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 1286da27f796SRik van Riel /* 1287da27f796SRik van Riel * Make a mount point inaccessible to new lookups. 1288da27f796SRik van Riel * Because there may still be current users, the caller MUST WAIT 1289da27f796SRik van Riel * for an RCU grace period before destroying the mount point. 1290da27f796SRik van Riel */ 1291da27f796SRik van Riel void mnt_make_shortterm(struct vfsmount *mnt) 1292da27f796SRik van Riel { 1293da27f796SRik van Riel if (mnt) 1294da27f796SRik van Riel real_mount(mnt)->mnt_ns = NULL; 1295da27f796SRik van Riel } 1296da27f796SRik van Riel 12971f287bc4SRandy Dunlap /** 12981f287bc4SRandy Dunlap * path_is_mountpoint() - Check if path is a mount in the current namespace. 12991f287bc4SRandy Dunlap * @path: path to check 1300c6609c0aSIan Kent * 1301c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1302c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1303c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1304c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1305c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1306c6609c0aSIan Kent * alone. 1307c6609c0aSIan Kent */ 1308c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1309c6609c0aSIan Kent { 1310c6609c0aSIan Kent unsigned seq; 1311c6609c0aSIan Kent bool res; 1312c6609c0aSIan Kent 1313c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1314c6609c0aSIan Kent return false; 1315c6609c0aSIan Kent 1316c6609c0aSIan Kent rcu_read_lock(); 1317c6609c0aSIan Kent do { 1318c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1319c6609c0aSIan Kent res = __path_is_mountpoint(path); 1320c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1321c6609c0aSIan Kent rcu_read_unlock(); 1322c6609c0aSIan Kent 1323c6609c0aSIan Kent return res; 1324c6609c0aSIan Kent } 1325c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1326c6609c0aSIan Kent 1327ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 13287b7b1aceSAl Viro { 13293064c356SAl Viro struct mount *p; 13303064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 13313064c356SAl Viro if (IS_ERR(p)) 13323064c356SAl Viro return ERR_CAST(p); 13333064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 13343064c356SAl Viro return &p->mnt; 13357b7b1aceSAl Viro } 13361da177e4SLinus Torvalds 1337a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 13389f6c61f9SMiklos Szeredi static struct mount *mnt_list_next(struct mnt_namespace *ns, 13399f6c61f9SMiklos Szeredi struct list_head *p) 13409f6c61f9SMiklos Szeredi { 13419f6c61f9SMiklos Szeredi struct mount *mnt, *ret = NULL; 13429f6c61f9SMiklos Szeredi 13439f6c61f9SMiklos Szeredi lock_ns_list(ns); 13449f6c61f9SMiklos Szeredi list_for_each_continue(p, &ns->list) { 13459f6c61f9SMiklos Szeredi mnt = list_entry(p, typeof(*mnt), mnt_list); 13469f6c61f9SMiklos Szeredi if (!mnt_is_cursor(mnt)) { 13479f6c61f9SMiklos Szeredi ret = mnt; 13489f6c61f9SMiklos Szeredi break; 13499f6c61f9SMiklos Szeredi } 13509f6c61f9SMiklos Szeredi } 13519f6c61f9SMiklos Szeredi unlock_ns_list(ns); 13529f6c61f9SMiklos Szeredi 13539f6c61f9SMiklos Szeredi return ret; 13549f6c61f9SMiklos Szeredi } 13559f6c61f9SMiklos Szeredi 13560226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 13571da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 13581da177e4SLinus Torvalds { 1359ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13609f6c61f9SMiklos Szeredi struct list_head *prev; 13611da177e4SLinus Torvalds 1362390c6843SRam Pai down_read(&namespace_sem); 13639f6c61f9SMiklos Szeredi if (!*pos) { 13649f6c61f9SMiklos Szeredi prev = &p->ns->list; 13659f6c61f9SMiklos Szeredi } else { 13669f6c61f9SMiklos Szeredi prev = &p->cursor.mnt_list; 13679f6c61f9SMiklos Szeredi 13689f6c61f9SMiklos Szeredi /* Read after we'd reached the end? */ 13699f6c61f9SMiklos Szeredi if (list_empty(prev)) 13709f6c61f9SMiklos Szeredi return NULL; 1371c7999c36SAl Viro } 1372c7999c36SAl Viro 13739f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, prev); 13741da177e4SLinus Torvalds } 13751da177e4SLinus Torvalds 13761da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13771da177e4SLinus Torvalds { 1378ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13799f6c61f9SMiklos Szeredi struct mount *mnt = v; 1380b0765fb8SPavel Emelianov 13819f6c61f9SMiklos Szeredi ++*pos; 13829f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, &mnt->mnt_list); 13831da177e4SLinus Torvalds } 13841da177e4SLinus Torvalds 13851da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13861da177e4SLinus Torvalds { 13879f6c61f9SMiklos Szeredi struct proc_mounts *p = m->private; 13889f6c61f9SMiklos Szeredi struct mount *mnt = v; 13899f6c61f9SMiklos Szeredi 13909f6c61f9SMiklos Szeredi lock_ns_list(p->ns); 13919f6c61f9SMiklos Szeredi if (mnt) 13929f6c61f9SMiklos Szeredi list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); 13939f6c61f9SMiklos Szeredi else 13949f6c61f9SMiklos Szeredi list_del_init(&p->cursor.mnt_list); 13959f6c61f9SMiklos Szeredi unlock_ns_list(p->ns); 1396390c6843SRam Pai up_read(&namespace_sem); 13971da177e4SLinus Torvalds } 13981da177e4SLinus Torvalds 13990226f492SAl Viro static int m_show(struct seq_file *m, void *v) 14009f5596afSAl Viro { 1401ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 14029f6c61f9SMiklos Szeredi struct mount *r = v; 14030226f492SAl Viro return p->show(m, &r->mnt); 14041da177e4SLinus Torvalds } 14051da177e4SLinus Torvalds 1406a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 14071da177e4SLinus Torvalds .start = m_start, 14081da177e4SLinus Torvalds .next = m_next, 14091da177e4SLinus Torvalds .stop = m_stop, 14100226f492SAl Viro .show = m_show, 1411b4629fe2SChuck Lever }; 14129f6c61f9SMiklos Szeredi 14139f6c61f9SMiklos Szeredi void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) 14149f6c61f9SMiklos Szeredi { 14159f6c61f9SMiklos Szeredi down_read(&namespace_sem); 14169f6c61f9SMiklos Szeredi lock_ns_list(ns); 14179f6c61f9SMiklos Szeredi list_del(&cursor->mnt_list); 14189f6c61f9SMiklos Szeredi unlock_ns_list(ns); 14199f6c61f9SMiklos Szeredi up_read(&namespace_sem); 14209f6c61f9SMiklos Szeredi } 1421a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1422b4629fe2SChuck Lever 14231da177e4SLinus Torvalds /** 14241da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 14251f287bc4SRandy Dunlap * @m: root of mount tree 14261da177e4SLinus Torvalds * 14271da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 14281da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 14291da177e4SLinus Torvalds * busy. 14301da177e4SLinus Torvalds */ 1431909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 14321da177e4SLinus Torvalds { 1433909b0a88SAl Viro struct mount *mnt = real_mount(m); 143436341f64SRam Pai int actual_refs = 0; 143536341f64SRam Pai int minimum_refs = 0; 1436315fc83eSAl Viro struct mount *p; 1437909b0a88SAl Viro BUG_ON(!m); 14381da177e4SLinus Torvalds 1439b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1440719ea2fbSAl Viro lock_mount_hash(); 1441909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 144283adc753SAl Viro actual_refs += mnt_get_count(p); 14431da177e4SLinus Torvalds minimum_refs += 2; 14441da177e4SLinus Torvalds } 1445719ea2fbSAl Viro unlock_mount_hash(); 14461da177e4SLinus Torvalds 14471da177e4SLinus Torvalds if (actual_refs > minimum_refs) 14481da177e4SLinus Torvalds return 0; 1449e3474a8eSIan Kent 1450e3474a8eSIan Kent return 1; 14511da177e4SLinus Torvalds } 14521da177e4SLinus Torvalds 14531da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 14541da177e4SLinus Torvalds 14551da177e4SLinus Torvalds /** 14561da177e4SLinus Torvalds * may_umount - check if a mount point is busy 14571da177e4SLinus Torvalds * @mnt: root of mount 14581da177e4SLinus Torvalds * 14591da177e4SLinus Torvalds * This is called to check if a mount point has any 14601da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 14611da177e4SLinus Torvalds * mount has sub mounts this will return busy 14621da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 14631da177e4SLinus Torvalds * 14641da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 14651da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 14661da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 14671da177e4SLinus Torvalds */ 14681da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 14691da177e4SLinus Torvalds { 1470e3474a8eSIan Kent int ret = 1; 14718ad08d8aSAl Viro down_read(&namespace_sem); 1472719ea2fbSAl Viro lock_mount_hash(); 14731ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1474e3474a8eSIan Kent ret = 0; 1475719ea2fbSAl Viro unlock_mount_hash(); 14768ad08d8aSAl Viro up_read(&namespace_sem); 1477a05964f3SRam Pai return ret; 14781da177e4SLinus Torvalds } 14791da177e4SLinus Torvalds 14801da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14811da177e4SLinus Torvalds 148297216be0SAl Viro static void namespace_unlock(void) 14831da177e4SLinus Torvalds { 1484a3b3c562SEric W. Biederman struct hlist_head head; 148556cbb429SAl Viro struct hlist_node *p; 148656cbb429SAl Viro struct mount *m; 14874edbe133SAl Viro LIST_HEAD(list); 148897216be0SAl Viro 1489a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 14904edbe133SAl Viro list_splice_init(&ex_mountpoints, &list); 1491a3b3c562SEric W. Biederman 149297216be0SAl Viro up_write(&namespace_sem); 1493a3b3c562SEric W. Biederman 14944edbe133SAl Viro shrink_dentry_list(&list); 14954edbe133SAl Viro 1496a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 149797216be0SAl Viro return; 149897216be0SAl Viro 149922cb7405SNeilBrown synchronize_rcu_expedited(); 150048a066e7SAl Viro 150156cbb429SAl Viro hlist_for_each_entry_safe(m, p, &head, mnt_umount) { 150256cbb429SAl Viro hlist_del(&m->mnt_umount); 150356cbb429SAl Viro mntput(&m->mnt); 150456cbb429SAl Viro } 150570fbcdf4SRam Pai } 150670fbcdf4SRam Pai 150797216be0SAl Viro static inline void namespace_lock(void) 1508e3197d83SAl Viro { 150997216be0SAl Viro down_write(&namespace_sem); 1510e3197d83SAl Viro } 1511e3197d83SAl Viro 1512e819f152SEric W. Biederman enum umount_tree_flags { 1513e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1514e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1515e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1516e819f152SEric W. Biederman }; 1517f2d0a123SEric W. Biederman 1518f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1519f2d0a123SEric W. Biederman { 1520f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1521f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1522f2d0a123SEric W. Biederman return true; 1523f2d0a123SEric W. Biederman 1524f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1525f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1526f2d0a123SEric W. Biederman return true; 1527f2d0a123SEric W. Biederman 1528f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1529f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1530f2d0a123SEric W. Biederman * connected to mounted mounts. 1531f2d0a123SEric W. Biederman */ 1532f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1533f2d0a123SEric W. Biederman return true; 1534f2d0a123SEric W. Biederman 1535f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1536f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1537f2d0a123SEric W. Biederman return false; 1538f2d0a123SEric W. Biederman 1539f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1540f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1541f2d0a123SEric W. Biederman return false; 1542f2d0a123SEric W. Biederman 1543f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1544f2d0a123SEric W. Biederman return true; 1545f2d0a123SEric W. Biederman } 1546f2d0a123SEric W. Biederman 154799b7db7bSNick Piggin /* 154848a066e7SAl Viro * mount_lock must be held 154999b7db7bSNick Piggin * namespace_sem must be held for write 155099b7db7bSNick Piggin */ 1551e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 155270fbcdf4SRam Pai { 1553c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1554315fc83eSAl Viro struct mount *p; 155570fbcdf4SRam Pai 15565d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 15575d88457eSEric W. Biederman propagate_mount_unlock(mnt); 15585d88457eSEric W. Biederman 1559c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1560590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1561590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1562c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1563590ce4bcSEric W. Biederman } 1564c003b26fSEric W. Biederman 1565411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1566c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1567c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 156838129a13SAl Viro } 156970fbcdf4SRam Pai 1570c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1571e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 15727b8a53fdSAl Viro propagate_umount(&tmp_list); 1573a05964f3SRam Pai 1574c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1575d2921684SEric W. Biederman struct mnt_namespace *ns; 1576ce07d891SEric W. Biederman bool disconnect; 1577c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 15786776db3dSAl Viro list_del_init(&p->mnt_expire); 15791a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1580d2921684SEric W. Biederman ns = p->mnt_ns; 1581d2921684SEric W. Biederman if (ns) { 1582d2921684SEric W. Biederman ns->mounts--; 1583d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1584d2921684SEric W. Biederman } 158563d37a84SAl Viro p->mnt_ns = NULL; 1586e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 158748a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 158887b95ce0SAl Viro 1589f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1590676da58dSAl Viro if (mnt_has_parent(p)) { 159181b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1592ce07d891SEric W. Biederman if (!disconnect) { 1593ce07d891SEric W. Biederman /* Don't forget about p */ 1594ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1595ce07d891SEric W. Biederman } else { 15966a46c573SEric W. Biederman umount_mnt(p); 15977c4b93d8SAl Viro } 1598ce07d891SEric W. Biederman } 15990f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 160019a1c409SAl Viro if (disconnect) 160119a1c409SAl Viro hlist_add_head(&p->mnt_umount, &unmounted); 160238129a13SAl Viro } 16031da177e4SLinus Torvalds } 16041da177e4SLinus Torvalds 1605b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1606c35038beSAl Viro 16078d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 16088d0347f6SDavid Howells { 16098d0347f6SDavid Howells int ret = 0; 16108d0347f6SDavid Howells 16118d0347f6SDavid Howells down_write(&sb->s_umount); 16128d0347f6SDavid Howells if (!sb_rdonly(sb)) { 16138d0347f6SDavid Howells struct fs_context *fc; 16148d0347f6SDavid Howells 16158d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 16168d0347f6SDavid Howells SB_RDONLY); 16178d0347f6SDavid Howells if (IS_ERR(fc)) { 16188d0347f6SDavid Howells ret = PTR_ERR(fc); 16198d0347f6SDavid Howells } else { 16208d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 16218d0347f6SDavid Howells if (!ret) 16228d0347f6SDavid Howells ret = reconfigure_super(fc); 16238d0347f6SDavid Howells put_fs_context(fc); 16248d0347f6SDavid Howells } 16258d0347f6SDavid Howells } 16268d0347f6SDavid Howells up_write(&sb->s_umount); 16278d0347f6SDavid Howells return ret; 16288d0347f6SDavid Howells } 16298d0347f6SDavid Howells 16301ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 16311da177e4SLinus Torvalds { 16321ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 16331da177e4SLinus Torvalds int retval; 16341da177e4SLinus Torvalds 16351ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 16361da177e4SLinus Torvalds if (retval) 16371da177e4SLinus Torvalds return retval; 16381da177e4SLinus Torvalds 16391da177e4SLinus Torvalds /* 16401da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 16411da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 16421da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 16431da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 16441da177e4SLinus Torvalds */ 16451da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 16461ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 16471da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 16481da177e4SLinus Torvalds return -EINVAL; 16491da177e4SLinus Torvalds 1650b3e19d92SNick Piggin /* 1651b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1652b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1653b3e19d92SNick Piggin */ 1654719ea2fbSAl Viro lock_mount_hash(); 165583adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1656719ea2fbSAl Viro unlock_mount_hash(); 16571da177e4SLinus Torvalds return -EBUSY; 1658b3e19d92SNick Piggin } 1659719ea2fbSAl Viro unlock_mount_hash(); 16601da177e4SLinus Torvalds 1661863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 16621da177e4SLinus Torvalds return -EAGAIN; 16631da177e4SLinus Torvalds } 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds /* 16661da177e4SLinus Torvalds * If we may have to abort operations to get out of this 16671da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 16681da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 16691da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 16701da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 16711da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 16721da177e4SLinus Torvalds * about for the moment. 16731da177e4SLinus Torvalds */ 16741da177e4SLinus Torvalds 167542faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 167642faad99SAl Viro sb->s_op->umount_begin(sb); 167742faad99SAl Viro } 16781da177e4SLinus Torvalds 16791da177e4SLinus Torvalds /* 16801da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 16811da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 16821da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 16831da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 16841da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 16851da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 16861da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 16871da177e4SLinus Torvalds */ 16881ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 16891da177e4SLinus Torvalds /* 16901da177e4SLinus Torvalds * Special case for "unmounting" root ... 16911da177e4SLinus Torvalds * we just try to remount it readonly. 16921da177e4SLinus Torvalds */ 1693bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1694a1480dccSAndy Lutomirski return -EPERM; 16958d0347f6SDavid Howells return do_umount_root(sb); 16961da177e4SLinus Torvalds } 16971da177e4SLinus Torvalds 169897216be0SAl Viro namespace_lock(); 1699719ea2fbSAl Viro lock_mount_hash(); 17001da177e4SLinus Torvalds 170125d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 170225d202edSEric W. Biederman retval = -EINVAL; 170325d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 170425d202edSEric W. Biederman goto out; 170525d202edSEric W. Biederman 170625d202edSEric W. Biederman event++; 170748a066e7SAl Viro if (flags & MNT_DETACH) { 170848a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1709e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 171048a066e7SAl Viro retval = 0; 171148a066e7SAl Viro } else { 1712b54b9be7SAl Viro shrink_submounts(mnt); 17131da177e4SLinus Torvalds retval = -EBUSY; 171448a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 17151a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1716e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 17171da177e4SLinus Torvalds retval = 0; 17181da177e4SLinus Torvalds } 171948a066e7SAl Viro } 172025d202edSEric W. Biederman out: 1721719ea2fbSAl Viro unlock_mount_hash(); 1722e3197d83SAl Viro namespace_unlock(); 17231da177e4SLinus Torvalds return retval; 17241da177e4SLinus Torvalds } 17251da177e4SLinus Torvalds 17261da177e4SLinus Torvalds /* 172780b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 172880b5dce8SEric W. Biederman * 172980b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 173080b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 173180b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 173280b5dce8SEric W. Biederman * leaking them. 173380b5dce8SEric W. Biederman * 173480b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 173580b5dce8SEric W. Biederman */ 173680b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 173780b5dce8SEric W. Biederman { 173880b5dce8SEric W. Biederman struct mountpoint *mp; 173980b5dce8SEric W. Biederman struct mount *mnt; 174080b5dce8SEric W. Biederman 174180b5dce8SEric W. Biederman namespace_lock(); 17423895dbf8SEric W. Biederman lock_mount_hash(); 174380b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1744adc9b5c0SAl Viro if (!mp) 174580b5dce8SEric W. Biederman goto out_unlock; 174680b5dce8SEric W. Biederman 1747e06b933eSAndrey Ulanov event++; 174880b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 174980b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1750ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1751fe78fcc8SEric W. Biederman umount_mnt(mnt); 175256cbb429SAl Viro hlist_add_head(&mnt->mnt_umount, &unmounted); 1753ce07d891SEric W. Biederman } 1754e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 175580b5dce8SEric W. Biederman } 175680b5dce8SEric W. Biederman put_mountpoint(mp); 175780b5dce8SEric W. Biederman out_unlock: 17583895dbf8SEric W. Biederman unlock_mount_hash(); 175980b5dce8SEric W. Biederman namespace_unlock(); 176080b5dce8SEric W. Biederman } 176180b5dce8SEric W. Biederman 176280b5dce8SEric W. Biederman /* 17639b40bc90SAl Viro * Is the caller allowed to modify his namespace? 17649b40bc90SAl Viro */ 1765a5f85d78SAl Viro bool may_mount(void) 17669b40bc90SAl Viro { 17679b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 17689b40bc90SAl Viro } 17699b40bc90SAl Viro 1770f7e33bdbSJeff Layton static void warn_mandlock(void) 17719e8925b6SJeff Layton { 1772f7e33bdbSJeff Layton pr_warn_once("=======================================================\n" 1773f7e33bdbSJeff Layton "WARNING: The mand mount option has been deprecated and\n" 1774f7e33bdbSJeff Layton " and is ignored by this kernel. Remove the mand\n" 1775f7e33bdbSJeff Layton " option from the mount to silence this warning.\n" 1776f7e33bdbSJeff Layton "=======================================================\n"); 17779e8925b6SJeff Layton } 17789e8925b6SJeff Layton 177925ccd24fSChristoph Hellwig static int can_umount(const struct path *path, int flags) 17801da177e4SLinus Torvalds { 178125ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 17821da177e4SLinus Torvalds 17839b40bc90SAl Viro if (!may_mount()) 17849b40bc90SAl Viro return -EPERM; 178541525f56SChristoph Hellwig if (path->dentry != path->mnt->mnt_root) 178625ccd24fSChristoph Hellwig return -EINVAL; 1787143c8c91SAl Viro if (!check_mnt(mnt)) 178825ccd24fSChristoph Hellwig return -EINVAL; 178925d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 179025ccd24fSChristoph Hellwig return -EINVAL; 1791b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 179225ccd24fSChristoph Hellwig return -EPERM; 179325ccd24fSChristoph Hellwig return 0; 179425ccd24fSChristoph Hellwig } 17951da177e4SLinus Torvalds 1796a0a6df9aSAl Viro // caller is responsible for flags being sane 179725ccd24fSChristoph Hellwig int path_umount(struct path *path, int flags) 179825ccd24fSChristoph Hellwig { 179925ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 180025ccd24fSChristoph Hellwig int ret; 180125ccd24fSChristoph Hellwig 180225ccd24fSChristoph Hellwig ret = can_umount(path, flags); 180325ccd24fSChristoph Hellwig if (!ret) 180425ccd24fSChristoph Hellwig ret = do_umount(mnt, flags); 180525ccd24fSChristoph Hellwig 1806429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 180741525f56SChristoph Hellwig dput(path->dentry); 1808900148dcSAl Viro mntput_no_expire(mnt); 180925ccd24fSChristoph Hellwig return ret; 18101da177e4SLinus Torvalds } 18111da177e4SLinus Torvalds 181209267defSChristoph Hellwig static int ksys_umount(char __user *name, int flags) 181341525f56SChristoph Hellwig { 181441525f56SChristoph Hellwig int lookup_flags = LOOKUP_MOUNTPOINT; 181541525f56SChristoph Hellwig struct path path; 181641525f56SChristoph Hellwig int ret; 181741525f56SChristoph Hellwig 1818a0a6df9aSAl Viro // basic validity checks done first 1819a0a6df9aSAl Viro if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1820a0a6df9aSAl Viro return -EINVAL; 1821a0a6df9aSAl Viro 182241525f56SChristoph Hellwig if (!(flags & UMOUNT_NOFOLLOW)) 182341525f56SChristoph Hellwig lookup_flags |= LOOKUP_FOLLOW; 182441525f56SChristoph Hellwig ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); 182541525f56SChristoph Hellwig if (ret) 182641525f56SChristoph Hellwig return ret; 182741525f56SChristoph Hellwig return path_umount(&path, flags); 182841525f56SChristoph Hellwig } 182941525f56SChristoph Hellwig 18303a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 18313a18ef5cSDominik Brodowski { 18323a18ef5cSDominik Brodowski return ksys_umount(name, flags); 18333a18ef5cSDominik Brodowski } 18343a18ef5cSDominik Brodowski 18351da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 18361da177e4SLinus Torvalds 18371da177e4SLinus Torvalds /* 18381da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 18391da177e4SLinus Torvalds */ 1840bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 18411da177e4SLinus Torvalds { 18423a18ef5cSDominik Brodowski return ksys_umount(name, 0); 18431da177e4SLinus Torvalds } 18441da177e4SLinus Torvalds 18451da177e4SLinus Torvalds #endif 18461da177e4SLinus Torvalds 18474ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 18488823c079SEric W. Biederman { 18494ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1850e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1851e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 18524ce5d2b1SEric W. Biederman } 18534ce5d2b1SEric W. Biederman 1854213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 185558be2825SAl Viro { 185658be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 185758be2825SAl Viro } 185858be2825SAl Viro 1859303cc571SChristian Brauner struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) 1860303cc571SChristian Brauner { 1861303cc571SChristian Brauner return &mnt->ns; 1862303cc571SChristian Brauner } 1863303cc571SChristian Brauner 18644ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 18654ce5d2b1SEric W. Biederman { 18664ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 18674ce5d2b1SEric W. Biederman * mount namespace loop? 18684ce5d2b1SEric W. Biederman */ 18694ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 18704ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 18714ce5d2b1SEric W. Biederman return false; 18724ce5d2b1SEric W. Biederman 1873f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 18748823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 18758823c079SEric W. Biederman } 18768823c079SEric W. Biederman 187787129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 187836341f64SRam Pai int flag) 18791da177e4SLinus Torvalds { 188084d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 18811da177e4SLinus Torvalds 18824ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 18834ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 18844ce5d2b1SEric W. Biederman 18854ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1886be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 18879676f0c6SRam Pai 188836341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1889be34d1a3SDavid Howells if (IS_ERR(q)) 1890be34d1a3SDavid Howells return q; 1891be34d1a3SDavid Howells 1892a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 18931da177e4SLinus Torvalds 18941da177e4SLinus Torvalds p = mnt; 18956b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1896315fc83eSAl Viro struct mount *s; 18977ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 18981da177e4SLinus Torvalds continue; 18991da177e4SLinus Torvalds 1900909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 19014ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 19024ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1903df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1904df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1905df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1906df7342b2SEric W. Biederman goto out; 1907df7342b2SEric W. Biederman } else { 19084ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 19094ce5d2b1SEric W. Biederman continue; 19104ce5d2b1SEric W. Biederman } 1911df7342b2SEric W. Biederman } 19124ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 19134ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 19149676f0c6SRam Pai s = skip_mnt_tree(s); 19159676f0c6SRam Pai continue; 19169676f0c6SRam Pai } 19170714a533SAl Viro while (p != s->mnt_parent) { 19180714a533SAl Viro p = p->mnt_parent; 19190714a533SAl Viro q = q->mnt_parent; 19201da177e4SLinus Torvalds } 192187129cc0SAl Viro p = s; 192284d17192SAl Viro parent = q; 192387129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1924be34d1a3SDavid Howells if (IS_ERR(q)) 1925be34d1a3SDavid Howells goto out; 1926719ea2fbSAl Viro lock_mount_hash(); 19271a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 19281064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1929719ea2fbSAl Viro unlock_mount_hash(); 19301da177e4SLinus Torvalds } 19311da177e4SLinus Torvalds } 19321da177e4SLinus Torvalds return res; 1933be34d1a3SDavid Howells out: 19341da177e4SLinus Torvalds if (res) { 1935719ea2fbSAl Viro lock_mount_hash(); 1936e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1937719ea2fbSAl Viro unlock_mount_hash(); 19381da177e4SLinus Torvalds } 1939be34d1a3SDavid Howells return q; 19401da177e4SLinus Torvalds } 19411da177e4SLinus Torvalds 1942be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1943be34d1a3SDavid Howells 1944ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 19458aec0809SAl Viro { 1946cb338d06SAl Viro struct mount *tree; 194797216be0SAl Viro namespace_lock(); 1948cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1949cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1950cd4a4017SEric W. Biederman else 195187129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 195287129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1953328e6d90SAl Viro namespace_unlock(); 1954be34d1a3SDavid Howells if (IS_ERR(tree)) 195552e220d3SDan Carpenter return ERR_CAST(tree); 1956be34d1a3SDavid Howells return &tree->mnt; 19578aec0809SAl Viro } 19588aec0809SAl Viro 1959a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1960a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1961a07b2000SAl Viro 1962a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1963a07b2000SAl Viro { 1964a07b2000SAl Viro struct mnt_namespace *ns; 1965a07b2000SAl Viro namespace_lock(); 1966a07b2000SAl Viro lock_mount_hash(); 1967a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 196844dfd84aSDavid Howells if (ns) { 196944dfd84aSDavid Howells if (is_anon_ns(ns)) 1970a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 197144dfd84aSDavid Howells else 197244dfd84aSDavid Howells ns = NULL; 197344dfd84aSDavid Howells } 1974a07b2000SAl Viro unlock_mount_hash(); 1975a07b2000SAl Viro namespace_unlock(); 197644dfd84aSDavid Howells if (ns) 1977a07b2000SAl Viro free_mnt_ns(ns); 1978a07b2000SAl Viro } 1979a07b2000SAl Viro 19808aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 19818aec0809SAl Viro { 198297216be0SAl Viro namespace_lock(); 1983719ea2fbSAl Viro lock_mount_hash(); 19849c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1985719ea2fbSAl Viro unlock_mount_hash(); 19863ab6abeeSAl Viro namespace_unlock(); 19878aec0809SAl Viro } 19888aec0809SAl Viro 1989427215d8SMiklos Szeredi static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 1990427215d8SMiklos Szeredi { 1991427215d8SMiklos Szeredi struct mount *child; 1992427215d8SMiklos Szeredi 1993427215d8SMiklos Szeredi list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 1994427215d8SMiklos Szeredi if (!is_subdir(child->mnt_mountpoint, dentry)) 1995427215d8SMiklos Szeredi continue; 1996427215d8SMiklos Szeredi 1997427215d8SMiklos Szeredi if (child->mnt.mnt_flags & MNT_LOCKED) 1998427215d8SMiklos Szeredi return true; 1999427215d8SMiklos Szeredi } 2000427215d8SMiklos Szeredi return false; 2001427215d8SMiklos Szeredi } 2002427215d8SMiklos Szeredi 2003c771d683SMiklos Szeredi /** 2004c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 20051f287bc4SRandy Dunlap * @path: path to clone 2006c771d683SMiklos Szeredi * 20071f287bc4SRandy Dunlap * This creates a new vfsmount, which will be the clone of @path. The new mount 20081f287bc4SRandy Dunlap * will not be attached anywhere in the namespace and will be private (i.e. 20091f287bc4SRandy Dunlap * changes to the originating mount won't be propagated into this). 2010c771d683SMiklos Szeredi * 2011c771d683SMiklos Szeredi * Release with mntput(). 2012c771d683SMiklos Szeredi */ 2013ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 2014c771d683SMiklos Szeredi { 2015c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 2016c771d683SMiklos Szeredi struct mount *new_mnt; 2017c771d683SMiklos Szeredi 2018427215d8SMiklos Szeredi down_read(&namespace_sem); 2019c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 2020427215d8SMiklos Szeredi goto invalid; 2021427215d8SMiklos Szeredi 2022427215d8SMiklos Szeredi if (!check_mnt(old_mnt)) 2023427215d8SMiklos Szeredi goto invalid; 2024427215d8SMiklos Szeredi 2025427215d8SMiklos Szeredi if (has_locked_children(old_mnt, path->dentry)) 2026427215d8SMiklos Szeredi goto invalid; 2027c771d683SMiklos Szeredi 2028c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 2029427215d8SMiklos Szeredi up_read(&namespace_sem); 2030427215d8SMiklos Szeredi 2031c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 2032c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 2033c771d683SMiklos Szeredi 2034df820f8dSMiklos Szeredi /* Longterm mount to be removed by kern_unmount*() */ 2035df820f8dSMiklos Szeredi new_mnt->mnt_ns = MNT_NS_INTERNAL; 2036df820f8dSMiklos Szeredi 2037c771d683SMiklos Szeredi return &new_mnt->mnt; 2038427215d8SMiklos Szeredi 2039427215d8SMiklos Szeredi invalid: 2040427215d8SMiklos Szeredi up_read(&namespace_sem); 2041427215d8SMiklos Szeredi return ERR_PTR(-EINVAL); 2042c771d683SMiklos Szeredi } 2043c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 2044c771d683SMiklos Szeredi 20451f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 20461f707137SAl Viro struct vfsmount *root) 20471f707137SAl Viro { 20481a4eeaf2SAl Viro struct mount *mnt; 20491f707137SAl Viro int res = f(root, arg); 20501f707137SAl Viro if (res) 20511f707137SAl Viro return res; 20521a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 20531a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 20541f707137SAl Viro if (res) 20551f707137SAl Viro return res; 20561f707137SAl Viro } 20571f707137SAl Viro return 0; 20581f707137SAl Viro } 20591f707137SAl Viro 20603bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 20613bd045ccSAl Viro { 20623bd045ccSAl Viro struct mount *p; 20633bd045ccSAl Viro 20643bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 20653bd045ccSAl Viro int flags = p->mnt.mnt_flags; 20663bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 20673bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 20683bd045ccSAl Viro 20693bd045ccSAl Viro if (flags & MNT_READONLY) 20703bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 20713bd045ccSAl Viro 20723bd045ccSAl Viro if (flags & MNT_NODEV) 20733bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 20743bd045ccSAl Viro 20753bd045ccSAl Viro if (flags & MNT_NOSUID) 20763bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 20773bd045ccSAl Viro 20783bd045ccSAl Viro if (flags & MNT_NOEXEC) 20793bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 20803bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 20813bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 20823bd045ccSAl Viro flags |= MNT_LOCKED; 20833bd045ccSAl Viro p->mnt.mnt_flags = flags; 20843bd045ccSAl Viro } 20853bd045ccSAl Viro } 20863bd045ccSAl Viro 20874b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 2088719f5d7fSMiklos Szeredi { 2089315fc83eSAl Viro struct mount *p; 2090719f5d7fSMiklos Szeredi 2091909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 2092fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 20934b8b21f4SAl Viro mnt_release_group_id(p); 2094719f5d7fSMiklos Szeredi } 2095719f5d7fSMiklos Szeredi } 2096719f5d7fSMiklos Szeredi 20974b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 2098719f5d7fSMiklos Szeredi { 2099315fc83eSAl Viro struct mount *p; 2100719f5d7fSMiklos Szeredi 2101909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 2102fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 21034b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 2104719f5d7fSMiklos Szeredi if (err) { 21054b8b21f4SAl Viro cleanup_group_ids(mnt, p); 2106719f5d7fSMiklos Szeredi return err; 2107719f5d7fSMiklos Szeredi } 2108719f5d7fSMiklos Szeredi } 2109719f5d7fSMiklos Szeredi } 2110719f5d7fSMiklos Szeredi 2111719f5d7fSMiklos Szeredi return 0; 2112719f5d7fSMiklos Szeredi } 2113719f5d7fSMiklos Szeredi 2114d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 2115d2921684SEric W. Biederman { 2116d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 2117124f75f8SAl Viro unsigned int mounts = 0; 2118d2921684SEric W. Biederman struct mount *p; 2119d2921684SEric W. Biederman 2120124f75f8SAl Viro if (ns->mounts >= max) 2121124f75f8SAl Viro return -ENOSPC; 2122124f75f8SAl Viro max -= ns->mounts; 2123124f75f8SAl Viro if (ns->pending_mounts >= max) 2124124f75f8SAl Viro return -ENOSPC; 2125124f75f8SAl Viro max -= ns->pending_mounts; 2126124f75f8SAl Viro 2127d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 2128d2921684SEric W. Biederman mounts++; 2129d2921684SEric W. Biederman 2130124f75f8SAl Viro if (mounts > max) 2131d2921684SEric W. Biederman return -ENOSPC; 2132d2921684SEric W. Biederman 2133124f75f8SAl Viro ns->pending_mounts += mounts; 2134d2921684SEric W. Biederman return 0; 2135d2921684SEric W. Biederman } 2136d2921684SEric W. Biederman 2137b90fa9aeSRam Pai /* 2138b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 2139b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 214021444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 214121444403SRam Pai * store the parent mount and mountpoint dentry. 214221444403SRam Pai * (done when source_mnt is moved) 2143b90fa9aeSRam Pai * 2144b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 2145b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 21469676f0c6SRam Pai * --------------------------------------------------------------------------- 2147b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 21489676f0c6SRam Pai * |************************************************************************** 21499676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21509676f0c6SRam Pai * | dest | | | | | 21519676f0c6SRam Pai * | | | | | | | 21529676f0c6SRam Pai * | v | | | | | 21539676f0c6SRam Pai * |************************************************************************** 21549676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 21555afe0022SRam Pai * | | | | | | 21569676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 21579676f0c6SRam Pai * *************************************************************************** 2158b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 2159b90fa9aeSRam Pai * destination mount. 2160b90fa9aeSRam Pai * 2161b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2162b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2163b90fa9aeSRam Pai * the peer group of the source mount. 2164b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2165b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2166b90fa9aeSRam Pai * mount. 21675afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 21685afe0022SRam Pai * of the destination mount and the cloned mount is made slave 21695afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 21705afe0022SRam Pai * is marked as 'shared and slave'. 21715afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 21725afe0022SRam Pai * source mount. 21735afe0022SRam Pai * 21749676f0c6SRam Pai * --------------------------------------------------------------------------- 217521444403SRam Pai * | MOVE MOUNT OPERATION | 21769676f0c6SRam Pai * |************************************************************************** 21779676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21789676f0c6SRam Pai * | dest | | | | | 21799676f0c6SRam Pai * | | | | | | | 21809676f0c6SRam Pai * | v | | | | | 21819676f0c6SRam Pai * |************************************************************************** 21829676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 21835afe0022SRam Pai * | | | | | | 21849676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 21859676f0c6SRam Pai * *************************************************************************** 21865afe0022SRam Pai * 21875afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 21885afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 218921444403SRam Pai * (+*) the mount is moved to the destination. 21905afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 21915afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 21925afe0022SRam Pai * the mount is marked as 'shared and slave'. 21935afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2194b90fa9aeSRam Pai * 2195b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2196b90fa9aeSRam Pai * applied to each mount in the tree. 2197b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2198b90fa9aeSRam Pai * in allocations. 2199b90fa9aeSRam Pai */ 22000fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 220184d17192SAl Viro struct mount *dest_mnt, 220284d17192SAl Viro struct mountpoint *dest_mp, 22032763d119SAl Viro bool moving) 2204b90fa9aeSRam Pai { 22053bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 220638129a13SAl Viro HLIST_HEAD(tree_list); 2207d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 22081064f874SEric W. Biederman struct mountpoint *smp; 2209315fc83eSAl Viro struct mount *child, *p; 221038129a13SAl Viro struct hlist_node *n; 2211719f5d7fSMiklos Szeredi int err; 2212b90fa9aeSRam Pai 22131064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 22141064f874SEric W. Biederman * to be tucked under other mounts. 22151064f874SEric W. Biederman */ 22161064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 22171064f874SEric W. Biederman if (IS_ERR(smp)) 22181064f874SEric W. Biederman return PTR_ERR(smp); 22191064f874SEric W. Biederman 2220d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 22212763d119SAl Viro if (!moving) { 2222d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2223d2921684SEric W. Biederman if (err) 2224d2921684SEric W. Biederman goto out; 2225d2921684SEric W. Biederman } 2226d2921684SEric W. Biederman 2227fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 22280fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2229719f5d7fSMiklos Szeredi if (err) 2230719f5d7fSMiklos Szeredi goto out; 223184d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2232f2ebb3a9SAl Viro lock_mount_hash(); 2233719f5d7fSMiklos Szeredi if (err) 2234719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2235909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 22360f0afb1dSAl Viro set_mnt_shared(p); 22370b1b901bSAl Viro } else { 22380b1b901bSAl Viro lock_mount_hash(); 2239b90fa9aeSRam Pai } 22402763d119SAl Viro if (moving) { 22412763d119SAl Viro unhash_mnt(source_mnt); 224284d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2243143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 224421444403SRam Pai } else { 224544dfd84aSDavid Howells if (source_mnt->mnt_ns) { 224644dfd84aSDavid Howells /* move from anon - the caller will destroy */ 224744dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 224844dfd84aSDavid Howells } 224984d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 22501064f874SEric W. Biederman commit_tree(source_mnt); 225121444403SRam Pai } 2252b90fa9aeSRam Pai 225338129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 22541d6a32acSAl Viro struct mount *q; 225538129a13SAl Viro hlist_del_init(&child->mnt_hash); 22561064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 22571d6a32acSAl Viro child->mnt_mountpoint); 22581064f874SEric W. Biederman if (q) 22591064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 22603bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 22613bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 22623bd045ccSAl Viro lock_mnt_tree(child); 2263d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 22641064f874SEric W. Biederman commit_tree(child); 2265b90fa9aeSRam Pai } 22661064f874SEric W. Biederman put_mountpoint(smp); 2267719ea2fbSAl Viro unlock_mount_hash(); 226899b7db7bSNick Piggin 2269b90fa9aeSRam Pai return 0; 2270719f5d7fSMiklos Szeredi 2271719f5d7fSMiklos Szeredi out_cleanup_ids: 2272f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2273f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2274d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2275e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2276f2ebb3a9SAl Viro } 2277f2ebb3a9SAl Viro unlock_mount_hash(); 22780fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2279719f5d7fSMiklos Szeredi out: 2280d2921684SEric W. Biederman ns->pending_mounts = 0; 22811064f874SEric W. Biederman 22821064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 22831064f874SEric W. Biederman put_mountpoint(smp); 22841064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 22851064f874SEric W. Biederman 2286719f5d7fSMiklos Szeredi return err; 2287b90fa9aeSRam Pai } 2288b90fa9aeSRam Pai 228984d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2290b12cea91SAl Viro { 2291b12cea91SAl Viro struct vfsmount *mnt; 229284d17192SAl Viro struct dentry *dentry = path->dentry; 2293b12cea91SAl Viro retry: 22945955102cSAl Viro inode_lock(dentry->d_inode); 229584d17192SAl Viro if (unlikely(cant_mount(dentry))) { 22965955102cSAl Viro inode_unlock(dentry->d_inode); 229784d17192SAl Viro return ERR_PTR(-ENOENT); 2298b12cea91SAl Viro } 229997216be0SAl Viro namespace_lock(); 2300b12cea91SAl Viro mnt = lookup_mnt(path); 230184d17192SAl Viro if (likely(!mnt)) { 23023895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 230384d17192SAl Viro if (IS_ERR(mp)) { 230497216be0SAl Viro namespace_unlock(); 23055955102cSAl Viro inode_unlock(dentry->d_inode); 230684d17192SAl Viro return mp; 230784d17192SAl Viro } 230884d17192SAl Viro return mp; 230984d17192SAl Viro } 231097216be0SAl Viro namespace_unlock(); 23115955102cSAl Viro inode_unlock(path->dentry->d_inode); 2312b12cea91SAl Viro path_put(path); 2313b12cea91SAl Viro path->mnt = mnt; 231484d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2315b12cea91SAl Viro goto retry; 2316b12cea91SAl Viro } 2317b12cea91SAl Viro 231884d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2319b12cea91SAl Viro { 232084d17192SAl Viro struct dentry *dentry = where->m_dentry; 23213895dbf8SEric W. Biederman 23223895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 232384d17192SAl Viro put_mountpoint(where); 23243895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 23253895dbf8SEric W. Biederman 2326328e6d90SAl Viro namespace_unlock(); 23275955102cSAl Viro inode_unlock(dentry->d_inode); 2328b12cea91SAl Viro } 2329b12cea91SAl Viro 233084d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 23311da177e4SLinus Torvalds { 2332e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 23331da177e4SLinus Torvalds return -EINVAL; 23341da177e4SLinus Torvalds 2335e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2336e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 23371da177e4SLinus Torvalds return -ENOTDIR; 23381da177e4SLinus Torvalds 23392763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 23401da177e4SLinus Torvalds } 23411da177e4SLinus Torvalds 23421da177e4SLinus Torvalds /* 23437a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 23447a2e8a8fSValerie Aurora */ 23457a2e8a8fSValerie Aurora 2346e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 23477a2e8a8fSValerie Aurora { 2348e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 23497a2e8a8fSValerie Aurora 23507a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 23517a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 23527a2e8a8fSValerie Aurora return 0; 23537a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 23547a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 23557a2e8a8fSValerie Aurora return 0; 23567a2e8a8fSValerie Aurora return type; 23577a2e8a8fSValerie Aurora } 23587a2e8a8fSValerie Aurora 23597a2e8a8fSValerie Aurora /* 236007b20889SRam Pai * recursively change the type of the mountpoint. 236107b20889SRam Pai */ 2362e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 236307b20889SRam Pai { 2364315fc83eSAl Viro struct mount *m; 23654b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2366e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 23677a2e8a8fSValerie Aurora int type; 2368719f5d7fSMiklos Szeredi int err = 0; 236907b20889SRam Pai 23702d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 237107b20889SRam Pai return -EINVAL; 237207b20889SRam Pai 2373e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 23747a2e8a8fSValerie Aurora if (!type) 23757a2e8a8fSValerie Aurora return -EINVAL; 23767a2e8a8fSValerie Aurora 237797216be0SAl Viro namespace_lock(); 2378719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2379719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2380719f5d7fSMiklos Szeredi if (err) 2381719f5d7fSMiklos Szeredi goto out_unlock; 2382719f5d7fSMiklos Szeredi } 2383719f5d7fSMiklos Szeredi 2384719ea2fbSAl Viro lock_mount_hash(); 2385909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 23860f0afb1dSAl Viro change_mnt_propagation(m, type); 2387719ea2fbSAl Viro unlock_mount_hash(); 2388719f5d7fSMiklos Szeredi 2389719f5d7fSMiklos Szeredi out_unlock: 239097216be0SAl Viro namespace_unlock(); 2391719f5d7fSMiklos Szeredi return err; 239207b20889SRam Pai } 239307b20889SRam Pai 2394a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2395a07b2000SAl Viro { 2396a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2397a07b2000SAl Viro 2398a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2399a07b2000SAl Viro return mnt; 2400a07b2000SAl Viro 2401a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2402a07b2000SAl Viro return mnt; 2403a07b2000SAl Viro 2404a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2405a07b2000SAl Viro return mnt; 2406a07b2000SAl Viro 2407a07b2000SAl Viro if (recurse) 2408a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2409a07b2000SAl Viro else 2410a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2411a07b2000SAl Viro 2412a07b2000SAl Viro if (!IS_ERR(mnt)) 2413a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2414a07b2000SAl Viro 2415a07b2000SAl Viro return mnt; 2416a07b2000SAl Viro } 2417a07b2000SAl Viro 241807b20889SRam Pai /* 24191da177e4SLinus Torvalds * do loopback mount. 24201da177e4SLinus Torvalds */ 2421808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 24222dafe1c4SEric Sandeen int recurse) 24231da177e4SLinus Torvalds { 24242d92ab3cSAl Viro struct path old_path; 2425a07b2000SAl Viro struct mount *mnt = NULL, *parent; 242684d17192SAl Viro struct mountpoint *mp; 242757eccb83SAl Viro int err; 24281da177e4SLinus Torvalds if (!old_name || !*old_name) 24291da177e4SLinus Torvalds return -EINVAL; 2430815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 24311da177e4SLinus Torvalds if (err) 24321da177e4SLinus Torvalds return err; 24331da177e4SLinus Torvalds 24348823c079SEric W. Biederman err = -EINVAL; 24354ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 24368823c079SEric W. Biederman goto out; 24378823c079SEric W. Biederman 243884d17192SAl Viro mp = lock_mount(path); 2439a07b2000SAl Viro if (IS_ERR(mp)) { 244084d17192SAl Viro err = PTR_ERR(mp); 24419676f0c6SRam Pai goto out; 2442a07b2000SAl Viro } 24439676f0c6SRam Pai 244484d17192SAl Viro parent = real_mount(path->mnt); 2445e149ed2bSAl Viro if (!check_mnt(parent)) 2446e149ed2bSAl Viro goto out2; 2447e149ed2bSAl Viro 2448a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2449be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2450be34d1a3SDavid Howells err = PTR_ERR(mnt); 2451e9c5d8a5SAndrey Vagin goto out2; 2452be34d1a3SDavid Howells } 2453ccd48bc7SAl Viro 245484d17192SAl Viro err = graft_tree(mnt, parent, mp); 24551da177e4SLinus Torvalds if (err) { 2456719ea2fbSAl Viro lock_mount_hash(); 2457e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2458719ea2fbSAl Viro unlock_mount_hash(); 24595b83d2c5SRam Pai } 2460b12cea91SAl Viro out2: 246184d17192SAl Viro unlock_mount(mp); 2462ccd48bc7SAl Viro out: 24632d92ab3cSAl Viro path_put(&old_path); 24641da177e4SLinus Torvalds return err; 24651da177e4SLinus Torvalds } 24661da177e4SLinus Torvalds 2467a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2468a07b2000SAl Viro { 2469a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2470a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2471a07b2000SAl Viro struct mount *mnt, *p; 2472a07b2000SAl Viro struct file *file; 2473a07b2000SAl Viro 2474a07b2000SAl Viro if (IS_ERR(ns)) 2475a07b2000SAl Viro return ERR_CAST(ns); 2476a07b2000SAl Viro 2477a07b2000SAl Viro namespace_lock(); 2478a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2479a07b2000SAl Viro if (IS_ERR(mnt)) { 2480a07b2000SAl Viro namespace_unlock(); 2481a07b2000SAl Viro free_mnt_ns(ns); 2482a07b2000SAl Viro return ERR_CAST(mnt); 2483a07b2000SAl Viro } 2484a07b2000SAl Viro 2485a07b2000SAl Viro lock_mount_hash(); 2486a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2487a07b2000SAl Viro p->mnt_ns = ns; 2488a07b2000SAl Viro ns->mounts++; 2489a07b2000SAl Viro } 2490a07b2000SAl Viro ns->root = mnt; 2491a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2492a07b2000SAl Viro mntget(&mnt->mnt); 2493a07b2000SAl Viro unlock_mount_hash(); 2494a07b2000SAl Viro namespace_unlock(); 2495a07b2000SAl Viro 2496a07b2000SAl Viro mntput(path->mnt); 2497a07b2000SAl Viro path->mnt = &mnt->mnt; 2498a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2499a07b2000SAl Viro if (IS_ERR(file)) 2500a07b2000SAl Viro dissolve_on_fput(path->mnt); 2501a07b2000SAl Viro else 2502a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2503a07b2000SAl Viro return file; 2504a07b2000SAl Viro } 2505a07b2000SAl Viro 25062658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2507a07b2000SAl Viro { 2508a07b2000SAl Viro struct file *file; 2509a07b2000SAl Viro struct path path; 2510a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2511a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2512a07b2000SAl Viro int error; 2513a07b2000SAl Viro int fd; 2514a07b2000SAl Viro 2515a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2516a07b2000SAl Viro 2517a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2518a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2519a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2520a07b2000SAl Viro return -EINVAL; 2521a07b2000SAl Viro 2522a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2523a07b2000SAl Viro return -EINVAL; 2524a07b2000SAl Viro 2525a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2526a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2527a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2528a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2529a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2530a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2531a07b2000SAl Viro 2532a07b2000SAl Viro if (detached && !may_mount()) 2533a07b2000SAl Viro return -EPERM; 2534a07b2000SAl Viro 2535a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2536a07b2000SAl Viro if (fd < 0) 2537a07b2000SAl Viro return fd; 2538a07b2000SAl Viro 2539a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2540a07b2000SAl Viro if (unlikely(error)) { 2541a07b2000SAl Viro file = ERR_PTR(error); 2542a07b2000SAl Viro } else { 2543a07b2000SAl Viro if (detached) 2544a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2545a07b2000SAl Viro else 2546a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2547a07b2000SAl Viro path_put(&path); 2548a07b2000SAl Viro } 2549a07b2000SAl Viro if (IS_ERR(file)) { 2550a07b2000SAl Viro put_unused_fd(fd); 2551a07b2000SAl Viro return PTR_ERR(file); 2552a07b2000SAl Viro } 2553a07b2000SAl Viro fd_install(fd, file); 2554a07b2000SAl Viro return fd; 2555a07b2000SAl Viro } 2556a07b2000SAl Viro 255743f5e655SDavid Howells /* 255843f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 255943f5e655SDavid Howells * 256043f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 256143f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 256243f5e655SDavid Howells */ 256343f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 25642e4b7fcdSDave Hansen { 256543f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 25662e4b7fcdSDave Hansen 256743f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 256843f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 256943f5e655SDavid Howells return false; 257043f5e655SDavid Howells 257143f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 257243f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 257343f5e655SDavid Howells return false; 257443f5e655SDavid Howells 257543f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 257643f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 257743f5e655SDavid Howells return false; 257843f5e655SDavid Howells 257943f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 258043f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 258143f5e655SDavid Howells return false; 258243f5e655SDavid Howells 258343f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 258443f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 258543f5e655SDavid Howells return false; 258643f5e655SDavid Howells 258743f5e655SDavid Howells return true; 258843f5e655SDavid Howells } 258943f5e655SDavid Howells 259043f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 259143f5e655SDavid Howells { 259243f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 259343f5e655SDavid Howells 259443f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 25952e4b7fcdSDave Hansen return 0; 25962e4b7fcdSDave Hansen 25972e4b7fcdSDave Hansen if (readonly_request) 259843f5e655SDavid Howells return mnt_make_readonly(mnt); 259943f5e655SDavid Howells 260068847c94SChristian Brauner mnt->mnt.mnt_flags &= ~MNT_READONLY; 260168847c94SChristian Brauner return 0; 260243f5e655SDavid Howells } 260343f5e655SDavid Howells 260443f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 260543f5e655SDavid Howells { 260643f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 260743f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 260843f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 260943f5e655SDavid Howells } 261043f5e655SDavid Howells 2611f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2612f8b92ba6SDeepa Dinamani { 2613f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2614f8b92ba6SDeepa Dinamani 2615f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2616a128b054SAnthony Iliopoulos (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) && 2617f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2618f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2619f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2620f8b92ba6SDeepa Dinamani struct tm tm; 2621f8b92ba6SDeepa Dinamani 2622f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2623f8b92ba6SDeepa Dinamani 26240ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 26250ecee669SEric Biggers sb->s_type->name, 26260ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 26270ecee669SEric Biggers mntpath, 2628f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2629f8b92ba6SDeepa Dinamani 2630f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2631a128b054SAnthony Iliopoulos sb->s_iflags |= SB_I_TS_EXPIRY_WARNED; 2632f8b92ba6SDeepa Dinamani } 2633f8b92ba6SDeepa Dinamani } 2634f8b92ba6SDeepa Dinamani 263543f5e655SDavid Howells /* 263643f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 263743f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 263843f5e655SDavid Howells * to mount(2). 263943f5e655SDavid Howells */ 264043f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 264143f5e655SDavid Howells { 264243f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 264343f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 264443f5e655SDavid Howells int ret; 264543f5e655SDavid Howells 264643f5e655SDavid Howells if (!check_mnt(mnt)) 264743f5e655SDavid Howells return -EINVAL; 264843f5e655SDavid Howells 264943f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 265043f5e655SDavid Howells return -EINVAL; 265143f5e655SDavid Howells 265243f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 265343f5e655SDavid Howells return -EPERM; 265443f5e655SDavid Howells 2655e58ace1aSChristian Brauner /* 2656e58ace1aSChristian Brauner * We're only checking whether the superblock is read-only not 2657e58ace1aSChristian Brauner * changing it, so only take down_read(&sb->s_umount). 2658e58ace1aSChristian Brauner */ 2659e58ace1aSChristian Brauner down_read(&sb->s_umount); 266068847c94SChristian Brauner lock_mount_hash(); 266143f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 266243f5e655SDavid Howells if (ret == 0) 266343f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 266468847c94SChristian Brauner unlock_mount_hash(); 2665e58ace1aSChristian Brauner up_read(&sb->s_umount); 2666f8b92ba6SDeepa Dinamani 2667f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2668f8b92ba6SDeepa Dinamani 266943f5e655SDavid Howells return ret; 26702e4b7fcdSDave Hansen } 26712e4b7fcdSDave Hansen 26721da177e4SLinus Torvalds /* 26731da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 26741da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 26751da177e4SLinus Torvalds * on it - tough luck. 26761da177e4SLinus Torvalds */ 2677e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2678e462ec50SDavid Howells int mnt_flags, void *data) 26791da177e4SLinus Torvalds { 26801da177e4SLinus Torvalds int err; 26812d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2682143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 26838d0347f6SDavid Howells struct fs_context *fc; 26841da177e4SLinus Torvalds 2685143c8c91SAl Viro if (!check_mnt(mnt)) 26861da177e4SLinus Torvalds return -EINVAL; 26871da177e4SLinus Torvalds 26882d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 26891da177e4SLinus Torvalds return -EINVAL; 26901da177e4SLinus Torvalds 269143f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 269207b64558SEric W. Biederman return -EPERM; 26939566d674SEric W. Biederman 26948d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 26958d0347f6SDavid Howells if (IS_ERR(fc)) 26968d0347f6SDavid Howells return PTR_ERR(fc); 2697ff36fe2cSEric Paris 2698b330966fSMiklos Szeredi fc->oldapi = true; 26998d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 27008d0347f6SDavid Howells if (!err) { 27011da177e4SLinus Torvalds down_write(&sb->s_umount); 270257eccb83SAl Viro err = -EPERM; 270343f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 27048d0347f6SDavid Howells err = reconfigure_super(fc); 270568847c94SChristian Brauner if (!err) { 270668847c94SChristian Brauner lock_mount_hash(); 270743f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 270868847c94SChristian Brauner unlock_mount_hash(); 270968847c94SChristian Brauner } 27100e55a7ccSDan Williams } 27116339dab8SAl Viro up_write(&sb->s_umount); 27128d0347f6SDavid Howells } 2713f8b92ba6SDeepa Dinamani 2714f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2715f8b92ba6SDeepa Dinamani 27168d0347f6SDavid Howells put_fs_context(fc); 27171da177e4SLinus Torvalds return err; 27181da177e4SLinus Torvalds } 27191da177e4SLinus Torvalds 2720cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 27219676f0c6SRam Pai { 2722315fc83eSAl Viro struct mount *p; 2723909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2724fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 27259676f0c6SRam Pai return 1; 27269676f0c6SRam Pai } 27279676f0c6SRam Pai return 0; 27289676f0c6SRam Pai } 27299676f0c6SRam Pai 273044dfd84aSDavid Howells /* 273144dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 273244dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 273344dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 273444dfd84aSDavid Howells * cycles to be made. 273544dfd84aSDavid Howells */ 273644dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 273744dfd84aSDavid Howells { 273844dfd84aSDavid Howells struct mount *p; 273944dfd84aSDavid Howells bool ret = false; 274044dfd84aSDavid Howells 274144dfd84aSDavid Howells lock_mount_hash(); 274244dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 274344dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 274444dfd84aSDavid Howells goto out; 274544dfd84aSDavid Howells 274644dfd84aSDavid Howells ret = true; 274744dfd84aSDavid Howells out: 274844dfd84aSDavid Howells unlock_mount_hash(); 274944dfd84aSDavid Howells return ret; 275044dfd84aSDavid Howells } 275144dfd84aSDavid Howells 27529ffb14efSPavel Tikhomirov static int do_set_group(struct path *from_path, struct path *to_path) 27539ffb14efSPavel Tikhomirov { 27549ffb14efSPavel Tikhomirov struct mount *from, *to; 27559ffb14efSPavel Tikhomirov int err; 27569ffb14efSPavel Tikhomirov 27579ffb14efSPavel Tikhomirov from = real_mount(from_path->mnt); 27589ffb14efSPavel Tikhomirov to = real_mount(to_path->mnt); 27599ffb14efSPavel Tikhomirov 27609ffb14efSPavel Tikhomirov namespace_lock(); 27619ffb14efSPavel Tikhomirov 27629ffb14efSPavel Tikhomirov err = -EINVAL; 27639ffb14efSPavel Tikhomirov /* To and From must be mounted */ 27649ffb14efSPavel Tikhomirov if (!is_mounted(&from->mnt)) 27659ffb14efSPavel Tikhomirov goto out; 27669ffb14efSPavel Tikhomirov if (!is_mounted(&to->mnt)) 27679ffb14efSPavel Tikhomirov goto out; 27689ffb14efSPavel Tikhomirov 27699ffb14efSPavel Tikhomirov err = -EPERM; 27709ffb14efSPavel Tikhomirov /* We should be allowed to modify mount namespaces of both mounts */ 27719ffb14efSPavel Tikhomirov if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27729ffb14efSPavel Tikhomirov goto out; 27739ffb14efSPavel Tikhomirov if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27749ffb14efSPavel Tikhomirov goto out; 27759ffb14efSPavel Tikhomirov 27769ffb14efSPavel Tikhomirov err = -EINVAL; 27779ffb14efSPavel Tikhomirov /* To and From paths should be mount roots */ 27789ffb14efSPavel Tikhomirov if (from_path->dentry != from_path->mnt->mnt_root) 27799ffb14efSPavel Tikhomirov goto out; 27809ffb14efSPavel Tikhomirov if (to_path->dentry != to_path->mnt->mnt_root) 27819ffb14efSPavel Tikhomirov goto out; 27829ffb14efSPavel Tikhomirov 27839ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed across same superblock */ 27849ffb14efSPavel Tikhomirov if (from->mnt.mnt_sb != to->mnt.mnt_sb) 27859ffb14efSPavel Tikhomirov goto out; 27869ffb14efSPavel Tikhomirov 27879ffb14efSPavel Tikhomirov /* From mount root should be wider than To mount root */ 27889ffb14efSPavel Tikhomirov if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) 27899ffb14efSPavel Tikhomirov goto out; 27909ffb14efSPavel Tikhomirov 27919ffb14efSPavel Tikhomirov /* From mount should not have locked children in place of To's root */ 27929ffb14efSPavel Tikhomirov if (has_locked_children(from, to->mnt.mnt_root)) 27939ffb14efSPavel Tikhomirov goto out; 27949ffb14efSPavel Tikhomirov 27959ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed on private mounts */ 27969ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) 27979ffb14efSPavel Tikhomirov goto out; 27989ffb14efSPavel Tikhomirov 27999ffb14efSPavel Tikhomirov /* From should not be private */ 28009ffb14efSPavel Tikhomirov if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) 28019ffb14efSPavel Tikhomirov goto out; 28029ffb14efSPavel Tikhomirov 28039ffb14efSPavel Tikhomirov if (IS_MNT_SLAVE(from)) { 28049ffb14efSPavel Tikhomirov struct mount *m = from->mnt_master; 28059ffb14efSPavel Tikhomirov 28069ffb14efSPavel Tikhomirov list_add(&to->mnt_slave, &m->mnt_slave_list); 28079ffb14efSPavel Tikhomirov to->mnt_master = m; 28089ffb14efSPavel Tikhomirov } 28099ffb14efSPavel Tikhomirov 28109ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(from)) { 28119ffb14efSPavel Tikhomirov to->mnt_group_id = from->mnt_group_id; 28129ffb14efSPavel Tikhomirov list_add(&to->mnt_share, &from->mnt_share); 28139ffb14efSPavel Tikhomirov lock_mount_hash(); 28149ffb14efSPavel Tikhomirov set_mnt_shared(to); 28159ffb14efSPavel Tikhomirov unlock_mount_hash(); 28169ffb14efSPavel Tikhomirov } 28179ffb14efSPavel Tikhomirov 28189ffb14efSPavel Tikhomirov err = 0; 28199ffb14efSPavel Tikhomirov out: 28209ffb14efSPavel Tikhomirov namespace_unlock(); 28219ffb14efSPavel Tikhomirov return err; 28229ffb14efSPavel Tikhomirov } 28239ffb14efSPavel Tikhomirov 28242db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 28251da177e4SLinus Torvalds { 282644dfd84aSDavid Howells struct mnt_namespace *ns; 2827676da58dSAl Viro struct mount *p; 28280fb54e50SAl Viro struct mount *old; 28292763d119SAl Viro struct mount *parent; 28302763d119SAl Viro struct mountpoint *mp, *old_mp; 283157eccb83SAl Viro int err; 283244dfd84aSDavid Howells bool attached; 28331da177e4SLinus Torvalds 28342db154b3SDavid Howells mp = lock_mount(new_path); 283584d17192SAl Viro if (IS_ERR(mp)) 28362db154b3SDavid Howells return PTR_ERR(mp); 2837cc53ce53SDavid Howells 28382db154b3SDavid Howells old = real_mount(old_path->mnt); 28392db154b3SDavid Howells p = real_mount(new_path->mnt); 28402763d119SAl Viro parent = old->mnt_parent; 284144dfd84aSDavid Howells attached = mnt_has_parent(old); 28422763d119SAl Viro old_mp = old->mnt_mp; 284344dfd84aSDavid Howells ns = old->mnt_ns; 2844143c8c91SAl Viro 28451da177e4SLinus Torvalds err = -EINVAL; 284644dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 284744dfd84aSDavid Howells if (!check_mnt(p)) 28482db154b3SDavid Howells goto out; 28491da177e4SLinus Torvalds 2850570d7a98SEric Biggers /* The thing moved must be mounted... */ 2851570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 285244dfd84aSDavid Howells goto out; 285344dfd84aSDavid Howells 2854570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2855570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 28562db154b3SDavid Howells goto out; 28571da177e4SLinus Torvalds 28582db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 28592db154b3SDavid Howells goto out; 28602db154b3SDavid Howells 28612db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 28622db154b3SDavid Howells goto out; 28632db154b3SDavid Howells 28642db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 28652db154b3SDavid Howells d_is_dir(old_path->dentry)) 28662db154b3SDavid Howells goto out; 286721444403SRam Pai /* 286821444403SRam Pai * Don't move a mount residing in a shared parent. 286921444403SRam Pai */ 28702763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 28712db154b3SDavid Howells goto out; 28729676f0c6SRam Pai /* 28739676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 28749676f0c6SRam Pai * mount which is shared. 28759676f0c6SRam Pai */ 2876fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 28772db154b3SDavid Howells goto out; 28781da177e4SLinus Torvalds err = -ELOOP; 287944dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 288044dfd84aSDavid Howells goto out; 2881fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2882676da58dSAl Viro if (p == old) 28832db154b3SDavid Howells goto out; 28841da177e4SLinus Torvalds 28852db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 28862763d119SAl Viro attached); 28874ac91378SJan Blunck if (err) 28882db154b3SDavid Howells goto out; 28891da177e4SLinus Torvalds 28901da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 28911da177e4SLinus Torvalds * automatically */ 28926776db3dSAl Viro list_del_init(&old->mnt_expire); 28932763d119SAl Viro if (attached) 28942763d119SAl Viro put_mountpoint(old_mp); 28951da177e4SLinus Torvalds out: 28962db154b3SDavid Howells unlock_mount(mp); 289744dfd84aSDavid Howells if (!err) { 28982763d119SAl Viro if (attached) 28992763d119SAl Viro mntput_no_expire(parent); 29002763d119SAl Viro else 290144dfd84aSDavid Howells free_mnt_ns(ns); 290244dfd84aSDavid Howells } 29032db154b3SDavid Howells return err; 29042db154b3SDavid Howells } 29052db154b3SDavid Howells 29062db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 29072db154b3SDavid Howells { 29082db154b3SDavid Howells struct path old_path; 29092db154b3SDavid Howells int err; 29102db154b3SDavid Howells 29112db154b3SDavid Howells if (!old_name || !*old_name) 29122db154b3SDavid Howells return -EINVAL; 29132db154b3SDavid Howells 29142db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 29152db154b3SDavid Howells if (err) 29162db154b3SDavid Howells return err; 29172db154b3SDavid Howells 29182db154b3SDavid Howells err = do_move_mount(&old_path, path); 29192d92ab3cSAl Viro path_put(&old_path); 29201da177e4SLinus Torvalds return err; 29211da177e4SLinus Torvalds } 29221da177e4SLinus Torvalds 29239d412a43SAl Viro /* 29249d412a43SAl Viro * add a mount into a namespace's mount tree 29259d412a43SAl Viro */ 29268f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 29271e2d8464SAl Viro const struct path *path, int mnt_flags) 29289d412a43SAl Viro { 29298f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 29309d412a43SAl Viro 2931f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 29329d412a43SAl Viro 293384d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2934156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2935156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 29368f11538eSAl Viro return -EINVAL; 2937156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 293884d17192SAl Viro if (!parent->mnt_ns) 29398f11538eSAl Viro return -EINVAL; 2940156cacb1SAl Viro } 29419d412a43SAl Viro 29429d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 294395bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 29449d412a43SAl Viro path->mnt->mnt_root == path->dentry) 29458f11538eSAl Viro return -EBUSY; 29469d412a43SAl Viro 2947e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 29488f11538eSAl Viro return -EINVAL; 29499d412a43SAl Viro 295095bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 29518f11538eSAl Viro return graft_tree(newmnt, parent, mp); 29529d412a43SAl Viro } 2953b1e75df4SAl Viro 2954132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2955132e4608SDavid Howells 2956132e4608SDavid Howells /* 2957132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2958132e4608SDavid Howells * be added to the namespace tree. 2959132e4608SDavid Howells */ 2960132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2961132e4608SDavid Howells unsigned int mnt_flags) 2962132e4608SDavid Howells { 2963132e4608SDavid Howells struct vfsmount *mnt; 29648f11538eSAl Viro struct mountpoint *mp; 2965132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2966132e4608SDavid Howells int error; 2967132e4608SDavid Howells 2968c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2969c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2970c9ce29edSAl Viro error = -EPERM; 2971c9ce29edSAl Viro 2972c9ce29edSAl Viro if (unlikely(error)) { 2973c9ce29edSAl Viro fc_drop_locked(fc); 2974c9ce29edSAl Viro return error; 2975132e4608SDavid Howells } 2976132e4608SDavid Howells 2977132e4608SDavid Howells up_write(&sb->s_umount); 2978132e4608SDavid Howells 2979132e4608SDavid Howells mnt = vfs_create_mount(fc); 2980132e4608SDavid Howells if (IS_ERR(mnt)) 2981132e4608SDavid Howells return PTR_ERR(mnt); 2982132e4608SDavid Howells 2983f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 2984f8b92ba6SDeepa Dinamani 29858f11538eSAl Viro mp = lock_mount(mountpoint); 29868f11538eSAl Viro if (IS_ERR(mp)) { 29878f11538eSAl Viro mntput(mnt); 29888f11538eSAl Viro return PTR_ERR(mp); 29898f11538eSAl Viro } 29908f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 29918f11538eSAl Viro unlock_mount(mp); 29920ecee669SEric Biggers if (error < 0) 29930ecee669SEric Biggers mntput(mnt); 2994f8b92ba6SDeepa Dinamani return error; 2995f8b92ba6SDeepa Dinamani } 2996f8b92ba6SDeepa Dinamani 29971da177e4SLinus Torvalds /* 29981da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 29991da177e4SLinus Torvalds * namespace's tree 30001da177e4SLinus Torvalds */ 3001e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 3002808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 30031da177e4SLinus Torvalds { 30040c55cfc4SEric W. Biederman struct file_system_type *type; 3005a0c9a8b8SAl Viro struct fs_context *fc; 3006a0c9a8b8SAl Viro const char *subtype = NULL; 3007a0c9a8b8SAl Viro int err = 0; 30081da177e4SLinus Torvalds 30090c55cfc4SEric W. Biederman if (!fstype) 30101da177e4SLinus Torvalds return -EINVAL; 30111da177e4SLinus Torvalds 30120c55cfc4SEric W. Biederman type = get_fs_type(fstype); 30130c55cfc4SEric W. Biederman if (!type) 30140c55cfc4SEric W. Biederman return -ENODEV; 30150c55cfc4SEric W. Biederman 3016a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 3017a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 3018a0c9a8b8SAl Viro if (subtype) { 3019a0c9a8b8SAl Viro subtype++; 3020a0c9a8b8SAl Viro if (!*subtype) { 30210c55cfc4SEric W. Biederman put_filesystem(type); 3022a0c9a8b8SAl Viro return -EINVAL; 3023a0c9a8b8SAl Viro } 3024a0c9a8b8SAl Viro } 30258654df4eSEric W. Biederman } 30268654df4eSEric W. Biederman 3027a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 3028a0c9a8b8SAl Viro put_filesystem(type); 3029a0c9a8b8SAl Viro if (IS_ERR(fc)) 3030a0c9a8b8SAl Viro return PTR_ERR(fc); 3031a0c9a8b8SAl Viro 30323e1aeb00SDavid Howells if (subtype) 30333e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 30343e1aeb00SDavid Howells subtype, strlen(subtype)); 30353e1aeb00SDavid Howells if (!err && name) 30363e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 3037a0c9a8b8SAl Viro if (!err) 3038a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 3039c3aabf07SAl Viro if (!err && !mount_capable(fc)) 3040c3aabf07SAl Viro err = -EPERM; 3041a0c9a8b8SAl Viro if (!err) 3042a0c9a8b8SAl Viro err = vfs_get_tree(fc); 3043132e4608SDavid Howells if (!err) 3044132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 3045a0c9a8b8SAl Viro 3046a0c9a8b8SAl Viro put_fs_context(fc); 304715f9a3f3SAl Viro return err; 30481da177e4SLinus Torvalds } 30491da177e4SLinus Torvalds 30501e2d8464SAl Viro int finish_automount(struct vfsmount *m, const struct path *path) 305119a167afSAl Viro { 305226df6034SAl Viro struct dentry *dentry = path->dentry; 30538f11538eSAl Viro struct mountpoint *mp; 305425e195aaSAl Viro struct mount *mnt; 305519a167afSAl Viro int err; 305625e195aaSAl Viro 305725e195aaSAl Viro if (!m) 305825e195aaSAl Viro return 0; 305925e195aaSAl Viro if (IS_ERR(m)) 306025e195aaSAl Viro return PTR_ERR(m); 306125e195aaSAl Viro 306225e195aaSAl Viro mnt = real_mount(m); 306319a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 306419a167afSAl Viro * expired before we get a chance to add it 306519a167afSAl Viro */ 30666776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 306719a167afSAl Viro 306819a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 306926df6034SAl Viro m->mnt_root == dentry) { 3070b1e75df4SAl Viro err = -ELOOP; 307126df6034SAl Viro goto discard; 307219a167afSAl Viro } 307319a167afSAl Viro 307426df6034SAl Viro /* 307526df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 307626df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 307726df6034SAl Viro * got", not "try to mount it on top". 307826df6034SAl Viro */ 307926df6034SAl Viro inode_lock(dentry->d_inode); 308026df6034SAl Viro namespace_lock(); 308126df6034SAl Viro if (unlikely(cant_mount(dentry))) { 308226df6034SAl Viro err = -ENOENT; 308326df6034SAl Viro goto discard_locked; 308426df6034SAl Viro } 308526df6034SAl Viro rcu_read_lock(); 308626df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 308726df6034SAl Viro rcu_read_unlock(); 308826df6034SAl Viro err = 0; 308926df6034SAl Viro goto discard_locked; 309026df6034SAl Viro } 309126df6034SAl Viro rcu_read_unlock(); 309226df6034SAl Viro mp = get_mountpoint(dentry); 30938f11538eSAl Viro if (IS_ERR(mp)) { 30948f11538eSAl Viro err = PTR_ERR(mp); 309526df6034SAl Viro goto discard_locked; 30968f11538eSAl Viro } 309726df6034SAl Viro 30988f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 30998f11538eSAl Viro unlock_mount(mp); 310026df6034SAl Viro if (unlikely(err)) 310126df6034SAl Viro goto discard; 310226df6034SAl Viro mntput(m); 3103b1e75df4SAl Viro return 0; 310426df6034SAl Viro 310526df6034SAl Viro discard_locked: 310626df6034SAl Viro namespace_unlock(); 310726df6034SAl Viro inode_unlock(dentry->d_inode); 310826df6034SAl Viro discard: 3109b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 31106776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 311197216be0SAl Viro namespace_lock(); 31126776db3dSAl Viro list_del_init(&mnt->mnt_expire); 311397216be0SAl Viro namespace_unlock(); 311419a167afSAl Viro } 3115b1e75df4SAl Viro mntput(m); 3116b1e75df4SAl Viro mntput(m); 311719a167afSAl Viro return err; 311819a167afSAl Viro } 311919a167afSAl Viro 3120ea5b778aSDavid Howells /** 3121ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 3122ea5b778aSDavid Howells * @mnt: The mount to list. 3123ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 3124ea5b778aSDavid Howells */ 3125ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 3126ea5b778aSDavid Howells { 312797216be0SAl Viro namespace_lock(); 3128ea5b778aSDavid Howells 31296776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 3130ea5b778aSDavid Howells 313197216be0SAl Viro namespace_unlock(); 3132ea5b778aSDavid Howells } 3133ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 3134ea5b778aSDavid Howells 3135ea5b778aSDavid Howells /* 31361da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 31371da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 31381da177e4SLinus Torvalds * here 31391da177e4SLinus Torvalds */ 31401da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 31411da177e4SLinus Torvalds { 3142761d5c38SAl Viro struct mount *mnt, *next; 31431da177e4SLinus Torvalds LIST_HEAD(graveyard); 31441da177e4SLinus Torvalds 31451da177e4SLinus Torvalds if (list_empty(mounts)) 31461da177e4SLinus Torvalds return; 31471da177e4SLinus Torvalds 314897216be0SAl Viro namespace_lock(); 3149719ea2fbSAl Viro lock_mount_hash(); 31501da177e4SLinus Torvalds 31511da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 31521da177e4SLinus Torvalds * following criteria: 31531da177e4SLinus Torvalds * - only referenced by its parent vfsmount 31541da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 31551da177e4SLinus Torvalds * cleared by mntput()) 31561da177e4SLinus Torvalds */ 31576776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 3158863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 31591ab59738SAl Viro propagate_mount_busy(mnt, 1)) 31601da177e4SLinus Torvalds continue; 31616776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 31621da177e4SLinus Torvalds } 3163bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 31646776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 3165143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 3166e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3167bcc5c7d2SAl Viro } 3168719ea2fbSAl Viro unlock_mount_hash(); 31693ab6abeeSAl Viro namespace_unlock(); 31701da177e4SLinus Torvalds } 31711da177e4SLinus Torvalds 31721da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 31731da177e4SLinus Torvalds 31741da177e4SLinus Torvalds /* 31755528f911STrond Myklebust * Ripoff of 'select_parent()' 31765528f911STrond Myklebust * 31775528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 31785528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 31795528f911STrond Myklebust */ 3180692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 31815528f911STrond Myklebust { 3182692afc31SAl Viro struct mount *this_parent = parent; 31835528f911STrond Myklebust struct list_head *next; 31845528f911STrond Myklebust int found = 0; 31855528f911STrond Myklebust 31865528f911STrond Myklebust repeat: 31876b41d536SAl Viro next = this_parent->mnt_mounts.next; 31885528f911STrond Myklebust resume: 31896b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 31905528f911STrond Myklebust struct list_head *tmp = next; 31916b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 31925528f911STrond Myklebust 31935528f911STrond Myklebust next = tmp->next; 3194692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 31955528f911STrond Myklebust continue; 31965528f911STrond Myklebust /* 31975528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 31985528f911STrond Myklebust */ 31996b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 32005528f911STrond Myklebust this_parent = mnt; 32015528f911STrond Myklebust goto repeat; 32025528f911STrond Myklebust } 32035528f911STrond Myklebust 32041ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 32056776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 32065528f911STrond Myklebust found++; 32075528f911STrond Myklebust } 32085528f911STrond Myklebust } 32095528f911STrond Myklebust /* 32105528f911STrond Myklebust * All done at this level ... ascend and resume the search 32115528f911STrond Myklebust */ 32125528f911STrond Myklebust if (this_parent != parent) { 32136b41d536SAl Viro next = this_parent->mnt_child.next; 32140714a533SAl Viro this_parent = this_parent->mnt_parent; 32155528f911STrond Myklebust goto resume; 32165528f911STrond Myklebust } 32175528f911STrond Myklebust return found; 32185528f911STrond Myklebust } 32195528f911STrond Myklebust 32205528f911STrond Myklebust /* 32215528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 32225528f911STrond Myklebust * submounts of a specific parent mountpoint 322399b7db7bSNick Piggin * 322448a066e7SAl Viro * mount_lock must be held for write 32255528f911STrond Myklebust */ 3226b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 32275528f911STrond Myklebust { 32285528f911STrond Myklebust LIST_HEAD(graveyard); 3229761d5c38SAl Viro struct mount *m; 32305528f911STrond Myklebust 32315528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 3232c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3233bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3234761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 32356776db3dSAl Viro mnt_expire); 3236143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3237e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3238bcc5c7d2SAl Viro } 3239bcc5c7d2SAl Viro } 32405528f911STrond Myklebust } 32415528f911STrond Myklebust 3242028abd92SChristoph Hellwig static void *copy_mount_options(const void __user * data) 32431da177e4SLinus Torvalds { 3244b40ef869SAl Viro char *copy; 3245d563d678SCatalin Marinas unsigned left, offset; 32461da177e4SLinus Torvalds 32471da177e4SLinus Torvalds if (!data) 3248b40ef869SAl Viro return NULL; 32491da177e4SLinus Torvalds 3250b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3251b40ef869SAl Viro if (!copy) 3252b40ef869SAl Viro return ERR_PTR(-ENOMEM); 32531da177e4SLinus Torvalds 3254d563d678SCatalin Marinas left = copy_from_user(copy, data, PAGE_SIZE); 32551da177e4SLinus Torvalds 3256d563d678SCatalin Marinas /* 3257d563d678SCatalin Marinas * Not all architectures have an exact copy_from_user(). Resort to 3258d563d678SCatalin Marinas * byte at a time. 3259d563d678SCatalin Marinas */ 3260d563d678SCatalin Marinas offset = PAGE_SIZE - left; 3261d563d678SCatalin Marinas while (left) { 3262d563d678SCatalin Marinas char c; 3263d563d678SCatalin Marinas if (get_user(c, (const char __user *)data + offset)) 3264d563d678SCatalin Marinas break; 3265d563d678SCatalin Marinas copy[offset] = c; 3266d563d678SCatalin Marinas left--; 3267d563d678SCatalin Marinas offset++; 3268d563d678SCatalin Marinas } 3269d563d678SCatalin Marinas 3270d563d678SCatalin Marinas if (left == PAGE_SIZE) { 3271b40ef869SAl Viro kfree(copy); 3272b40ef869SAl Viro return ERR_PTR(-EFAULT); 32731da177e4SLinus Torvalds } 3274d563d678SCatalin Marinas 3275b40ef869SAl Viro return copy; 32761da177e4SLinus Torvalds } 32771da177e4SLinus Torvalds 3278028abd92SChristoph Hellwig static char *copy_mount_string(const void __user *data) 3279eca6f534SVegard Nossum { 3280fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3281eca6f534SVegard Nossum } 3282eca6f534SVegard Nossum 32831da177e4SLinus Torvalds /* 32841da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 32851da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 32861da177e4SLinus Torvalds * 32871da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 32881da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 32891da177e4SLinus Torvalds * information (or be NULL). 32901da177e4SLinus Torvalds * 32911da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 32921da177e4SLinus Torvalds * When the flags word was introduced its top half was required 32931da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 32941da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 32951da177e4SLinus Torvalds * and must be discarded. 32961da177e4SLinus Torvalds */ 3297c60166f0SChristoph Hellwig int path_mount(const char *dev_name, struct path *path, 3298808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 32991da177e4SLinus Torvalds { 3300e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 3301a1e6aaa3SChristoph Hellwig int ret; 33021da177e4SLinus Torvalds 33031da177e4SLinus Torvalds /* Discard magic */ 33041da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 33051da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 33061da177e4SLinus Torvalds 33071da177e4SLinus Torvalds /* Basic sanity checks */ 33081da177e4SLinus Torvalds if (data_page) 33091da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 33101da177e4SLinus Torvalds 3311e462ec50SDavid Howells if (flags & MS_NOUSER) 3312e462ec50SDavid Howells return -EINVAL; 3313e462ec50SDavid Howells 3314a1e6aaa3SChristoph Hellwig ret = security_sb_mount(dev_name, path, type_page, flags, data_page); 3315a1e6aaa3SChristoph Hellwig if (ret) 3316a1e6aaa3SChristoph Hellwig return ret; 3317a1e6aaa3SChristoph Hellwig if (!may_mount()) 3318a1e6aaa3SChristoph Hellwig return -EPERM; 3319f7e33bdbSJeff Layton if (flags & SB_MANDLOCK) 3320f7e33bdbSJeff Layton warn_mandlock(); 3321a27ab9f2STetsuo Handa 3322613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3323613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 33240a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 33250a1c01c9SMatthew Garrett 33261da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 33271da177e4SLinus Torvalds if (flags & MS_NOSUID) 33281da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 33291da177e4SLinus Torvalds if (flags & MS_NODEV) 33301da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 33311da177e4SLinus Torvalds if (flags & MS_NOEXEC) 33321da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3333fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3334fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3335fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3336fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3337d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3338d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3339a9e5b732SDavid Howells if (flags & MS_RDONLY) 33402e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3341dab741e0SMattias Nissler if (flags & MS_NOSYMFOLLOW) 3342dab741e0SMattias Nissler mnt_flags |= MNT_NOSYMFOLLOW; 3343fc33a7bbSChristoph Hellwig 3344ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3345ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3346ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3347ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3348ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3349a1e6aaa3SChristoph Hellwig mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; 3350ffbc6f0eSEric W. Biederman } 3351ffbc6f0eSEric W. Biederman 3352e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3353e462ec50SDavid Howells SB_SYNCHRONOUS | 3354e462ec50SDavid Howells SB_MANDLOCK | 3355e462ec50SDavid Howells SB_DIRSYNC | 3356e462ec50SDavid Howells SB_SILENT | 3357917086ffSMimi Zohar SB_POSIXACL | 3358d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3359917086ffSMimi Zohar SB_I_VERSION); 33601da177e4SLinus Torvalds 336143f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 3362a1e6aaa3SChristoph Hellwig return do_reconfigure_mnt(path, mnt_flags); 3363a1e6aaa3SChristoph Hellwig if (flags & MS_REMOUNT) 3364a1e6aaa3SChristoph Hellwig return do_remount(path, flags, sb_flags, mnt_flags, data_page); 3365a1e6aaa3SChristoph Hellwig if (flags & MS_BIND) 3366a1e6aaa3SChristoph Hellwig return do_loopback(path, dev_name, flags & MS_REC); 3367a1e6aaa3SChristoph Hellwig if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 3368a1e6aaa3SChristoph Hellwig return do_change_type(path, flags); 3369a1e6aaa3SChristoph Hellwig if (flags & MS_MOVE) 3370a1e6aaa3SChristoph Hellwig return do_move_mount_old(path, dev_name); 3371a1e6aaa3SChristoph Hellwig 3372a1e6aaa3SChristoph Hellwig return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, 33731da177e4SLinus Torvalds data_page); 3374a1e6aaa3SChristoph Hellwig } 3375a1e6aaa3SChristoph Hellwig 3376a1e6aaa3SChristoph Hellwig long do_mount(const char *dev_name, const char __user *dir_name, 3377a1e6aaa3SChristoph Hellwig const char *type_page, unsigned long flags, void *data_page) 3378a1e6aaa3SChristoph Hellwig { 3379a1e6aaa3SChristoph Hellwig struct path path; 3380a1e6aaa3SChristoph Hellwig int ret; 3381a1e6aaa3SChristoph Hellwig 3382a1e6aaa3SChristoph Hellwig ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3383a1e6aaa3SChristoph Hellwig if (ret) 3384a1e6aaa3SChristoph Hellwig return ret; 3385a1e6aaa3SChristoph Hellwig ret = path_mount(dev_name, &path, type_page, flags, data_page); 33862d92ab3cSAl Viro path_put(&path); 3387a1e6aaa3SChristoph Hellwig return ret; 33881da177e4SLinus Torvalds } 33891da177e4SLinus Torvalds 3390537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3391537f7ccbSEric W. Biederman { 3392537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3393537f7ccbSEric W. Biederman } 3394537f7ccbSEric W. Biederman 3395537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3396537f7ccbSEric W. Biederman { 3397537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3398537f7ccbSEric W. Biederman } 3399537f7ccbSEric W. Biederman 3400771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3401771b1371SEric W. Biederman { 340274e83122SAl Viro if (!is_anon_ns(ns)) 34036344c433SAl Viro ns_free_inum(&ns->ns); 3404537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3405771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3406771b1371SEric W. Biederman kfree(ns); 3407771b1371SEric W. Biederman } 3408771b1371SEric W. Biederman 34098823c079SEric W. Biederman /* 34108823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 34118823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 34128823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 34138823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 34148823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 34158823c079SEric W. Biederman */ 34168823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 34178823c079SEric W. Biederman 341874e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3419cf8d2c11STrond Myklebust { 3420cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3421537f7ccbSEric W. Biederman struct ucounts *ucounts; 342298f842e6SEric W. Biederman int ret; 3423cf8d2c11STrond Myklebust 3424537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3425537f7ccbSEric W. Biederman if (!ucounts) 3426df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3427537f7ccbSEric W. Biederman 342830acd0bdSVasily Averin new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); 3429537f7ccbSEric W. Biederman if (!new_ns) { 3430537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3431cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3432537f7ccbSEric W. Biederman } 343374e83122SAl Viro if (!anon) { 34346344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 343598f842e6SEric W. Biederman if (ret) { 343698f842e6SEric W. Biederman kfree(new_ns); 3437537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 343898f842e6SEric W. Biederman return ERR_PTR(ret); 343998f842e6SEric W. Biederman } 344074e83122SAl Viro } 344133c42940SAl Viro new_ns->ns.ops = &mntns_operations; 344274e83122SAl Viro if (!anon) 34438823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 34441a7b8969SKirill Tkhai refcount_set(&new_ns->ns.count, 1); 3445cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3446cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 34479f6c61f9SMiklos Szeredi spin_lock_init(&new_ns->ns_lock); 3448771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3449537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3450cf8d2c11STrond Myklebust return new_ns; 3451cf8d2c11STrond Myklebust } 3452cf8d2c11STrond Myklebust 34530766f788SEmese Revfy __latent_entropy 34549559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 34559559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 34561da177e4SLinus Torvalds { 34576b3286edSKirill Korotaev struct mnt_namespace *new_ns; 34587f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3459315fc83eSAl Viro struct mount *p, *q; 34609559f689SAl Viro struct mount *old; 3461cb338d06SAl Viro struct mount *new; 34627a472ef4SEric W. Biederman int copy_flags; 34631da177e4SLinus Torvalds 34649559f689SAl Viro BUG_ON(!ns); 34659559f689SAl Viro 34669559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 34679559f689SAl Viro get_mnt_ns(ns); 34689559f689SAl Viro return ns; 34699559f689SAl Viro } 34709559f689SAl Viro 34719559f689SAl Viro old = ns->root; 34729559f689SAl Viro 347374e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3474cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3475cf8d2c11STrond Myklebust return new_ns; 34761da177e4SLinus Torvalds 347797216be0SAl Viro namespace_lock(); 34781da177e4SLinus Torvalds /* First pass: copy the tree topology */ 34794ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 34809559f689SAl Viro if (user_ns != ns->user_ns) 34813bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 34827a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3483be34d1a3SDavid Howells if (IS_ERR(new)) { 3484328e6d90SAl Viro namespace_unlock(); 3485771b1371SEric W. Biederman free_mnt_ns(new_ns); 3486be34d1a3SDavid Howells return ERR_CAST(new); 34871da177e4SLinus Torvalds } 34883bd045ccSAl Viro if (user_ns != ns->user_ns) { 34893bd045ccSAl Viro lock_mount_hash(); 34903bd045ccSAl Viro lock_mnt_tree(new); 34913bd045ccSAl Viro unlock_mount_hash(); 34923bd045ccSAl Viro } 3493be08d6d2SAl Viro new_ns->root = new; 34941a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 34951da177e4SLinus Torvalds 34961da177e4SLinus Torvalds /* 34971da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 34981da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 34991da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 35001da177e4SLinus Torvalds */ 3501909b0a88SAl Viro p = old; 3502cb338d06SAl Viro q = new; 35031da177e4SLinus Torvalds while (p) { 3504143c8c91SAl Viro q->mnt_ns = new_ns; 3505d2921684SEric W. Biederman new_ns->mounts++; 35069559f689SAl Viro if (new_fs) { 35079559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 35089559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3509315fc83eSAl Viro rootmnt = &p->mnt; 35101da177e4SLinus Torvalds } 35119559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 35129559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3513315fc83eSAl Viro pwdmnt = &p->mnt; 35141da177e4SLinus Torvalds } 35151da177e4SLinus Torvalds } 3516909b0a88SAl Viro p = next_mnt(p, old); 3517909b0a88SAl Viro q = next_mnt(q, new); 35184ce5d2b1SEric W. Biederman if (!q) 35194ce5d2b1SEric W. Biederman break; 352061d8e426SAl Viro // an mntns binding we'd skipped? 35214ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 352261d8e426SAl Viro p = next_mnt(skip_mnt_tree(p), old); 35231da177e4SLinus Torvalds } 3524328e6d90SAl Viro namespace_unlock(); 35251da177e4SLinus Torvalds 35261da177e4SLinus Torvalds if (rootmnt) 3527f03c6599SAl Viro mntput(rootmnt); 35281da177e4SLinus Torvalds if (pwdmnt) 3529f03c6599SAl Viro mntput(pwdmnt); 35301da177e4SLinus Torvalds 3531741a2951SJANAK DESAI return new_ns; 3532741a2951SJANAK DESAI } 3533741a2951SJANAK DESAI 353474e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3535cf8d2c11STrond Myklebust { 35361a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3537ea441d11SAl Viro struct mnt_namespace *ns; 3538d31da0f0SAl Viro struct super_block *s; 3539ea441d11SAl Viro struct path path; 3540ea441d11SAl Viro int err; 3541ea441d11SAl Viro 354274e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 354374e83122SAl Viro if (IS_ERR(ns)) { 354474e83122SAl Viro mntput(m); 3545ea441d11SAl Viro return ERR_CAST(ns); 354674e83122SAl Viro } 354774e83122SAl Viro mnt->mnt_ns = ns; 354874e83122SAl Viro ns->root = mnt; 354974e83122SAl Viro ns->mounts++; 355074e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3551ea441d11SAl Viro 355274e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3553ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3554ea441d11SAl Viro 3555ea441d11SAl Viro put_mnt_ns(ns); 3556ea441d11SAl Viro 3557ea441d11SAl Viro if (err) 3558ea441d11SAl Viro return ERR_PTR(err); 3559ea441d11SAl Viro 3560ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3561d31da0f0SAl Viro s = path.mnt->mnt_sb; 3562d31da0f0SAl Viro atomic_inc(&s->s_active); 3563ea441d11SAl Viro mntput(path.mnt); 3564ea441d11SAl Viro /* lock the sucker */ 3565d31da0f0SAl Viro down_write(&s->s_umount); 3566ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3567ea441d11SAl Viro return path.dentry; 3568ea441d11SAl Viro } 3569ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3570ea441d11SAl Viro 3571cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3572cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 35731da177e4SLinus Torvalds { 3574eca6f534SVegard Nossum int ret; 3575eca6f534SVegard Nossum char *kernel_type; 3576eca6f534SVegard Nossum char *kernel_dev; 3577b40ef869SAl Viro void *options; 35781da177e4SLinus Torvalds 3579b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3580b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3581b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3582eca6f534SVegard Nossum goto out_type; 35831da177e4SLinus Torvalds 3584b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3585b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3586b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3587eca6f534SVegard Nossum goto out_dev; 35881da177e4SLinus Torvalds 3589b40ef869SAl Viro options = copy_mount_options(data); 3590b40ef869SAl Viro ret = PTR_ERR(options); 3591b40ef869SAl Viro if (IS_ERR(options)) 3592eca6f534SVegard Nossum goto out_data; 35931da177e4SLinus Torvalds 3594b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3595eca6f534SVegard Nossum 3596b40ef869SAl Viro kfree(options); 3597eca6f534SVegard Nossum out_data: 3598eca6f534SVegard Nossum kfree(kernel_dev); 3599eca6f534SVegard Nossum out_dev: 3600eca6f534SVegard Nossum kfree(kernel_type); 3601eca6f534SVegard Nossum out_type: 3602eca6f534SVegard Nossum return ret; 36031da177e4SLinus Torvalds } 36041da177e4SLinus Torvalds 36055b490500SChristian Brauner #define FSMOUNT_VALID_FLAGS \ 36065b490500SChristian Brauner (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ 3607dd8b477fSChristian Brauner MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ 3608dd8b477fSChristian Brauner MOUNT_ATTR_NOSYMFOLLOW) 36095b490500SChristian Brauner 36109caccd41SChristian Brauner #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) 36112a186721SChristian Brauner 36122a186721SChristian Brauner #define MOUNT_SETATTR_PROPAGATION_FLAGS \ 36132a186721SChristian Brauner (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) 36142a186721SChristian Brauner 36155b490500SChristian Brauner static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) 36165b490500SChristian Brauner { 36175b490500SChristian Brauner unsigned int mnt_flags = 0; 36185b490500SChristian Brauner 36195b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_RDONLY) 36205b490500SChristian Brauner mnt_flags |= MNT_READONLY; 36215b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOSUID) 36225b490500SChristian Brauner mnt_flags |= MNT_NOSUID; 36235b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODEV) 36245b490500SChristian Brauner mnt_flags |= MNT_NODEV; 36255b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOEXEC) 36265b490500SChristian Brauner mnt_flags |= MNT_NOEXEC; 36275b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODIRATIME) 36285b490500SChristian Brauner mnt_flags |= MNT_NODIRATIME; 3629dd8b477fSChristian Brauner if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) 3630dd8b477fSChristian Brauner mnt_flags |= MNT_NOSYMFOLLOW; 36315b490500SChristian Brauner 36325b490500SChristian Brauner return mnt_flags; 36335b490500SChristian Brauner } 36345b490500SChristian Brauner 36351da177e4SLinus Torvalds /* 363693766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 363793766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 363893766fbdSDavid Howells */ 363993766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 364093766fbdSDavid Howells unsigned int, attr_flags) 364193766fbdSDavid Howells { 364293766fbdSDavid Howells struct mnt_namespace *ns; 364393766fbdSDavid Howells struct fs_context *fc; 364493766fbdSDavid Howells struct file *file; 364593766fbdSDavid Howells struct path newmount; 364693766fbdSDavid Howells struct mount *mnt; 364793766fbdSDavid Howells struct fd f; 364893766fbdSDavid Howells unsigned int mnt_flags = 0; 364993766fbdSDavid Howells long ret; 365093766fbdSDavid Howells 365193766fbdSDavid Howells if (!may_mount()) 365293766fbdSDavid Howells return -EPERM; 365393766fbdSDavid Howells 365493766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 365593766fbdSDavid Howells return -EINVAL; 365693766fbdSDavid Howells 36575b490500SChristian Brauner if (attr_flags & ~FSMOUNT_VALID_FLAGS) 365893766fbdSDavid Howells return -EINVAL; 365993766fbdSDavid Howells 36605b490500SChristian Brauner mnt_flags = attr_flags_to_mnt_flags(attr_flags); 366193766fbdSDavid Howells 366293766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 366393766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 366493766fbdSDavid Howells break; 366593766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 366693766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 366793766fbdSDavid Howells break; 366893766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 366993766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 367093766fbdSDavid Howells break; 367193766fbdSDavid Howells default: 367293766fbdSDavid Howells return -EINVAL; 367393766fbdSDavid Howells } 367493766fbdSDavid Howells 367593766fbdSDavid Howells f = fdget(fs_fd); 367693766fbdSDavid Howells if (!f.file) 367793766fbdSDavid Howells return -EBADF; 367893766fbdSDavid Howells 367993766fbdSDavid Howells ret = -EINVAL; 368093766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 368193766fbdSDavid Howells goto err_fsfd; 368293766fbdSDavid Howells 368393766fbdSDavid Howells fc = f.file->private_data; 368493766fbdSDavid Howells 368593766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 368693766fbdSDavid Howells if (ret < 0) 368793766fbdSDavid Howells goto err_fsfd; 368893766fbdSDavid Howells 368993766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 369093766fbdSDavid Howells ret = -EINVAL; 369193766fbdSDavid Howells if (!fc->root) 369293766fbdSDavid Howells goto err_unlock; 369393766fbdSDavid Howells 369493766fbdSDavid Howells ret = -EPERM; 369593766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 369693766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 369793766fbdSDavid Howells goto err_unlock; 369893766fbdSDavid Howells } 369993766fbdSDavid Howells 370093766fbdSDavid Howells ret = -EBUSY; 370193766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 370293766fbdSDavid Howells goto err_unlock; 370393766fbdSDavid Howells 3704f7e33bdbSJeff Layton if (fc->sb_flags & SB_MANDLOCK) 3705f7e33bdbSJeff Layton warn_mandlock(); 370693766fbdSDavid Howells 370793766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 370893766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 370993766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 371093766fbdSDavid Howells goto err_unlock; 371193766fbdSDavid Howells } 371293766fbdSDavid Howells newmount.dentry = dget(fc->root); 371393766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 371493766fbdSDavid Howells 371593766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 371693766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 371793766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 371893766fbdSDavid Howells * don't want to have to handle any errors incurred. 371993766fbdSDavid Howells */ 372093766fbdSDavid Howells vfs_clean_context(fc); 372193766fbdSDavid Howells 372293766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 372393766fbdSDavid Howells if (IS_ERR(ns)) { 372493766fbdSDavid Howells ret = PTR_ERR(ns); 372593766fbdSDavid Howells goto err_path; 372693766fbdSDavid Howells } 372793766fbdSDavid Howells mnt = real_mount(newmount.mnt); 372893766fbdSDavid Howells mnt->mnt_ns = ns; 372993766fbdSDavid Howells ns->root = mnt; 373093766fbdSDavid Howells ns->mounts = 1; 373193766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 37321b0b9cc8SEric Biggers mntget(newmount.mnt); 373393766fbdSDavid Howells 373493766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 373593766fbdSDavid Howells * it, not just simply put it. 373693766fbdSDavid Howells */ 373793766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 373893766fbdSDavid Howells if (IS_ERR(file)) { 373993766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 374093766fbdSDavid Howells ret = PTR_ERR(file); 374193766fbdSDavid Howells goto err_path; 374293766fbdSDavid Howells } 374393766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 374493766fbdSDavid Howells 374593766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 374693766fbdSDavid Howells if (ret >= 0) 374793766fbdSDavid Howells fd_install(ret, file); 374893766fbdSDavid Howells else 374993766fbdSDavid Howells fput(file); 375093766fbdSDavid Howells 375193766fbdSDavid Howells err_path: 375293766fbdSDavid Howells path_put(&newmount); 375393766fbdSDavid Howells err_unlock: 375493766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 375593766fbdSDavid Howells err_fsfd: 375693766fbdSDavid Howells fdput(f); 375793766fbdSDavid Howells return ret; 375893766fbdSDavid Howells } 375993766fbdSDavid Howells 376093766fbdSDavid Howells /* 376193766fbdSDavid Howells * Move a mount from one place to another. In combination with 376293766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 376393766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 376493766fbdSDavid Howells * a mount subtree. 37652db154b3SDavid Howells * 37662db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 37672db154b3SDavid Howells */ 37682db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 37692658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 37702658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 37712db154b3SDavid Howells unsigned int, flags) 37722db154b3SDavid Howells { 37732db154b3SDavid Howells struct path from_path, to_path; 37742db154b3SDavid Howells unsigned int lflags; 37752db154b3SDavid Howells int ret = 0; 37762db154b3SDavid Howells 37772db154b3SDavid Howells if (!may_mount()) 37782db154b3SDavid Howells return -EPERM; 37792db154b3SDavid Howells 37802db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 37812db154b3SDavid Howells return -EINVAL; 37822db154b3SDavid Howells 37832db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 37842db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 37852db154b3SDavid Howells * to clear it afterwards. 37862db154b3SDavid Howells */ 37872db154b3SDavid Howells lflags = 0; 37882db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37892db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37902db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 37912db154b3SDavid Howells 37922db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 37932db154b3SDavid Howells if (ret < 0) 37942db154b3SDavid Howells return ret; 37952db154b3SDavid Howells 37962db154b3SDavid Howells lflags = 0; 37972db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37982db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37992db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 38002db154b3SDavid Howells 38012db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 38022db154b3SDavid Howells if (ret < 0) 38032db154b3SDavid Howells goto out_from; 38042db154b3SDavid Howells 38052db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 38062db154b3SDavid Howells if (ret < 0) 38072db154b3SDavid Howells goto out_to; 38082db154b3SDavid Howells 38099ffb14efSPavel Tikhomirov if (flags & MOVE_MOUNT_SET_GROUP) 38109ffb14efSPavel Tikhomirov ret = do_set_group(&from_path, &to_path); 38119ffb14efSPavel Tikhomirov else 38122db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 38132db154b3SDavid Howells 38142db154b3SDavid Howells out_to: 38152db154b3SDavid Howells path_put(&to_path); 38162db154b3SDavid Howells out_from: 38172db154b3SDavid Howells path_put(&from_path); 38182db154b3SDavid Howells return ret; 38192db154b3SDavid Howells } 38202db154b3SDavid Howells 38212db154b3SDavid Howells /* 3822afac7cbaSAl Viro * Return true if path is reachable from root 3823afac7cbaSAl Viro * 382448a066e7SAl Viro * namespace_sem or mount_lock is held 3825afac7cbaSAl Viro */ 3826643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3827afac7cbaSAl Viro const struct path *root) 3828afac7cbaSAl Viro { 3829643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3830a73324daSAl Viro dentry = mnt->mnt_mountpoint; 38310714a533SAl Viro mnt = mnt->mnt_parent; 3832afac7cbaSAl Viro } 3833643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3834afac7cbaSAl Viro } 3835afac7cbaSAl Viro 3836640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3837afac7cbaSAl Viro { 383825ab4c9bSYaowei Bai bool res; 383948a066e7SAl Viro read_seqlock_excl(&mount_lock); 3840643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 384148a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3842afac7cbaSAl Viro return res; 3843afac7cbaSAl Viro } 3844afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3845afac7cbaSAl Viro 3846afac7cbaSAl Viro /* 38471da177e4SLinus Torvalds * pivot_root Semantics: 38481da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 38491da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 38501da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 38511da177e4SLinus Torvalds * 38521da177e4SLinus Torvalds * Restrictions: 38531da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 38541da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 38551da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 38561da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 38571da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 38581da177e4SLinus Torvalds * 38594a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 38600c1bc6b8SMauro Carvalho Chehab * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives 38614a0d11faSNeil Brown * in this situation. 38624a0d11faSNeil Brown * 38631da177e4SLinus Torvalds * Notes: 38641da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 38651da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 38661da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 38671da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 38681da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 38691da177e4SLinus Torvalds * first. 38701da177e4SLinus Torvalds */ 38713480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 38723480b257SHeiko Carstens const char __user *, put_old) 38731da177e4SLinus Torvalds { 38742763d119SAl Viro struct path new, old, root; 38752763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 387684d17192SAl Viro struct mountpoint *old_mp, *root_mp; 38771da177e4SLinus Torvalds int error; 38781da177e4SLinus Torvalds 38799b40bc90SAl Viro if (!may_mount()) 38801da177e4SLinus Torvalds return -EPERM; 38811da177e4SLinus Torvalds 3882ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3883ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 38841da177e4SLinus Torvalds if (error) 38851da177e4SLinus Torvalds goto out0; 38861da177e4SLinus Torvalds 3887ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 3888ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 38891da177e4SLinus Torvalds if (error) 38901da177e4SLinus Torvalds goto out1; 38911da177e4SLinus Torvalds 38922d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3893b12cea91SAl Viro if (error) 3894b12cea91SAl Viro goto out2; 38951da177e4SLinus Torvalds 3896f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 389784d17192SAl Viro old_mp = lock_mount(&old); 389884d17192SAl Viro error = PTR_ERR(old_mp); 389984d17192SAl Viro if (IS_ERR(old_mp)) 3900b12cea91SAl Viro goto out3; 3901b12cea91SAl Viro 39021da177e4SLinus Torvalds error = -EINVAL; 3903419148daSAl Viro new_mnt = real_mount(new.mnt); 3904419148daSAl Viro root_mnt = real_mount(root.mnt); 390584d17192SAl Viro old_mnt = real_mount(old.mnt); 39062763d119SAl Viro ex_parent = new_mnt->mnt_parent; 39072763d119SAl Viro root_parent = root_mnt->mnt_parent; 390884d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 39092763d119SAl Viro IS_MNT_SHARED(ex_parent) || 39102763d119SAl Viro IS_MNT_SHARED(root_parent)) 3911b12cea91SAl Viro goto out4; 3912143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3913b12cea91SAl Viro goto out4; 39145ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 39155ff9d8a6SEric W. Biederman goto out4; 39161da177e4SLinus Torvalds error = -ENOENT; 3917f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3918b12cea91SAl Viro goto out4; 39191da177e4SLinus Torvalds error = -EBUSY; 392084d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3921b12cea91SAl Viro goto out4; /* loop, on the same file system */ 39221da177e4SLinus Torvalds error = -EINVAL; 39238c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3924b12cea91SAl Viro goto out4; /* not a mountpoint */ 3925676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3926b12cea91SAl Viro goto out4; /* not attached */ 39272d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3928b12cea91SAl Viro goto out4; /* not a mountpoint */ 3929676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3930b12cea91SAl Viro goto out4; /* not attached */ 39314ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 393284d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3933b12cea91SAl Viro goto out4; 39340d082601SEric W. Biederman /* make certain new is below the root */ 39350d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 39360d082601SEric W. Biederman goto out4; 3937719ea2fbSAl Viro lock_mount_hash(); 39382763d119SAl Viro umount_mnt(new_mnt); 39392763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 39405ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 39415ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 39425ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 39435ff9d8a6SEric W. Biederman } 39444ac91378SJan Blunck /* mount old root on put_old */ 394584d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 39464ac91378SJan Blunck /* mount new_root on / */ 39472763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 39482763d119SAl Viro mnt_add_count(root_parent, -1); 39496b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 39504fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 39514fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 39523895dbf8SEric W. Biederman put_mountpoint(root_mp); 3953719ea2fbSAl Viro unlock_mount_hash(); 39542d8f3038SAl Viro chroot_fs_refs(&root, &new); 39551da177e4SLinus Torvalds error = 0; 3956b12cea91SAl Viro out4: 395784d17192SAl Viro unlock_mount(old_mp); 39582763d119SAl Viro if (!error) 39592763d119SAl Viro mntput_no_expire(ex_parent); 3960b12cea91SAl Viro out3: 39618c3ee42eSAl Viro path_put(&root); 3962b12cea91SAl Viro out2: 39632d8f3038SAl Viro path_put(&old); 39641da177e4SLinus Torvalds out1: 39652d8f3038SAl Viro path_put(&new); 39661da177e4SLinus Torvalds out0: 39671da177e4SLinus Torvalds return error; 39681da177e4SLinus Torvalds } 39691da177e4SLinus Torvalds 39702a186721SChristian Brauner static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) 39712a186721SChristian Brauner { 39722a186721SChristian Brauner unsigned int flags = mnt->mnt.mnt_flags; 39732a186721SChristian Brauner 39742a186721SChristian Brauner /* flags to clear */ 39752a186721SChristian Brauner flags &= ~kattr->attr_clr; 39762a186721SChristian Brauner /* flags to raise */ 39772a186721SChristian Brauner flags |= kattr->attr_set; 39782a186721SChristian Brauner 39792a186721SChristian Brauner return flags; 39802a186721SChristian Brauner } 39812a186721SChristian Brauner 39829caccd41SChristian Brauner static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 39839caccd41SChristian Brauner { 39849caccd41SChristian Brauner struct vfsmount *m = &mnt->mnt; 3985bd303368SChristian Brauner struct user_namespace *fs_userns = m->mnt_sb->s_user_ns; 39869caccd41SChristian Brauner 3987256c8aedSChristian Brauner if (!kattr->mnt_idmap) 39889caccd41SChristian Brauner return 0; 39899caccd41SChristian Brauner 39909caccd41SChristian Brauner /* 3991bd303368SChristian Brauner * Creating an idmapped mount with the filesystem wide idmapping 3992bd303368SChristian Brauner * doesn't make sense so block that. We don't allow mushy semantics. 3993bd303368SChristian Brauner */ 39943707d84cSChristian Brauner if (!check_fsmapping(kattr->mnt_idmap, m->mnt_sb)) 3995bd303368SChristian Brauner return -EINVAL; 3996bd303368SChristian Brauner 3997bd303368SChristian Brauner /* 39989caccd41SChristian Brauner * Once a mount has been idmapped we don't allow it to change its 39999caccd41SChristian Brauner * mapping. It makes things simpler and callers can just create 40009caccd41SChristian Brauner * another bind-mount they can idmap if they want to. 40019caccd41SChristian Brauner */ 4002bb49e9e7SChristian Brauner if (is_idmapped_mnt(m)) 40039caccd41SChristian Brauner return -EPERM; 40049caccd41SChristian Brauner 40059caccd41SChristian Brauner /* The underlying filesystem doesn't support idmapped mounts yet. */ 40069caccd41SChristian Brauner if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) 40079caccd41SChristian Brauner return -EINVAL; 40089caccd41SChristian Brauner 40099caccd41SChristian Brauner /* We're not controlling the superblock. */ 4010bd303368SChristian Brauner if (!ns_capable(fs_userns, CAP_SYS_ADMIN)) 40119caccd41SChristian Brauner return -EPERM; 40129caccd41SChristian Brauner 40139caccd41SChristian Brauner /* Mount has already been visible in the filesystem hierarchy. */ 40149caccd41SChristian Brauner if (!is_anon_ns(mnt->mnt_ns)) 40159caccd41SChristian Brauner return -EINVAL; 40169caccd41SChristian Brauner 40179caccd41SChristian Brauner return 0; 40189caccd41SChristian Brauner } 40199caccd41SChristian Brauner 4020a26f788bSChristian Brauner /** 4021a26f788bSChristian Brauner * mnt_allow_writers() - check whether the attribute change allows writers 4022a26f788bSChristian Brauner * @kattr: the new mount attributes 4023a26f788bSChristian Brauner * @mnt: the mount to which @kattr will be applied 4024a26f788bSChristian Brauner * 4025a26f788bSChristian Brauner * Check whether thew new mount attributes in @kattr allow concurrent writers. 4026a26f788bSChristian Brauner * 4027a26f788bSChristian Brauner * Return: true if writers need to be held, false if not 4028a26f788bSChristian Brauner */ 4029a26f788bSChristian Brauner static inline bool mnt_allow_writers(const struct mount_kattr *kattr, 4030a26f788bSChristian Brauner const struct mount *mnt) 40312a186721SChristian Brauner { 4032e1bbcd27SChristian Brauner return (!(kattr->attr_set & MNT_READONLY) || 4033e1bbcd27SChristian Brauner (mnt->mnt.mnt_flags & MNT_READONLY)) && 4034256c8aedSChristian Brauner !kattr->mnt_idmap; 40352a186721SChristian Brauner } 40362a186721SChristian Brauner 403787bb5b60SChristian Brauner static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt) 40382a186721SChristian Brauner { 4039e257039fSAl Viro struct mount *m; 4040e257039fSAl Viro int err; 40412a186721SChristian Brauner 4042e257039fSAl Viro for (m = mnt; m; m = next_mnt(m, mnt)) { 4043e257039fSAl Viro if (!can_change_locked_flags(m, recalc_flags(kattr, m))) { 4044e257039fSAl Viro err = -EPERM; 4045e257039fSAl Viro break; 40462a186721SChristian Brauner } 40472a186721SChristian Brauner 404887bb5b60SChristian Brauner err = can_idmap_mount(kattr, m); 404987bb5b60SChristian Brauner if (err) 4050e257039fSAl Viro break; 40512a186721SChristian Brauner 4052e257039fSAl Viro if (!mnt_allow_writers(kattr, m)) { 405387bb5b60SChristian Brauner err = mnt_hold_writers(m); 405487bb5b60SChristian Brauner if (err) 4055e257039fSAl Viro break; 40562a186721SChristian Brauner } 40572a186721SChristian Brauner 4058e257039fSAl Viro if (!kattr->recurse) 405987bb5b60SChristian Brauner return 0; 40602a186721SChristian Brauner } 40612a186721SChristian Brauner 4062e257039fSAl Viro if (err) { 4063e257039fSAl Viro struct mount *p; 4064e257039fSAl Viro 40650014edaeSChristian Brauner /* 40660014edaeSChristian Brauner * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will 40670014edaeSChristian Brauner * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all 40680014edaeSChristian Brauner * mounts and needs to take care to include the first mount. 40690014edaeSChristian Brauner */ 40700014edaeSChristian Brauner for (p = mnt; p; p = next_mnt(p, mnt)) { 4071e257039fSAl Viro /* If we had to hold writers unblock them. */ 4072e257039fSAl Viro if (p->mnt.mnt_flags & MNT_WRITE_HOLD) 4073e257039fSAl Viro mnt_unhold_writers(p); 40740014edaeSChristian Brauner 40750014edaeSChristian Brauner /* 40760014edaeSChristian Brauner * We're done once the first mount we changed got 40770014edaeSChristian Brauner * MNT_WRITE_HOLD unset. 40780014edaeSChristian Brauner */ 40790014edaeSChristian Brauner if (p == m) 40800014edaeSChristian Brauner break; 4081e257039fSAl Viro } 4082e257039fSAl Viro } 4083e257039fSAl Viro return err; 40842a186721SChristian Brauner } 40852a186721SChristian Brauner 40869caccd41SChristian Brauner static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 40879caccd41SChristian Brauner { 4088256c8aedSChristian Brauner if (!kattr->mnt_idmap) 40899caccd41SChristian Brauner return; 40909caccd41SChristian Brauner 4091bd303368SChristian Brauner /* 4092256c8aedSChristian Brauner * Pairs with smp_load_acquire() in mnt_idmap(). 4093256c8aedSChristian Brauner * 4094256c8aedSChristian Brauner * Since we only allow a mount to change the idmapping once and 4095256c8aedSChristian Brauner * verified this in can_idmap_mount() we know that the mount has 4096256c8aedSChristian Brauner * @nop_mnt_idmap attached to it. So there's no need to drop any 4097256c8aedSChristian Brauner * references. 4098bd303368SChristian Brauner */ 4099256c8aedSChristian Brauner smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap)); 41009caccd41SChristian Brauner } 41019caccd41SChristian Brauner 4102e257039fSAl Viro static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt) 41032a186721SChristian Brauner { 4104e257039fSAl Viro struct mount *m; 41052a186721SChristian Brauner 4106e257039fSAl Viro for (m = mnt; m; m = next_mnt(m, mnt)) { 41072a186721SChristian Brauner unsigned int flags; 41082a186721SChristian Brauner 41099caccd41SChristian Brauner do_idmap_mount(kattr, m); 41102a186721SChristian Brauner flags = recalc_flags(kattr, m); 41112a186721SChristian Brauner WRITE_ONCE(m->mnt.mnt_flags, flags); 41122a186721SChristian Brauner 411303b6abeeSChristian Brauner /* If we had to hold writers unblock them. */ 411403b6abeeSChristian Brauner if (m->mnt.mnt_flags & MNT_WRITE_HOLD) 41152a186721SChristian Brauner mnt_unhold_writers(m); 41162a186721SChristian Brauner 4117e257039fSAl Viro if (kattr->propagation) 41182a186721SChristian Brauner change_mnt_propagation(m, kattr->propagation); 4119e257039fSAl Viro if (!kattr->recurse) 41202a186721SChristian Brauner break; 4121e257039fSAl Viro } 41222a186721SChristian Brauner touch_mnt_namespace(mnt->mnt_ns); 41232a186721SChristian Brauner } 41242a186721SChristian Brauner 41252a186721SChristian Brauner static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) 41262a186721SChristian Brauner { 412787bb5b60SChristian Brauner struct mount *mnt = real_mount(path->mnt); 41282a186721SChristian Brauner int err = 0; 41292a186721SChristian Brauner 41302a186721SChristian Brauner if (path->dentry != mnt->mnt.mnt_root) 41312a186721SChristian Brauner return -EINVAL; 41322a186721SChristian Brauner 4133256c8aedSChristian Brauner if (kattr->mnt_userns) { 4134256c8aedSChristian Brauner struct mnt_idmap *mnt_idmap; 4135256c8aedSChristian Brauner 4136256c8aedSChristian Brauner mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns); 4137256c8aedSChristian Brauner if (IS_ERR(mnt_idmap)) 4138256c8aedSChristian Brauner return PTR_ERR(mnt_idmap); 4139256c8aedSChristian Brauner kattr->mnt_idmap = mnt_idmap; 4140256c8aedSChristian Brauner } 4141256c8aedSChristian Brauner 41422a186721SChristian Brauner if (kattr->propagation) { 41432a186721SChristian Brauner /* 41442a186721SChristian Brauner * Only take namespace_lock() if we're actually changing 41452a186721SChristian Brauner * propagation. 41462a186721SChristian Brauner */ 41472a186721SChristian Brauner namespace_lock(); 41482a186721SChristian Brauner if (kattr->propagation == MS_SHARED) { 41492a186721SChristian Brauner err = invent_group_ids(mnt, kattr->recurse); 41502a186721SChristian Brauner if (err) { 41512a186721SChristian Brauner namespace_unlock(); 41522a186721SChristian Brauner return err; 41532a186721SChristian Brauner } 41542a186721SChristian Brauner } 41552a186721SChristian Brauner } 41562a186721SChristian Brauner 415787bb5b60SChristian Brauner err = -EINVAL; 41582a186721SChristian Brauner lock_mount_hash(); 41592a186721SChristian Brauner 416087bb5b60SChristian Brauner /* Ensure that this isn't anything purely vfs internal. */ 416187bb5b60SChristian Brauner if (!is_mounted(&mnt->mnt)) 416287bb5b60SChristian Brauner goto out; 41632a186721SChristian Brauner 416487bb5b60SChristian Brauner /* 416587bb5b60SChristian Brauner * If this is an attached mount make sure it's located in the callers 416687bb5b60SChristian Brauner * mount namespace. If it's not don't let the caller interact with it. 416787bb5b60SChristian Brauner * If this is a detached mount make sure it has an anonymous mount 416887bb5b60SChristian Brauner * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE. 416987bb5b60SChristian Brauner */ 417087bb5b60SChristian Brauner if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns))) 417187bb5b60SChristian Brauner goto out; 417287bb5b60SChristian Brauner 417387bb5b60SChristian Brauner /* 417487bb5b60SChristian Brauner * First, we get the mount tree in a shape where we can change mount 417587bb5b60SChristian Brauner * properties without failure. If we succeeded to do so we commit all 417687bb5b60SChristian Brauner * changes and if we failed we clean up. 417787bb5b60SChristian Brauner */ 417887bb5b60SChristian Brauner err = mount_setattr_prepare(kattr, mnt); 4179e257039fSAl Viro if (!err) 4180e257039fSAl Viro mount_setattr_commit(kattr, mnt); 418187bb5b60SChristian Brauner 418287bb5b60SChristian Brauner out: 41832a186721SChristian Brauner unlock_mount_hash(); 41842a186721SChristian Brauner 41852a186721SChristian Brauner if (kattr->propagation) { 41862a186721SChristian Brauner if (err) 41872a186721SChristian Brauner cleanup_group_ids(mnt, NULL); 4188*cb2239c1SChristian Brauner namespace_unlock(); 41892a186721SChristian Brauner } 41902a186721SChristian Brauner 41912a186721SChristian Brauner return err; 41922a186721SChristian Brauner } 41932a186721SChristian Brauner 41949caccd41SChristian Brauner static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, 41959caccd41SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 41969caccd41SChristian Brauner { 41979caccd41SChristian Brauner int err = 0; 41989caccd41SChristian Brauner struct ns_common *ns; 41999caccd41SChristian Brauner struct user_namespace *mnt_userns; 42009caccd41SChristian Brauner struct file *file; 42019caccd41SChristian Brauner 42029caccd41SChristian Brauner if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) 42039caccd41SChristian Brauner return 0; 42049caccd41SChristian Brauner 42059caccd41SChristian Brauner /* 42069caccd41SChristian Brauner * We currently do not support clearing an idmapped mount. If this ever 42079caccd41SChristian Brauner * is a use-case we can revisit this but for now let's keep it simple 42089caccd41SChristian Brauner * and not allow it. 42099caccd41SChristian Brauner */ 42109caccd41SChristian Brauner if (attr->attr_clr & MOUNT_ATTR_IDMAP) 42119caccd41SChristian Brauner return -EINVAL; 42129caccd41SChristian Brauner 42139caccd41SChristian Brauner if (attr->userns_fd > INT_MAX) 42149caccd41SChristian Brauner return -EINVAL; 42159caccd41SChristian Brauner 42169caccd41SChristian Brauner file = fget(attr->userns_fd); 42179caccd41SChristian Brauner if (!file) 42189caccd41SChristian Brauner return -EBADF; 42199caccd41SChristian Brauner 42209caccd41SChristian Brauner if (!proc_ns_file(file)) { 42219caccd41SChristian Brauner err = -EINVAL; 42229caccd41SChristian Brauner goto out_fput; 42239caccd41SChristian Brauner } 42249caccd41SChristian Brauner 42259caccd41SChristian Brauner ns = get_proc_ns(file_inode(file)); 42269caccd41SChristian Brauner if (ns->ops->type != CLONE_NEWUSER) { 42279caccd41SChristian Brauner err = -EINVAL; 42289caccd41SChristian Brauner goto out_fput; 42299caccd41SChristian Brauner } 42309caccd41SChristian Brauner 42319caccd41SChristian Brauner /* 4232bd303368SChristian Brauner * The initial idmapping cannot be used to create an idmapped 4233bd303368SChristian Brauner * mount. We use the initial idmapping as an indicator of a mount 4234bd303368SChristian Brauner * that is not idmapped. It can simply be passed into helpers that 4235bd303368SChristian Brauner * are aware of idmapped mounts as a convenient shortcut. A user 4236bd303368SChristian Brauner * can just create a dedicated identity mapping to achieve the same 4237bd303368SChristian Brauner * result. 42389caccd41SChristian Brauner */ 42399caccd41SChristian Brauner mnt_userns = container_of(ns, struct user_namespace, ns); 42403707d84cSChristian Brauner if (mnt_userns == &init_user_ns) { 42419caccd41SChristian Brauner err = -EPERM; 42429caccd41SChristian Brauner goto out_fput; 42439caccd41SChristian Brauner } 4244bf1ac16eSSeth Forshee 4245bf1ac16eSSeth Forshee /* We're not controlling the target namespace. */ 4246bf1ac16eSSeth Forshee if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) { 4247bf1ac16eSSeth Forshee err = -EPERM; 4248bf1ac16eSSeth Forshee goto out_fput; 4249bf1ac16eSSeth Forshee } 4250bf1ac16eSSeth Forshee 42519caccd41SChristian Brauner kattr->mnt_userns = get_user_ns(mnt_userns); 42529caccd41SChristian Brauner 42539caccd41SChristian Brauner out_fput: 42549caccd41SChristian Brauner fput(file); 42559caccd41SChristian Brauner return err; 42569caccd41SChristian Brauner } 42579caccd41SChristian Brauner 42589caccd41SChristian Brauner static int build_mount_kattr(const struct mount_attr *attr, size_t usize, 42592a186721SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 42602a186721SChristian Brauner { 42612a186721SChristian Brauner unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 42622a186721SChristian Brauner 42632a186721SChristian Brauner if (flags & AT_NO_AUTOMOUNT) 42642a186721SChristian Brauner lookup_flags &= ~LOOKUP_AUTOMOUNT; 42652a186721SChristian Brauner if (flags & AT_SYMLINK_NOFOLLOW) 42662a186721SChristian Brauner lookup_flags &= ~LOOKUP_FOLLOW; 42672a186721SChristian Brauner if (flags & AT_EMPTY_PATH) 42682a186721SChristian Brauner lookup_flags |= LOOKUP_EMPTY; 42692a186721SChristian Brauner 42702a186721SChristian Brauner *kattr = (struct mount_kattr) { 42712a186721SChristian Brauner .lookup_flags = lookup_flags, 42722a186721SChristian Brauner .recurse = !!(flags & AT_RECURSIVE), 42732a186721SChristian Brauner }; 42742a186721SChristian Brauner 42752a186721SChristian Brauner if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) 42762a186721SChristian Brauner return -EINVAL; 42772a186721SChristian Brauner if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) 42782a186721SChristian Brauner return -EINVAL; 42792a186721SChristian Brauner kattr->propagation = attr->propagation; 42802a186721SChristian Brauner 42812a186721SChristian Brauner if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) 42822a186721SChristian Brauner return -EINVAL; 42832a186721SChristian Brauner 42842a186721SChristian Brauner kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); 42852a186721SChristian Brauner kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); 42862a186721SChristian Brauner 42872a186721SChristian Brauner /* 42882a186721SChristian Brauner * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, 42892a186721SChristian Brauner * users wanting to transition to a different atime setting cannot 42902a186721SChristian Brauner * simply specify the atime setting in @attr_set, but must also 42912a186721SChristian Brauner * specify MOUNT_ATTR__ATIME in the @attr_clr field. 42922a186721SChristian Brauner * So ensure that MOUNT_ATTR__ATIME can't be partially set in 42932a186721SChristian Brauner * @attr_clr and that @attr_set can't have any atime bits set if 42942a186721SChristian Brauner * MOUNT_ATTR__ATIME isn't set in @attr_clr. 42952a186721SChristian Brauner */ 42962a186721SChristian Brauner if (attr->attr_clr & MOUNT_ATTR__ATIME) { 42972a186721SChristian Brauner if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) 42982a186721SChristian Brauner return -EINVAL; 42992a186721SChristian Brauner 43002a186721SChristian Brauner /* 43012a186721SChristian Brauner * Clear all previous time settings as they are mutually 43022a186721SChristian Brauner * exclusive. 43032a186721SChristian Brauner */ 43042a186721SChristian Brauner kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; 43052a186721SChristian Brauner switch (attr->attr_set & MOUNT_ATTR__ATIME) { 43062a186721SChristian Brauner case MOUNT_ATTR_RELATIME: 43072a186721SChristian Brauner kattr->attr_set |= MNT_RELATIME; 43082a186721SChristian Brauner break; 43092a186721SChristian Brauner case MOUNT_ATTR_NOATIME: 43102a186721SChristian Brauner kattr->attr_set |= MNT_NOATIME; 43112a186721SChristian Brauner break; 43122a186721SChristian Brauner case MOUNT_ATTR_STRICTATIME: 43132a186721SChristian Brauner break; 43142a186721SChristian Brauner default: 43152a186721SChristian Brauner return -EINVAL; 43162a186721SChristian Brauner } 43172a186721SChristian Brauner } else { 43182a186721SChristian Brauner if (attr->attr_set & MOUNT_ATTR__ATIME) 43192a186721SChristian Brauner return -EINVAL; 43202a186721SChristian Brauner } 43212a186721SChristian Brauner 43229caccd41SChristian Brauner return build_mount_idmapped(attr, usize, kattr, flags); 43239caccd41SChristian Brauner } 43249caccd41SChristian Brauner 43259caccd41SChristian Brauner static void finish_mount_kattr(struct mount_kattr *kattr) 43269caccd41SChristian Brauner { 43279caccd41SChristian Brauner put_user_ns(kattr->mnt_userns); 43289caccd41SChristian Brauner kattr->mnt_userns = NULL; 4329256c8aedSChristian Brauner 4330256c8aedSChristian Brauner if (kattr->mnt_idmap) 4331256c8aedSChristian Brauner mnt_idmap_put(kattr->mnt_idmap); 43322a186721SChristian Brauner } 43332a186721SChristian Brauner 43342a186721SChristian Brauner SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, 43352a186721SChristian Brauner unsigned int, flags, struct mount_attr __user *, uattr, 43362a186721SChristian Brauner size_t, usize) 43372a186721SChristian Brauner { 43382a186721SChristian Brauner int err; 43392a186721SChristian Brauner struct path target; 43402a186721SChristian Brauner struct mount_attr attr; 43412a186721SChristian Brauner struct mount_kattr kattr; 43422a186721SChristian Brauner 43432a186721SChristian Brauner BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); 43442a186721SChristian Brauner 43452a186721SChristian Brauner if (flags & ~(AT_EMPTY_PATH | 43462a186721SChristian Brauner AT_RECURSIVE | 43472a186721SChristian Brauner AT_SYMLINK_NOFOLLOW | 43482a186721SChristian Brauner AT_NO_AUTOMOUNT)) 43492a186721SChristian Brauner return -EINVAL; 43502a186721SChristian Brauner 43512a186721SChristian Brauner if (unlikely(usize > PAGE_SIZE)) 43522a186721SChristian Brauner return -E2BIG; 43532a186721SChristian Brauner if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) 43542a186721SChristian Brauner return -EINVAL; 43552a186721SChristian Brauner 43562a186721SChristian Brauner if (!may_mount()) 43572a186721SChristian Brauner return -EPERM; 43582a186721SChristian Brauner 43592a186721SChristian Brauner err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); 43602a186721SChristian Brauner if (err) 43612a186721SChristian Brauner return err; 43622a186721SChristian Brauner 43632a186721SChristian Brauner /* Don't bother walking through the mounts if this is a nop. */ 43642a186721SChristian Brauner if (attr.attr_set == 0 && 43652a186721SChristian Brauner attr.attr_clr == 0 && 43662a186721SChristian Brauner attr.propagation == 0) 43672a186721SChristian Brauner return 0; 43682a186721SChristian Brauner 43699caccd41SChristian Brauner err = build_mount_kattr(&attr, usize, &kattr, flags); 43702a186721SChristian Brauner if (err) 43712a186721SChristian Brauner return err; 43722a186721SChristian Brauner 43732a186721SChristian Brauner err = user_path_at(dfd, path, kattr.lookup_flags, &target); 4374012e3322SChristian Brauner if (!err) { 43752a186721SChristian Brauner err = do_mount_setattr(&target, &kattr); 43762a186721SChristian Brauner path_put(&target); 4377012e3322SChristian Brauner } 4378012e3322SChristian Brauner finish_mount_kattr(&kattr); 43792a186721SChristian Brauner return err; 43802a186721SChristian Brauner } 43812a186721SChristian Brauner 43821da177e4SLinus Torvalds static void __init init_mount_tree(void) 43831da177e4SLinus Torvalds { 43841da177e4SLinus Torvalds struct vfsmount *mnt; 438574e83122SAl Viro struct mount *m; 43866b3286edSKirill Korotaev struct mnt_namespace *ns; 4387ac748a09SJan Blunck struct path root; 43881da177e4SLinus Torvalds 4389fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 43901da177e4SLinus Torvalds if (IS_ERR(mnt)) 43911da177e4SLinus Torvalds panic("Can't create rootfs"); 4392b3e19d92SNick Piggin 439374e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 43943b22edc5STrond Myklebust if (IS_ERR(ns)) 43951da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 439674e83122SAl Viro m = real_mount(mnt); 439774e83122SAl Viro m->mnt_ns = ns; 439874e83122SAl Viro ns->root = m; 439974e83122SAl Viro ns->mounts = 1; 440074e83122SAl Viro list_add(&m->mnt_list, &ns->list); 44016b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 44026b3286edSKirill Korotaev get_mnt_ns(ns); 44031da177e4SLinus Torvalds 4404be08d6d2SAl Viro root.mnt = mnt; 4405be08d6d2SAl Viro root.dentry = mnt->mnt_root; 4406da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 4407ac748a09SJan Blunck 4408ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 4409ac748a09SJan Blunck set_fs_root(current->fs, &root); 44101da177e4SLinus Torvalds } 44111da177e4SLinus Torvalds 441274bf17cfSDenis Cheng void __init mnt_init(void) 44131da177e4SLinus Torvalds { 441415a67dd8SRandy Dunlap int err; 44151da177e4SLinus Torvalds 44167d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 441779f6540bSVasily Averin 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); 44181da177e4SLinus Torvalds 44190818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 442038129a13SAl Viro sizeof(struct hlist_head), 44210818bf27SAl Viro mhash_entries, 19, 44223d375d78SPavel Tatashin HASH_ZERO, 44230818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 44240818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 44250818bf27SAl Viro sizeof(struct hlist_head), 44260818bf27SAl Viro mphash_entries, 19, 44273d375d78SPavel Tatashin HASH_ZERO, 44280818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 44291da177e4SLinus Torvalds 443084d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 44311da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 44321da177e4SLinus Torvalds 44334b93dc9bSTejun Heo kernfs_init(); 44344b93dc9bSTejun Heo 443515a67dd8SRandy Dunlap err = sysfs_init(); 443615a67dd8SRandy Dunlap if (err) 443715a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 44388e24eea7SHarvey Harrison __func__, err); 443900d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 444000d26666SGreg Kroah-Hartman if (!fs_kobj) 44418e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 4442037f11b4SAl Viro shmem_init(); 44431da177e4SLinus Torvalds init_rootfs(); 44441da177e4SLinus Torvalds init_mount_tree(); 44451da177e4SLinus Torvalds } 44461da177e4SLinus Torvalds 4447616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 44481da177e4SLinus Torvalds { 44491a7b8969SKirill Tkhai if (!refcount_dec_and_test(&ns->ns.count)) 4450616511d0STrond Myklebust return; 44517b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 4452771b1371SEric W. Biederman free_mnt_ns(ns); 44531da177e4SLinus Torvalds } 44549d412a43SAl Viro 4455d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 44569d412a43SAl Viro { 4457423e0ab0STim Chen struct vfsmount *mnt; 4458d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 4459423e0ab0STim Chen if (!IS_ERR(mnt)) { 4460423e0ab0STim Chen /* 4461423e0ab0STim Chen * it is a longterm mount, don't release mnt until 4462423e0ab0STim Chen * we unmount before file sys is unregistered 4463423e0ab0STim Chen */ 4464f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 4465423e0ab0STim Chen } 4466423e0ab0STim Chen return mnt; 44679d412a43SAl Viro } 4468d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 4469423e0ab0STim Chen 4470423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 4471423e0ab0STim Chen { 4472423e0ab0STim Chen /* release long term mount so mount point can be released */ 4473da27f796SRik van Riel if (!IS_ERR(mnt)) { 4474da27f796SRik van Riel mnt_make_shortterm(mnt); 447548a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 4476423e0ab0STim Chen mntput(mnt); 4477423e0ab0STim Chen } 4478423e0ab0STim Chen } 4479423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 448002125a82SAl Viro 4481df820f8dSMiklos Szeredi void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) 4482df820f8dSMiklos Szeredi { 4483df820f8dSMiklos Szeredi unsigned int i; 4484df820f8dSMiklos Szeredi 4485df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4486da27f796SRik van Riel mnt_make_shortterm(mnt[i]); 4487df820f8dSMiklos Szeredi synchronize_rcu_expedited(); 4488df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4489df820f8dSMiklos Szeredi mntput(mnt[i]); 4490df820f8dSMiklos Szeredi } 4491df820f8dSMiklos Szeredi EXPORT_SYMBOL(kern_unmount_array); 4492df820f8dSMiklos Szeredi 449302125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 449402125a82SAl Viro { 4495143c8c91SAl Viro return check_mnt(real_mount(mnt)); 449602125a82SAl Viro } 44978823c079SEric W. Biederman 44983151527eSEric W. Biederman bool current_chrooted(void) 44993151527eSEric W. Biederman { 45003151527eSEric W. Biederman /* Does the current process have a non-standard root */ 45013151527eSEric W. Biederman struct path ns_root; 45023151527eSEric W. Biederman struct path fs_root; 45033151527eSEric W. Biederman bool chrooted; 45043151527eSEric W. Biederman 45053151527eSEric W. Biederman /* Find the namespace root */ 45063151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 45073151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 45083151527eSEric W. Biederman path_get(&ns_root); 45093151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 45103151527eSEric W. Biederman ; 45113151527eSEric W. Biederman 45123151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 45133151527eSEric W. Biederman 45143151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 45153151527eSEric W. Biederman 45163151527eSEric W. Biederman path_put(&fs_root); 45173151527eSEric W. Biederman path_put(&ns_root); 45183151527eSEric W. Biederman 45193151527eSEric W. Biederman return chrooted; 45203151527eSEric W. Biederman } 45213151527eSEric W. Biederman 4522132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 4523132e4608SDavid Howells const struct super_block *sb, 45248654df4eSEric W. Biederman int *new_mnt_flags) 452587a8ebd6SEric W. Biederman { 45268c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 452787a8ebd6SEric W. Biederman struct mount *mnt; 4528e51db735SEric W. Biederman bool visible = false; 452987a8ebd6SEric W. Biederman 453044bb4385SAl Viro down_read(&namespace_sem); 45319f6c61f9SMiklos Szeredi lock_ns_list(ns); 453287a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 4533e51db735SEric W. Biederman struct mount *child; 453477b1a97dSEric W. Biederman int mnt_flags; 453577b1a97dSEric W. Biederman 45369f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 45379f6c61f9SMiklos Szeredi continue; 45389f6c61f9SMiklos Szeredi 4539132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 4540e51db735SEric W. Biederman continue; 4541e51db735SEric W. Biederman 45427e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 45437e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 45447e96c1b0SEric W. Biederman */ 45457e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 45467e96c1b0SEric W. Biederman continue; 45477e96c1b0SEric W. Biederman 4548a1935c17SEric W. Biederman /* A local view of the mount flags */ 454977b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 455077b1a97dSEric W. Biederman 4551695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 4552bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 4553695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 4554695e9df0SEric W. Biederman 45558c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 45568c6cf9ccSEric W. Biederman * than the proposed new mount. 45578c6cf9ccSEric W. Biederman */ 455877b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 45598c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 45608c6cf9ccSEric W. Biederman continue; 456177b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 456277b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 45638c6cf9ccSEric W. Biederman continue; 45648c6cf9ccSEric W. Biederman 4565ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 4566ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 4567ceeb0e5dSEric W. Biederman * empty directories. 4568e51db735SEric W. Biederman */ 4569e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 4570e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 4571ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 4572d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 4573ceeb0e5dSEric W. Biederman continue; 45747236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 45757236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 4576e51db735SEric W. Biederman goto next; 457787a8ebd6SEric W. Biederman } 45788c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 457977b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 45808c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 4581e51db735SEric W. Biederman visible = true; 4582e51db735SEric W. Biederman goto found; 4583e51db735SEric W. Biederman next: ; 458487a8ebd6SEric W. Biederman } 4585e51db735SEric W. Biederman found: 45869f6c61f9SMiklos Szeredi unlock_ns_list(ns); 458744bb4385SAl Viro up_read(&namespace_sem); 4588e51db735SEric W. Biederman return visible; 458987a8ebd6SEric W. Biederman } 459087a8ebd6SEric W. Biederman 4591132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 45928654df4eSEric W. Biederman { 4593a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 45948654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 45958654df4eSEric W. Biederman unsigned long s_iflags; 45968654df4eSEric W. Biederman 45978654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 45988654df4eSEric W. Biederman return false; 45998654df4eSEric W. Biederman 46008654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 4601132e4608SDavid Howells s_iflags = sb->s_iflags; 46028654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 46038654df4eSEric W. Biederman return false; 46048654df4eSEric W. Biederman 4605a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 4606a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 4607a1935c17SEric W. Biederman required_iflags); 4608a1935c17SEric W. Biederman return true; 4609a1935c17SEric W. Biederman } 4610a1935c17SEric W. Biederman 4611132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 46128654df4eSEric W. Biederman } 46138654df4eSEric W. Biederman 4614380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 4615380cf5baSAndy Lutomirski { 4616380cf5baSAndy Lutomirski /* 4617380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 4618380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 4619380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 4620380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 4621380cf5baSAndy Lutomirski * in other namespaces. 4622380cf5baSAndy Lutomirski */ 4623380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 4624380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 4625380cf5baSAndy Lutomirski } 4626380cf5baSAndy Lutomirski 462764964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 46288823c079SEric W. Biederman { 462958be2825SAl Viro struct ns_common *ns = NULL; 46308823c079SEric W. Biederman struct nsproxy *nsproxy; 46318823c079SEric W. Biederman 4632728dba3aSEric W. Biederman task_lock(task); 4633728dba3aSEric W. Biederman nsproxy = task->nsproxy; 46348823c079SEric W. Biederman if (nsproxy) { 463558be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 463658be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 46378823c079SEric W. Biederman } 4638728dba3aSEric W. Biederman task_unlock(task); 46398823c079SEric W. Biederman 46408823c079SEric W. Biederman return ns; 46418823c079SEric W. Biederman } 46428823c079SEric W. Biederman 464364964528SAl Viro static void mntns_put(struct ns_common *ns) 46448823c079SEric W. Biederman { 464558be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 46468823c079SEric W. Biederman } 46478823c079SEric W. Biederman 4648f2a8d52eSChristian Brauner static int mntns_install(struct nsset *nsset, struct ns_common *ns) 46498823c079SEric W. Biederman { 4650f2a8d52eSChristian Brauner struct nsproxy *nsproxy = nsset->nsproxy; 4651f2a8d52eSChristian Brauner struct fs_struct *fs = nsset->fs; 46524f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 4653f2a8d52eSChristian Brauner struct user_namespace *user_ns = nsset->cred->user_ns; 46548823c079SEric W. Biederman struct path root; 46554f757f3cSAl Viro int err; 46568823c079SEric W. Biederman 46570c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 4658f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_CHROOT) || 4659f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_ADMIN)) 4660ae11e0f1SZhao Hongjiang return -EPERM; 46618823c079SEric W. Biederman 466274e83122SAl Viro if (is_anon_ns(mnt_ns)) 466374e83122SAl Viro return -EINVAL; 466474e83122SAl Viro 46658823c079SEric W. Biederman if (fs->users != 1) 46668823c079SEric W. Biederman return -EINVAL; 46678823c079SEric W. Biederman 46688823c079SEric W. Biederman get_mnt_ns(mnt_ns); 46694f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 46708823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 46718823c079SEric W. Biederman 46728823c079SEric W. Biederman /* Find the root */ 46734f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 46744f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 46754f757f3cSAl Viro if (err) { 46764f757f3cSAl Viro /* revert to old namespace */ 46774f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 46784f757f3cSAl Viro put_mnt_ns(mnt_ns); 46794f757f3cSAl Viro return err; 46804f757f3cSAl Viro } 46818823c079SEric W. Biederman 46824068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 46834068367cSAndrei Vagin 46848823c079SEric W. Biederman /* Update the pwd and root */ 46858823c079SEric W. Biederman set_fs_pwd(fs, &root); 46868823c079SEric W. Biederman set_fs_root(fs, &root); 46878823c079SEric W. Biederman 46888823c079SEric W. Biederman path_put(&root); 46898823c079SEric W. Biederman return 0; 46908823c079SEric W. Biederman } 46918823c079SEric W. Biederman 4692bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 4693bcac25a5SAndrey Vagin { 4694bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 4695bcac25a5SAndrey Vagin } 4696bcac25a5SAndrey Vagin 46978823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 46988823c079SEric W. Biederman .name = "mnt", 46998823c079SEric W. Biederman .type = CLONE_NEWNS, 47008823c079SEric W. Biederman .get = mntns_get, 47018823c079SEric W. Biederman .put = mntns_put, 47028823c079SEric W. Biederman .install = mntns_install, 4703bcac25a5SAndrey Vagin .owner = mntns_owner, 47048823c079SEric W. Biederman }; 4705ab171b95SLuis Chamberlain 4706ab171b95SLuis Chamberlain #ifdef CONFIG_SYSCTL 4707ab171b95SLuis Chamberlain static struct ctl_table fs_namespace_sysctls[] = { 4708ab171b95SLuis Chamberlain { 4709ab171b95SLuis Chamberlain .procname = "mount-max", 4710ab171b95SLuis Chamberlain .data = &sysctl_mount_max, 4711ab171b95SLuis Chamberlain .maxlen = sizeof(unsigned int), 4712ab171b95SLuis Chamberlain .mode = 0644, 4713ab171b95SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 4714ab171b95SLuis Chamberlain .extra1 = SYSCTL_ONE, 4715ab171b95SLuis Chamberlain }, 4716ab171b95SLuis Chamberlain { } 4717ab171b95SLuis Chamberlain }; 4718ab171b95SLuis Chamberlain 4719ab171b95SLuis Chamberlain static int __init init_fs_namespace_sysctls(void) 4720ab171b95SLuis Chamberlain { 4721ab171b95SLuis Chamberlain register_sysctl_init("fs", fs_namespace_sysctls); 4722ab171b95SLuis Chamberlain return 0; 4723ab171b95SLuis Chamberlain } 4724ab171b95SLuis Chamberlain fs_initcall(init_fs_namespace_sysctls); 4725ab171b95SLuis Chamberlain 4726ab171b95SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 4727