W. Biederman } 1248ce07d891SEric W. Biederman } 1249719ea2fbSAl Viro unlock_mount_hash(); 12504edbe133SAl Viro shrink_dentry_list(&list); 1251649a795aSAl Viro 12529ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 12539ea459e1SAl Viro struct task_struct *task = current; 12549ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 12559ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 125691989c70SJens Axboe if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) 12579ea459e1SAl Viro return; 12589ea459e1SAl Viro } 12599ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 12609ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 12619ea459e1SAl Viro return; 12629ea459e1SAl Viro } 12639ea459e1SAl Viro cleanup_mnt(mnt); 1264b3e19d92SNick Piggin } 1265b3e19d92SNick Piggin 1266b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1267b3e19d92SNick Piggin { 1268b3e19d92SNick Piggin if (mnt) { 1269863d684fSAl Viro struct mount *m = real_mount(mnt); 1270b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1271863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1272863d684fSAl Viro m->mnt_expiry_mark = 0; 1273863d684fSAl Viro mntput_no_expire(m); 1274b3e19d92SNick Piggin } 1275b3e19d92SNick Piggin } 1276b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1277b3e19d92SNick Piggin 1278b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1279b3e19d92SNick Piggin { 1280b3e19d92SNick Piggin if (mnt) 128183adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1282b3e19d92SNick Piggin return mnt; 1283b3e19d92SNick Piggin } 1284b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1285b3e19d92SNick Piggin 12861f287bc4SRandy Dunlap /** 12871f287bc4SRandy Dunlap * path_is_mountpoint() - Check if path is a mount in the current namespace. 12881f287bc4SRandy Dunlap * @path: path to check 1289c6609c0aSIan Kent * 1290c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1291c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1292c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1293c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1294c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1295c6609c0aSIan Kent * alone. 1296c6609c0aSIan Kent */ 1297c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1298c6609c0aSIan Kent { 1299c6609c0aSIan Kent unsigned seq; 1300c6609c0aSIan Kent bool res; 1301c6609c0aSIan Kent 1302c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1303c6609c0aSIan Kent return false; 1304c6609c0aSIan Kent 1305c6609c0aSIan Kent rcu_read_lock(); 1306c6609c0aSIan Kent do { 1307c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1308c6609c0aSIan Kent res = __path_is_mountpoint(path); 1309c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1310c6609c0aSIan Kent rcu_read_unlock(); 1311c6609c0aSIan Kent 1312c6609c0aSIan Kent return res; 1313c6609c0aSIan Kent } 1314c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1315c6609c0aSIan Kent 1316ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 13177b7b1aceSAl Viro { 13183064c356SAl Viro struct mount *p; 13193064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 13203064c356SAl Viro if (IS_ERR(p)) 13213064c356SAl Viro return ERR_CAST(p); 13223064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 13233064c356SAl Viro return &p->mnt; 13247b7b1aceSAl Viro } 13251da177e4SLinus Torvalds 1326a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 13279f6c61f9SMiklos Szeredi static struct mount *mnt_list_next(struct mnt_namespace *ns, 13289f6c61f9SMiklos Szeredi struct list_head *p) 13299f6c61f9SMiklos Szeredi { 13309f6c61f9SMiklos Szeredi struct mount *mnt, *ret = NULL; 13319f6c61f9SMiklos Szeredi 13329f6c61f9SMiklos Szeredi lock_ns_list(ns); 13339f6c61f9SMiklos Szeredi list_for_each_continue(p, &ns->list) { 13349f6c61f9SMiklos Szeredi mnt = list_entry(p, typeof(*mnt), mnt_list); 13359f6c61f9SMiklos Szeredi if (!mnt_is_cursor(mnt)) { 13369f6c61f9SMiklos Szeredi ret = mnt; 13379f6c61f9SMiklos Szeredi break; 13389f6c61f9SMiklos Szeredi } 13399f6c61f9SMiklos Szeredi } 13409f6c61f9SMiklos Szeredi unlock_ns_list(ns); 13419f6c61f9SMiklos Szeredi 13429f6c61f9SMiklos Szeredi return ret; 13439f6c61f9SMiklos Szeredi } 13449f6c61f9SMiklos Szeredi 13450226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 13461da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 13471da177e4SLinus Torvalds { 1348ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13499f6c61f9SMiklos Szeredi struct list_head *prev; 13501da177e4SLinus Torvalds 1351390c6843SRam Pai down_read(&namespace_sem); 13529f6c61f9SMiklos Szeredi if (!*pos) { 13539f6c61f9SMiklos Szeredi prev = &p->ns->list; 13549f6c61f9SMiklos Szeredi } else { 13559f6c61f9SMiklos Szeredi prev = &p->cursor.mnt_list; 13569f6c61f9SMiklos Szeredi 13579f6c61f9SMiklos Szeredi /* Read after we'd reached the end? */ 13589f6c61f9SMiklos Szeredi if (list_empty(prev)) 13599f6c61f9SMiklos Szeredi return NULL; 1360c7999c36SAl Viro } 1361c7999c36SAl Viro 13629f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, prev); 13631da177e4SLinus Torvalds } 13641da177e4SLinus Torvalds 13651da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13661da177e4SLinus Torvalds { 1367ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13689f6c61f9SMiklos Szeredi struct mount *mnt = v; 1369b0765fb8SPavel Emelianov 13709f6c61f9SMiklos Szeredi ++*pos; 13719f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, &mnt->mnt_list); 13721da177e4SLinus Torvalds } 13731da177e4SLinus Torvalds 13741da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13751da177e4SLinus Torvalds { 13769f6c61f9SMiklos Szeredi struct proc_mounts *p = m->private; 13779f6c61f9SMiklos Szeredi struct mount *mnt = v; 13789f6c61f9SMiklos Szeredi 13799f6c61f9SMiklos Szeredi lock_ns_list(p->ns); 13809f6c61f9SMiklos Szeredi if (mnt) 13819f6c61f9SMiklos Szeredi list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); 13829f6c61f9SMiklos Szeredi else 13839f6c61f9SMiklos Szeredi list_del_init(&p->cursor.mnt_list); 13849f6c61f9SMiklos Szeredi unlock_ns_list(p->ns); 1385390c6843SRam Pai up_read(&namespace_sem); 13861da177e4SLinus Torvalds } 13871da177e4SLinus Torvalds 13880226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13899f5596afSAl Viro { 1390ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13919f6c61f9SMiklos Szeredi struct mount *r = v; 13920226f492SAl Viro return p->show(m, &r->mnt); 13931da177e4SLinus Torvalds } 13941da177e4SLinus Torvalds 1395a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13961da177e4SLinus Torvalds .start = m_start, 13971da177e4SLinus Torvalds .next = m_next, 13981da177e4SLinus Torvalds .stop = m_stop, 13990226f492SAl Viro .show = m_show, 1400b4629fe2SChuck Lever }; 14019f6c61f9SMiklos Szeredi 14029f6c61f9SMiklos Szeredi void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) 14039f6c61f9SMiklos Szeredi { 14049f6c61f9SMiklos Szeredi down_read(&namespace_sem); 14059f6c61f9SMiklos Szeredi lock_ns_list(ns); 14069f6c61f9SMiklos Szeredi list_del(&cursor->mnt_list); 14079f6c61f9SMiklos Szeredi unlock_ns_list(ns); 14089f6c61f9SMiklos Szeredi up_read(&namespace_sem); 14099f6c61f9SMiklos Szeredi } 1410a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1411b4629fe2SChuck Lever 14121da177e4SLinus Torvalds /** 14131da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 14141f287bc4SRandy Dunlap * @m: root of mount tree 14151da177e4SLinus Torvalds * 14161da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 14171da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 14181da177e4SLinus Torvalds * busy. 14191da177e4SLinus Torvalds */ 1420909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 14211da177e4SLinus Torvalds { 1422909b0a88SAl Viro struct mount *mnt = real_mount(m); 142336341f64SRam Pai int actual_refs = 0; 142436341f64SRam Pai int minimum_refs = 0; 1425315fc83eSAl Viro struct mount *p; 1426909b0a88SAl Viro BUG_ON(!m); 14271da177e4SLinus Torvalds 1428b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1429719ea2fbSAl Viro lock_mount_hash(); 1430909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 143183adc753SAl Viro actual_refs += mnt_get_count(p); 14321da177e4SLinus Torvalds minimum_refs += 2; 14331da177e4SLinus Torvalds } 1434719ea2fbSAl Viro unlock_mount_hash(); 14351da177e4SLinus Torvalds 14361da177e4SLinus Torvalds if (actual_refs > minimum_refs) 14371da177e4SLinus Torvalds return 0; 1438e3474a8eSIan Kent 1439e3474a8eSIan Kent return 1; 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds 14421da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 14431da177e4SLinus Torvalds 14441da177e4SLinus Torvalds /** 14451da177e4SLinus Torvalds * may_umount - check if a mount point is busy 14461da177e4SLinus Torvalds * @mnt: root of mount 14471da177e4SLinus Torvalds * 14481da177e4SLinus Torvalds * This is called to check if a mount point has any 14491da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 14501da177e4SLinus Torvalds * mount has sub mounts this will return busy 14511da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 14521da177e4SLinus Torvalds * 14531da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 14541da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 14551da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 14561da177e4SLinus Torvalds */ 14571da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 14581da177e4SLinus Torvalds { 1459e3474a8eSIan Kent int ret = 1; 14608ad08d8aSAl Viro down_read(&namespace_sem); 1461719ea2fbSAl Viro lock_mount_hash(); 14621ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1463e3474a8eSIan Kent ret = 0; 1464719ea2fbSAl Viro unlock_mount_hash(); 14658ad08d8aSAl Viro up_read(&namespace_sem); 1466a05964f3SRam Pai return ret; 14671da177e4SLinus Torvalds } 14681da177e4SLinus Torvalds 14691da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14701da177e4SLinus Torvalds 147197216be0SAl Viro static void namespace_unlock(void) 14721da177e4SLinus Torvalds { 1473a3b3c562SEric W. Biederman struct hlist_head head; 147456cbb429SAl Viro struct hlist_node *p; 147556cbb429SAl Viro struct mount *m; 14764edbe133SAl Viro LIST_HEAD(list); 147797216be0SAl Viro 1478a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 14794edbe133SAl Viro list_splice_init(&ex_mountpoints, &list); 1480a3b3c562SEric W. Biederman 148197216be0SAl Viro up_write(&namespace_sem); 1482a3b3c562SEric W. Biederman 14834edbe133SAl Viro shrink_dentry_list(&list); 14844edbe133SAl Viro 1485a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 148697216be0SAl Viro return; 148797216be0SAl Viro 148822cb7405SNeilBrown synchronize_rcu_expedited(); 148948a066e7SAl Viro 149056cbb429SAl Viro hlist_for_each_entry_safe(m, p, &head, mnt_umount) { 149156cbb429SAl Viro hlist_del(&m->mnt_umount); 149256cbb429SAl Viro mntput(&m->mnt); 149356cbb429SAl Viro } 149470fbcdf4SRam Pai } 149570fbcdf4SRam Pai 149697216be0SAl Viro static inline void namespace_lock(void) 1497e3197d83SAl Viro { 149897216be0SAl Viro down_write(&namespace_sem); 1499e3197d83SAl Viro } 1500e3197d83SAl Viro 1501e819f152SEric W. Biederman enum umount_tree_flags { 1502e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1503e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1504e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1505e819f152SEric W. Biederman }; 1506f2d0a123SEric W. Biederman 1507f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1508f2d0a123SEric W. Biederman { 1509f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1510f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1511f2d0a123SEric W. Biederman return true; 1512f2d0a123SEric W. Biederman 1513f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1514f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1515f2d0a123SEric W. Biederman return true; 1516f2d0a123SEric W. Biederman 1517f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1518f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1519f2d0a123SEric W. Biederman * connected to mounted mounts. 1520f2d0a123SEric W. Biederman */ 1521f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1522f2d0a123SEric W. Biederman return true; 1523f2d0a123SEric W. Biederman 1524f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1525f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1526f2d0a123SEric W. Biederman return false; 1527f2d0a123SEric W. Biederman 1528f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1529f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1530f2d0a123SEric W. Biederman return false; 1531f2d0a123SEric W. Biederman 1532f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1533f2d0a123SEric W. Biederman return true; 1534f2d0a123SEric W. Biederman } 1535f2d0a123SEric W. Biederman 153699b7db7bSNick Piggin /* 153748a066e7SAl Viro * mount_lock must be held 153899b7db7bSNick Piggin * namespace_sem must be held for write 153999b7db7bSNick Piggin */ 1540e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 154170fbcdf4SRam Pai { 1542c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1543315fc83eSAl Viro struct mount *p; 154470fbcdf4SRam Pai 15455d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 15465d88457eSEric W. Biederman propagate_mount_unlock(mnt); 15475d88457eSEric W. Biederman 1548c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1549590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1550590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1551c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1552590ce4bcSEric W. Biederman } 1553c003b26fSEric W. Biederman 1554411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1555c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1556c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 155738129a13SAl Viro } 155870fbcdf4SRam Pai 1559c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1560e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 15617b8a53fdSAl Viro propagate_umount(&tmp_list); 1562a05964f3SRam Pai 1563c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1564d2921684SEric W. Biederman struct mnt_namespace *ns; 1565ce07d891SEric W. Biederman bool disconnect; 1566c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 15676776db3dSAl Viro list_del_init(&p->mnt_expire); 15681a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1569d2921684SEric W. Biederman ns = p->mnt_ns; 1570d2921684SEric W. Biederman if (ns) { 1571d2921684SEric W. Biederman ns->mounts--; 1572d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1573d2921684SEric W. Biederman } 157463d37a84SAl Viro p->mnt_ns = NULL; 1575e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 157648a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 157787b95ce0SAl Viro 1578f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1579676da58dSAl Viro if (mnt_has_parent(p)) { 158081b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1581ce07d891SEric W. Biederman if (!disconnect) { 1582ce07d891SEric W. Biederman /* Don't forget about p */ 1583ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1584ce07d891SEric W. Biederman } else { 15856a46c573SEric W. Biederman umount_mnt(p); 15867c4b93d8SAl Viro } 1587ce07d891SEric W. Biederman } 15880f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 158919a1c409SAl Viro if (disconnect) 159019a1c409SAl Viro hlist_add_head(&p->mnt_umount, &unmounted); 159138129a13SAl Viro } 15921da177e4SLinus Torvalds } 15931da177e4SLinus Torvalds 1594b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1595c35038beSAl Viro 15968d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 15978d0347f6SDavid Howells { 15988d0347f6SDavid Howells int ret = 0; 15998d0347f6SDavid Howells 16008d0347f6SDavid Howells down_write(&sb->s_umount); 16018d0347f6SDavid Howells if (!sb_rdonly(sb)) { 16028d0347f6SDavid Howells struct fs_context *fc; 16038d0347f6SDavid Howells 16048d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 16058d0347f6SDavid Howells SB_RDONLY); 16068d0347f6SDavid Howells if (IS_ERR(fc)) { 16078d0347f6SDavid Howells ret = PTR_ERR(fc); 16088d0347f6SDavid Howells } else { 16098d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 16108d0347f6SDavid Howells if (!ret) 16118d0347f6SDavid Howells ret = reconfigure_super(fc); 16128d0347f6SDavid Howells put_fs_context(fc); 16138d0347f6SDavid Howells } 16148d0347f6SDavid Howells } 16158d0347f6SDavid Howells up_write(&sb->s_umount); 16168d0347f6SDavid Howells return ret; 16178d0347f6SDavid Howells } 16188d0347f6SDavid Howells 16191ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 16201da177e4SLinus Torvalds { 16211ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 16221da177e4SLinus Torvalds int retval; 16231da177e4SLinus Torvalds 16241ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 16251da177e4SLinus Torvalds if (retval) 16261da177e4SLinus Torvalds return retval; 16271da177e4SLinus Torvalds 16281da177e4SLinus Torvalds /* 16291da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 16301da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 16311da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 16321da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 16331da177e4SLinus Torvalds */ 16341da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 16351ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 16361da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 16371da177e4SLinus Torvalds return -EINVAL; 16381da177e4SLinus Torvalds 1639b3e19d92SNick Piggin /* 1640b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1641b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1642b3e19d92SNick Piggin */ 1643719ea2fbSAl Viro lock_mount_hash(); 164483adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1645719ea2fbSAl Viro unlock_mount_hash(); 16461da177e4SLinus Torvalds return -EBUSY; 1647b3e19d92SNick Piggin } 1648719ea2fbSAl Viro unlock_mount_hash(); 16491da177e4SLinus Torvalds 1650863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 16511da177e4SLinus Torvalds return -EAGAIN; 16521da177e4SLinus Torvalds } 16531da177e4SLinus Torvalds 16541da177e4SLinus Torvalds /* 16551da177e4SLinus Torvalds * If we may have to abort operations to get out of this 16561da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 16571da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 16581da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 16591da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 16601da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 16611da177e4SLinus Torvalds * about for the moment. 16621da177e4SLinus Torvalds */ 16631da177e4SLinus Torvalds 166442faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 166542faad99SAl Viro sb->s_op->umount_begin(sb); 166642faad99SAl Viro } 16671da177e4SLinus Torvalds 16681da177e4SLinus Torvalds /* 16691da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 16701da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 16711da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 16721da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 16731da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 16741da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 16751da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 16761da177e4SLinus Torvalds */ 16771ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 16781da177e4SLinus Torvalds /* 16791da177e4SLinus Torvalds * Special case for "unmounting" root ... 16801da177e4SLinus Torvalds * we just try to remount it readonly. 16811da177e4SLinus Torvalds */ 1682bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1683a1480dccSAndy Lutomirski return -EPERM; 16848d0347f6SDavid Howells return do_umount_root(sb); 16851da177e4SLinus Torvalds } 16861da177e4SLinus Torvalds 168797216be0SAl Viro namespace_lock(); 1688719ea2fbSAl Viro lock_mount_hash(); 16891da177e4SLinus Torvalds 169025d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 169125d202edSEric W. Biederman retval = -EINVAL; 169225d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 169325d202edSEric W. Biederman goto out; 169425d202edSEric W. Biederman 169525d202edSEric W. Biederman event++; 169648a066e7SAl Viro if (flags & MNT_DETACH) { 169748a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1698e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 169948a066e7SAl Viro retval = 0; 170048a066e7SAl Viro } else { 1701b54b9be7SAl Viro shrink_submounts(mnt); 17021da177e4SLinus Torvalds retval = -EBUSY; 170348a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 17041a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1705e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 17061da177e4SLinus Torvalds retval = 0; 17071da177e4SLinus Torvalds } 170848a066e7SAl Viro } 170925d202edSEric W. Biederman out: 1710719ea2fbSAl Viro unlock_mount_hash(); 1711e3197d83SAl Viro namespace_unlock(); 17121da177e4SLinus Torvalds return retval; 17131da177e4SLinus Torvalds } 17141da177e4SLinus Torvalds 17151da177e4SLinus Torvalds /* 171680b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 171780b5dce8SEric W. Biederman * 171880b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 171980b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 172080b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 172180b5dce8SEric W. Biederman * leaking them. 172280b5dce8SEric W. Biederman * 172380b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 172480b5dce8SEric W. Biederman */ 172580b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 172680b5dce8SEric W. Biederman { 172780b5dce8SEric W. Biederman struct mountpoint *mp; 172880b5dce8SEric W. Biederman struct mount *mnt; 172980b5dce8SEric W. Biederman 173080b5dce8SEric W. Biederman namespace_lock(); 17313895dbf8SEric W. Biederman lock_mount_hash(); 173280b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1733adc9b5c0SAl Viro if (!mp) 173480b5dce8SEric W. Biederman goto out_unlock; 173580b5dce8SEric W. Biederman 1736e06b933eSAndrey Ulanov event++; 173780b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 173880b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1739ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1740fe78fcc8SEric W. Biederman umount_mnt(mnt); 174156cbb429SAl Viro hlist_add_head(&mnt->mnt_umount, &unmounted); 1742ce07d891SEric W. Biederman } 1743e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 174480b5dce8SEric W. Biederman } 174580b5dce8SEric W. Biederman put_mountpoint(mp); 174680b5dce8SEric W. Biederman out_unlock: 17473895dbf8SEric W. Biederman unlock_mount_hash(); 174880b5dce8SEric W. Biederman namespace_unlock(); 174980b5dce8SEric W. Biederman } 175080b5dce8SEric W. Biederman 175180b5dce8SEric W. Biederman /* 17529b40bc90SAl Viro * Is the caller allowed to modify his namespace? 17539b40bc90SAl Viro */ 1754a5f85d78SAl Viro bool may_mount(void) 17559b40bc90SAl Viro { 17569b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 17579b40bc90SAl Viro } 17589b40bc90SAl Viro 1759f7e33bdbSJeff Layton static void warn_mandlock(void) 17609e8925b6SJeff Layton { 1761f7e33bdbSJeff Layton pr_warn_once("=======================================================\n" 1762f7e33bdbSJeff Layton "WARNING: The mand mount option has been deprecated and\n" 1763f7e33bdbSJeff Layton " and is ignored by this kernel. Remove the mand\n" 1764f7e33bdbSJeff Layton " option from the mount to silence this warning.\n" 1765f7e33bdbSJeff Layton "=======================================================\n"); 17669e8925b6SJeff Layton } 17679e8925b6SJeff Layton 176825ccd24fSChristoph Hellwig static int can_umount(const struct path *path, int flags) 17691da177e4SLinus Torvalds { 177025ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 17711da177e4SLinus Torvalds 17729b40bc90SAl Viro if (!may_mount()) 17739b40bc90SAl Viro return -EPERM; 177441525f56SChristoph Hellwig if (path->dentry != path->mnt->mnt_root) 177525ccd24fSChristoph Hellwig return -EINVAL; 1776143c8c91SAl Viro if (!check_mnt(mnt)) 177725ccd24fSChristoph Hellwig return -EINVAL; 177825d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 177925ccd24fSChristoph Hellwig return -EINVAL; 1780b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 178125ccd24fSChristoph Hellwig return -EPERM; 178225ccd24fSChristoph Hellwig return 0; 178325ccd24fSChristoph Hellwig } 17841da177e4SLinus Torvalds 1785a0a6df9aSAl Viro // caller is responsible for flags being sane 178625ccd24fSChristoph Hellwig int path_umount(struct path *path, int flags) 178725ccd24fSChristoph Hellwig { 178825ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 178925ccd24fSChristoph Hellwig int ret; 179025ccd24fSChristoph Hellwig 179125ccd24fSChristoph Hellwig ret = can_umount(path, flags); 179225ccd24fSChristoph Hellwig if (!ret) 179325ccd24fSChristoph Hellwig ret = do_umount(mnt, flags); 179425ccd24fSChristoph Hellwig 1795429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 179641525f56SChristoph Hellwig dput(path->dentry); 1797900148dcSAl Viro mntput_no_expire(mnt); 179825ccd24fSChristoph Hellwig return ret; 17991da177e4SLinus Torvalds } 18001da177e4SLinus Torvalds 180109267defSChristoph Hellwig static int ksys_umount(char __user *name, int flags) 180241525f56SChristoph Hellwig { 180341525f56SChristoph Hellwig int lookup_flags = LOOKUP_MOUNTPOINT; 180441525f56SChristoph Hellwig struct path path; 180541525f56SChristoph Hellwig int ret; 180641525f56SChristoph Hellwig 1807a0a6df9aSAl Viro // basic validity checks done first 1808a0a6df9aSAl Viro if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1809a0a6df9aSAl Viro return -EINVAL; 1810a0a6df9aSAl Viro 181141525f56SChristoph Hellwig if (!(flags & UMOUNT_NOFOLLOW)) 181241525f56SChristoph Hellwig lookup_flags |= LOOKUP_FOLLOW; 181341525f56SChristoph Hellwig ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); 181441525f56SChristoph Hellwig if (ret) 181541525f56SChristoph Hellwig return ret; 181641525f56SChristoph Hellwig return path_umount(&path, flags); 181741525f56SChristoph Hellwig } 181841525f56SChristoph Hellwig 18193a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 18203a18ef5cSDominik Brodowski { 18213a18ef5cSDominik Brodowski return ksys_umount(name, flags); 18223a18ef5cSDominik Brodowski } 18233a18ef5cSDominik Brodowski 18241da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 18251da177e4SLinus Torvalds 18261da177e4SLinus Torvalds /* 18271da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 18281da177e4SLinus Torvalds */ 1829bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 18301da177e4SLinus Torvalds { 18313a18ef5cSDominik Brodowski return ksys_umount(name, 0); 18321da177e4SLinus Torvalds } 18331da177e4SLinus Torvalds 18341da177e4SLinus Torvalds #endif 18351da177e4SLinus Torvalds 18364ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 18378823c079SEric W. Biederman { 18384ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1839e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1840e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 18414ce5d2b1SEric W. Biederman } 18424ce5d2b1SEric W. Biederman 1843213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 184458be2825SAl Viro { 184558be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 184658be2825SAl Viro } 184758be2825SAl Viro 1848303cc571SChristian Brauner struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) 1849303cc571SChristian Brauner { 1850303cc571SChristian Brauner return &mnt->ns; 1851303cc571SChristian Brauner } 1852303cc571SChristian Brauner 18534ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 18544ce5d2b1SEric W. Biederman { 18554ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 18564ce5d2b1SEric W. Biederman * mount namespace loop? 18574ce5d2b1SEric W. Biederman */ 18584ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 18594ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 18604ce5d2b1SEric W. Biederman return false; 18614ce5d2b1SEric W. Biederman 1862f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 18638823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 18648823c079SEric W. Biederman } 18658823c079SEric W. Biederman 186687129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 186736341f64SRam Pai int flag) 18681da177e4SLinus Torvalds { 186984d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 18701da177e4SLinus Torvalds 18714ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 18724ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 18734ce5d2b1SEric W. Biederman 18744ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1875be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 18769676f0c6SRam Pai 187736341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1878be34d1a3SDavid Howells if (IS_ERR(q)) 1879be34d1a3SDavid Howells return q; 1880be34d1a3SDavid Howells 1881a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 18821da177e4SLinus Torvalds 18831da177e4SLinus Torvalds p = mnt; 18846b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1885315fc83eSAl Viro struct mount *s; 18867ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 18871da177e4SLinus Torvalds continue; 18881da177e4SLinus Torvalds 1889909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 18904ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 18914ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1892df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1893df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1894df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1895df7342b2SEric W. Biederman goto out; 1896df7342b2SEric W. Biederman } else { 18974ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 18984ce5d2b1SEric W. Biederman continue; 18994ce5d2b1SEric W. Biederman } 1900df7342b2SEric W. Biederman } 19014ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 19024ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 19039676f0c6SRam Pai s = skip_mnt_tree(s); 19049676f0c6SRam Pai continue; 19059676f0c6SRam Pai } 19060714a533SAl Viro while (p != s->mnt_parent) { 19070714a533SAl Viro p = p->mnt_parent; 19080714a533SAl Viro q = q->mnt_parent; 19091da177e4SLinus Torvalds } 191087129cc0SAl Viro p = s; 191184d17192SAl Viro parent = q; 191287129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1913be34d1a3SDavid Howells if (IS_ERR(q)) 1914be34d1a3SDavid Howells goto out; 1915719ea2fbSAl Viro lock_mount_hash(); 19161a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 19171064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1918719ea2fbSAl Viro unlock_mount_hash(); 19191da177e4SLinus Torvalds } 19201da177e4SLinus Torvalds } 19211da177e4SLinus Torvalds return res; 1922be34d1a3SDavid Howells out: 19231da177e4SLinus Torvalds if (res) { 1924719ea2fbSAl Viro lock_mount_hash(); 1925e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1926719ea2fbSAl Viro unlock_mount_hash(); 19271da177e4SLinus Torvalds } 1928be34d1a3SDavid Howells return q; 19291da177e4SLinus Torvalds } 19301da177e4SLinus Torvalds 1931be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1932be34d1a3SDavid Howells 1933ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 19348aec0809SAl Viro { 1935cb338d06SAl Viro struct mount *tree; 193697216be0SAl Viro namespace_lock(); 1937cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1938cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1939cd4a4017SEric W. Biederman else 194087129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 194187129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1942328e6d90SAl Viro namespace_unlock(); 1943be34d1a3SDavid Howells if (IS_ERR(tree)) 194452e220d3SDan Carpenter return ERR_CAST(tree); 1945be34d1a3SDavid Howells return &tree->mnt; 19468aec0809SAl Viro } 19478aec0809SAl Viro 1948a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1949a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1950a07b2000SAl Viro 1951a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1952a07b2000SAl Viro { 1953a07b2000SAl Viro struct mnt_namespace *ns; 1954a07b2000SAl Viro namespace_lock(); 1955a07b2000SAl Viro lock_mount_hash(); 1956a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 195744dfd84aSDavid Howells if (ns) { 195844dfd84aSDavid Howells if (is_anon_ns(ns)) 1959a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 196044dfd84aSDavid Howells else 196144dfd84aSDavid Howells ns = NULL; 196244dfd84aSDavid Howells } 1963a07b2000SAl Viro unlock_mount_hash(); 1964a07b2000SAl Viro namespace_unlock(); 196544dfd84aSDavid Howells if (ns) 1966a07b2000SAl Viro free_mnt_ns(ns); 1967a07b2000SAl Viro } 1968a07b2000SAl Viro 19698aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 19708aec0809SAl Viro { 197197216be0SAl Viro namespace_lock(); 1972719ea2fbSAl Viro lock_mount_hash(); 19739c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1974719ea2fbSAl Viro unlock_mount_hash(); 19753ab6abeeSAl Viro namespace_unlock(); 19768aec0809SAl Viro } 19778aec0809SAl Viro 1978427215d8SMiklos Szeredi static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 1979427215d8SMiklos Szeredi { 1980427215d8SMiklos Szeredi struct mount *child; 1981427215d8SMiklos Szeredi 1982427215d8SMiklos Szeredi list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 1983427215d8SMiklos Szeredi if (!is_subdir(child->mnt_mountpoint, dentry)) 1984427215d8SMiklos Szeredi continue; 1985427215d8SMiklos Szeredi 1986427215d8SMiklos Szeredi if (child->mnt.mnt_flags & MNT_LOCKED) 1987427215d8SMiklos Szeredi return true; 1988427215d8SMiklos Szeredi } 1989427215d8SMiklos Szeredi return false; 1990427215d8SMiklos Szeredi } 1991427215d8SMiklos Szeredi 1992c771d683SMiklos Szeredi /** 1993c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 19941f287bc4SRandy Dunlap * @path: path to clone 1995c771d683SMiklos Szeredi * 19961f287bc4SRandy Dunlap * This creates a new vfsmount, which will be the clone of @path. The new mount 19971f287bc4SRandy Dunlap * will not be attached anywhere in the namespace and will be private (i.e. 19981f287bc4SRandy Dunlap * changes to the originating mount won't be propagated into this). 1999c771d683SMiklos Szeredi * 2000c771d683SMiklos Szeredi * Release with mntput(). 2001c771d683SMiklos Szeredi */ 2002ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 2003c771d683SMiklos Szeredi { 2004c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 2005c771d683SMiklos Szeredi struct mount *new_mnt; 2006c771d683SMiklos Szeredi 2007427215d8SMiklos Szeredi down_read(&namespace_sem); 2008c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 2009427215d8SMiklos Szeredi goto invalid; 2010427215d8SMiklos Szeredi 2011427215d8SMiklos Szeredi if (!check_mnt(old_mnt)) 2012427215d8SMiklos Szeredi goto invalid; 2013427215d8SMiklos Szeredi 2014427215d8SMiklos Szeredi if (has_locked_children(old_mnt, path->dentry)) 2015427215d8SMiklos Szeredi goto invalid; 2016c771d683SMiklos Szeredi 2017c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 2018427215d8SMiklos Szeredi up_read(&namespace_sem); 2019427215d8SMiklos Szeredi 2020c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 2021c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 2022c771d683SMiklos Szeredi 2023df820f8dSMiklos Szeredi /* Longterm mount to be removed by kern_unmount*() */ 2024df820f8dSMiklos Szeredi new_mnt->mnt_ns = MNT_NS_INTERNAL; 2025df820f8dSMiklos Szeredi 2026c771d683SMiklos Szeredi return &new_mnt->mnt; 2027427215d8SMiklos Szeredi 2028427215d8SMiklos Szeredi invalid: 2029427215d8SMiklos Szeredi up_read(&namespace_sem); 2030427215d8SMiklos Szeredi return ERR_PTR(-EINVAL); 2031c771d683SMiklos Szeredi } 2032c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 2033c771d683SMiklos Szeredi 20341f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 20351f707137SAl Viro struct vfsmount *root) 20361f707137SAl Viro { 20371a4eeaf2SAl Viro struct mount *mnt; 20381f707137SAl Viro int res = f(root, arg); 20391f707137SAl Viro if (res) 20401f707137SAl Viro return res; 20411a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 20421a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 20431f707137SAl Viro if (res) 20441f707137SAl Viro return res; 20451f707137SAl Viro } 20461f707137SAl Viro return 0; 20471f707137SAl Viro } 20481f707137SAl Viro 20493bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 20503bd045ccSAl Viro { 20513bd045ccSAl Viro struct mount *p; 20523bd045ccSAl Viro 20533bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 20543bd045ccSAl Viro int flags = p->mnt.mnt_flags; 20553bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 20563bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 20573bd045ccSAl Viro 20583bd045ccSAl Viro if (flags & MNT_READONLY) 20593bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 20603bd045ccSAl Viro 20613bd045ccSAl Viro if (flags & MNT_NODEV) 20623bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 20633bd045ccSAl Viro 20643bd045ccSAl Viro if (flags & MNT_NOSUID) 20653bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 20663bd045ccSAl Viro 20673bd045ccSAl Viro if (flags & MNT_NOEXEC) 20683bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 20693bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 20703bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 20713bd045ccSAl Viro flags |= MNT_LOCKED; 20723bd045ccSAl Viro p->mnt.mnt_flags = flags; 20733bd045ccSAl Viro } 20743bd045ccSAl Viro } 20753bd045ccSAl Viro 20764b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 2077719f5d7fSMiklos Szeredi { 2078315fc83eSAl Viro struct mount *p; 2079719f5d7fSMiklos Szeredi 2080909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 2081fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 20824b8b21f4SAl Viro mnt_release_group_id(p); 2083719f5d7fSMiklos Szeredi } 2084719f5d7fSMiklos Szeredi } 2085719f5d7fSMiklos Szeredi 20864b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 2087719f5d7fSMiklos Szeredi { 2088315fc83eSAl Viro struct mount *p; 2089719f5d7fSMiklos Szeredi 2090909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 2091fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 20924b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 2093719f5d7fSMiklos Szeredi if (err) { 20944b8b21f4SAl Viro cleanup_group_ids(mnt, p); 2095719f5d7fSMiklos Szeredi return err; 2096719f5d7fSMiklos Szeredi } 2097719f5d7fSMiklos Szeredi } 2098719f5d7fSMiklos Szeredi } 2099719f5d7fSMiklos Szeredi 2100719f5d7fSMiklos Szeredi return 0; 2101719f5d7fSMiklos Szeredi } 2102719f5d7fSMiklos Szeredi 2103d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 2104d2921684SEric W. Biederman { 2105d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 2106124f75f8SAl Viro unsigned int mounts = 0; 2107d2921684SEric W. Biederman struct mount *p; 2108d2921684SEric W. Biederman 2109124f75f8SAl Viro if (ns->mounts >= max) 2110124f75f8SAl Viro return -ENOSPC; 2111124f75f8SAl Viro max -= ns->mounts; 2112124f75f8SAl Viro if (ns->pending_mounts >= max) 2113124f75f8SAl Viro return -ENOSPC; 2114124f75f8SAl Viro max -= ns->pending_mounts; 2115124f75f8SAl Viro 2116d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 2117d2921684SEric W. Biederman mounts++; 2118d2921684SEric W. Biederman 2119124f75f8SAl Viro if (mounts > max) 2120d2921684SEric W. Biederman return -ENOSPC; 2121d2921684SEric W. Biederman 2122124f75f8SAl Viro ns->pending_mounts += mounts; 2123d2921684SEric W. Biederman return 0; 2124d2921684SEric W. Biederman } 2125d2921684SEric W. Biederman 2126b90fa9aeSRam Pai /* 2127b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 2128b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 212921444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 213021444403SRam Pai * store the parent mount and mountpoint dentry. 213121444403SRam Pai * (done when source_mnt is moved) 2132b90fa9aeSRam Pai * 2133b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 2134b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 21359676f0c6SRam Pai * --------------------------------------------------------------------------- 2136b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 21379676f0c6SRam Pai * |************************************************************************** 21389676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21399676f0c6SRam Pai * | dest | | | | | 21409676f0c6SRam Pai * | | | | | | | 21419676f0c6SRam Pai * | v | | | | | 21429676f0c6SRam Pai * |************************************************************************** 21439676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 21445afe0022SRam Pai * | | | | | | 21459676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 21469676f0c6SRam Pai * *************************************************************************** 2147b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 2148b90fa9aeSRam Pai * destination mount. 2149b90fa9aeSRam Pai * 2150b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2151b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2152b90fa9aeSRam Pai * the peer group of the source mount. 2153b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2154b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2155b90fa9aeSRam Pai * mount. 21565afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 21575afe0022SRam Pai * of the destination mount and the cloned mount is made slave 21585afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 21595afe0022SRam Pai * is marked as 'shared and slave'. 21605afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 21615afe0022SRam Pai * source mount. 21625afe0022SRam Pai * 21639676f0c6SRam Pai * --------------------------------------------------------------------------- 216421444403SRam Pai * | MOVE MOUNT OPERATION | 21659676f0c6SRam Pai * |************************************************************************** 21669676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21679676f0c6SRam Pai * | dest | | | | | 21689676f0c6SRam Pai * | | | | | | | 21699676f0c6SRam Pai * | v | | | | | 21709676f0c6SRam Pai * |************************************************************************** 21719676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 21725afe0022SRam Pai * | | | | | | 21739676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 21749676f0c6SRam Pai * *************************************************************************** 21755afe0022SRam Pai * 21765afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 21775afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 217821444403SRam Pai * (+*) the mount is moved to the destination. 21795afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 21805afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 21815afe0022SRam Pai * the mount is marked as 'shared and slave'. 21825afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2183b90fa9aeSRam Pai * 2184b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2185b90fa9aeSRam Pai * applied to each mount in the tree. 2186b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2187b90fa9aeSRam Pai * in allocations. 2188b90fa9aeSRam Pai */ 21890fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 219084d17192SAl Viro struct mount *dest_mnt, 219184d17192SAl Viro struct mountpoint *dest_mp, 21922763d119SAl Viro bool moving) 2193b90fa9aeSRam Pai { 21943bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 219538129a13SAl Viro HLIST_HEAD(tree_list); 2196d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 21971064f874SEric W. Biederman struct mountpoint *smp; 2198315fc83eSAl Viro struct mount *child, *p; 219938129a13SAl Viro struct hlist_node *n; 2200719f5d7fSMiklos Szeredi int err; 2201b90fa9aeSRam Pai 22021064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 22031064f874SEric W. Biederman * to be tucked under other mounts. 22041064f874SEric W. Biederman */ 22051064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 22061064f874SEric W. Biederman if (IS_ERR(smp)) 22071064f874SEric W. Biederman return PTR_ERR(smp); 22081064f874SEric W. Biederman 2209d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 22102763d119SAl Viro if (!moving) { 2211d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2212d2921684SEric W. Biederman if (err) 2213d2921684SEric W. Biederman goto out; 2214d2921684SEric W. Biederman } 2215d2921684SEric W. Biederman 2216fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 22170fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2218719f5d7fSMiklos Szeredi if (err) 2219719f5d7fSMiklos Szeredi goto out; 222084d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2221f2ebb3a9SAl Viro lock_mount_hash(); 2222719f5d7fSMiklos Szeredi if (err) 2223719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2224909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 22250f0afb1dSAl Viro set_mnt_shared(p); 22260b1b901bSAl Viro } else { 22270b1b901bSAl Viro lock_mount_hash(); 2228b90fa9aeSRam Pai } 22292763d119SAl Viro if (moving) { 22302763d119SAl Viro unhash_mnt(source_mnt); 223184d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2232143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 223321444403SRam Pai } else { 223444dfd84aSDavid Howells if (source_mnt->mnt_ns) { 223544dfd84aSDavid Howells /* move from anon - the caller will destroy */ 223644dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 223744dfd84aSDavid Howells } 223884d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 22391064f874SEric W. Biederman commit_tree(source_mnt); 224021444403SRam Pai } 2241b90fa9aeSRam Pai 224238129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 22431d6a32acSAl Viro struct mount *q; 224438129a13SAl Viro hlist_del_init(&child->mnt_hash); 22451064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 22461d6a32acSAl Viro child->mnt_mountpoint); 22471064f874SEric W. Biederman if (q) 22481064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 22493bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 22503bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 22513bd045ccSAl Viro lock_mnt_tree(child); 2252d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 22531064f874SEric W. Biederman commit_tree(child); 2254b90fa9aeSRam Pai } 22551064f874SEric W. Biederman put_mountpoint(smp); 2256719ea2fbSAl Viro unlock_mount_hash(); 225799b7db7bSNick Piggin 2258b90fa9aeSRam Pai return 0; 2259719f5d7fSMiklos Szeredi 2260719f5d7fSMiklos Szeredi out_cleanup_ids: 2261f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2262f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2263d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2264e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2265f2ebb3a9SAl Viro } 2266f2ebb3a9SAl Viro unlock_mount_hash(); 22670fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2268719f5d7fSMiklos Szeredi out: 2269d2921684SEric W. Biederman ns->pending_mounts = 0; 22701064f874SEric W. Biederman 22711064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 22721064f874SEric W. Biederman put_mountpoint(smp); 22731064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 22741064f874SEric W. Biederman 2275719f5d7fSMiklos Szeredi return err; 2276b90fa9aeSRam Pai } 2277b90fa9aeSRam Pai 227884d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2279b12cea91SAl Viro { 2280b12cea91SAl Viro struct vfsmount *mnt; 228184d17192SAl Viro struct dentry *dentry = path->dentry; 2282b12cea91SAl Viro retry: 22835955102cSAl Viro inode_lock(dentry->d_inode); 228484d17192SAl Viro if (unlikely(cant_mount(dentry))) { 22855955102cSAl Viro inode_unlock(dentry->d_inode); 228684d17192SAl Viro return ERR_PTR(-ENOENT); 2287b12cea91SAl Viro } 228897216be0SAl Viro namespace_lock(); 2289b12cea91SAl Viro mnt = lookup_mnt(path); 229084d17192SAl Viro if (likely(!mnt)) { 22913895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 229284d17192SAl Viro if (IS_ERR(mp)) { 229397216be0SAl Viro namespace_unlock(); 22945955102cSAl Viro inode_unlock(dentry->d_inode); 229584d17192SAl Viro return mp; 229684d17192SAl Viro } 229784d17192SAl Viro return mp; 229884d17192SAl Viro } 229997216be0SAl Viro namespace_unlock(); 23005955102cSAl Viro inode_unlock(path->dentry->d_inode); 2301b12cea91SAl Viro path_put(path); 2302b12cea91SAl Viro path->mnt = mnt; 230384d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2304b12cea91SAl Viro goto retry; 2305b12cea91SAl Viro } 2306b12cea91SAl Viro 230784d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2308b12cea91SAl Viro { 230984d17192SAl Viro struct dentry *dentry = where->m_dentry; 23103895dbf8SEric W. Biederman 23113895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 231284d17192SAl Viro put_mountpoint(where); 23133895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 23143895dbf8SEric W. Biederman 2315328e6d90SAl Viro namespace_unlock(); 23165955102cSAl Viro inode_unlock(dentry->d_inode); 2317b12cea91SAl Viro } 2318b12cea91SAl Viro 231984d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 23201da177e4SLinus Torvalds { 2321e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 23221da177e4SLinus Torvalds return -EINVAL; 23231da177e4SLinus Torvalds 2324e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2325e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 23261da177e4SLinus Torvalds return -ENOTDIR; 23271da177e4SLinus Torvalds 23282763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 23291da177e4SLinus Torvalds } 23301da177e4SLinus Torvalds 23311da177e4SLinus Torvalds /* 23327a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 23337a2e8a8fSValerie Aurora */ 23347a2e8a8fSValerie Aurora 2335e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 23367a2e8a8fSValerie Aurora { 2337e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 23387a2e8a8fSValerie Aurora 23397a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 23407a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 23417a2e8a8fSValerie Aurora return 0; 23427a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 23437a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 23447a2e8a8fSValerie Aurora return 0; 23457a2e8a8fSValerie Aurora return type; 23467a2e8a8fSValerie Aurora } 23477a2e8a8fSValerie Aurora 23487a2e8a8fSValerie Aurora /* 234907b20889SRam Pai * recursively change the type of the mountpoint. 235007b20889SRam Pai */ 2351e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 235207b20889SRam Pai { 2353315fc83eSAl Viro struct mount *m; 23544b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2355e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 23567a2e8a8fSValerie Aurora int type; 2357719f5d7fSMiklos Szeredi int err = 0; 235807b20889SRam Pai 23592d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 236007b20889SRam Pai return -EINVAL; 236107b20889SRam Pai 2362e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 23637a2e8a8fSValerie Aurora if (!type) 23647a2e8a8fSValerie Aurora return -EINVAL; 23657a2e8a8fSValerie Aurora 236697216be0SAl Viro namespace_lock(); 2367719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2368719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2369719f5d7fSMiklos Szeredi if (err) 2370719f5d7fSMiklos Szeredi goto out_unlock; 2371719f5d7fSMiklos Szeredi } 2372719f5d7fSMiklos Szeredi 2373719ea2fbSAl Viro lock_mount_hash(); 2374909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 23750f0afb1dSAl Viro change_mnt_propagation(m, type); 2376719ea2fbSAl Viro unlock_mount_hash(); 2377719f5d7fSMiklos Szeredi 2378719f5d7fSMiklos Szeredi out_unlock: 237997216be0SAl Viro namespace_unlock(); 2380719f5d7fSMiklos Szeredi return err; 238107b20889SRam Pai } 238207b20889SRam Pai 2383a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2384a07b2000SAl Viro { 2385a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2386a07b2000SAl Viro 2387a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2388a07b2000SAl Viro return mnt; 2389a07b2000SAl Viro 2390a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2391a07b2000SAl Viro return mnt; 2392a07b2000SAl Viro 2393a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2394a07b2000SAl Viro return mnt; 2395a07b2000SAl Viro 2396a07b2000SAl Viro if (recurse) 2397a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2398a07b2000SAl Viro else 2399a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2400a07b2000SAl Viro 2401a07b2000SAl Viro if (!IS_ERR(mnt)) 2402a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2403a07b2000SAl Viro 2404a07b2000SAl Viro return mnt; 2405a07b2000SAl Viro } 2406a07b2000SAl Viro 240707b20889SRam Pai /* 24081da177e4SLinus Torvalds * do loopback mount. 24091da177e4SLinus Torvalds */ 2410808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 24112dafe1c4SEric Sandeen int recurse) 24121da177e4SLinus Torvalds { 24132d92ab3cSAl Viro struct path old_path; 2414a07b2000SAl Viro struct mount *mnt = NULL, *parent; 241584d17192SAl Viro struct mountpoint *mp; 241657eccb83SAl Viro int err; 24171da177e4SLinus Torvalds if (!old_name || !*old_name) 24181da177e4SLinus Torvalds return -EINVAL; 2419815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 24201da177e4SLinus Torvalds if (err) 24211da177e4SLinus Torvalds return err; 24221da177e4SLinus Torvalds 24238823c079SEric W. Biederman err = -EINVAL; 24244ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 24258823c079SEric W. Biederman goto out; 24268823c079SEric W. Biederman 242784d17192SAl Viro mp = lock_mount(path); 2428a07b2000SAl Viro if (IS_ERR(mp)) { 242984d17192SAl Viro err = PTR_ERR(mp); 24309676f0c6SRam Pai goto out; 2431a07b2000SAl Viro } 24329676f0c6SRam Pai 243384d17192SAl Viro parent = real_mount(path->mnt); 2434e149ed2bSAl Viro if (!check_mnt(parent)) 2435e149ed2bSAl Viro goto out2; 2436e149ed2bSAl Viro 2437a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2438be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2439be34d1a3SDavid Howells err = PTR_ERR(mnt); 2440e9c5d8a5SAndrey Vagin goto out2; 2441be34d1a3SDavid Howells } 2442ccd48bc7SAl Viro 244384d17192SAl Viro err = graft_tree(mnt, parent, mp); 24441da177e4SLinus Torvalds if (err) { 2445719ea2fbSAl Viro lock_mount_hash(); 2446e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2447719ea2fbSAl Viro unlock_mount_hash(); 24485b83d2c5SRam Pai } 2449b12cea91SAl Viro out2: 245084d17192SAl Viro unlock_mount(mp); 2451ccd48bc7SAl Viro out: 24522d92ab3cSAl Viro path_put(&old_path); 24531da177e4SLinus Torvalds return err; 24541da177e4SLinus Torvalds } 24551da177e4SLinus Torvalds 2456a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2457a07b2000SAl Viro { 2458a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2459a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2460a07b2000SAl Viro struct mount *mnt, *p; 2461a07b2000SAl Viro struct file *file; 2462a07b2000SAl Viro 2463a07b2000SAl Viro if (IS_ERR(ns)) 2464a07b2000SAl Viro return ERR_CAST(ns); 2465a07b2000SAl Viro 2466a07b2000SAl Viro namespace_lock(); 2467a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2468a07b2000SAl Viro if (IS_ERR(mnt)) { 2469a07b2000SAl Viro namespace_unlock(); 2470a07b2000SAl Viro free_mnt_ns(ns); 2471a07b2000SAl Viro return ERR_CAST(mnt); 2472a07b2000SAl Viro } 2473a07b2000SAl Viro 2474a07b2000SAl Viro lock_mount_hash(); 2475a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2476a07b2000SAl Viro p->mnt_ns = ns; 2477a07b2000SAl Viro ns->mounts++; 2478a07b2000SAl Viro } 2479a07b2000SAl Viro ns->root = mnt; 2480a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2481a07b2000SAl Viro mntget(&mnt->mnt); 2482a07b2000SAl Viro unlock_mount_hash(); 2483a07b2000SAl Viro namespace_unlock(); 2484a07b2000SAl Viro 2485a07b2000SAl Viro mntput(path->mnt); 2486a07b2000SAl Viro path->mnt = &mnt->mnt; 2487a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2488a07b2000SAl Viro if (IS_ERR(file)) 2489a07b2000SAl Viro dissolve_on_fput(path->mnt); 2490a07b2000SAl Viro else 2491a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2492a07b2000SAl Viro return file; 2493a07b2000SAl Viro } 2494a07b2000SAl Viro 24952658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2496a07b2000SAl Viro { 2497a07b2000SAl Viro struct file *file; 2498a07b2000SAl Viro struct path path; 2499a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2500a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2501a07b2000SAl Viro int error; 2502a07b2000SAl Viro int fd; 2503a07b2000SAl Viro 2504a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2505a07b2000SAl Viro 2506a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2507a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2508a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2509a07b2000SAl Viro return -EINVAL; 2510a07b2000SAl Viro 2511a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2512a07b2000SAl Viro return -EINVAL; 2513a07b2000SAl Viro 2514a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2515a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2516a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2517a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2518a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2519a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2520a07b2000SAl Viro 2521a07b2000SAl Viro if (detached && !may_mount()) 2522a07b2000SAl Viro return -EPERM; 2523a07b2000SAl Viro 2524a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2525a07b2000SAl Viro if (fd < 0) 2526a07b2000SAl Viro return fd; 2527a07b2000SAl Viro 2528a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2529a07b2000SAl Viro if (unlikely(error)) { 2530a07b2000SAl Viro file = ERR_PTR(error); 2531a07b2000SAl Viro } else { 2532a07b2000SAl Viro if (detached) 2533a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2534a07b2000SAl Viro else 2535a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2536a07b2000SAl Viro path_put(&path); 2537a07b2000SAl Viro } 2538a07b2000SAl Viro if (IS_ERR(file)) { 2539a07b2000SAl Viro put_unused_fd(fd); 2540a07b2000SAl Viro return PTR_ERR(file); 2541a07b2000SAl Viro } 2542a07b2000SAl Viro fd_install(fd, file); 2543a07b2000SAl Viro return fd; 2544a07b2000SAl Viro } 2545a07b2000SAl Viro 254643f5e655SDavid Howells /* 254743f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 254843f5e655SDavid Howells * 254943f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 255043f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 255143f5e655SDavid Howells */ 255243f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 25532e4b7fcdSDave Hansen { 255443f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 25552e4b7fcdSDave Hansen 255643f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 255743f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 255843f5e655SDavid Howells return false; 255943f5e655SDavid Howells 256043f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 256143f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 256243f5e655SDavid Howells return false; 256343f5e655SDavid Howells 256443f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 256543f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 256643f5e655SDavid Howells return false; 256743f5e655SDavid Howells 256843f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 256943f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 257043f5e655SDavid Howells return false; 257143f5e655SDavid Howells 257243f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 257343f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 257443f5e655SDavid Howells return false; 257543f5e655SDavid Howells 257643f5e655SDavid Howells return true; 257743f5e655SDavid Howells } 257843f5e655SDavid Howells 257943f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 258043f5e655SDavid Howells { 258143f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 258243f5e655SDavid Howells 258343f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 25842e4b7fcdSDave Hansen return 0; 25852e4b7fcdSDave Hansen 25862e4b7fcdSDave Hansen if (readonly_request) 258743f5e655SDavid Howells return mnt_make_readonly(mnt); 258843f5e655SDavid Howells 258968847c94SChristian Brauner mnt->mnt.mnt_flags &= ~MNT_READONLY; 259068847c94SChristian Brauner return 0; 259143f5e655SDavid Howells } 259243f5e655SDavid Howells 259343f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 259443f5e655SDavid Howells { 259543f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 259643f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 259743f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 259843f5e655SDavid Howells } 259943f5e655SDavid Howells 2600f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2601f8b92ba6SDeepa Dinamani { 2602f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2603f8b92ba6SDeepa Dinamani 2604f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2605a128b054SAnthony Iliopoulos (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) && 2606f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2607f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2608f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2609f8b92ba6SDeepa Dinamani struct tm tm; 2610f8b92ba6SDeepa Dinamani 2611f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2612f8b92ba6SDeepa Dinamani 26130ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 26140ecee669SEric Biggers sb->s_type->name, 26150ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 26160ecee669SEric Biggers mntpath, 2617f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2618f8b92ba6SDeepa Dinamani 2619f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2620a128b054SAnthony Iliopoulos sb->s_iflags |= SB_I_TS_EXPIRY_WARNED; 2621f8b92ba6SDeepa Dinamani } 2622f8b92ba6SDeepa Dinamani } 2623f8b92ba6SDeepa Dinamani 262443f5e655SDavid Howells /* 262543f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 262643f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 262743f5e655SDavid Howells * to mount(2). 262843f5e655SDavid Howells */ 262943f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 263043f5e655SDavid Howells { 263143f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 263243f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 263343f5e655SDavid Howells int ret; 263443f5e655SDavid Howells 263543f5e655SDavid Howells if (!check_mnt(mnt)) 263643f5e655SDavid Howells return -EINVAL; 263743f5e655SDavid Howells 263843f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 263943f5e655SDavid Howells return -EINVAL; 264043f5e655SDavid Howells 264143f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 264243f5e655SDavid Howells return -EPERM; 264343f5e655SDavid Howells 2644e58ace1aSChristian Brauner /* 2645e58ace1aSChristian Brauner * We're only checking whether the superblock is read-only not 2646e58ace1aSChristian Brauner * changing it, so only take down_read(&sb->s_umount). 2647e58ace1aSChristian Brauner */ 2648e58ace1aSChristian Brauner down_read(&sb->s_umount); 264968847c94SChristian Brauner lock_mount_hash(); 265043f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 265143f5e655SDavid Howells if (ret == 0) 265243f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 265368847c94SChristian Brauner unlock_mount_hash(); 2654e58ace1aSChristian Brauner up_read(&sb->s_umount); 2655f8b92ba6SDeepa Dinamani 2656f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2657f8b92ba6SDeepa Dinamani 265843f5e655SDavid Howells return ret; 26592e4b7fcdSDave Hansen } 26602e4b7fcdSDave Hansen 26611da177e4SLinus Torvalds /* 26621da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 26631da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 26641da177e4SLinus Torvalds * on it - tough luck. 26651da177e4SLinus Torvalds */ 2666e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2667e462ec50SDavid Howells int mnt_flags, void *data) 26681da177e4SLinus Torvalds { 26691da177e4SLinus Torvalds int err; 26702d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2671143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 26728d0347f6SDavid Howells struct fs_context *fc; 26731da177e4SLinus Torvalds 2674143c8c91SAl Viro if (!check_mnt(mnt)) 26751da177e4SLinus Torvalds return -EINVAL; 26761da177e4SLinus Torvalds 26772d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 26781da177e4SLinus Torvalds return -EINVAL; 26791da177e4SLinus Torvalds 268043f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 268107b64558SEric W. Biederman return -EPERM; 26829566d674SEric W. Biederman 26838d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 26848d0347f6SDavid Howells if (IS_ERR(fc)) 26858d0347f6SDavid Howells return PTR_ERR(fc); 2686ff36fe2cSEric Paris 2687b330966fSMiklos Szeredi fc->oldapi = true; 26888d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 26898d0347f6SDavid Howells if (!err) { 26901da177e4SLinus Torvalds down_write(&sb->s_umount); 269157eccb83SAl Viro err = -EPERM; 269243f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 26938d0347f6SDavid Howells err = reconfigure_super(fc); 269468847c94SChristian Brauner if (!err) { 269568847c94SChristian Brauner lock_mount_hash(); 269643f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 269768847c94SChristian Brauner unlock_mount_hash(); 269868847c94SChristian Brauner } 26990e55a7ccSDan Williams } 27006339dab8SAl Viro up_write(&sb->s_umount); 27018d0347f6SDavid Howells } 2702f8b92ba6SDeepa Dinamani 2703f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2704f8b92ba6SDeepa Dinamani 27058d0347f6SDavid Howells put_fs_context(fc); 27061da177e4SLinus Torvalds return err; 27071da177e4SLinus Torvalds } 27081da177e4SLinus Torvalds 2709cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 27109676f0c6SRam Pai { 2711315fc83eSAl Viro struct mount *p; 2712909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2713fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 27149676f0c6SRam Pai return 1; 27159676f0c6SRam Pai } 27169676f0c6SRam Pai return 0; 27179676f0c6SRam Pai } 27189676f0c6SRam Pai 271944dfd84aSDavid Howells /* 272044dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 272144dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 272244dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 272344dfd84aSDavid Howells * cycles to be made. 272444dfd84aSDavid Howells */ 272544dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 272644dfd84aSDavid Howells { 272744dfd84aSDavid Howells struct mount *p; 272844dfd84aSDavid Howells bool ret = false; 272944dfd84aSDavid Howells 273044dfd84aSDavid Howells lock_mount_hash(); 273144dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 273244dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 273344dfd84aSDavid Howells goto out; 273444dfd84aSDavid Howells 273544dfd84aSDavid Howells ret = true; 273644dfd84aSDavid Howells out: 273744dfd84aSDavid Howells unlock_mount_hash(); 273844dfd84aSDavid Howells return ret; 273944dfd84aSDavid Howells } 274044dfd84aSDavid Howells 27419ffb14efSPavel Tikhomirov static int do_set_group(struct path *from_path, struct path *to_path) 27429ffb14efSPavel Tikhomirov { 27439ffb14efSPavel Tikhomirov struct mount *from, *to; 27449ffb14efSPavel Tikhomirov int err; 27459ffb14efSPavel Tikhomirov 27469ffb14efSPavel Tikhomirov from = real_mount(from_path->mnt); 27479ffb14efSPavel Tikhomirov to = real_mount(to_path->mnt); 27489ffb14efSPavel Tikhomirov 27499ffb14efSPavel Tikhomirov namespace_lock(); 27509ffb14efSPavel Tikhomirov 27519ffb14efSPavel Tikhomirov err = -EINVAL; 27529ffb14efSPavel Tikhomirov /* To and From must be mounted */ 27539ffb14efSPavel Tikhomirov if (!is_mounted(&from->mnt)) 27549ffb14efSPavel Tikhomirov goto out; 27559ffb14efSPavel Tikhomirov if (!is_mounted(&to->mnt)) 27569ffb14efSPavel Tikhomirov goto out; 27579ffb14efSPavel Tikhomirov 27589ffb14efSPavel Tikhomirov err = -EPERM; 27599ffb14efSPavel Tikhomirov /* We should be allowed to modify mount namespaces of both mounts */ 27609ffb14efSPavel Tikhomirov if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27619ffb14efSPavel Tikhomirov goto out; 27629ffb14efSPavel Tikhomirov if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27639ffb14efSPavel Tikhomirov goto out; 27649ffb14efSPavel Tikhomirov 27659ffb14efSPavel Tikhomirov err = -EINVAL; 27669ffb14efSPavel Tikhomirov /* To and From paths should be mount roots */ 27679ffb14efSPavel Tikhomirov if (from_path->dentry != from_path->mnt->mnt_root) 27689ffb14efSPavel Tikhomirov goto out; 27699ffb14efSPavel Tikhomirov if (to_path->dentry != to_path->mnt->mnt_root) 27709ffb14efSPavel Tikhomirov goto out; 27719ffb14efSPavel Tikhomirov 27729ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed across same superblock */ 27739ffb14efSPavel Tikhomirov if (from->mnt.mnt_sb != to->mnt.mnt_sb) 27749ffb14efSPavel Tikhomirov goto out; 27759ffb14efSPavel Tikhomirov 27769ffb14efSPavel Tikhomirov /* From mount root should be wider than To mount root */ 27779ffb14efSPavel Tikhomirov if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) 27789ffb14efSPavel Tikhomirov goto out; 27799ffb14efSPavel Tikhomirov 27809ffb14efSPavel Tikhomirov /* From mount should not have locked children in place of To's root */ 27819ffb14efSPavel Tikhomirov if (has_locked_children(from, to->mnt.mnt_root)) 27829ffb14efSPavel Tikhomirov goto out; 27839ffb14efSPavel Tikhomirov 27849ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed on private mounts */ 27859ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) 27869ffb14efSPavel Tikhomirov goto out; 27879ffb14efSPavel Tikhomirov 27889ffb14efSPavel Tikhomirov /* From should not be private */ 27899ffb14efSPavel Tikhomirov if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) 27909ffb14efSPavel Tikhomirov goto out; 27919ffb14efSPavel Tikhomirov 27929ffb14efSPavel Tikhomirov if (IS_MNT_SLAVE(from)) { 27939ffb14efSPavel Tikhomirov struct mount *m = from->mnt_master; 27949ffb14efSPavel Tikhomirov 27959ffb14efSPavel Tikhomirov list_add(&to->mnt_slave, &m->mnt_slave_list); 27969ffb14efSPavel Tikhomirov to->mnt_master = m; 27979ffb14efSPavel Tikhomirov } 27989ffb14efSPavel Tikhomirov 27999ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(from)) { 28009ffb14efSPavel Tikhomirov to->mnt_group_id = from->mnt_group_id; 28019ffb14efSPavel Tikhomirov list_add(&to->mnt_share, &from->mnt_share); 28029ffb14efSPavel Tikhomirov lock_mount_hash(); 28039ffb14efSPavel Tikhomirov set_mnt_shared(to); 28049ffb14efSPavel Tikhomirov unlock_mount_hash(); 28059ffb14efSPavel Tikhomirov } 28069ffb14efSPavel Tikhomirov 28079ffb14efSPavel Tikhomirov err = 0; 28089ffb14efSPavel Tikhomirov out: 28099ffb14efSPavel Tikhomirov namespace_unlock(); 28109ffb14efSPavel Tikhomirov return err; 28119ffb14efSPavel Tikhomirov } 28129ffb14efSPavel Tikhomirov 28132db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 28141da177e4SLinus Torvalds { 281544dfd84aSDavid Howells struct mnt_namespace *ns; 2816676da58dSAl Viro struct mount *p; 28170fb54e50SAl Viro struct mount *old; 28182763d119SAl Viro struct mount *parent; 28192763d119SAl Viro struct mountpoint *mp, *old_mp; 282057eccb83SAl Viro int err; 282144dfd84aSDavid Howells bool attached; 28221da177e4SLinus Torvalds 28232db154b3SDavid Howells mp = lock_mount(new_path); 282484d17192SAl Viro if (IS_ERR(mp)) 28252db154b3SDavid Howells return PTR_ERR(mp); 2826cc53ce53SDavid Howells 28272db154b3SDavid Howells old = real_mount(old_path->mnt); 28282db154b3SDavid Howells p = real_mount(new_path->mnt); 28292763d119SAl Viro parent = old->mnt_parent; 283044dfd84aSDavid Howells attached = mnt_has_parent(old); 28312763d119SAl Viro old_mp = old->mnt_mp; 283244dfd84aSDavid Howells ns = old->mnt_ns; 2833143c8c91SAl Viro 28341da177e4SLinus Torvalds err = -EINVAL; 283544dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 283644dfd84aSDavid Howells if (!check_mnt(p)) 28372db154b3SDavid Howells goto out; 28381da177e4SLinus Torvalds 2839570d7a98SEric Biggers /* The thing moved must be mounted... */ 2840570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 284144dfd84aSDavid Howells goto out; 284244dfd84aSDavid Howells 2843570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2844570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 28452db154b3SDavid Howells goto out; 28461da177e4SLinus Torvalds 28472db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 28482db154b3SDavid Howells goto out; 28492db154b3SDavid Howells 28502db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 28512db154b3SDavid Howells goto out; 28522db154b3SDavid Howells 28532db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 28542db154b3SDavid Howells d_is_dir(old_path->dentry)) 28552db154b3SDavid Howells goto out; 285621444403SRam Pai /* 285721444403SRam Pai * Don't move a mount residing in a shared parent. 285821444403SRam Pai */ 28592763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 28602db154b3SDavid Howells goto out; 28619676f0c6SRam Pai /* 28629676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 28639676f0c6SRam Pai * mount which is shared. 28649676f0c6SRam Pai */ 2865fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 28662db154b3SDavid Howells goto out; 28671da177e4SLinus Torvalds err = -ELOOP; 286844dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 286944dfd84aSDavid Howells goto out; 2870fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2871676da58dSAl Viro if (p == old) 28722db154b3SDavid Howells goto out; 28731da177e4SLinus Torvalds 28742db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 28752763d119SAl Viro attached); 28764ac91378SJan Blunck if (err) 28772db154b3SDavid Howells goto out; 28781da177e4SLinus Torvalds 28791da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 28801da177e4SLinus Torvalds * automatically */ 28816776db3dSAl Viro list_del_init(&old->mnt_expire); 28822763d119SAl Viro if (attached) 28832763d119SAl Viro put_mountpoint(old_mp); 28841da177e4SLinus Torvalds out: 28852db154b3SDavid Howells unlock_mount(mp); 288644dfd84aSDavid Howells if (!err) { 28872763d119SAl Viro if (attached) 28882763d119SAl Viro mntput_no_expire(parent); 28892763d119SAl Viro else 289044dfd84aSDavid Howells free_mnt_ns(ns); 289144dfd84aSDavid Howells } 28922db154b3SDavid Howells return err; 28932db154b3SDavid Howells } 28942db154b3SDavid Howells 28952db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 28962db154b3SDavid Howells { 28972db154b3SDavid Howells struct path old_path; 28982db154b3SDavid Howells int err; 28992db154b3SDavid Howells 29002db154b3SDavid Howells if (!old_name || !*old_name) 29012db154b3SDavid Howells return -EINVAL; 29022db154b3SDavid Howells 29032db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 29042db154b3SDavid Howells if (err) 29052db154b3SDavid Howells return err; 29062db154b3SDavid Howells 29072db154b3SDavid Howells err = do_move_mount(&old_path, path); 29082d92ab3cSAl Viro path_put(&old_path); 29091da177e4SLinus Torvalds return err; 29101da177e4SLinus Torvalds } 29111da177e4SLinus Torvalds 29129d412a43SAl Viro /* 29139d412a43SAl Viro * add a mount into a namespace's mount tree 29149d412a43SAl Viro */ 29158f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 29161e2d8464SAl Viro const struct path *path, int mnt_flags) 29179d412a43SAl Viro { 29188f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 29199d412a43SAl Viro 2920f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 29219d412a43SAl Viro 292284d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2923156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2924156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 29258f11538eSAl Viro return -EINVAL; 2926156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 292784d17192SAl Viro if (!parent->mnt_ns) 29288f11538eSAl Viro return -EINVAL; 2929156cacb1SAl Viro } 29309d412a43SAl Viro 29319d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 293295bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 29339d412a43SAl Viro path->mnt->mnt_root == path->dentry) 29348f11538eSAl Viro return -EBUSY; 29359d412a43SAl Viro 2936e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 29378f11538eSAl Viro return -EINVAL; 29389d412a43SAl Viro 293995bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 29408f11538eSAl Viro return graft_tree(newmnt, parent, mp); 29419d412a43SAl Viro } 2942b1e75df4SAl Viro 2943132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2944132e4608SDavid Howells 2945132e4608SDavid Howells /* 2946132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2947132e4608SDavid Howells * be added to the namespace tree. 2948132e4608SDavid Howells */ 2949132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2950132e4608SDavid Howells unsigned int mnt_flags) 2951132e4608SDavid Howells { 2952132e4608SDavid Howells struct vfsmount *mnt; 29538f11538eSAl Viro struct mountpoint *mp; 2954132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2955132e4608SDavid Howells int error; 2956132e4608SDavid Howells 2957c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2958c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2959c9ce29edSAl Viro error = -EPERM; 2960c9ce29edSAl Viro 2961c9ce29edSAl Viro if (unlikely(error)) { 2962c9ce29edSAl Viro fc_drop_locked(fc); 2963c9ce29edSAl Viro return error; 2964132e4608SDavid Howells } 2965132e4608SDavid Howells 2966132e4608SDavid Howells up_write(&sb->s_umount); 2967132e4608SDavid Howells 2968132e4608SDavid Howells mnt = vfs_create_mount(fc); 2969132e4608SDavid Howells if (IS_ERR(mnt)) 2970132e4608SDavid Howells return PTR_ERR(mnt); 2971132e4608SDavid Howells 2972f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 2973f8b92ba6SDeepa Dinamani 29748f11538eSAl Viro mp = lock_mount(mountpoint); 29758f11538eSAl Viro if (IS_ERR(mp)) { 29768f11538eSAl Viro mntput(mnt); 29778f11538eSAl Viro return PTR_ERR(mp); 29788f11538eSAl Viro } 29798f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 29808f11538eSAl Viro unlock_mount(mp); 29810ecee669SEric Biggers if (error < 0) 29820ecee669SEric Biggers mntput(mnt); 2983f8b92ba6SDeepa Dinamani return error; 2984f8b92ba6SDeepa Dinamani } 2985f8b92ba6SDeepa Dinamani 29861da177e4SLinus Torvalds /* 29871da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 29881da177e4SLinus Torvalds * namespace's tree 29891da177e4SLinus Torvalds */ 2990e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2991808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 29921da177e4SLinus Torvalds { 29930c55cfc4SEric W. Biederman struct file_system_type *type; 2994a0c9a8b8SAl Viro struct fs_context *fc; 2995a0c9a8b8SAl Viro const char *subtype = NULL; 2996a0c9a8b8SAl Viro int err = 0; 29971da177e4SLinus Torvalds 29980c55cfc4SEric W. Biederman if (!fstype) 29991da177e4SLinus Torvalds return -EINVAL; 30001da177e4SLinus Torvalds 30010c55cfc4SEric W. Biederman type = get_fs_type(fstype); 30020c55cfc4SEric W. Biederman if (!type) 30030c55cfc4SEric W. Biederman return -ENODEV; 30040c55cfc4SEric W. Biederman 3005a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 3006a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 3007a0c9a8b8SAl Viro if (subtype) { 3008a0c9a8b8SAl Viro subtype++; 3009a0c9a8b8SAl Viro if (!*subtype) { 30100c55cfc4SEric W. Biederman put_filesystem(type); 3011a0c9a8b8SAl Viro return -EINVAL; 3012a0c9a8b8SAl Viro } 3013a0c9a8b8SAl Viro } 30148654df4eSEric W. Biederman } 30158654df4eSEric W. Biederman 3016a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 3017a0c9a8b8SAl Viro put_filesystem(type); 3018a0c9a8b8SAl Viro if (IS_ERR(fc)) 3019a0c9a8b8SAl Viro return PTR_ERR(fc); 3020a0c9a8b8SAl Viro 30213e1aeb00SDavid Howells if (subtype) 30223e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 30233e1aeb00SDavid Howells subtype, strlen(subtype)); 30243e1aeb00SDavid Howells if (!err && name) 30253e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 3026a0c9a8b8SAl Viro if (!err) 3027a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 3028c3aabf07SAl Viro if (!err && !mount_capable(fc)) 3029c3aabf07SAl Viro err = -EPERM; 3030a0c9a8b8SAl Viro if (!err) 3031a0c9a8b8SAl Viro err = vfs_get_tree(fc); 3032132e4608SDavid Howells if (!err) 3033132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 3034a0c9a8b8SAl Viro 3035a0c9a8b8SAl Viro put_fs_context(fc); 303615f9a3f3SAl Viro return err; 30371da177e4SLinus Torvalds } 30381da177e4SLinus Torvalds 30391e2d8464SAl Viro int finish_automount(struct vfsmount *m, const struct path *path) 304019a167afSAl Viro { 304126df6034SAl Viro struct dentry *dentry = path->dentry; 30428f11538eSAl Viro struct mountpoint *mp; 304325e195aaSAl Viro struct mount *mnt; 304419a167afSAl Viro int err; 304525e195aaSAl Viro 304625e195aaSAl Viro if (!m) 304725e195aaSAl Viro return 0; 304825e195aaSAl Viro if (IS_ERR(m)) 304925e195aaSAl Viro return PTR_ERR(m); 305025e195aaSAl Viro 305125e195aaSAl Viro mnt = real_mount(m); 305219a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 305319a167afSAl Viro * expired before we get a chance to add it 305419a167afSAl Viro */ 30556776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 305619a167afSAl Viro 305719a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 305826df6034SAl Viro m->mnt_root == dentry) { 3059b1e75df4SAl Viro err = -ELOOP; 306026df6034SAl Viro goto discard; 306119a167afSAl Viro } 306219a167afSAl Viro 306326df6034SAl Viro /* 306426df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 306526df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 306626df6034SAl Viro * got", not "try to mount it on top". 306726df6034SAl Viro */ 306826df6034SAl Viro inode_lock(dentry->d_inode); 306926df6034SAl Viro namespace_lock(); 307026df6034SAl Viro if (unlikely(cant_mount(dentry))) { 307126df6034SAl Viro err = -ENOENT; 307226df6034SAl Viro goto discard_locked; 307326df6034SAl Viro } 307426df6034SAl Viro rcu_read_lock(); 307526df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 307626df6034SAl Viro rcu_read_unlock(); 307726df6034SAl Viro err = 0; 307826df6034SAl Viro goto discard_locked; 307926df6034SAl Viro } 308026df6034SAl Viro rcu_read_unlock(); 308126df6034SAl Viro mp = get_mountpoint(dentry); 30828f11538eSAl Viro if (IS_ERR(mp)) { 30838f11538eSAl Viro err = PTR_ERR(mp); 308426df6034SAl Viro goto discard_locked; 30858f11538eSAl Viro } 308626df6034SAl Viro 30878f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 30888f11538eSAl Viro unlock_mount(mp); 308926df6034SAl Viro if (unlikely(err)) 309026df6034SAl Viro goto discard; 309126df6034SAl Viro mntput(m); 3092b1e75df4SAl Viro return 0; 309326df6034SAl Viro 309426df6034SAl Viro discard_locked: 309526df6034SAl Viro namespace_unlock(); 309626df6034SAl Viro inode_unlock(dentry->d_inode); 309726df6034SAl Viro discard: 3098b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 30996776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 310097216be0SAl Viro namespace_lock(); 31016776db3dSAl Viro list_del_init(&mnt->mnt_expire); 310297216be0SAl Viro namespace_unlock(); 310319a167afSAl Viro } 3104b1e75df4SAl Viro mntput(m); 3105b1e75df4SAl Viro mntput(m); 310619a167afSAl Viro return err; 310719a167afSAl Viro } 310819a167afSAl Viro 3109ea5b778aSDavid Howells /** 3110ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 3111ea5b778aSDavid Howells * @mnt: The mount to list. 3112ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 3113ea5b778aSDavid Howells */ 3114ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 3115ea5b778aSDavid Howells { 311697216be0SAl Viro namespace_lock(); 3117ea5b778aSDavid Howells 31186776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 3119ea5b778aSDavid Howells 312097216be0SAl Viro namespace_unlock(); 3121ea5b778aSDavid Howells } 3122ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 3123ea5b778aSDavid Howells 3124ea5b778aSDavid Howells /* 31251da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 31261da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 31271da177e4SLinus Torvalds * here 31281da177e4SLinus Torvalds */ 31291da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 31301da177e4SLinus Torvalds { 3131761d5c38SAl Viro struct mount *mnt, *next; 31321da177e4SLinus Torvalds LIST_HEAD(graveyard); 31331da177e4SLinus Torvalds 31341da177e4SLinus Torvalds if (list_empty(mounts)) 31351da177e4SLinus Torvalds return; 31361da177e4SLinus Torvalds 313797216be0SAl Viro namespace_lock(); 3138719ea2fbSAl Viro lock_mount_hash(); 31391da177e4SLinus Torvalds 31401da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 31411da177e4SLinus Torvalds * following criteria: 31421da177e4SLinus Torvalds * - only referenced by its parent vfsmount 31431da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 31441da177e4SLinus Torvalds * cleared by mntput()) 31451da177e4SLinus Torvalds */ 31466776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 3147863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 31481ab59738SAl Viro propagate_mount_busy(mnt, 1)) 31491da177e4SLinus Torvalds continue; 31506776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 31511da177e4SLinus Torvalds } 3152bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 31536776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 3154143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 3155e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3156bcc5c7d2SAl Viro } 3157719ea2fbSAl Viro unlock_mount_hash(); 31583ab6abeeSAl Viro namespace_unlock(); 31591da177e4SLinus Torvalds } 31601da177e4SLinus Torvalds 31611da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 31621da177e4SLinus Torvalds 31631da177e4SLinus Torvalds /* 31645528f911STrond Myklebust * Ripoff of 'select_parent()' 31655528f911STrond Myklebust * 31665528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 31675528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 31685528f911STrond Myklebust */ 3169692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 31705528f911STrond Myklebust { 3171692afc31SAl Viro struct mount *this_parent = parent; 31725528f911STrond Myklebust struct list_head *next; 31735528f911STrond Myklebust int found = 0; 31745528f911STrond Myklebust 31755528f911STrond Myklebust repeat: 31766b41d536SAl Viro next = this_parent->mnt_mounts.next; 31775528f911STrond Myklebust resume: 31786b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 31795528f911STrond Myklebust struct list_head *tmp = next; 31806b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 31815528f911STrond Myklebust 31825528f911STrond Myklebust next = tmp->next; 3183692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 31845528f911STrond Myklebust continue; 31855528f911STrond Myklebust /* 31865528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 31875528f911STrond Myklebust */ 31886b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 31895528f911STrond Myklebust this_parent = mnt; 31905528f911STrond Myklebust goto repeat; 31915528f911STrond Myklebust } 31925528f911STrond Myklebust 31931ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 31946776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 31955528f911STrond Myklebust found++; 31965528f911STrond Myklebust } 31975528f911STrond Myklebust } 31985528f911STrond Myklebust /* 31995528f911STrond Myklebust * All done at this level ... ascend and resume the search 32005528f911STrond Myklebust */ 32015528f911STrond Myklebust if (this_parent != parent) { 32026b41d536SAl Viro next = this_parent->mnt_child.next; 32030714a533SAl Viro this_parent = this_parent->mnt_parent; 32045528f911STrond Myklebust goto resume; 32055528f911STrond Myklebust } 32065528f911STrond Myklebust return found; 32075528f911STrond Myklebust } 32085528f911STrond Myklebust 32095528f911STrond Myklebust /* 32105528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 32115528f911STrond Myklebust * submounts of a specific parent mountpoint 321299b7db7bSNick Piggin * 321348a066e7SAl Viro * mount_lock must be held for write 32145528f911STrond Myklebust */ 3215b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 32165528f911STrond Myklebust { 32175528f911STrond Myklebust LIST_HEAD(graveyard); 3218761d5c38SAl Viro struct mount *m; 32195528f911STrond Myklebust 32205528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 3221c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3222bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3223761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 32246776db3dSAl Viro mnt_expire); 3225143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3226e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3227bcc5c7d2SAl Viro } 3228bcc5c7d2SAl Viro } 32295528f911STrond Myklebust } 32305528f911STrond Myklebust 3231028abd92SChristoph Hellwig static void *copy_mount_options(const void __user * data) 32321da177e4SLinus Torvalds { 3233b40ef869SAl Viro char *copy; 3234d563d678SCatalin Marinas unsigned left, offset; 32351da177e4SLinus Torvalds 32361da177e4SLinus Torvalds if (!data) 3237b40ef869SAl Viro return NULL; 32381da177e4SLinus Torvalds 3239b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3240b40ef869SAl Viro if (!copy) 3241b40ef869SAl Viro return ERR_PTR(-ENOMEM); 32421da177e4SLinus Torvalds 3243d563d678SCatalin Marinas left = copy_from_user(copy, data, PAGE_SIZE); 32441da177e4SLinus Torvalds 3245d563d678SCatalin Marinas /* 3246d563d678SCatalin Marinas * Not all architectures have an exact copy_from_user(). Resort to 3247d563d678SCatalin Marinas * byte at a time. 3248d563d678SCatalin Marinas */ 3249d563d678SCatalin Marinas offset = PAGE_SIZE - left; 3250d563d678SCatalin Marinas while (left) { 3251d563d678SCatalin Marinas char c; 3252d563d678SCatalin Marinas if (get_user(c, (const char __user *)data + offset)) 3253d563d678SCatalin Marinas break; 3254d563d678SCatalin Marinas copy[offset] = c; 3255d563d678SCatalin Marinas left--; 3256d563d678SCatalin Marinas offset++; 3257d563d678SCatalin Marinas } 3258d563d678SCatalin Marinas 3259d563d678SCatalin Marinas if (left == PAGE_SIZE) { 3260b40ef869SAl Viro kfree(copy); 3261b40ef869SAl Viro return ERR_PTR(-EFAULT); 32621da177e4SLinus Torvalds } 3263d563d678SCatalin Marinas 3264b40ef869SAl Viro return copy; 32651da177e4SLinus Torvalds } 32661da177e4SLinus Torvalds 3267028abd92SChristoph Hellwig static char *copy_mount_string(const void __user *data) 3268eca6f534SVegard Nossum { 3269fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3270eca6f534SVegard Nossum } 3271eca6f534SVegard Nossum 32721da177e4SLinus Torvalds /* 32731da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 32741da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 32751da177e4SLinus Torvalds * 32761da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 32771da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 32781da177e4SLinus Torvalds * information (or be NULL). 32791da177e4SLinus Torvalds * 32801da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 32811da177e4SLinus Torvalds * When the flags word was introduced its top half was required 32821da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 32831da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 32841da177e4SLinus Torvalds * and must be discarded. 32851da177e4SLinus Torvalds */ 3286c60166f0SChristoph Hellwig int path_mount(const char *dev_name, struct path *path, 3287808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 32881da177e4SLinus Torvalds { 3289e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 3290a1e6aaa3SChristoph Hellwig int ret; 32911da177e4SLinus Torvalds 32921da177e4SLinus Torvalds /* Discard magic */ 32931da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 32941da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 32951da177e4SLinus Torvalds 32961da177e4SLinus Torvalds /* Basic sanity checks */ 32971da177e4SLinus Torvalds if (data_page) 32981da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 32991da177e4SLinus Torvalds 3300e462ec50SDavid Howells if (flags & MS_NOUSER) 3301e462ec50SDavid Howells return -EINVAL; 3302e462ec50SDavid Howells 3303a1e6aaa3SChristoph Hellwig ret = security_sb_mount(dev_name, path, type_page, flags, data_page); 3304a1e6aaa3SChristoph Hellwig if (ret) 3305a1e6aaa3SChristoph Hellwig return ret; 3306a1e6aaa3SChristoph Hellwig if (!may_mount()) 3307a1e6aaa3SChristoph Hellwig return -EPERM; 3308f7e33bdbSJeff Layton if (flags & SB_MANDLOCK) 3309f7e33bdbSJeff Layton warn_mandlock(); 3310a27ab9f2STetsuo Handa 3311613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3312613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 33130a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 33140a1c01c9SMatthew Garrett 33151da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 33161da177e4SLinus Torvalds if (flags & MS_NOSUID) 33171da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 33181da177e4SLinus Torvalds if (flags & MS_NODEV) 33191da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 33201da177e4SLinus Torvalds if (flags & MS_NOEXEC) 33211da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3322fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3323fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3324fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3325fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3326d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3327d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3328a9e5b732SDavid Howells if (flags & MS_RDONLY) 33292e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3330dab741e0SMattias Nissler if (flags & MS_NOSYMFOLLOW) 3331dab741e0SMattias Nissler mnt_flags |= MNT_NOSYMFOLLOW; 3332fc33a7bbSChristoph Hellwig 3333ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3334ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3335ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3336ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3337ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3338a1e6aaa3SChristoph Hellwig mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; 3339ffbc6f0eSEric W. Biederman } 3340ffbc6f0eSEric W. Biederman 3341e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3342e462ec50SDavid Howells SB_SYNCHRONOUS | 3343e462ec50SDavid Howells SB_MANDLOCK | 3344e462ec50SDavid Howells SB_DIRSYNC | 3345e462ec50SDavid Howells SB_SILENT | 3346917086ffSMimi Zohar SB_POSIXACL | 3347d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3348917086ffSMimi Zohar SB_I_VERSION); 33491da177e4SLinus Torvalds 335043f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 3351a1e6aaa3SChristoph Hellwig return do_reconfigure_mnt(path, mnt_flags); 3352a1e6aaa3SChristoph Hellwig if (flags & MS_REMOUNT) 3353a1e6aaa3SChristoph Hellwig return do_remount(path, flags, sb_flags, mnt_flags, data_page); 3354a1e6aaa3SChristoph Hellwig if (flags & MS_BIND) 3355a1e6aaa3SChristoph Hellwig return do_loopback(path, dev_name, flags & MS_REC); 3356a1e6aaa3SChristoph Hellwig if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 3357a1e6aaa3SChristoph Hellwig return do_change_type(path, flags); 3358a1e6aaa3SChristoph Hellwig if (flags & MS_MOVE) 3359a1e6aaa3SChristoph Hellwig return do_move_mount_old(path, dev_name); 3360a1e6aaa3SChristoph Hellwig 3361a1e6aaa3SChristoph Hellwig return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, 33621da177e4SLinus Torvalds data_page); 3363a1e6aaa3SChristoph Hellwig } 3364a1e6aaa3SChristoph Hellwig 3365a1e6aaa3SChristoph Hellwig long do_mount(const char *dev_name, const char __user *dir_name, 3366a1e6aaa3SChristoph Hellwig const char *type_page, unsigned long flags, void *data_page) 3367a1e6aaa3SChristoph Hellwig { 3368a1e6aaa3SChristoph Hellwig struct path path; 3369a1e6aaa3SChristoph Hellwig int ret; 3370a1e6aaa3SChristoph Hellwig 3371a1e6aaa3SChristoph Hellwig ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3372a1e6aaa3SChristoph Hellwig if (ret) 3373a1e6aaa3SChristoph Hellwig return ret; 3374a1e6aaa3SChristoph Hellwig ret = path_mount(dev_name, &path, type_page, flags, data_page); 33752d92ab3cSAl Viro path_put(&path); 3376a1e6aaa3SChristoph Hellwig return ret; 33771da177e4SLinus Torvalds } 33781da177e4SLinus Torvalds 3379537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3380537f7ccbSEric W. Biederman { 3381537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3382537f7ccbSEric W. Biederman } 3383537f7ccbSEric W. Biederman 3384537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3385537f7ccbSEric W. Biederman { 3386537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3387537f7ccbSEric W. Biederman } 3388537f7ccbSEric W. Biederman 3389771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3390771b1371SEric W. Biederman { 339174e83122SAl Viro if (!is_anon_ns(ns)) 33926344c433SAl Viro ns_free_inum(&ns->ns); 3393537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3394771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3395771b1371SEric W. Biederman kfree(ns); 3396771b1371SEric W. Biederman } 3397771b1371SEric W. Biederman 33988823c079SEric W. Biederman /* 33998823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 34008823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 34018823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 34028823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 34038823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 34048823c079SEric W. Biederman */ 34058823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 34068823c079SEric W. Biederman 340774e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3408cf8d2c11STrond Myklebust { 3409cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3410537f7ccbSEric W. Biederman struct ucounts *ucounts; 341198f842e6SEric W. Biederman int ret; 3412cf8d2c11STrond Myklebust 3413537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3414537f7ccbSEric W. Biederman if (!ucounts) 3415df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3416537f7ccbSEric W. Biederman 341730acd0bdSVasily Averin new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); 3418537f7ccbSEric W. Biederman if (!new_ns) { 3419537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3420cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3421537f7ccbSEric W. Biederman } 342274e83122SAl Viro if (!anon) { 34236344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 342498f842e6SEric W. Biederman if (ret) { 342598f842e6SEric W. Biederman kfree(new_ns); 3426537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 342798f842e6SEric W. Biederman return ERR_PTR(ret); 342898f842e6SEric W. Biederman } 342974e83122SAl Viro } 343033c42940SAl Viro new_ns->ns.ops = &mntns_operations; 343174e83122SAl Viro if (!anon) 34328823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 34331a7b8969SKirill Tkhai refcount_set(&new_ns->ns.count, 1); 3434cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3435cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 34369f6c61f9SMiklos Szeredi spin_lock_init(&new_ns->ns_lock); 3437771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3438537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3439cf8d2c11STrond Myklebust return new_ns; 3440cf8d2c11STrond Myklebust } 3441cf8d2c11STrond Myklebust 34420766f788SEmese Revfy __latent_entropy 34439559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 34449559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 34451da177e4SLinus Torvalds { 34466b3286edSKirill Korotaev struct mnt_namespace *new_ns; 34477f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3448315fc83eSAl Viro struct mount *p, *q; 34499559f689SAl Viro struct mount *old; 3450cb338d06SAl Viro struct mount *new; 34517a472ef4SEric W. Biederman int copy_flags; 34521da177e4SLinus Torvalds 34539559f689SAl Viro BUG_ON(!ns); 34549559f689SAl Viro 34559559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 34569559f689SAl Viro get_mnt_ns(ns); 34579559f689SAl Viro return ns; 34589559f689SAl Viro } 34599559f689SAl Viro 34609559f689SAl Viro old = ns->root; 34619559f689SAl Viro 346274e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3463cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3464cf8d2c11STrond Myklebust return new_ns; 34651da177e4SLinus Torvalds 346697216be0SAl Viro namespace_lock(); 34671da177e4SLinus Torvalds /* First pass: copy the tree topology */ 34684ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 34699559f689SAl Viro if (user_ns != ns->user_ns) 34703bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 34717a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3472be34d1a3SDavid Howells if (IS_ERR(new)) { 3473328e6d90SAl Viro namespace_unlock(); 3474771b1371SEric W. Biederman free_mnt_ns(new_ns); 3475be34d1a3SDavid Howells return ERR_CAST(new); 34761da177e4SLinus Torvalds } 34773bd045ccSAl Viro if (user_ns != ns->user_ns) { 34783bd045ccSAl Viro lock_mount_hash(); 34793bd045ccSAl Viro lock_mnt_tree(new); 34803bd045ccSAl Viro unlock_mount_hash(); 34813bd045ccSAl Viro } 3482be08d6d2SAl Viro new_ns->root = new; 34831a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 34841da177e4SLinus Torvalds 34851da177e4SLinus Torvalds /* 34861da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 34871da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 34881da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 34891da177e4SLinus Torvalds */ 3490909b0a88SAl Viro p = old; 3491cb338d06SAl Viro q = new; 34921da177e4SLinus Torvalds while (p) { 3493143c8c91SAl Viro q->mnt_ns = new_ns; 3494d2921684SEric W. Biederman new_ns->mounts++; 34959559f689SAl Viro if (new_fs) { 34969559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 34979559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3498315fc83eSAl Viro rootmnt = &p->mnt; 34991da177e4SLinus Torvalds } 35009559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 35019559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3502315fc83eSAl Viro pwdmnt = &p->mnt; 35031da177e4SLinus Torvalds } 35041da177e4SLinus Torvalds } 3505909b0a88SAl Viro p = next_mnt(p, old); 3506909b0a88SAl Viro q = next_mnt(q, new); 35074ce5d2b1SEric W. Biederman if (!q) 35084ce5d2b1SEric W. Biederman break; 350961d8e426SAl Viro // an mntns binding we'd skipped? 35104ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 351161d8e426SAl Viro p = next_mnt(skip_mnt_tree(p), old); 35121da177e4SLinus Torvalds } 3513328e6d90SAl Viro namespace_unlock(); 35141da177e4SLinus Torvalds 35151da177e4SLinus Torvalds if (rootmnt) 3516f03c6599SAl Viro mntput(rootmnt); 35171da177e4SLinus Torvalds if (pwdmnt) 3518f03c6599SAl Viro mntput(pwdmnt); 35191da177e4SLinus Torvalds 3520741a2951SJANAK DESAI return new_ns; 3521741a2951SJANAK DESAI } 3522741a2951SJANAK DESAI 352374e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3524cf8d2c11STrond Myklebust { 35251a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3526ea441d11SAl Viro struct mnt_namespace *ns; 3527d31da0f0SAl Viro struct super_block *s; 3528ea441d11SAl Viro struct path path; 3529ea441d11SAl Viro int err; 3530ea441d11SAl Viro 353174e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 353274e83122SAl Viro if (IS_ERR(ns)) { 353374e83122SAl Viro mntput(m); 3534ea441d11SAl Viro return ERR_CAST(ns); 353574e83122SAl Viro } 353674e83122SAl Viro mnt->mnt_ns = ns; 353774e83122SAl Viro ns->root = mnt; 353874e83122SAl Viro ns->mounts++; 353974e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3540ea441d11SAl Viro 354174e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3542ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3543ea441d11SAl Viro 3544ea441d11SAl Viro put_mnt_ns(ns); 3545ea441d11SAl Viro 3546ea441d11SAl Viro if (err) 3547ea441d11SAl Viro return ERR_PTR(err); 3548ea441d11SAl Viro 3549ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3550d31da0f0SAl Viro s = path.mnt->mnt_sb; 3551d31da0f0SAl Viro atomic_inc(&s->s_active); 3552ea441d11SAl Viro mntput(path.mnt); 3553ea441d11SAl Viro /* lock the sucker */ 3554d31da0f0SAl Viro down_write(&s->s_umount); 3555ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3556ea441d11SAl Viro return path.dentry; 3557ea441d11SAl Viro } 3558ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3559ea441d11SAl Viro 3560cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3561cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 35621da177e4SLinus Torvalds { 3563eca6f534SVegard Nossum int ret; 3564eca6f534SVegard Nossum char *kernel_type; 3565eca6f534SVegard Nossum char *kernel_dev; 3566b40ef869SAl Viro void *options; 35671da177e4SLinus Torvalds 3568b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3569b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3570b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3571eca6f534SVegard Nossum goto out_type; 35721da177e4SLinus Torvalds 3573b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3574b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3575b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3576eca6f534SVegard Nossum goto out_dev; 35771da177e4SLinus Torvalds 3578b40ef869SAl Viro options = copy_mount_options(data); 3579b40ef869SAl Viro ret = PTR_ERR(options); 3580b40ef869SAl Viro if (IS_ERR(options)) 3581eca6f534SVegard Nossum goto out_data; 35821da177e4SLinus Torvalds 3583b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3584eca6f534SVegard Nossum 3585b40ef869SAl Viro kfree(options); 3586eca6f534SVegard Nossum out_data: 3587eca6f534SVegard Nossum kfree(kernel_dev); 3588eca6f534SVegard Nossum out_dev: 3589eca6f534SVegard Nossum kfree(kernel_type); 3590eca6f534SVegard Nossum out_type: 3591eca6f534SVegard Nossum return ret; 35921da177e4SLinus Torvalds } 35931da177e4SLinus Torvalds 35945b490500SChristian Brauner #define FSMOUNT_VALID_FLAGS \ 35955b490500SChristian Brauner (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ 3596dd8b477fSChristian Brauner MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ 3597dd8b477fSChristian Brauner MOUNT_ATTR_NOSYMFOLLOW) 35985b490500SChristian Brauner 35999caccd41SChristian Brauner #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) 36002a186721SChristian Brauner 36012a186721SChristian Brauner #define MOUNT_SETATTR_PROPAGATION_FLAGS \ 36022a186721SChristian Brauner (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) 36032a186721SChristian Brauner 36045b490500SChristian Brauner static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) 36055b490500SChristian Brauner { 36065b490500SChristian Brauner unsigned int mnt_flags = 0; 36075b490500SChristian Brauner 36085b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_RDONLY) 36095b490500SChristian Brauner mnt_flags |= MNT_READONLY; 36105b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOSUID) 36115b490500SChristian Brauner mnt_flags |= MNT_NOSUID; 36125b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODEV) 36135b490500SChristian Brauner mnt_flags |= MNT_NODEV; 36145b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOEXEC) 36155b490500SChristian Brauner mnt_flags |= MNT_NOEXEC; 36165b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODIRATIME) 36175b490500SChristian Brauner mnt_flags |= MNT_NODIRATIME; 3618dd8b477fSChristian Brauner if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) 3619dd8b477fSChristian Brauner mnt_flags |= MNT_NOSYMFOLLOW; 36205b490500SChristian Brauner 36215b490500SChristian Brauner return mnt_flags; 36225b490500SChristian Brauner } 36235b490500SChristian Brauner 36241da177e4SLinus Torvalds /* 362593766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 362693766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 362793766fbdSDavid Howells */ 362893766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 362993766fbdSDavid Howells unsigned int, attr_flags) 363093766fbdSDavid Howells { 363193766fbdSDavid Howells struct mnt_namespace *ns; 363293766fbdSDavid Howells struct fs_context *fc; 363393766fbdSDavid Howells struct file *file; 363493766fbdSDavid Howells struct path newmount; 363593766fbdSDavid Howells struct mount *mnt; 363693766fbdSDavid Howells struct fd f; 363793766fbdSDavid Howells unsigned int mnt_flags = 0; 363893766fbdSDavid Howells long ret; 363993766fbdSDavid Howells 364093766fbdSDavid Howells if (!may_mount()) 364193766fbdSDavid Howells return -EPERM; 364293766fbdSDavid Howells 364393766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 364493766fbdSDavid Howells return -EINVAL; 364593766fbdSDavid Howells 36465b490500SChristian Brauner if (attr_flags & ~FSMOUNT_VALID_FLAGS) 364793766fbdSDavid Howells return -EINVAL; 364893766fbdSDavid Howells 36495b490500SChristian Brauner mnt_flags = attr_flags_to_mnt_flags(attr_flags); 365093766fbdSDavid Howells 365193766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 365293766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 365393766fbdSDavid Howells break; 365493766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 365593766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 365693766fbdSDavid Howells break; 365793766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 365893766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 365993766fbdSDavid Howells break; 366093766fbdSDavid Howells default: 366193766fbdSDavid Howells return -EINVAL; 366293766fbdSDavid Howells } 366393766fbdSDavid Howells 366493766fbdSDavid Howells f = fdget(fs_fd); 366593766fbdSDavid Howells if (!f.file) 366693766fbdSDavid Howells return -EBADF; 366793766fbdSDavid Howells 366893766fbdSDavid Howells ret = -EINVAL; 366993766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 367093766fbdSDavid Howells goto err_fsfd; 367193766fbdSDavid Howells 367293766fbdSDavid Howells fc = f.file->private_data; 367393766fbdSDavid Howells 367493766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 367593766fbdSDavid Howells if (ret < 0) 367693766fbdSDavid Howells goto err_fsfd; 367793766fbdSDavid Howells 367893766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 367993766fbdSDavid Howells ret = -EINVAL; 368093766fbdSDavid Howells if (!fc->root) 368193766fbdSDavid Howells goto err_unlock; 368293766fbdSDavid Howells 368393766fbdSDavid Howells ret = -EPERM; 368493766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 368593766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 368693766fbdSDavid Howells goto err_unlock; 368793766fbdSDavid Howells } 368893766fbdSDavid Howells 368993766fbdSDavid Howells ret = -EBUSY; 369093766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 369193766fbdSDavid Howells goto err_unlock; 369293766fbdSDavid Howells 3693f7e33bdbSJeff Layton if (fc->sb_flags & SB_MANDLOCK) 3694f7e33bdbSJeff Layton warn_mandlock(); 369593766fbdSDavid Howells 369693766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 369793766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 369893766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 369993766fbdSDavid Howells goto err_unlock; 370093766fbdSDavid Howells } 370193766fbdSDavid Howells newmount.dentry = dget(fc->root); 370293766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 370393766fbdSDavid Howells 370493766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 370593766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 370693766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 370793766fbdSDavid Howells * don't want to have to handle any errors incurred. 370893766fbdSDavid Howells */ 370993766fbdSDavid Howells vfs_clean_context(fc); 371093766fbdSDavid Howells 371193766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 371293766fbdSDavid Howells if (IS_ERR(ns)) { 371393766fbdSDavid Howells ret = PTR_ERR(ns); 371493766fbdSDavid Howells goto err_path; 371593766fbdSDavid Howells } 371693766fbdSDavid Howells mnt = real_mount(newmount.mnt); 371793766fbdSDavid Howells mnt->mnt_ns = ns; 371893766fbdSDavid Howells ns->root = mnt; 371993766fbdSDavid Howells ns->mounts = 1; 372093766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 37211b0b9cc8SEric Biggers mntget(newmount.mnt); 372293766fbdSDavid Howells 372393766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 372493766fbdSDavid Howells * it, not just simply put it. 372593766fbdSDavid Howells */ 372693766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 372793766fbdSDavid Howells if (IS_ERR(file)) { 372893766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 372993766fbdSDavid Howells ret = PTR_ERR(file); 373093766fbdSDavid Howells goto err_path; 373193766fbdSDavid Howells } 373293766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 373393766fbdSDavid Howells 373493766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 373593766fbdSDavid Howells if (ret >= 0) 373693766fbdSDavid Howells fd_install(ret, file); 373793766fbdSDavid Howells else 373893766fbdSDavid Howells fput(file); 373993766fbdSDavid Howells 374093766fbdSDavid Howells err_path: 374193766fbdSDavid Howells path_put(&newmount); 374293766fbdSDavid Howells err_unlock: 374393766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 374493766fbdSDavid Howells err_fsfd: 374593766fbdSDavid Howells fdput(f); 374693766fbdSDavid Howells return ret; 374793766fbdSDavid Howells } 374893766fbdSDavid Howells 374993766fbdSDavid Howells /* 375093766fbdSDavid Howells * Move a mount from one place to another. In combination with 375193766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 375293766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 375393766fbdSDavid Howells * a mount subtree. 37542db154b3SDavid Howells * 37552db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 37562db154b3SDavid Howells */ 37572db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 37582658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 37592658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 37602db154b3SDavid Howells unsigned int, flags) 37612db154b3SDavid Howells { 37622db154b3SDavid Howells struct path from_path, to_path; 37632db154b3SDavid Howells unsigned int lflags; 37642db154b3SDavid Howells int ret = 0; 37652db154b3SDavid Howells 37662db154b3SDavid Howells if (!may_mount()) 37672db154b3SDavid Howells return -EPERM; 37682db154b3SDavid Howells 37692db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 37702db154b3SDavid Howells return -EINVAL; 37712db154b3SDavid Howells 37722db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 37732db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 37742db154b3SDavid Howells * to clear it afterwards. 37752db154b3SDavid Howells */ 37762db154b3SDavid Howells lflags = 0; 37772db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37782db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37792db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 37802db154b3SDavid Howells 37812db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 37822db154b3SDavid Howells if (ret < 0) 37832db154b3SDavid Howells return ret; 37842db154b3SDavid Howells 37852db154b3SDavid Howells lflags = 0; 37862db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37872db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37882db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 37892db154b3SDavid Howells 37902db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 37912db154b3SDavid Howells if (ret < 0) 37922db154b3SDavid Howells goto out_from; 37932db154b3SDavid Howells 37942db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 37952db154b3SDavid Howells if (ret < 0) 37962db154b3SDavid Howells goto out_to; 37972db154b3SDavid Howells 37989ffb14efSPavel Tikhomirov if (flags & MOVE_MOUNT_SET_GROUP) 37999ffb14efSPavel Tikhomirov ret = do_set_group(&from_path, &to_path); 38009ffb14efSPavel Tikhomirov else 38012db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 38022db154b3SDavid Howells 38032db154b3SDavid Howells out_to: 38042db154b3SDavid Howells path_put(&to_path); 38052db154b3SDavid Howells out_from: 38062db154b3SDavid Howells path_put(&from_path); 38072db154b3SDavid Howells return ret; 38082db154b3SDavid Howells } 38092db154b3SDavid Howells 38102db154b3SDavid Howells /* 3811afac7cbaSAl Viro * Return true if path is reachable from root 3812afac7cbaSAl Viro * 381348a066e7SAl Viro * namespace_sem or mount_lock is held 3814afac7cbaSAl Viro */ 3815643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3816afac7cbaSAl Viro const struct path *root) 3817afac7cbaSAl Viro { 3818643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3819a73324daSAl Viro dentry = mnt->mnt_mountpoint; 38200714a533SAl Viro mnt = mnt->mnt_parent; 3821afac7cbaSAl Viro } 3822643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3823afac7cbaSAl Viro } 3824afac7cbaSAl Viro 3825640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3826afac7cbaSAl Viro { 382725ab4c9bSYaowei Bai bool res; 382848a066e7SAl Viro read_seqlock_excl(&mount_lock); 3829643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 383048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3831afac7cbaSAl Viro return res; 3832afac7cbaSAl Viro } 3833afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3834afac7cbaSAl Viro 3835afac7cbaSAl Viro /* 38361da177e4SLinus Torvalds * pivot_root Semantics: 38371da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 38381da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 38391da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 38401da177e4SLinus Torvalds * 38411da177e4SLinus Torvalds * Restrictions: 38421da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 38431da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 38441da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 38451da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 38461da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 38471da177e4SLinus Torvalds * 38484a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 38490c1bc6b8SMauro Carvalho Chehab * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives 38504a0d11faSNeil Brown * in this situation. 38514a0d11faSNeil Brown * 38521da177e4SLinus Torvalds * Notes: 38531da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 38541da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 38551da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 38561da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 38571da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 38581da177e4SLinus Torvalds * first. 38591da177e4SLinus Torvalds */ 38603480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 38613480b257SHeiko Carstens const char __user *, put_old) 38621da177e4SLinus Torvalds { 38632763d119SAl Viro struct path new, old, root; 38642763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 386584d17192SAl Viro struct mountpoint *old_mp, *root_mp; 38661da177e4SLinus Torvalds int error; 38671da177e4SLinus Torvalds 38689b40bc90SAl Viro if (!may_mount()) 38691da177e4SLinus Torvalds return -EPERM; 38701da177e4SLinus Torvalds 3871ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3872ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 38731da177e4SLinus Torvalds if (error) 38741da177e4SLinus Torvalds goto out0; 38751da177e4SLinus Torvalds 3876ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 3877ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 38781da177e4SLinus Torvalds if (error) 38791da177e4SLinus Torvalds goto out1; 38801da177e4SLinus Torvalds 38812d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3882b12cea91SAl Viro if (error) 3883b12cea91SAl Viro goto out2; 38841da177e4SLinus Torvalds 3885f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 388684d17192SAl Viro old_mp = lock_mount(&old); 388784d17192SAl Viro error = PTR_ERR(old_mp); 388884d17192SAl Viro if (IS_ERR(old_mp)) 3889b12cea91SAl Viro goto out3; 3890b12cea91SAl Viro 38911da177e4SLinus Torvalds error = -EINVAL; 3892419148daSAl Viro new_mnt = real_mount(new.mnt); 3893419148daSAl Viro root_mnt = real_mount(root.mnt); 389484d17192SAl Viro old_mnt = real_mount(old.mnt); 38952763d119SAl Viro ex_parent = new_mnt->mnt_parent; 38962763d119SAl Viro root_parent = root_mnt->mnt_parent; 389784d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 38982763d119SAl Viro IS_MNT_SHARED(ex_parent) || 38992763d119SAl Viro IS_MNT_SHARED(root_parent)) 3900b12cea91SAl Viro goto out4; 3901143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3902b12cea91SAl Viro goto out4; 39035ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 39045ff9d8a6SEric W. Biederman goto out4; 39051da177e4SLinus Torvalds error = -ENOENT; 3906f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3907b12cea91SAl Viro goto out4; 39081da177e4SLinus Torvalds error = -EBUSY; 390984d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3910b12cea91SAl Viro goto out4; /* loop, on the same file system */ 39111da177e4SLinus Torvalds error = -EINVAL; 39128c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3913b12cea91SAl Viro goto out4; /* not a mountpoint */ 3914676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3915b12cea91SAl Viro goto out4; /* not attached */ 39162d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3917b12cea91SAl Viro goto out4; /* not a mountpoint */ 3918676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3919b12cea91SAl Viro goto out4; /* not attached */ 39204ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 392184d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3922b12cea91SAl Viro goto out4; 39230d082601SEric W. Biederman /* make certain new is below the root */ 39240d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 39250d082601SEric W. Biederman goto out4; 3926719ea2fbSAl Viro lock_mount_hash(); 39272763d119SAl Viro umount_mnt(new_mnt); 39282763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 39295ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 39305ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 39315ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 39325ff9d8a6SEric W. Biederman } 39334ac91378SJan Blunck /* mount old root on put_old */ 393484d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 39354ac91378SJan Blunck /* mount new_root on / */ 39362763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 39372763d119SAl Viro mnt_add_count(root_parent, -1); 39386b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 39394fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 39404fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 39413895dbf8SEric W. Biederman put_mountpoint(root_mp); 3942719ea2fbSAl Viro unlock_mount_hash(); 39432d8f3038SAl Viro chroot_fs_refs(&root, &new); 39441da177e4SLinus Torvalds error = 0; 3945b12cea91SAl Viro out4: 394684d17192SAl Viro unlock_mount(old_mp); 39472763d119SAl Viro if (!error) 39482763d119SAl Viro mntput_no_expire(ex_parent); 3949b12cea91SAl Viro out3: 39508c3ee42eSAl Viro path_put(&root); 3951b12cea91SAl Viro out2: 39522d8f3038SAl Viro path_put(&old); 39531da177e4SLinus Torvalds out1: 39542d8f3038SAl Viro path_put(&new); 39551da177e4SLinus Torvalds out0: 39561da177e4SLinus Torvalds return error; 39571da177e4SLinus Torvalds } 39581da177e4SLinus Torvalds 39592a186721SChristian Brauner static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) 39602a186721SChristian Brauner { 39612a186721SChristian Brauner unsigned int flags = mnt->mnt.mnt_flags; 39622a186721SChristian Brauner 39632a186721SChristian Brauner /* flags to clear */ 39642a186721SChristian Brauner flags &= ~kattr->attr_clr; 39652a186721SChristian Brauner /* flags to raise */ 39662a186721SChristian Brauner flags |= kattr->attr_set; 39672a186721SChristian Brauner 39682a186721SChristian Brauner return flags; 39692a186721SChristian Brauner } 39702a186721SChristian Brauner 39719caccd41SChristian Brauner static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 39729caccd41SChristian Brauner { 39739caccd41SChristian Brauner struct vfsmount *m = &mnt->mnt; 3974bd303368SChristian Brauner struct user_namespace *fs_userns = m->mnt_sb->s_user_ns; 39759caccd41SChristian Brauner 3976256c8aedSChristian Brauner if (!kattr->mnt_idmap) 39779caccd41SChristian Brauner return 0; 39789caccd41SChristian Brauner 39799caccd41SChristian Brauner /* 3980bd303368SChristian Brauner * Creating an idmapped mount with the filesystem wide idmapping 3981bd303368SChristian Brauner * doesn't make sense so block that. We don't allow mushy semantics. 3982bd303368SChristian Brauner */ 3983*3707d84cSChristian Brauner if (!check_fsmapping(kattr->mnt_idmap, m->mnt_sb)) 3984bd303368SChristian Brauner return -EINVAL; 3985bd303368SChristian Brauner 3986bd303368SChristian Brauner /* 39879caccd41SChristian Brauner * Once a mount has been idmapped we don't allow it to change its 39889caccd41SChristian Brauner * mapping. It makes things simpler and callers can just create 39899caccd41SChristian Brauner * another bind-mount they can idmap if they want to. 39909caccd41SChristian Brauner */ 3991bb49e9e7SChristian Brauner if (is_idmapped_mnt(m)) 39929caccd41SChristian Brauner return -EPERM; 39939caccd41SChristian Brauner 39949caccd41SChristian Brauner /* The underlying filesystem doesn't support idmapped mounts yet. */ 39959caccd41SChristian Brauner if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) 39969caccd41SChristian Brauner return -EINVAL; 39979caccd41SChristian Brauner 39989caccd41SChristian Brauner /* We're not controlling the superblock. */ 3999bd303368SChristian Brauner if (!ns_capable(fs_userns, CAP_SYS_ADMIN)) 40009caccd41SChristian Brauner return -EPERM; 40019caccd41SChristian Brauner 40029caccd41SChristian Brauner /* Mount has already been visible in the filesystem hierarchy. */ 40039caccd41SChristian Brauner if (!is_anon_ns(mnt->mnt_ns)) 40049caccd41SChristian Brauner return -EINVAL; 40059caccd41SChristian Brauner 40069caccd41SChristian Brauner return 0; 40079caccd41SChristian Brauner } 40089caccd41SChristian Brauner 4009a26f788bSChristian Brauner /** 4010a26f788bSChristian Brauner * mnt_allow_writers() - check whether the attribute change allows writers 4011a26f788bSChristian Brauner * @kattr: the new mount attributes 4012a26f788bSChristian Brauner * @mnt: the mount to which @kattr will be applied 4013a26f788bSChristian Brauner * 4014a26f788bSChristian Brauner * Check whether thew new mount attributes in @kattr allow concurrent writers. 4015a26f788bSChristian Brauner * 4016a26f788bSChristian Brauner * Return: true if writers need to be held, false if not 4017a26f788bSChristian Brauner */ 4018a26f788bSChristian Brauner static inline bool mnt_allow_writers(const struct mount_kattr *kattr, 4019a26f788bSChristian Brauner const struct mount *mnt) 40202a186721SChristian Brauner { 4021e1bbcd27SChristian Brauner return (!(kattr->attr_set & MNT_READONLY) || 4022e1bbcd27SChristian Brauner (mnt->mnt.mnt_flags & MNT_READONLY)) && 4023256c8aedSChristian Brauner !kattr->mnt_idmap; 40242a186721SChristian Brauner } 40252a186721SChristian Brauner 402687bb5b60SChristian Brauner static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt) 40272a186721SChristian Brauner { 4028e257039fSAl Viro struct mount *m; 4029e257039fSAl Viro int err; 40302a186721SChristian Brauner 4031e257039fSAl Viro for (m = mnt; m; m = next_mnt(m, mnt)) { 4032e257039fSAl Viro if (!can_change_locked_flags(m, recalc_flags(kattr, m))) { 4033e257039fSAl Viro err = -EPERM; 4034e257039fSAl Viro break; 40352a186721SChristian Brauner } 40362a186721SChristian Brauner 403787bb5b60SChristian Brauner err = can_idmap_mount(kattr, m); 403887bb5b60SChristian Brauner if (err) 4039e257039fSAl Viro break; 40402a186721SChristian Brauner 4041e257039fSAl Viro if (!mnt_allow_writers(kattr, m)) { 404287bb5b60SChristian Brauner err = mnt_hold_writers(m); 404387bb5b60SChristian Brauner if (err) 4044e257039fSAl Viro break; 40452a186721SChristian Brauner } 40462a186721SChristian Brauner 4047e257039fSAl Viro if (!kattr->recurse) 404887bb5b60SChristian Brauner return 0; 40492a186721SChristian Brauner } 40502a186721SChristian Brauner 4051e257039fSAl Viro if (err) { 4052e257039fSAl Viro struct mount *p; 4053e257039fSAl Viro 40540014edaeSChristian Brauner /* 40550014edaeSChristian Brauner * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will 40560014edaeSChristian Brauner * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all 40570014edaeSChristian Brauner * mounts and needs to take care to include the first mount. 40580014edaeSChristian Brauner */ 40590014edaeSChristian Brauner for (p = mnt; p; p = next_mnt(p, mnt)) { 4060e257039fSAl Viro /* If we had to hold writers unblock them. */ 4061e257039fSAl Viro if (p->mnt.mnt_flags & MNT_WRITE_HOLD) 4062e257039fSAl Viro mnt_unhold_writers(p); 40630014edaeSChristian Brauner 40640014edaeSChristian Brauner /* 40650014edaeSChristian Brauner * We're done once the first mount we changed got 40660014edaeSChristian Brauner * MNT_WRITE_HOLD unset. 40670014edaeSChristian Brauner */ 40680014edaeSChristian Brauner if (p == m) 40690014edaeSChristian Brauner break; 4070e257039fSAl Viro } 4071e257039fSAl Viro } 4072e257039fSAl Viro return err; 40732a186721SChristian Brauner } 40742a186721SChristian Brauner 40759caccd41SChristian Brauner static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 40769caccd41SChristian Brauner { 4077256c8aedSChristian Brauner if (!kattr->mnt_idmap) 40789caccd41SChristian Brauner return; 40799caccd41SChristian Brauner 4080bd303368SChristian Brauner /* 4081256c8aedSChristian Brauner * Pairs with smp_load_acquire() in mnt_idmap(). 4082256c8aedSChristian Brauner * 4083256c8aedSChristian Brauner * Since we only allow a mount to change the idmapping once and 4084256c8aedSChristian Brauner * verified this in can_idmap_mount() we know that the mount has 4085256c8aedSChristian Brauner * @nop_mnt_idmap attached to it. So there's no need to drop any 4086256c8aedSChristian Brauner * references. 4087bd303368SChristian Brauner */ 4088256c8aedSChristian Brauner smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap)); 40899caccd41SChristian Brauner } 40909caccd41SChristian Brauner 4091e257039fSAl Viro static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt) 40922a186721SChristian Brauner { 4093e257039fSAl Viro struct mount *m; 40942a186721SChristian Brauner 4095e257039fSAl Viro for (m = mnt; m; m = next_mnt(m, mnt)) { 40962a186721SChristian Brauner unsigned int flags; 40972a186721SChristian Brauner 40989caccd41SChristian Brauner do_idmap_mount(kattr, m); 40992a186721SChristian Brauner flags = recalc_flags(kattr, m); 41002a186721SChristian Brauner WRITE_ONCE(m->mnt.mnt_flags, flags); 41012a186721SChristian Brauner 410203b6abeeSChristian Brauner /* If we had to hold writers unblock them. */ 410303b6abeeSChristian Brauner if (m->mnt.mnt_flags & MNT_WRITE_HOLD) 41042a186721SChristian Brauner mnt_unhold_writers(m); 41052a186721SChristian Brauner 4106e257039fSAl Viro if (kattr->propagation) 41072a186721SChristian Brauner change_mnt_propagation(m, kattr->propagation); 4108e257039fSAl Viro if (!kattr->recurse) 41092a186721SChristian Brauner break; 4110e257039fSAl Viro } 41112a186721SChristian Brauner touch_mnt_namespace(mnt->mnt_ns); 41122a186721SChristian Brauner } 41132a186721SChristian Brauner 41142a186721SChristian Brauner static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) 41152a186721SChristian Brauner { 411687bb5b60SChristian Brauner struct mount *mnt = real_mount(path->mnt); 41172a186721SChristian Brauner int err = 0; 41182a186721SChristian Brauner 41192a186721SChristian Brauner if (path->dentry != mnt->mnt.mnt_root) 41202a186721SChristian Brauner return -EINVAL; 41212a186721SChristian Brauner 4122256c8aedSChristian Brauner if (kattr->mnt_userns) { 4123256c8aedSChristian Brauner struct mnt_idmap *mnt_idmap; 4124256c8aedSChristian Brauner 4125256c8aedSChristian Brauner mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns); 4126256c8aedSChristian Brauner if (IS_ERR(mnt_idmap)) 4127256c8aedSChristian Brauner return PTR_ERR(mnt_idmap); 4128256c8aedSChristian Brauner kattr->mnt_idmap = mnt_idmap; 4129256c8aedSChristian Brauner } 4130256c8aedSChristian Brauner 41312a186721SChristian Brauner if (kattr->propagation) { 41322a186721SChristian Brauner /* 41332a186721SChristian Brauner * Only take namespace_lock() if we're actually changing 41342a186721SChristian Brauner * propagation. 41352a186721SChristian Brauner */ 41362a186721SChristian Brauner namespace_lock(); 41372a186721SChristian Brauner if (kattr->propagation == MS_SHARED) { 41382a186721SChristian Brauner err = invent_group_ids(mnt, kattr->recurse); 41392a186721SChristian Brauner if (err) { 41402a186721SChristian Brauner namespace_unlock(); 41412a186721SChristian Brauner return err; 41422a186721SChristian Brauner } 41432a186721SChristian Brauner } 41442a186721SChristian Brauner } 41452a186721SChristian Brauner 414687bb5b60SChristian Brauner err = -EINVAL; 41472a186721SChristian Brauner lock_mount_hash(); 41482a186721SChristian Brauner 414987bb5b60SChristian Brauner /* Ensure that this isn't anything purely vfs internal. */ 415087bb5b60SChristian Brauner if (!is_mounted(&mnt->mnt)) 415187bb5b60SChristian Brauner goto out; 41522a186721SChristian Brauner 415387bb5b60SChristian Brauner /* 415487bb5b60SChristian Brauner * If this is an attached mount make sure it's located in the callers 415587bb5b60SChristian Brauner * mount namespace. If it's not don't let the caller interact with it. 415687bb5b60SChristian Brauner * If this is a detached mount make sure it has an anonymous mount 415787bb5b60SChristian Brauner * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE. 415887bb5b60SChristian Brauner */ 415987bb5b60SChristian Brauner if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns))) 416087bb5b60SChristian Brauner goto out; 416187bb5b60SChristian Brauner 416287bb5b60SChristian Brauner /* 416387bb5b60SChristian Brauner * First, we get the mount tree in a shape where we can change mount 416487bb5b60SChristian Brauner * properties without failure. If we succeeded to do so we commit all 416587bb5b60SChristian Brauner * changes and if we failed we clean up. 416687bb5b60SChristian Brauner */ 416787bb5b60SChristian Brauner err = mount_setattr_prepare(kattr, mnt); 4168e257039fSAl Viro if (!err) 4169e257039fSAl Viro mount_setattr_commit(kattr, mnt); 417087bb5b60SChristian Brauner 417187bb5b60SChristian Brauner out: 41722a186721SChristian Brauner unlock_mount_hash(); 41732a186721SChristian Brauner 41742a186721SChristian Brauner if (kattr->propagation) { 41752a186721SChristian Brauner namespace_unlock(); 41762a186721SChristian Brauner if (err) 41772a186721SChristian Brauner cleanup_group_ids(mnt, NULL); 41782a186721SChristian Brauner } 41792a186721SChristian Brauner 41802a186721SChristian Brauner return err; 41812a186721SChristian Brauner } 41822a186721SChristian Brauner 41839caccd41SChristian Brauner static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, 41849caccd41SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 41859caccd41SChristian Brauner { 41869caccd41SChristian Brauner int err = 0; 41879caccd41SChristian Brauner struct ns_common *ns; 41889caccd41SChristian Brauner struct user_namespace *mnt_userns; 41899caccd41SChristian Brauner struct file *file; 41909caccd41SChristian Brauner 41919caccd41SChristian Brauner if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) 41929caccd41SChristian Brauner return 0; 41939caccd41SChristian Brauner 41949caccd41SChristian Brauner /* 41959caccd41SChristian Brauner * We currently do not support clearing an idmapped mount. If this ever 41969caccd41SChristian Brauner * is a use-case we can revisit this but for now let's keep it simple 41979caccd41SChristian Brauner * and not allow it. 41989caccd41SChristian Brauner */ 41999caccd41SChristian Brauner if (attr->attr_clr & MOUNT_ATTR_IDMAP) 42009caccd41SChristian Brauner return -EINVAL; 42019caccd41SChristian Brauner 42029caccd41SChristian Brauner if (attr->userns_fd > INT_MAX) 42039caccd41SChristian Brauner return -EINVAL; 42049caccd41SChristian Brauner 42059caccd41SChristian Brauner file = fget(attr->userns_fd); 42069caccd41SChristian Brauner if (!file) 42079caccd41SChristian Brauner return -EBADF; 42089caccd41SChristian Brauner 42099caccd41SChristian Brauner if (!proc_ns_file(file)) { 42109caccd41SChristian Brauner err = -EINVAL; 42119caccd41SChristian Brauner goto out_fput; 42129caccd41SChristian Brauner } 42139caccd41SChristian Brauner 42149caccd41SChristian Brauner ns = get_proc_ns(file_inode(file)); 42159caccd41SChristian Brauner if (ns->ops->type != CLONE_NEWUSER) { 42169caccd41SChristian Brauner err = -EINVAL; 42179caccd41SChristian Brauner goto out_fput; 42189caccd41SChristian Brauner } 42199caccd41SChristian Brauner 42209caccd41SChristian Brauner /* 4221bd303368SChristian Brauner * The initial idmapping cannot be used to create an idmapped 4222bd303368SChristian Brauner * mount. We use the initial idmapping as an indicator of a mount 4223bd303368SChristian Brauner * that is not idmapped. It can simply be passed into helpers that 4224bd303368SChristian Brauner * are aware of idmapped mounts as a convenient shortcut. A user 4225bd303368SChristian Brauner * can just create a dedicated identity mapping to achieve the same 4226bd303368SChristian Brauner * result. 42279caccd41SChristian Brauner */ 42289caccd41SChristian Brauner mnt_userns = container_of(ns, struct user_namespace, ns); 4229*3707d84cSChristian Brauner if (mnt_userns == &init_user_ns) { 42309caccd41SChristian Brauner err = -EPERM; 42319caccd41SChristian Brauner goto out_fput; 42329caccd41SChristian Brauner } 4233bf1ac16eSSeth Forshee 4234bf1ac16eSSeth Forshee /* We're not controlling the target namespace. */ 4235bf1ac16eSSeth Forshee if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) { 4236bf1ac16eSSeth Forshee err = -EPERM; 4237bf1ac16eSSeth Forshee goto out_fput; 4238bf1ac16eSSeth Forshee } 4239bf1ac16eSSeth Forshee 42409caccd41SChristian Brauner kattr->mnt_userns = get_user_ns(mnt_userns); 42419caccd41SChristian Brauner 42429caccd41SChristian Brauner out_fput: 42439caccd41SChristian Brauner fput(file); 42449caccd41SChristian Brauner return err; 42459caccd41SChristian Brauner } 42469caccd41SChristian Brauner 42479caccd41SChristian Brauner static int build_mount_kattr(const struct mount_attr *attr, size_t usize, 42482a186721SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 42492a186721SChristian Brauner { 42502a186721SChristian Brauner unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 42512a186721SChristian Brauner 42522a186721SChristian Brauner if (flags & AT_NO_AUTOMOUNT) 42532a186721SChristian Brauner lookup_flags &= ~LOOKUP_AUTOMOUNT; 42542a186721SChristian Brauner if (flags & AT_SYMLINK_NOFOLLOW) 42552a186721SChristian Brauner lookup_flags &= ~LOOKUP_FOLLOW; 42562a186721SChristian Brauner if (flags & AT_EMPTY_PATH) 42572a186721SChristian Brauner lookup_flags |= LOOKUP_EMPTY; 42582a186721SChristian Brauner 42592a186721SChristian Brauner *kattr = (struct mount_kattr) { 42602a186721SChristian Brauner .lookup_flags = lookup_flags, 42612a186721SChristian Brauner .recurse = !!(flags & AT_RECURSIVE), 42622a186721SChristian Brauner }; 42632a186721SChristian Brauner 42642a186721SChristian Brauner if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) 42652a186721SChristian Brauner return -EINVAL; 42662a186721SChristian Brauner if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) 42672a186721SChristian Brauner return -EINVAL; 42682a186721SChristian Brauner kattr->propagation = attr->propagation; 42692a186721SChristian Brauner 42702a186721SChristian Brauner if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) 42712a186721SChristian Brauner return -EINVAL; 42722a186721SChristian Brauner 42732a186721SChristian Brauner kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); 42742a186721SChristian Brauner kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); 42752a186721SChristian Brauner 42762a186721SChristian Brauner /* 42772a186721SChristian Brauner * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, 42782a186721SChristian Brauner * users wanting to transition to a different atime setting cannot 42792a186721SChristian Brauner * simply specify the atime setting in @attr_set, but must also 42802a186721SChristian Brauner * specify MOUNT_ATTR__ATIME in the @attr_clr field. 42812a186721SChristian Brauner * So ensure that MOUNT_ATTR__ATIME can't be partially set in 42822a186721SChristian Brauner * @attr_clr and that @attr_set can't have any atime bits set if 42832a186721SChristian Brauner * MOUNT_ATTR__ATIME isn't set in @attr_clr. 42842a186721SChristian Brauner */ 42852a186721SChristian Brauner if (attr->attr_clr & MOUNT_ATTR__ATIME) { 42862a186721SChristian Brauner if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) 42872a186721SChristian Brauner return -EINVAL; 42882a186721SChristian Brauner 42892a186721SChristian Brauner /* 42902a186721SChristian Brauner * Clear all previous time settings as they are mutually 42912a186721SChristian Brauner * exclusive. 42922a186721SChristian Brauner */ 42932a186721SChristian Brauner kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; 42942a186721SChristian Brauner switch (attr->attr_set & MOUNT_ATTR__ATIME) { 42952a186721SChristian Brauner case MOUNT_ATTR_RELATIME: 42962a186721SChristian Brauner kattr->attr_set |= MNT_RELATIME; 42972a186721SChristian Brauner break; 42982a186721SChristian Brauner case MOUNT_ATTR_NOATIME: 42992a186721SChristian Brauner kattr->attr_set |= MNT_NOATIME; 43002a186721SChristian Brauner break; 43012a186721SChristian Brauner case MOUNT_ATTR_STRICTATIME: 43022a186721SChristian Brauner break; 43032a186721SChristian Brauner default: 43042a186721SChristian Brauner return -EINVAL; 43052a186721SChristian Brauner } 43062a186721SChristian Brauner } else { 43072a186721SChristian Brauner if (attr->attr_set & MOUNT_ATTR__ATIME) 43082a186721SChristian Brauner return -EINVAL; 43092a186721SChristian Brauner } 43102a186721SChristian Brauner 43119caccd41SChristian Brauner return build_mount_idmapped(attr, usize, kattr, flags); 43129caccd41SChristian Brauner } 43139caccd41SChristian Brauner 43149caccd41SChristian Brauner static void finish_mount_kattr(struct mount_kattr *kattr) 43159caccd41SChristian Brauner { 43169caccd41SChristian Brauner put_user_ns(kattr->mnt_userns); 43179caccd41SChristian Brauner kattr->mnt_userns = NULL; 4318256c8aedSChristian Brauner 4319256c8aedSChristian Brauner if (kattr->mnt_idmap) 4320256c8aedSChristian Brauner mnt_idmap_put(kattr->mnt_idmap); 43212a186721SChristian Brauner } 43222a186721SChristian Brauner 43232a186721SChristian Brauner SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, 43242a186721SChristian Brauner unsigned int, flags, struct mount_attr __user *, uattr, 43252a186721SChristian Brauner size_t, usize) 43262a186721SChristian Brauner { 43272a186721SChristian Brauner int err; 43282a186721SChristian Brauner struct path target; 43292a186721SChristian Brauner struct mount_attr attr; 43302a186721SChristian Brauner struct mount_kattr kattr; 43312a186721SChristian Brauner 43322a186721SChristian Brauner BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); 43332a186721SChristian Brauner 43342a186721SChristian Brauner if (flags & ~(AT_EMPTY_PATH | 43352a186721SChristian Brauner AT_RECURSIVE | 43362a186721SChristian Brauner AT_SYMLINK_NOFOLLOW | 43372a186721SChristian Brauner AT_NO_AUTOMOUNT)) 43382a186721SChristian Brauner return -EINVAL; 43392a186721SChristian Brauner 43402a186721SChristian Brauner if (unlikely(usize > PAGE_SIZE)) 43412a186721SChristian Brauner return -E2BIG; 43422a186721SChristian Brauner if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) 43432a186721SChristian Brauner return -EINVAL; 43442a186721SChristian Brauner 43452a186721SChristian Brauner if (!may_mount()) 43462a186721SChristian Brauner return -EPERM; 43472a186721SChristian Brauner 43482a186721SChristian Brauner err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); 43492a186721SChristian Brauner if (err) 43502a186721SChristian Brauner return err; 43512a186721SChristian Brauner 43522a186721SChristian Brauner /* Don't bother walking through the mounts if this is a nop. */ 43532a186721SChristian Brauner if (attr.attr_set == 0 && 43542a186721SChristian Brauner attr.attr_clr == 0 && 43552a186721SChristian Brauner attr.propagation == 0) 43562a186721SChristian Brauner return 0; 43572a186721SChristian Brauner 43589caccd41SChristian Brauner err = build_mount_kattr(&attr, usize, &kattr, flags); 43592a186721SChristian Brauner if (err) 43602a186721SChristian Brauner return err; 43612a186721SChristian Brauner 43622a186721SChristian Brauner err = user_path_at(dfd, path, kattr.lookup_flags, &target); 4363012e3322SChristian Brauner if (!err) { 43642a186721SChristian Brauner err = do_mount_setattr(&target, &kattr); 43652a186721SChristian Brauner path_put(&target); 4366012e3322SChristian Brauner } 4367012e3322SChristian Brauner finish_mount_kattr(&kattr); 43682a186721SChristian Brauner return err; 43692a186721SChristian Brauner } 43702a186721SChristian Brauner 43711da177e4SLinus Torvalds static void __init init_mount_tree(void) 43721da177e4SLinus Torvalds { 43731da177e4SLinus Torvalds struct vfsmount *mnt; 437474e83122SAl Viro struct mount *m; 43756b3286edSKirill Korotaev struct mnt_namespace *ns; 4376ac748a09SJan Blunck struct path root; 43771da177e4SLinus Torvalds 4378fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 43791da177e4SLinus Torvalds if (IS_ERR(mnt)) 43801da177e4SLinus Torvalds panic("Can't create rootfs"); 4381b3e19d92SNick Piggin 438274e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 43833b22edc5STrond Myklebust if (IS_ERR(ns)) 43841da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 438574e83122SAl Viro m = real_mount(mnt); 438674e83122SAl Viro m->mnt_ns = ns; 438774e83122SAl Viro ns->root = m; 438874e83122SAl Viro ns->mounts = 1; 438974e83122SAl Viro list_add(&m->mnt_list, &ns->list); 43906b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 43916b3286edSKirill Korotaev get_mnt_ns(ns); 43921da177e4SLinus Torvalds 4393be08d6d2SAl Viro root.mnt = mnt; 4394be08d6d2SAl Viro root.dentry = mnt->mnt_root; 4395da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 4396ac748a09SJan Blunck 4397ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 4398ac748a09SJan Blunck set_fs_root(current->fs, &root); 43991da177e4SLinus Torvalds } 44001da177e4SLinus Torvalds 440174bf17cfSDenis Cheng void __init mnt_init(void) 44021da177e4SLinus Torvalds { 440315a67dd8SRandy Dunlap int err; 44041da177e4SLinus Torvalds 44057d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 440679f6540bSVasily Averin 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); 44071da177e4SLinus Torvalds 44080818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 440938129a13SAl Viro sizeof(struct hlist_head), 44100818bf27SAl Viro mhash_entries, 19, 44113d375d78SPavel Tatashin HASH_ZERO, 44120818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 44130818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 44140818bf27SAl Viro sizeof(struct hlist_head), 44150818bf27SAl Viro mphash_entries, 19, 44163d375d78SPavel Tatashin HASH_ZERO, 44170818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 44181da177e4SLinus Torvalds 441984d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 44201da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 44211da177e4SLinus Torvalds 44224b93dc9bSTejun Heo kernfs_init(); 44234b93dc9bSTejun Heo 442415a67dd8SRandy Dunlap err = sysfs_init(); 442515a67dd8SRandy Dunlap if (err) 442615a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 44278e24eea7SHarvey Harrison __func__, err); 442800d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 442900d26666SGreg Kroah-Hartman if (!fs_kobj) 44308e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 4431037f11b4SAl Viro shmem_init(); 44321da177e4SLinus Torvalds init_rootfs(); 44331da177e4SLinus Torvalds init_mount_tree(); 44341da177e4SLinus Torvalds } 44351da177e4SLinus Torvalds 4436616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 44371da177e4SLinus Torvalds { 44381a7b8969SKirill Tkhai if (!refcount_dec_and_test(&ns->ns.count)) 4439616511d0STrond Myklebust return; 44407b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 4441771b1371SEric W. Biederman free_mnt_ns(ns); 44421da177e4SLinus Torvalds } 44439d412a43SAl Viro 4444d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 44459d412a43SAl Viro { 4446423e0ab0STim Chen struct vfsmount *mnt; 4447d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 4448423e0ab0STim Chen if (!IS_ERR(mnt)) { 4449423e0ab0STim Chen /* 4450423e0ab0STim Chen * it is a longterm mount, don't release mnt until 4451423e0ab0STim Chen * we unmount before file sys is unregistered 4452423e0ab0STim Chen */ 4453f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 4454423e0ab0STim Chen } 4455423e0ab0STim Chen return mnt; 44569d412a43SAl Viro } 4457d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 4458423e0ab0STim Chen 4459423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 4460423e0ab0STim Chen { 4461423e0ab0STim Chen /* release long term mount so mount point can be released */ 4462423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 4463f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 446448a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 4465423e0ab0STim Chen mntput(mnt); 4466423e0ab0STim Chen } 4467423e0ab0STim Chen } 4468423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 446902125a82SAl Viro 4470df820f8dSMiklos Szeredi void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) 4471df820f8dSMiklos Szeredi { 4472df820f8dSMiklos Szeredi unsigned int i; 4473df820f8dSMiklos Szeredi 4474df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4475df820f8dSMiklos Szeredi if (mnt[i]) 4476df820f8dSMiklos Szeredi real_mount(mnt[i])->mnt_ns = NULL; 4477df820f8dSMiklos Szeredi synchronize_rcu_expedited(); 4478df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4479df820f8dSMiklos Szeredi mntput(mnt[i]); 4480df820f8dSMiklos Szeredi } 4481df820f8dSMiklos Szeredi EXPORT_SYMBOL(kern_unmount_array); 4482df820f8dSMiklos Szeredi 448302125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 448402125a82SAl Viro { 4485143c8c91SAl Viro return check_mnt(real_mount(mnt)); 448602125a82SAl Viro } 44878823c079SEric W. Biederman 44883151527eSEric W. Biederman bool current_chrooted(void) 44893151527eSEric W. Biederman { 44903151527eSEric W. Biederman /* Does the current process have a non-standard root */ 44913151527eSEric W. Biederman struct path ns_root; 44923151527eSEric W. Biederman struct path fs_root; 44933151527eSEric W. Biederman bool chrooted; 44943151527eSEric W. Biederman 44953151527eSEric W. Biederman /* Find the namespace root */ 44963151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 44973151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 44983151527eSEric W. Biederman path_get(&ns_root); 44993151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 45003151527eSEric W. Biederman ; 45013151527eSEric W. Biederman 45023151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 45033151527eSEric W. Biederman 45043151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 45053151527eSEric W. Biederman 45063151527eSEric W. Biederman path_put(&fs_root); 45073151527eSEric W. Biederman path_put(&ns_root); 45083151527eSEric W. Biederman 45093151527eSEric W. Biederman return chrooted; 45103151527eSEric W. Biederman } 45113151527eSEric W. Biederman 4512132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 4513132e4608SDavid Howells const struct super_block *sb, 45148654df4eSEric W. Biederman int *new_mnt_flags) 451587a8ebd6SEric W. Biederman { 45168c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 451787a8ebd6SEric W. Biederman struct mount *mnt; 4518e51db735SEric W. Biederman bool visible = false; 451987a8ebd6SEric W. Biederman 452044bb4385SAl Viro down_read(&namespace_sem); 45219f6c61f9SMiklos Szeredi lock_ns_list(ns); 452287a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 4523e51db735SEric W. Biederman struct mount *child; 452477b1a97dSEric W. Biederman int mnt_flags; 452577b1a97dSEric W. Biederman 45269f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 45279f6c61f9SMiklos Szeredi continue; 45289f6c61f9SMiklos Szeredi 4529132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 4530e51db735SEric W. Biederman continue; 4531e51db735SEric W. Biederman 45327e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 45337e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 45347e96c1b0SEric W. Biederman */ 45357e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 45367e96c1b0SEric W. Biederman continue; 45377e96c1b0SEric W. Biederman 4538a1935c17SEric W. Biederman /* A local view of the mount flags */ 453977b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 454077b1a97dSEric W. Biederman 4541695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 4542bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 4543695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 4544695e9df0SEric W. Biederman 45458c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 45468c6cf9ccSEric W. Biederman * than the proposed new mount. 45478c6cf9ccSEric W. Biederman */ 454877b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 45498c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 45508c6cf9ccSEric W. Biederman continue; 455177b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 455277b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 45538c6cf9ccSEric W. Biederman continue; 45548c6cf9ccSEric W. Biederman 4555ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 4556ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 4557ceeb0e5dSEric W. Biederman * empty directories. 4558e51db735SEric W. Biederman */ 4559e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 4560e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 4561ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 4562d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 4563ceeb0e5dSEric W. Biederman continue; 45647236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 45657236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 4566e51db735SEric W. Biederman goto next; 456787a8ebd6SEric W. Biederman } 45688c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 456977b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 45708c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 4571e51db735SEric W. Biederman visible = true; 4572e51db735SEric W. Biederman goto found; 4573e51db735SEric W. Biederman next: ; 457487a8ebd6SEric W. Biederman } 4575e51db735SEric W. Biederman found: 45769f6c61f9SMiklos Szeredi unlock_ns_list(ns); 457744bb4385SAl Viro up_read(&namespace_sem); 4578e51db735SEric W. Biederman return visible; 457987a8ebd6SEric W. Biederman } 458087a8ebd6SEric W. Biederman 4581132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 45828654df4eSEric W. Biederman { 4583a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 45848654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 45858654df4eSEric W. Biederman unsigned long s_iflags; 45868654df4eSEric W. Biederman 45878654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 45888654df4eSEric W. Biederman return false; 45898654df4eSEric W. Biederman 45908654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 4591132e4608SDavid Howells s_iflags = sb->s_iflags; 45928654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 45938654df4eSEric W. Biederman return false; 45948654df4eSEric W. Biederman 4595a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 4596a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 4597a1935c17SEric W. Biederman required_iflags); 4598a1935c17SEric W. Biederman return true; 4599a1935c17SEric W. Biederman } 4600a1935c17SEric W. Biederman 4601132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 46028654df4eSEric W. Biederman } 46038654df4eSEric W. Biederman 4604380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 4605380cf5baSAndy Lutomirski { 4606380cf5baSAndy Lutomirski /* 4607380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 4608380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 4609380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 4610380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 4611380cf5baSAndy Lutomirski * in other namespaces. 4612380cf5baSAndy Lutomirski */ 4613380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 4614380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 4615380cf5baSAndy Lutomirski } 4616380cf5baSAndy Lutomirski 461764964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 46188823c079SEric W. Biederman { 461958be2825SAl Viro struct ns_common *ns = NULL; 46208823c079SEric W. Biederman struct nsproxy *nsproxy; 46218823c079SEric W. Biederman 4622728dba3aSEric W. Biederman task_lock(task); 4623728dba3aSEric W. Biederman nsproxy = task->nsproxy; 46248823c079SEric W. Biederman if (nsproxy) { 462558be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 462658be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 46278823c079SEric W. Biederman } 4628728dba3aSEric W. Biederman task_unlock(task); 46298823c079SEric W. Biederman 46308823c079SEric W. Biederman return ns; 46318823c079SEric W. Biederman } 46328823c079SEric W. Biederman 463364964528SAl Viro static void mntns_put(struct ns_common *ns) 46348823c079SEric W. Biederman { 463558be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 46368823c079SEric W. Biederman } 46378823c079SEric W. Biederman 4638f2a8d52eSChristian Brauner static int mntns_install(struct nsset *nsset, struct ns_common *ns) 46398823c079SEric W. Biederman { 4640f2a8d52eSChristian Brauner struct nsproxy *nsproxy = nsset->nsproxy; 4641f2a8d52eSChristian Brauner struct fs_struct *fs = nsset->fs; 46424f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 4643f2a8d52eSChristian Brauner struct user_namespace *user_ns = nsset->cred->user_ns; 46448823c079SEric W. Biederman struct path root; 46454f757f3cSAl Viro int err; 46468823c079SEric W. Biederman 46470c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 4648f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_CHROOT) || 4649f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_ADMIN)) 4650ae11e0f1SZhao Hongjiang return -EPERM; 46518823c079SEric W. Biederman 465274e83122SAl Viro if (is_anon_ns(mnt_ns)) 465374e83122SAl Viro return -EINVAL; 465474e83122SAl Viro 46558823c079SEric W. Biederman if (fs->users != 1) 46568823c079SEric W. Biederman return -EINVAL; 46578823c079SEric W. Biederman 46588823c079SEric W. Biederman get_mnt_ns(mnt_ns); 46594f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 46608823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 46618823c079SEric W. Biederman 46628823c079SEric W. Biederman /* Find the root */ 46634f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 46644f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 46654f757f3cSAl Viro if (err) { 46664f757f3cSAl Viro /* revert to old namespace */ 46674f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 46684f757f3cSAl Viro put_mnt_ns(mnt_ns); 46694f757f3cSAl Viro return err; 46704f757f3cSAl Viro } 46718823c079SEric W. Biederman 46724068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 46734068367cSAndrei Vagin 46748823c079SEric W. Biederman /* Update the pwd and root */ 46758823c079SEric W. Biederman set_fs_pwd(fs, &root); 46768823c079SEric W. Biederman set_fs_root(fs, &root); 46778823c079SEric W. Biederman 46788823c079SEric W. Biederman path_put(&root); 46798823c079SEric W. Biederman return 0; 46808823c079SEric W. Biederman } 46818823c079SEric W. Biederman 4682bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 4683bcac25a5SAndrey Vagin { 4684bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 4685bcac25a5SAndrey Vagin } 4686bcac25a5SAndrey Vagin 46878823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 46888823c079SEric W. Biederman .name = "mnt", 46898823c079SEric W. Biederman .type = CLONE_NEWNS, 46908823c079SEric W. Biederman .get = mntns_get, 46918823c079SEric W. Biederman .put = mntns_put, 46928823c079SEric W. Biederman .install = mntns_install, 4693bcac25a5SAndrey Vagin .owner = mntns_owner, 46948823c079SEric W. Biederman }; 4695ab171b95SLuis Chamberlain 4696ab171b95SLuis Chamberlain #ifdef CONFIG_SYSCTL 4697ab171b95SLuis Chamberlain static struct ctl_table fs_namespace_sysctls[] = { 4698ab171b95SLuis Chamberlain { 4699ab171b95SLuis Chamberlain .procname = "mount-max", 4700ab171b95SLuis Chamberlain .data = &sysctl_mount_max, 4701ab171b95SLuis Chamberlain .maxlen = sizeof(unsigned int), 4702ab171b95SLuis Chamberlain .mode = 0644, 4703ab171b95SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 4704ab171b95SLuis Chamberlain .extra1 = SYSCTL_ONE, 4705ab171b95SLuis Chamberlain }, 4706ab171b95SLuis Chamberlain { } 4707ab171b95SLuis Chamberlain }; 4708ab171b95SLuis Chamberlain 4709ab171b95SLuis Chamberlain static int __init init_fs_namespace_sysctls(void) 4710ab171b95SLuis Chamberlain { 4711ab171b95SLuis Chamberlain register_sysctl_init("fs", fs_namespace_sysctls); 4712ab171b95SLuis Chamberlain return 0; 4713ab171b95SLuis Chamberlain } 4714ab171b95SLuis Chamberlain fs_initcall(init_fs_namespace_sysctls); 4715ab171b95SLuis Chamberlain 4716ab171b95SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 4717