11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 185b825c3aSIngo Molnar #include <linux/cred.h> 1973cd49ecSMiklos Szeredi #include <linux/idr.h> 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23d10577a8SAl Viro #include <linux/uaccess.h> 240bb80f24SDavid Howells #include <linux/proc_ns.h> 2520b4fb48SLinus Torvalds #include <linux/magic.h> 2657c8a661SMike Rapoport #include <linux/memblock.h> 279ea459e1SAl Viro #include <linux/task_work.h> 289164bb4aSIngo Molnar #include <linux/sched/task.h> 299164bb4aSIngo Molnar 3007b20889SRam Pai #include "pnode.h" 31948730b0SAdrian Bunk #include "internal.h" 321da177e4SLinus Torvalds 33d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 34d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 35d2921684SEric W. Biederman 360818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 370818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 380818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 390818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 400818bf27SAl Viro 410818bf27SAl Viro static __initdata unsigned long mhash_entries; 420818bf27SAl Viro static int __init set_mhash_entries(char *str) 430818bf27SAl Viro { 440818bf27SAl Viro if (!str) 450818bf27SAl Viro return 0; 460818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 470818bf27SAl Viro return 1; 480818bf27SAl Viro } 490818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 500818bf27SAl Viro 510818bf27SAl Viro static __initdata unsigned long mphash_entries; 520818bf27SAl Viro static int __init set_mphash_entries(char *str) 530818bf27SAl Viro { 540818bf27SAl Viro if (!str) 550818bf27SAl Viro return 0; 560818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 570818bf27SAl Viro return 1; 580818bf27SAl Viro } 590818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6013f14b4dSEric Dumazet 61c7999c36SAl Viro static u64 event; 6273cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 63719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 645addc5ddSAl Viro 6538129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 660818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 67e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 6859aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 691da177e4SLinus Torvalds 70f87fd4c2SMiklos Szeredi /* /sys/fs */ 7100d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7200d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 73f87fd4c2SMiklos Szeredi 7499b7db7bSNick Piggin /* 7599b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7699b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 7799b7db7bSNick Piggin * up the tree. 7899b7db7bSNick Piggin * 7999b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8099b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8199b7db7bSNick Piggin */ 8248a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8399b7db7bSNick Piggin 8438129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 851da177e4SLinus Torvalds { 861da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 871da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 880818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 890818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 900818bf27SAl Viro } 910818bf27SAl Viro 920818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 930818bf27SAl Viro { 940818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 950818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 960818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 971da177e4SLinus Torvalds } 981da177e4SLinus Torvalds 99b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10073cd49ecSMiklos Szeredi { 101169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 10273cd49ecSMiklos Szeredi 103169b480eSMatthew Wilcox if (res < 0) 10473cd49ecSMiklos Szeredi return res; 105169b480eSMatthew Wilcox mnt->mnt_id = res; 106169b480eSMatthew Wilcox return 0; 10773cd49ecSMiklos Szeredi } 10873cd49ecSMiklos Szeredi 109b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11073cd49ecSMiklos Szeredi { 111169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 11273cd49ecSMiklos Szeredi } 11373cd49ecSMiklos Szeredi 114719f5d7fSMiklos Szeredi /* 115719f5d7fSMiklos Szeredi * Allocate a new peer group ID 116719f5d7fSMiklos Szeredi */ 1174b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 118719f5d7fSMiklos Szeredi { 119169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 120f21f6220SAl Viro 121169b480eSMatthew Wilcox if (res < 0) 122f21f6220SAl Viro return res; 123169b480eSMatthew Wilcox mnt->mnt_group_id = res; 124169b480eSMatthew Wilcox return 0; 125719f5d7fSMiklos Szeredi } 126719f5d7fSMiklos Szeredi 127719f5d7fSMiklos Szeredi /* 128719f5d7fSMiklos Szeredi * Release a peer group ID 129719f5d7fSMiklos Szeredi */ 1304b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 131719f5d7fSMiklos Szeredi { 132169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 13315169fe7SAl Viro mnt->mnt_group_id = 0; 134719f5d7fSMiklos Szeredi } 135719f5d7fSMiklos Szeredi 136b3e19d92SNick Piggin /* 137b3e19d92SNick Piggin * vfsmount lock must be held for read 138b3e19d92SNick Piggin */ 13983adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 140b3e19d92SNick Piggin { 141b3e19d92SNick Piggin #ifdef CONFIG_SMP 14268e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 143b3e19d92SNick Piggin #else 144b3e19d92SNick Piggin preempt_disable(); 14568e8a9feSAl Viro mnt->mnt_count += n; 146b3e19d92SNick Piggin preempt_enable(); 147b3e19d92SNick Piggin #endif 148b3e19d92SNick Piggin } 149b3e19d92SNick Piggin 150b3e19d92SNick Piggin /* 151b3e19d92SNick Piggin * vfsmount lock must be held for write 152b3e19d92SNick Piggin */ 15383adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 154b3e19d92SNick Piggin { 155b3e19d92SNick Piggin #ifdef CONFIG_SMP 156f03c6599SAl Viro unsigned int count = 0; 157b3e19d92SNick Piggin int cpu; 158b3e19d92SNick Piggin 159b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 16068e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 161b3e19d92SNick Piggin } 162b3e19d92SNick Piggin 163b3e19d92SNick Piggin return count; 164b3e19d92SNick Piggin #else 16568e8a9feSAl Viro return mnt->mnt_count; 166b3e19d92SNick Piggin #endif 167b3e19d92SNick Piggin } 168b3e19d92SNick Piggin 16987b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 17087b95ce0SAl Viro { 17187b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 17287b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 17387b95ce0SAl Viro pin_remove(p); 17487b95ce0SAl Viro mntput(&m->mnt); 17587b95ce0SAl Viro } 17687b95ce0SAl Viro 177b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1781da177e4SLinus Torvalds { 179c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 180c63181e6SAl Viro if (mnt) { 18173cd49ecSMiklos Szeredi int err; 18273cd49ecSMiklos Szeredi 183c63181e6SAl Viro err = mnt_alloc_id(mnt); 18488b38782SLi Zefan if (err) 18588b38782SLi Zefan goto out_free_cache; 18688b38782SLi Zefan 18788b38782SLi Zefan if (name) { 188fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 189c63181e6SAl Viro if (!mnt->mnt_devname) 19088b38782SLi Zefan goto out_free_id; 19173cd49ecSMiklos Szeredi } 19273cd49ecSMiklos Szeredi 193b3e19d92SNick Piggin #ifdef CONFIG_SMP 194c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 195c63181e6SAl Viro if (!mnt->mnt_pcp) 196b3e19d92SNick Piggin goto out_free_devname; 197b3e19d92SNick Piggin 198c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 199b3e19d92SNick Piggin #else 200c63181e6SAl Viro mnt->mnt_count = 1; 201c63181e6SAl Viro mnt->mnt_writers = 0; 202b3e19d92SNick Piggin #endif 203b3e19d92SNick Piggin 20438129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 205c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 206c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 207c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 208c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 209c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 210c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 211c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2120a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 21399b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 21487b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2151da177e4SLinus Torvalds } 216c63181e6SAl Viro return mnt; 21788b38782SLi Zefan 218d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 219d3ef3d73Snpiggin@suse.de out_free_devname: 220fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 221d3ef3d73Snpiggin@suse.de #endif 22288b38782SLi Zefan out_free_id: 223c63181e6SAl Viro mnt_free_id(mnt); 22488b38782SLi Zefan out_free_cache: 225c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 22688b38782SLi Zefan return NULL; 2271da177e4SLinus Torvalds } 2281da177e4SLinus Torvalds 2298366025eSDave Hansen /* 2308366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2318366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2328366025eSDave Hansen * We must keep track of when those operations start 2338366025eSDave Hansen * (for permission checks) and when they end, so that 2348366025eSDave Hansen * we can determine when writes are able to occur to 2358366025eSDave Hansen * a filesystem. 2368366025eSDave Hansen */ 2373d733633SDave Hansen /* 2383d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2393d733633SDave Hansen * @mnt: the mount to check for its write status 2403d733633SDave Hansen * 2413d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2423d733633SDave Hansen * It does not guarantee that the filesystem will stay 2433d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2443d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2453d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2463d733633SDave Hansen * r/w. 2473d733633SDave Hansen */ 2483d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2493d733633SDave Hansen { 2502e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2512e4b7fcdSDave Hansen return 1; 252bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt_sb)) 2532e4b7fcdSDave Hansen return 1; 2542e4b7fcdSDave Hansen return 0; 2553d733633SDave Hansen } 2563d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2573d733633SDave Hansen 25883adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2593d733633SDave Hansen { 260d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 26168e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 262d3ef3d73Snpiggin@suse.de #else 26368e8a9feSAl Viro mnt->mnt_writers++; 264d3ef3d73Snpiggin@suse.de #endif 2653d733633SDave Hansen } 2663d733633SDave Hansen 26783adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2683d733633SDave Hansen { 269d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 27068e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 271d3ef3d73Snpiggin@suse.de #else 27268e8a9feSAl Viro mnt->mnt_writers--; 273d3ef3d73Snpiggin@suse.de #endif 274d3ef3d73Snpiggin@suse.de } 275d3ef3d73Snpiggin@suse.de 27683adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 277d3ef3d73Snpiggin@suse.de { 278d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 279d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2803d733633SDave Hansen int cpu; 2813d733633SDave Hansen 2823d733633SDave Hansen for_each_possible_cpu(cpu) { 28368e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2843d733633SDave Hansen } 2853d733633SDave Hansen 286d3ef3d73Snpiggin@suse.de return count; 287d3ef3d73Snpiggin@suse.de #else 288d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 289d3ef3d73Snpiggin@suse.de #endif 2903d733633SDave Hansen } 2913d733633SDave Hansen 2924ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2934ed5e82fSMiklos Szeredi { 2944ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2954ed5e82fSMiklos Szeredi return 1; 2964ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2974ed5e82fSMiklos Szeredi smp_rmb(); 2984ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2994ed5e82fSMiklos Szeredi } 3004ed5e82fSMiklos Szeredi 3013d733633SDave Hansen /* 302eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 303eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 304eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 305eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3063d733633SDave Hansen */ 3078366025eSDave Hansen /** 308eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 30983adc753SAl Viro * @m: the mount on which to take a write 3108366025eSDave Hansen * 311eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 312eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 313eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 314eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 315eb04c282SJan Kara * called. This is effectively a refcount. 3168366025eSDave Hansen */ 317eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3188366025eSDave Hansen { 31983adc753SAl Viro struct mount *mnt = real_mount(m); 3203d733633SDave Hansen int ret = 0; 3213d733633SDave Hansen 322d3ef3d73Snpiggin@suse.de preempt_disable(); 323c6653a83SNick Piggin mnt_inc_writers(mnt); 324d3ef3d73Snpiggin@suse.de /* 325c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 326d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 327d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 328d3ef3d73Snpiggin@suse.de */ 329d3ef3d73Snpiggin@suse.de smp_mb(); 3306aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 331d3ef3d73Snpiggin@suse.de cpu_relax(); 332d3ef3d73Snpiggin@suse.de /* 333d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 334d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 335d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 336d3ef3d73Snpiggin@suse.de */ 337d3ef3d73Snpiggin@suse.de smp_rmb(); 3384ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 339c6653a83SNick Piggin mnt_dec_writers(mnt); 3403d733633SDave Hansen ret = -EROFS; 3413d733633SDave Hansen } 342d3ef3d73Snpiggin@suse.de preempt_enable(); 343eb04c282SJan Kara 344eb04c282SJan Kara return ret; 345eb04c282SJan Kara } 346eb04c282SJan Kara 347eb04c282SJan Kara /** 348eb04c282SJan Kara * mnt_want_write - get write access to a mount 349eb04c282SJan Kara * @m: the mount on which to take a write 350eb04c282SJan Kara * 351eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 352eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 353eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 354eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 355eb04c282SJan Kara */ 356eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 357eb04c282SJan Kara { 358eb04c282SJan Kara int ret; 359eb04c282SJan Kara 360eb04c282SJan Kara sb_start_write(m->mnt_sb); 361eb04c282SJan Kara ret = __mnt_want_write(m); 362eb04c282SJan Kara if (ret) 363eb04c282SJan Kara sb_end_write(m->mnt_sb); 3643d733633SDave Hansen return ret; 3658366025eSDave Hansen } 3668366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3678366025eSDave Hansen 3688366025eSDave Hansen /** 36996029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 37096029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 37196029c4eSnpiggin@suse.de * 37296029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 37396029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 37496029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 37596029c4eSnpiggin@suse.de * on it. This allows some optimisation. 37696029c4eSnpiggin@suse.de * 37796029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 37896029c4eSnpiggin@suse.de * drop the reference. 37996029c4eSnpiggin@suse.de */ 38096029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 38196029c4eSnpiggin@suse.de { 38296029c4eSnpiggin@suse.de /* superblock may be r/o */ 38396029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 38496029c4eSnpiggin@suse.de return -EROFS; 38596029c4eSnpiggin@suse.de preempt_disable(); 38683adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 38796029c4eSnpiggin@suse.de preempt_enable(); 38896029c4eSnpiggin@suse.de return 0; 38996029c4eSnpiggin@suse.de } 39096029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 39196029c4eSnpiggin@suse.de 39296029c4eSnpiggin@suse.de /** 393eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 394eb04c282SJan Kara * @file: the file who's mount on which to take a write 395eb04c282SJan Kara * 396eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 397eb04c282SJan Kara * do some optimisations if the file is open for write already 398eb04c282SJan Kara */ 399eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 400eb04c282SJan Kara { 40183f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 402eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 403eb04c282SJan Kara else 404eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 405eb04c282SJan Kara } 406eb04c282SJan Kara 407eb04c282SJan Kara /** 4087c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4097c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4107c6893e3SMiklos Szeredi * 4117c6893e3SMiklos Szeredi * This is like mnt_want_write, but it takes a file and can 4127c6893e3SMiklos Szeredi * do some optimisations if the file is open for write already 4137c6893e3SMiklos Szeredi */ 4147c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4157c6893e3SMiklos Szeredi { 4167c6893e3SMiklos Szeredi int ret; 4177c6893e3SMiklos Szeredi 4187c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4197c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4207c6893e3SMiklos Szeredi if (ret) 4217c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4227c6893e3SMiklos Szeredi return ret; 4237c6893e3SMiklos Szeredi } 42496029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 42596029c4eSnpiggin@suse.de 42696029c4eSnpiggin@suse.de /** 427eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4288366025eSDave Hansen * @mnt: the mount on which to give up write access 4298366025eSDave Hansen * 4308366025eSDave Hansen * Tells the low-level filesystem that we are done 4318366025eSDave Hansen * performing writes to it. Must be matched with 432eb04c282SJan Kara * __mnt_want_write() call above. 4338366025eSDave Hansen */ 434eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4358366025eSDave Hansen { 436d3ef3d73Snpiggin@suse.de preempt_disable(); 43783adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 438d3ef3d73Snpiggin@suse.de preempt_enable(); 4398366025eSDave Hansen } 440eb04c282SJan Kara 441eb04c282SJan Kara /** 442eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 443eb04c282SJan Kara * @mnt: the mount on which to give up write access 444eb04c282SJan Kara * 445eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 446eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 447eb04c282SJan Kara * mnt_want_write() call above. 448eb04c282SJan Kara */ 449eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 450eb04c282SJan Kara { 451eb04c282SJan Kara __mnt_drop_write(mnt); 452eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 453eb04c282SJan Kara } 4548366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4558366025eSDave Hansen 456eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 457eb04c282SJan Kara { 458eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 459eb04c282SJan Kara } 460eb04c282SJan Kara 4617c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4627c6893e3SMiklos Szeredi { 463a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4647c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4657c6893e3SMiklos Szeredi } 4662a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4672a79f17eSAl Viro 46883adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4698366025eSDave Hansen { 4703d733633SDave Hansen int ret = 0; 4713d733633SDave Hansen 472719ea2fbSAl Viro lock_mount_hash(); 47383adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 474d3ef3d73Snpiggin@suse.de /* 475d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 476d3ef3d73Snpiggin@suse.de * should be visible before we do. 477d3ef3d73Snpiggin@suse.de */ 478d3ef3d73Snpiggin@suse.de smp_mb(); 479d3ef3d73Snpiggin@suse.de 480d3ef3d73Snpiggin@suse.de /* 481d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 482d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 483d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 484d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 485d3ef3d73Snpiggin@suse.de * 486d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 487d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 488d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 489d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 490d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 491d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 492d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 493d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 494d3ef3d73Snpiggin@suse.de * we're counting up here. 495d3ef3d73Snpiggin@suse.de */ 496c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 497d3ef3d73Snpiggin@suse.de ret = -EBUSY; 498d3ef3d73Snpiggin@suse.de else 49983adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 500d3ef3d73Snpiggin@suse.de /* 501d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 502d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 503d3ef3d73Snpiggin@suse.de */ 504d3ef3d73Snpiggin@suse.de smp_wmb(); 50583adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 506719ea2fbSAl Viro unlock_mount_hash(); 5073d733633SDave Hansen return ret; 5083d733633SDave Hansen } 5098366025eSDave Hansen 51083adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5112e4b7fcdSDave Hansen { 512719ea2fbSAl Viro lock_mount_hash(); 51383adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 514719ea2fbSAl Viro unlock_mount_hash(); 5152e4b7fcdSDave Hansen } 5162e4b7fcdSDave Hansen 5174ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5184ed5e82fSMiklos Szeredi { 5194ed5e82fSMiklos Szeredi struct mount *mnt; 5204ed5e82fSMiklos Szeredi int err = 0; 5214ed5e82fSMiklos Szeredi 5228e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5238e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5248e8b8796SMiklos Szeredi return -EBUSY; 5258e8b8796SMiklos Szeredi 526719ea2fbSAl Viro lock_mount_hash(); 5274ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5284ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5294ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5304ed5e82fSMiklos Szeredi smp_mb(); 5314ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5324ed5e82fSMiklos Szeredi err = -EBUSY; 5334ed5e82fSMiklos Szeredi break; 5344ed5e82fSMiklos Szeredi } 5354ed5e82fSMiklos Szeredi } 5364ed5e82fSMiklos Szeredi } 5378e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5388e8b8796SMiklos Szeredi err = -EBUSY; 5398e8b8796SMiklos Szeredi 5404ed5e82fSMiklos Szeredi if (!err) { 5414ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5424ed5e82fSMiklos Szeredi smp_wmb(); 5434ed5e82fSMiklos Szeredi } 5444ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5454ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5464ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5474ed5e82fSMiklos Szeredi } 548719ea2fbSAl Viro unlock_mount_hash(); 5494ed5e82fSMiklos Szeredi 5504ed5e82fSMiklos Szeredi return err; 5514ed5e82fSMiklos Szeredi } 5524ed5e82fSMiklos Szeredi 553b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5541da177e4SLinus Torvalds { 555fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 556d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 55768e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 558d3ef3d73Snpiggin@suse.de #endif 559b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5601da177e4SLinus Torvalds } 5611da177e4SLinus Torvalds 5628ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5638ffcb32eSDavid Howells { 5648ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5658ffcb32eSDavid Howells } 5668ffcb32eSDavid Howells 56748a066e7SAl Viro /* call under rcu_read_lock */ 568294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 56948a066e7SAl Viro { 57048a066e7SAl Viro struct mount *mnt; 57148a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 572294d71ffSAl Viro return 1; 57348a066e7SAl Viro if (bastard == NULL) 574294d71ffSAl Viro return 0; 57548a066e7SAl Viro mnt = real_mount(bastard); 57648a066e7SAl Viro mnt_add_count(mnt, 1); 577119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 57848a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 579294d71ffSAl Viro return 0; 58048a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 58148a066e7SAl Viro mnt_add_count(mnt, -1); 582294d71ffSAl Viro return 1; 58348a066e7SAl Viro } 584119e1ef8SAl Viro lock_mount_hash(); 585119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 586119e1ef8SAl Viro mnt_add_count(mnt, -1); 587119e1ef8SAl Viro unlock_mount_hash(); 588119e1ef8SAl Viro return 1; 589119e1ef8SAl Viro } 590119e1ef8SAl Viro unlock_mount_hash(); 591119e1ef8SAl Viro /* caller will mntput() */ 592294d71ffSAl Viro return -1; 593294d71ffSAl Viro } 594294d71ffSAl Viro 595294d71ffSAl Viro /* call under rcu_read_lock */ 596294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 597294d71ffSAl Viro { 598294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 599294d71ffSAl Viro if (likely(!res)) 600294d71ffSAl Viro return true; 601294d71ffSAl Viro if (unlikely(res < 0)) { 60248a066e7SAl Viro rcu_read_unlock(); 60348a066e7SAl Viro mntput(bastard); 60448a066e7SAl Viro rcu_read_lock(); 605294d71ffSAl Viro } 60648a066e7SAl Viro return false; 60748a066e7SAl Viro } 60848a066e7SAl Viro 6091da177e4SLinus Torvalds /* 610474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 61148a066e7SAl Viro * call under rcu_read_lock() 6121da177e4SLinus Torvalds */ 613474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6141da177e4SLinus Torvalds { 61538129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 616474279dcSAl Viro struct mount *p; 6171da177e4SLinus Torvalds 61838129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 619474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 620474279dcSAl Viro return p; 621474279dcSAl Viro return NULL; 6221da177e4SLinus Torvalds } 623474279dcSAl Viro 624474279dcSAl Viro /* 625f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 626f015f126SDavid Howells * 627f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 628f015f126SDavid Howells * following mounts: 629f015f126SDavid Howells * 630f015f126SDavid Howells * mount /dev/sda1 /mnt 631f015f126SDavid Howells * mount /dev/sda2 /mnt 632f015f126SDavid Howells * mount /dev/sda3 /mnt 633f015f126SDavid Howells * 634f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 635f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 636f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 637f015f126SDavid Howells * 638f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 639a05964f3SRam Pai */ 640ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 641a05964f3SRam Pai { 642c7105365SAl Viro struct mount *child_mnt; 64348a066e7SAl Viro struct vfsmount *m; 64448a066e7SAl Viro unsigned seq; 64599b7db7bSNick Piggin 64648a066e7SAl Viro rcu_read_lock(); 64748a066e7SAl Viro do { 64848a066e7SAl Viro seq = read_seqbegin(&mount_lock); 649474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 65048a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 65148a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 65248a066e7SAl Viro rcu_read_unlock(); 65348a066e7SAl Viro return m; 654a05964f3SRam Pai } 655a05964f3SRam Pai 6567af1364fSEric W. Biederman /* 6577af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6587af1364fSEric W. Biederman * current mount namespace. 6597af1364fSEric W. Biederman * 6607af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6617af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6627af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6637af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6647af1364fSEric W. Biederman * is a mountpoint. 6657af1364fSEric W. Biederman * 6667af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6677af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6687af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6697af1364fSEric W. Biederman * parent mount. 6707af1364fSEric W. Biederman */ 6717af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6727af1364fSEric W. Biederman { 6737af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 6747af1364fSEric W. Biederman struct mount *mnt; 6757af1364fSEric W. Biederman bool is_covered = false; 6767af1364fSEric W. Biederman 6777af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 6787af1364fSEric W. Biederman goto out; 6797af1364fSEric W. Biederman 6807af1364fSEric W. Biederman down_read(&namespace_sem); 6817af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 6827af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 6837af1364fSEric W. Biederman if (is_covered) 6847af1364fSEric W. Biederman break; 6857af1364fSEric W. Biederman } 6867af1364fSEric W. Biederman up_read(&namespace_sem); 6877af1364fSEric W. Biederman out: 6887af1364fSEric W. Biederman return is_covered; 6897af1364fSEric W. Biederman } 6907af1364fSEric W. Biederman 691e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 69284d17192SAl Viro { 6930818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 69484d17192SAl Viro struct mountpoint *mp; 69584d17192SAl Viro 6960818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 69784d17192SAl Viro if (mp->m_dentry == dentry) { 69884d17192SAl Viro mp->m_count++; 69984d17192SAl Viro return mp; 70084d17192SAl Viro } 70184d17192SAl Viro } 702e2dfa935SEric W. Biederman return NULL; 703e2dfa935SEric W. Biederman } 704e2dfa935SEric W. Biederman 7053895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 706e2dfa935SEric W. Biederman { 7073895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 708e2dfa935SEric W. Biederman int ret; 70984d17192SAl Viro 7103895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7111e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7121e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7131e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7143895dbf8SEric W. Biederman mountpoint: 7153895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7163895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7173895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7183895dbf8SEric W. Biederman if (mp) 7193895dbf8SEric W. Biederman goto done; 72084d17192SAl Viro } 721eed81007SMiklos Szeredi 7223895dbf8SEric W. Biederman if (!new) 7233895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7243895dbf8SEric W. Biederman if (!new) 7253895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7263895dbf8SEric W. Biederman 7273895dbf8SEric W. Biederman 7283895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7293895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7303895dbf8SEric W. Biederman 7313895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7323895dbf8SEric W. Biederman if (ret == -EBUSY) 7333895dbf8SEric W. Biederman goto mountpoint; 7343895dbf8SEric W. Biederman 7353895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7363895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7373895dbf8SEric W. Biederman if (ret) 7383895dbf8SEric W. Biederman goto done; 7393895dbf8SEric W. Biederman 7403895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7413895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7423895dbf8SEric W. Biederman new->m_dentry = dentry; 7433895dbf8SEric W. Biederman new->m_count = 1; 7443895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7453895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7463895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7473895dbf8SEric W. Biederman 7483895dbf8SEric W. Biederman mp = new; 7493895dbf8SEric W. Biederman new = NULL; 7503895dbf8SEric W. Biederman done: 7513895dbf8SEric W. Biederman kfree(new); 75284d17192SAl Viro return mp; 75384d17192SAl Viro } 75484d17192SAl Viro 75584d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 75684d17192SAl Viro { 75784d17192SAl Viro if (!--mp->m_count) { 75884d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7590a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 76084d17192SAl Viro spin_lock(&dentry->d_lock); 76184d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 76284d17192SAl Viro spin_unlock(&dentry->d_lock); 7630818bf27SAl Viro hlist_del(&mp->m_hash); 76484d17192SAl Viro kfree(mp); 76584d17192SAl Viro } 76684d17192SAl Viro } 76784d17192SAl Viro 768143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7691da177e4SLinus Torvalds { 7706b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds 77399b7db7bSNick Piggin /* 77499b7db7bSNick Piggin * vfsmount lock must be held for write 77599b7db7bSNick Piggin */ 7766b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7775addc5ddSAl Viro { 7785addc5ddSAl Viro if (ns) { 7795addc5ddSAl Viro ns->event = ++event; 7805addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7815addc5ddSAl Viro } 7825addc5ddSAl Viro } 7835addc5ddSAl Viro 78499b7db7bSNick Piggin /* 78599b7db7bSNick Piggin * vfsmount lock must be held for write 78699b7db7bSNick Piggin */ 7876b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 7885addc5ddSAl Viro { 7895addc5ddSAl Viro if (ns && ns->event != event) { 7905addc5ddSAl Viro ns->event = event; 7915addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7925addc5ddSAl Viro } 7935addc5ddSAl Viro } 7945addc5ddSAl Viro 79599b7db7bSNick Piggin /* 79699b7db7bSNick Piggin * vfsmount lock must be held for write 79799b7db7bSNick Piggin */ 7987bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 7991da177e4SLinus Torvalds { 8000714a533SAl Viro mnt->mnt_parent = mnt; 801a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8026b41d536SAl Viro list_del_init(&mnt->mnt_child); 80338129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8040a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 80584d17192SAl Viro put_mountpoint(mnt->mnt_mp); 80684d17192SAl Viro mnt->mnt_mp = NULL; 8071da177e4SLinus Torvalds } 8081da177e4SLinus Torvalds 80999b7db7bSNick Piggin /* 81099b7db7bSNick Piggin * vfsmount lock must be held for write 81199b7db7bSNick Piggin */ 8127bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8137bdb11deSEric W. Biederman { 8147bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8157bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8167bdb11deSEric W. Biederman unhash_mnt(mnt); 8177bdb11deSEric W. Biederman } 8187bdb11deSEric W. Biederman 8197bdb11deSEric W. Biederman /* 8207bdb11deSEric W. Biederman * vfsmount lock must be held for write 8217bdb11deSEric W. Biederman */ 8226a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8236a46c573SEric W. Biederman { 8246a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8256a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8266a46c573SEric W. Biederman unhash_mnt(mnt); 8276a46c573SEric W. Biederman } 8286a46c573SEric W. Biederman 8296a46c573SEric W. Biederman /* 8306a46c573SEric W. Biederman * vfsmount lock must be held for write 8316a46c573SEric W. Biederman */ 83284d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 83384d17192SAl Viro struct mountpoint *mp, 83444d964d6SAl Viro struct mount *child_mnt) 835b90fa9aeSRam Pai { 83684d17192SAl Viro mp->m_count++; 8373a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 83884d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8393a2393d7SAl Viro child_mnt->mnt_parent = mnt; 84084d17192SAl Viro child_mnt->mnt_mp = mp; 8410a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 842b90fa9aeSRam Pai } 843b90fa9aeSRam Pai 8441064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8451064f874SEric W. Biederman { 8461064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8471064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8481064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8491064f874SEric W. Biederman } 8501064f874SEric W. Biederman 85199b7db7bSNick Piggin /* 85299b7db7bSNick Piggin * vfsmount lock must be held for write 85399b7db7bSNick Piggin */ 85484d17192SAl Viro static void attach_mnt(struct mount *mnt, 85584d17192SAl Viro struct mount *parent, 85684d17192SAl Viro struct mountpoint *mp) 8571da177e4SLinus Torvalds { 85884d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8591064f874SEric W. Biederman __attach_mnt(mnt, parent); 860b90fa9aeSRam Pai } 861b90fa9aeSRam Pai 8621064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 86312a5b529SAl Viro { 8641064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8651064f874SEric W. Biederman struct dentry *old_mountpoint = mnt->mnt_mountpoint; 8661064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8671064f874SEric W. Biederman 8681064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8691064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8701064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8711064f874SEric W. Biederman 8721064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 8731064f874SEric W. Biederman 8741064f874SEric W. Biederman put_mountpoint(old_mp); 8751064f874SEric W. Biederman 8761064f874SEric W. Biederman /* 8771064f874SEric W. Biederman * Safely avoid even the suggestion this code might sleep or 8781064f874SEric W. Biederman * lock the mount hash by taking advantage of the knowledge that 8791064f874SEric W. Biederman * mnt_change_mountpoint will not release the final reference 8801064f874SEric W. Biederman * to a mountpoint. 8811064f874SEric W. Biederman * 8821064f874SEric W. Biederman * During mounting, the mount passed in as the parent mount will 8831064f874SEric W. Biederman * continue to use the old mountpoint and during unmounting, the 8841064f874SEric W. Biederman * old mountpoint will continue to exist until namespace_unlock, 8851064f874SEric W. Biederman * which happens well after mnt_change_mountpoint. 8861064f874SEric W. Biederman */ 8871064f874SEric W. Biederman spin_lock(&old_mountpoint->d_lock); 8881064f874SEric W. Biederman old_mountpoint->d_lockref.count--; 8891064f874SEric W. Biederman spin_unlock(&old_mountpoint->d_lock); 8901064f874SEric W. Biederman 8911064f874SEric W. Biederman mnt_add_count(old_parent, -1); 89212a5b529SAl Viro } 89312a5b529SAl Viro 894b90fa9aeSRam Pai /* 89599b7db7bSNick Piggin * vfsmount lock must be held for write 896b90fa9aeSRam Pai */ 8971064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 898b90fa9aeSRam Pai { 8990714a533SAl Viro struct mount *parent = mnt->mnt_parent; 90083adc753SAl Viro struct mount *m; 901b90fa9aeSRam Pai LIST_HEAD(head); 902143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 903b90fa9aeSRam Pai 9040714a533SAl Viro BUG_ON(parent == mnt); 905b90fa9aeSRam Pai 9061a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 907f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 908143c8c91SAl Viro m->mnt_ns = n; 909f03c6599SAl Viro 910b90fa9aeSRam Pai list_splice(&head, n->list.prev); 911b90fa9aeSRam Pai 912d2921684SEric W. Biederman n->mounts += n->pending_mounts; 913d2921684SEric W. Biederman n->pending_mounts = 0; 914d2921684SEric W. Biederman 9151064f874SEric W. Biederman __attach_mnt(mnt, parent); 9166b3286edSKirill Korotaev touch_mnt_namespace(n); 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds 919909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9201da177e4SLinus Torvalds { 9216b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9226b41d536SAl Viro if (next == &p->mnt_mounts) { 9231da177e4SLinus Torvalds while (1) { 924909b0a88SAl Viro if (p == root) 9251da177e4SLinus Torvalds return NULL; 9266b41d536SAl Viro next = p->mnt_child.next; 9276b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9281da177e4SLinus Torvalds break; 9290714a533SAl Viro p = p->mnt_parent; 9301da177e4SLinus Torvalds } 9311da177e4SLinus Torvalds } 9326b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9331da177e4SLinus Torvalds } 9341da177e4SLinus Torvalds 935315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9369676f0c6SRam Pai { 9376b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9386b41d536SAl Viro while (prev != &p->mnt_mounts) { 9396b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9406b41d536SAl Viro prev = p->mnt_mounts.prev; 9419676f0c6SRam Pai } 9429676f0c6SRam Pai return p; 9439676f0c6SRam Pai } 9449676f0c6SRam Pai 9459d412a43SAl Viro struct vfsmount * 9469d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 9479d412a43SAl Viro { 948b105e270SAl Viro struct mount *mnt; 9499d412a43SAl Viro struct dentry *root; 9509d412a43SAl Viro 9519d412a43SAl Viro if (!type) 9529d412a43SAl Viro return ERR_PTR(-ENODEV); 9539d412a43SAl Viro 9549d412a43SAl Viro mnt = alloc_vfsmnt(name); 9559d412a43SAl Viro if (!mnt) 9569d412a43SAl Viro return ERR_PTR(-ENOMEM); 9579d412a43SAl Viro 958e462ec50SDavid Howells if (flags & SB_KERNMOUNT) 959b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9609d412a43SAl Viro 9619d412a43SAl Viro root = mount_fs(type, flags, name, data); 9629d412a43SAl Viro if (IS_ERR(root)) { 9638ffcb32eSDavid Howells mnt_free_id(mnt); 9649d412a43SAl Viro free_vfsmnt(mnt); 9659d412a43SAl Viro return ERR_CAST(root); 9669d412a43SAl Viro } 9679d412a43SAl Viro 968b105e270SAl Viro mnt->mnt.mnt_root = root; 969b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 970a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9710714a533SAl Viro mnt->mnt_parent = mnt; 972719ea2fbSAl Viro lock_mount_hash(); 97339f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 974719ea2fbSAl Viro unlock_mount_hash(); 975b105e270SAl Viro return &mnt->mnt; 9769d412a43SAl Viro } 9779d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 9789d412a43SAl Viro 97993faccbbSEric W. Biederman struct vfsmount * 98093faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 98193faccbbSEric W. Biederman const char *name, void *data) 98293faccbbSEric W. Biederman { 98393faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 98493faccbbSEric W. Biederman * through from the parent mount to the submount don't support 98593faccbbSEric W. Biederman * unprivileged mounts with submounts. 98693faccbbSEric W. Biederman */ 98793faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 98893faccbbSEric W. Biederman return ERR_PTR(-EPERM); 98993faccbbSEric W. Biederman 990e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 99193faccbbSEric W. Biederman } 99293faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 99393faccbbSEric W. Biederman 99487129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 99536341f64SRam Pai int flag) 9961da177e4SLinus Torvalds { 99787129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 998be34d1a3SDavid Howells struct mount *mnt; 999be34d1a3SDavid Howells int err; 10001da177e4SLinus Torvalds 1001be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1002be34d1a3SDavid Howells if (!mnt) 1003be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1004be34d1a3SDavid Howells 10057a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 100615169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1007719f5d7fSMiklos Szeredi else 100815169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1009719f5d7fSMiklos Szeredi 101015169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1011be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1012719f5d7fSMiklos Szeredi if (err) 1013719f5d7fSMiklos Szeredi goto out_free; 1014719f5d7fSMiklos Szeredi } 1015719f5d7fSMiklos Szeredi 101616a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 101716a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 1018132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 10199566d674SEric W. Biederman if (flag & CL_UNPRIVILEGED) { 10209566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; 10219566d674SEric W. Biederman 10229566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_READONLY) 1023132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 1024132c94e3SEric W. Biederman 10259566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NODEV) 10269566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; 10279566d674SEric W. Biederman 10289566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOSUID) 10299566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; 10309566d674SEric W. Biederman 10319566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOEXEC) 10329566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; 10339566d674SEric W. Biederman } 10349566d674SEric W. Biederman 10355ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 1036381cacb1SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && 1037381cacb1SEric W. Biederman (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) 10385ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 10395ff9d8a6SEric W. Biederman 10401da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1041b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1042b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1043a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10440714a533SAl Viro mnt->mnt_parent = mnt; 1045719ea2fbSAl Viro lock_mount_hash(); 104639f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1047719ea2fbSAl Viro unlock_mount_hash(); 1048b90fa9aeSRam Pai 10497a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10507a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10516776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 105232301920SAl Viro mnt->mnt_master = old; 1053fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10548aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1055fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10566776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1057d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10586776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1059d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10605235d448SAl Viro } else { 10615235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10625afe0022SRam Pai } 1063b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10640f0afb1dSAl Viro set_mnt_shared(mnt); 10651da177e4SLinus Torvalds 10661da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10671da177e4SLinus Torvalds * as the original if that was on one */ 106836341f64SRam Pai if (flag & CL_EXPIRE) { 10696776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10706776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10711da177e4SLinus Torvalds } 1072be34d1a3SDavid Howells 1073cb338d06SAl Viro return mnt; 1074719f5d7fSMiklos Szeredi 1075719f5d7fSMiklos Szeredi out_free: 10768ffcb32eSDavid Howells mnt_free_id(mnt); 1077719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1078be34d1a3SDavid Howells return ERR_PTR(err); 10791da177e4SLinus Torvalds } 10801da177e4SLinus Torvalds 10819ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10829ea459e1SAl Viro { 10839ea459e1SAl Viro /* 10849ea459e1SAl Viro * This probably indicates that somebody messed 10859ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 10869ea459e1SAl Viro * happens, the filesystem was probably unable 10879ea459e1SAl Viro * to make r/w->r/o transitions. 10889ea459e1SAl Viro */ 10899ea459e1SAl Viro /* 10909ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 10919ea459e1SAl Viro * so mnt_get_writers() below is safe. 10929ea459e1SAl Viro */ 10939ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 10949ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 10959ea459e1SAl Viro mnt_pin_kill(mnt); 10969ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 10979ea459e1SAl Viro dput(mnt->mnt.mnt_root); 10989ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 10999ea459e1SAl Viro mnt_free_id(mnt); 11009ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11019ea459e1SAl Viro } 11029ea459e1SAl Viro 11039ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11049ea459e1SAl Viro { 11059ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11069ea459e1SAl Viro } 11079ea459e1SAl Viro 11089ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11099ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11109ea459e1SAl Viro { 11119ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 111229785735SByungchul Park struct mount *m, *t; 11139ea459e1SAl Viro 111429785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 111529785735SByungchul Park cleanup_mnt(m); 11169ea459e1SAl Viro } 11179ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11189ea459e1SAl Viro 1119900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11207b7b1aceSAl Viro { 112148a066e7SAl Viro rcu_read_lock(); 11229ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11239ea0a46cSAl Viro /* 11249ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11259ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11269ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11279ea0a46cSAl Viro * be dropped until after an RCU delay done after 11289ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11299ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11309ea0a46cSAl Viro * we are dropping is not the final one. 11319ea0a46cSAl Viro */ 1132aa9c0e07SAl Viro mnt_add_count(mnt, -1); 113348a066e7SAl Viro rcu_read_unlock(); 113499b7db7bSNick Piggin return; 1135b3e19d92SNick Piggin } 1136719ea2fbSAl Viro lock_mount_hash(); 1137119e1ef8SAl Viro /* 1138119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1139119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1140119e1ef8SAl Viro */ 1141119e1ef8SAl Viro smp_mb(); 11429ea0a46cSAl Viro mnt_add_count(mnt, -1); 1143b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 114448a066e7SAl Viro rcu_read_unlock(); 1145719ea2fbSAl Viro unlock_mount_hash(); 114699b7db7bSNick Piggin return; 114799b7db7bSNick Piggin } 114848a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 114948a066e7SAl Viro rcu_read_unlock(); 115048a066e7SAl Viro unlock_mount_hash(); 115148a066e7SAl Viro return; 115248a066e7SAl Viro } 115348a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 115448a066e7SAl Viro rcu_read_unlock(); 1155962830dfSAndi Kleen 115639f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1157ce07d891SEric W. Biederman 1158ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1159ce07d891SEric W. Biederman struct mount *p, *tmp; 1160ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1161ce07d891SEric W. Biederman umount_mnt(p); 1162ce07d891SEric W. Biederman } 1163ce07d891SEric W. Biederman } 1164719ea2fbSAl Viro unlock_mount_hash(); 1165649a795aSAl Viro 11669ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11679ea459e1SAl Viro struct task_struct *task = current; 11689ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11699ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11709ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11719ea459e1SAl Viro return; 11729ea459e1SAl Viro } 11739ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11749ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11759ea459e1SAl Viro return; 11769ea459e1SAl Viro } 11779ea459e1SAl Viro cleanup_mnt(mnt); 1178b3e19d92SNick Piggin } 1179b3e19d92SNick Piggin 1180b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1181b3e19d92SNick Piggin { 1182b3e19d92SNick Piggin if (mnt) { 1183863d684fSAl Viro struct mount *m = real_mount(mnt); 1184b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1185863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1186863d684fSAl Viro m->mnt_expiry_mark = 0; 1187863d684fSAl Viro mntput_no_expire(m); 1188b3e19d92SNick Piggin } 1189b3e19d92SNick Piggin } 1190b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1191b3e19d92SNick Piggin 1192b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1193b3e19d92SNick Piggin { 1194b3e19d92SNick Piggin if (mnt) 119583adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1196b3e19d92SNick Piggin return mnt; 1197b3e19d92SNick Piggin } 1198b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1199b3e19d92SNick Piggin 1200c6609c0aSIan Kent /* path_is_mountpoint() - Check if path is a mount in the current 1201c6609c0aSIan Kent * namespace. 1202c6609c0aSIan Kent * 1203c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1204c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1205c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1206c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1207c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1208c6609c0aSIan Kent * alone. 1209c6609c0aSIan Kent */ 1210c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1211c6609c0aSIan Kent { 1212c6609c0aSIan Kent unsigned seq; 1213c6609c0aSIan Kent bool res; 1214c6609c0aSIan Kent 1215c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1216c6609c0aSIan Kent return false; 1217c6609c0aSIan Kent 1218c6609c0aSIan Kent rcu_read_lock(); 1219c6609c0aSIan Kent do { 1220c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1221c6609c0aSIan Kent res = __path_is_mountpoint(path); 1222c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1223c6609c0aSIan Kent rcu_read_unlock(); 1224c6609c0aSIan Kent 1225c6609c0aSIan Kent return res; 1226c6609c0aSIan Kent } 1227c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1228c6609c0aSIan Kent 1229ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12307b7b1aceSAl Viro { 12313064c356SAl Viro struct mount *p; 12323064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12333064c356SAl Viro if (IS_ERR(p)) 12343064c356SAl Viro return ERR_CAST(p); 12353064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12363064c356SAl Viro return &p->mnt; 12377b7b1aceSAl Viro } 12381da177e4SLinus Torvalds 1239a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12400226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12411da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12421da177e4SLinus Torvalds { 1243ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12441da177e4SLinus Torvalds 1245390c6843SRam Pai down_read(&namespace_sem); 1246c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1247c7999c36SAl Viro void *v = p->cached_mount; 1248c7999c36SAl Viro if (*pos == p->cached_index) 1249c7999c36SAl Viro return v; 1250c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1251c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1252c7999c36SAl Viro return p->cached_mount = v; 1253c7999c36SAl Viro } 1254c7999c36SAl Viro } 1255c7999c36SAl Viro 1256c7999c36SAl Viro p->cached_event = p->ns->event; 1257c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1258c7999c36SAl Viro p->cached_index = *pos; 1259c7999c36SAl Viro return p->cached_mount; 12601da177e4SLinus Torvalds } 12611da177e4SLinus Torvalds 12621da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12631da177e4SLinus Torvalds { 1264ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1265b0765fb8SPavel Emelianov 1266c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1267c7999c36SAl Viro p->cached_index = *pos; 1268c7999c36SAl Viro return p->cached_mount; 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 12711da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12721da177e4SLinus Torvalds { 1273390c6843SRam Pai up_read(&namespace_sem); 12741da177e4SLinus Torvalds } 12751da177e4SLinus Torvalds 12760226f492SAl Viro static int m_show(struct seq_file *m, void *v) 12779f5596afSAl Viro { 1278ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12791a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 12800226f492SAl Viro return p->show(m, &r->mnt); 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds 1283a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 12841da177e4SLinus Torvalds .start = m_start, 12851da177e4SLinus Torvalds .next = m_next, 12861da177e4SLinus Torvalds .stop = m_stop, 12870226f492SAl Viro .show = m_show, 1288b4629fe2SChuck Lever }; 1289a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1290b4629fe2SChuck Lever 12911da177e4SLinus Torvalds /** 12921da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 12931da177e4SLinus Torvalds * @mnt: root of mount tree 12941da177e4SLinus Torvalds * 12951da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 12961da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 12971da177e4SLinus Torvalds * busy. 12981da177e4SLinus Torvalds */ 1299909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13001da177e4SLinus Torvalds { 1301909b0a88SAl Viro struct mount *mnt = real_mount(m); 130236341f64SRam Pai int actual_refs = 0; 130336341f64SRam Pai int minimum_refs = 0; 1304315fc83eSAl Viro struct mount *p; 1305909b0a88SAl Viro BUG_ON(!m); 13061da177e4SLinus Torvalds 1307b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1308719ea2fbSAl Viro lock_mount_hash(); 1309909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 131083adc753SAl Viro actual_refs += mnt_get_count(p); 13111da177e4SLinus Torvalds minimum_refs += 2; 13121da177e4SLinus Torvalds } 1313719ea2fbSAl Viro unlock_mount_hash(); 13141da177e4SLinus Torvalds 13151da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13161da177e4SLinus Torvalds return 0; 1317e3474a8eSIan Kent 1318e3474a8eSIan Kent return 1; 13191da177e4SLinus Torvalds } 13201da177e4SLinus Torvalds 13211da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds /** 13241da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13251da177e4SLinus Torvalds * @mnt: root of mount 13261da177e4SLinus Torvalds * 13271da177e4SLinus Torvalds * This is called to check if a mount point has any 13281da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13291da177e4SLinus Torvalds * mount has sub mounts this will return busy 13301da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13311da177e4SLinus Torvalds * 13321da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13331da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13341da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13351da177e4SLinus Torvalds */ 13361da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13371da177e4SLinus Torvalds { 1338e3474a8eSIan Kent int ret = 1; 13398ad08d8aSAl Viro down_read(&namespace_sem); 1340719ea2fbSAl Viro lock_mount_hash(); 13411ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1342e3474a8eSIan Kent ret = 0; 1343719ea2fbSAl Viro unlock_mount_hash(); 13448ad08d8aSAl Viro up_read(&namespace_sem); 1345a05964f3SRam Pai return ret; 13461da177e4SLinus Torvalds } 13471da177e4SLinus Torvalds 13481da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13491da177e4SLinus Torvalds 135038129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1351e3197d83SAl Viro 135297216be0SAl Viro static void namespace_unlock(void) 13531da177e4SLinus Torvalds { 1354a3b3c562SEric W. Biederman struct hlist_head head; 135597216be0SAl Viro 1356a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1357a3b3c562SEric W. Biederman 135897216be0SAl Viro up_write(&namespace_sem); 1359a3b3c562SEric W. Biederman 1360a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 136197216be0SAl Viro return; 136297216be0SAl Viro 136348a066e7SAl Viro synchronize_rcu(); 136448a066e7SAl Viro 136587b95ce0SAl Viro group_pin_kill(&head); 136670fbcdf4SRam Pai } 136770fbcdf4SRam Pai 136897216be0SAl Viro static inline void namespace_lock(void) 1369e3197d83SAl Viro { 137097216be0SAl Viro down_write(&namespace_sem); 1371e3197d83SAl Viro } 1372e3197d83SAl Viro 1373e819f152SEric W. Biederman enum umount_tree_flags { 1374e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1375e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1376e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1377e819f152SEric W. Biederman }; 1378f2d0a123SEric W. Biederman 1379f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1380f2d0a123SEric W. Biederman { 1381f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1382f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1383f2d0a123SEric W. Biederman return true; 1384f2d0a123SEric W. Biederman 1385f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1386f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1387f2d0a123SEric W. Biederman return true; 1388f2d0a123SEric W. Biederman 1389f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1390f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1391f2d0a123SEric W. Biederman * connected to mounted mounts. 1392f2d0a123SEric W. Biederman */ 1393f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1394f2d0a123SEric W. Biederman return true; 1395f2d0a123SEric W. Biederman 1396f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1397f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1398f2d0a123SEric W. Biederman return false; 1399f2d0a123SEric W. Biederman 1400f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1401f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1402f2d0a123SEric W. Biederman return false; 1403f2d0a123SEric W. Biederman 1404f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1405f2d0a123SEric W. Biederman return true; 1406f2d0a123SEric W. Biederman } 1407f2d0a123SEric W. Biederman 140899b7db7bSNick Piggin /* 140948a066e7SAl Viro * mount_lock must be held 141099b7db7bSNick Piggin * namespace_sem must be held for write 141199b7db7bSNick Piggin */ 1412e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 141370fbcdf4SRam Pai { 1414c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1415315fc83eSAl Viro struct mount *p; 141670fbcdf4SRam Pai 14175d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14185d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14195d88457eSEric W. Biederman 1420c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1421590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1422590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1423c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1424590ce4bcSEric W. Biederman } 1425c003b26fSEric W. Biederman 1426411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1427c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1428c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 142938129a13SAl Viro } 143070fbcdf4SRam Pai 1431c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1432e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14337b8a53fdSAl Viro propagate_umount(&tmp_list); 1434a05964f3SRam Pai 1435c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1436d2921684SEric W. Biederman struct mnt_namespace *ns; 1437ce07d891SEric W. Biederman bool disconnect; 1438c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 14396776db3dSAl Viro list_del_init(&p->mnt_expire); 14401a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1441d2921684SEric W. Biederman ns = p->mnt_ns; 1442d2921684SEric W. Biederman if (ns) { 1443d2921684SEric W. Biederman ns->mounts--; 1444d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1445d2921684SEric W. Biederman } 144663d37a84SAl Viro p->mnt_ns = NULL; 1447e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 144848a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 144987b95ce0SAl Viro 1450f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1451ce07d891SEric W. Biederman 1452ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1453ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1454676da58dSAl Viro if (mnt_has_parent(p)) { 145581b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1456ce07d891SEric W. Biederman if (!disconnect) { 1457ce07d891SEric W. Biederman /* Don't forget about p */ 1458ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1459ce07d891SEric W. Biederman } else { 14606a46c573SEric W. Biederman umount_mnt(p); 14617c4b93d8SAl Viro } 1462ce07d891SEric W. Biederman } 14630f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 146438129a13SAl Viro } 14651da177e4SLinus Torvalds } 14661da177e4SLinus Torvalds 1467b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1468c35038beSAl Viro 14691ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 14701da177e4SLinus Torvalds { 14711ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 14721da177e4SLinus Torvalds int retval; 14731da177e4SLinus Torvalds 14741ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 14751da177e4SLinus Torvalds if (retval) 14761da177e4SLinus Torvalds return retval; 14771da177e4SLinus Torvalds 14781da177e4SLinus Torvalds /* 14791da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 14801da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 14811da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 14821da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 14831da177e4SLinus Torvalds */ 14841da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 14851ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 14861da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 14871da177e4SLinus Torvalds return -EINVAL; 14881da177e4SLinus Torvalds 1489b3e19d92SNick Piggin /* 1490b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1491b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1492b3e19d92SNick Piggin */ 1493719ea2fbSAl Viro lock_mount_hash(); 149483adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1495719ea2fbSAl Viro unlock_mount_hash(); 14961da177e4SLinus Torvalds return -EBUSY; 1497b3e19d92SNick Piggin } 1498719ea2fbSAl Viro unlock_mount_hash(); 14991da177e4SLinus Torvalds 1500863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 15011da177e4SLinus Torvalds return -EAGAIN; 15021da177e4SLinus Torvalds } 15031da177e4SLinus Torvalds 15041da177e4SLinus Torvalds /* 15051da177e4SLinus Torvalds * If we may have to abort operations to get out of this 15061da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 15071da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 15081da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 15091da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 15101da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 15111da177e4SLinus Torvalds * about for the moment. 15121da177e4SLinus Torvalds */ 15131da177e4SLinus Torvalds 151442faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 151542faad99SAl Viro sb->s_op->umount_begin(sb); 151642faad99SAl Viro } 15171da177e4SLinus Torvalds 15181da177e4SLinus Torvalds /* 15191da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15201da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15211da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15221da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15231da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15241da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15251da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15261da177e4SLinus Torvalds */ 15271ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15281da177e4SLinus Torvalds /* 15291da177e4SLinus Torvalds * Special case for "unmounting" root ... 15301da177e4SLinus Torvalds * we just try to remount it readonly. 15311da177e4SLinus Torvalds */ 1532bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1533a1480dccSAndy Lutomirski return -EPERM; 15341da177e4SLinus Torvalds down_write(&sb->s_umount); 1535bc98a42cSDavid Howells if (!sb_rdonly(sb)) 1536e462ec50SDavid Howells retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); 15371da177e4SLinus Torvalds up_write(&sb->s_umount); 15381da177e4SLinus Torvalds return retval; 15391da177e4SLinus Torvalds } 15401da177e4SLinus Torvalds 154197216be0SAl Viro namespace_lock(); 1542719ea2fbSAl Viro lock_mount_hash(); 15431da177e4SLinus Torvalds 154425d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 154525d202edSEric W. Biederman retval = -EINVAL; 154625d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 154725d202edSEric W. Biederman goto out; 154825d202edSEric W. Biederman 154925d202edSEric W. Biederman event++; 155048a066e7SAl Viro if (flags & MNT_DETACH) { 155148a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1552e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 155348a066e7SAl Viro retval = 0; 155448a066e7SAl Viro } else { 1555b54b9be7SAl Viro shrink_submounts(mnt); 15561da177e4SLinus Torvalds retval = -EBUSY; 155748a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 15581a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1559e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 15601da177e4SLinus Torvalds retval = 0; 15611da177e4SLinus Torvalds } 156248a066e7SAl Viro } 156325d202edSEric W. Biederman out: 1564719ea2fbSAl Viro unlock_mount_hash(); 1565e3197d83SAl Viro namespace_unlock(); 15661da177e4SLinus Torvalds return retval; 15671da177e4SLinus Torvalds } 15681da177e4SLinus Torvalds 15691da177e4SLinus Torvalds /* 157080b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 157180b5dce8SEric W. Biederman * 157280b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 157380b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 157480b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 157580b5dce8SEric W. Biederman * leaking them. 157680b5dce8SEric W. Biederman * 157780b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 157880b5dce8SEric W. Biederman */ 157980b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 158080b5dce8SEric W. Biederman { 158180b5dce8SEric W. Biederman struct mountpoint *mp; 158280b5dce8SEric W. Biederman struct mount *mnt; 158380b5dce8SEric W. Biederman 158480b5dce8SEric W. Biederman namespace_lock(); 15853895dbf8SEric W. Biederman lock_mount_hash(); 158680b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1587f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 158880b5dce8SEric W. Biederman goto out_unlock; 158980b5dce8SEric W. Biederman 1590e06b933eSAndrey Ulanov event++; 159180b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 159280b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1593ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1594fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1595fe78fcc8SEric W. Biederman umount_mnt(mnt); 1596ce07d891SEric W. Biederman } 1597e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 159880b5dce8SEric W. Biederman } 159980b5dce8SEric W. Biederman put_mountpoint(mp); 160080b5dce8SEric W. Biederman out_unlock: 16013895dbf8SEric W. Biederman unlock_mount_hash(); 160280b5dce8SEric W. Biederman namespace_unlock(); 160380b5dce8SEric W. Biederman } 160480b5dce8SEric W. Biederman 160580b5dce8SEric W. Biederman /* 16069b40bc90SAl Viro * Is the caller allowed to modify his namespace? 16079b40bc90SAl Viro */ 16089b40bc90SAl Viro static inline bool may_mount(void) 16099b40bc90SAl Viro { 16109b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 16119b40bc90SAl Viro } 16129b40bc90SAl Viro 16139e8925b6SJeff Layton static inline bool may_mandlock(void) 16149e8925b6SJeff Layton { 16159e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 16169e8925b6SJeff Layton return false; 16179e8925b6SJeff Layton #endif 161895ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 16199e8925b6SJeff Layton } 16209e8925b6SJeff Layton 16219b40bc90SAl Viro /* 16221da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16231da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16241da177e4SLinus Torvalds * 16251da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16261da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16271da177e4SLinus Torvalds */ 16281da177e4SLinus Torvalds 16293a18ef5cSDominik Brodowski int ksys_umount(char __user *name, int flags) 16301da177e4SLinus Torvalds { 16312d8f3038SAl Viro struct path path; 1632900148dcSAl Viro struct mount *mnt; 16331da177e4SLinus Torvalds int retval; 1634db1f05bbSMiklos Szeredi int lookup_flags = 0; 16351da177e4SLinus Torvalds 1636db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1637db1f05bbSMiklos Szeredi return -EINVAL; 1638db1f05bbSMiklos Szeredi 16399b40bc90SAl Viro if (!may_mount()) 16409b40bc90SAl Viro return -EPERM; 16419b40bc90SAl Viro 1642db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1643db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1644db1f05bbSMiklos Szeredi 1645197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 16461da177e4SLinus Torvalds if (retval) 16471da177e4SLinus Torvalds goto out; 1648900148dcSAl Viro mnt = real_mount(path.mnt); 16491da177e4SLinus Torvalds retval = -EINVAL; 16502d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 16511da177e4SLinus Torvalds goto dput_and_out; 1652143c8c91SAl Viro if (!check_mnt(mnt)) 16531da177e4SLinus Torvalds goto dput_and_out; 165425d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 16555ff9d8a6SEric W. Biederman goto dput_and_out; 1656b2f5d4dcSEric W. Biederman retval = -EPERM; 1657b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1658b2f5d4dcSEric W. Biederman goto dput_and_out; 16591da177e4SLinus Torvalds 1660900148dcSAl Viro retval = do_umount(mnt, flags); 16611da177e4SLinus Torvalds dput_and_out: 1662429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 16632d8f3038SAl Viro dput(path.dentry); 1664900148dcSAl Viro mntput_no_expire(mnt); 16651da177e4SLinus Torvalds out: 16661da177e4SLinus Torvalds return retval; 16671da177e4SLinus Torvalds } 16681da177e4SLinus Torvalds 16693a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 16703a18ef5cSDominik Brodowski { 16713a18ef5cSDominik Brodowski return ksys_umount(name, flags); 16723a18ef5cSDominik Brodowski } 16733a18ef5cSDominik Brodowski 16741da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 16751da177e4SLinus Torvalds 16761da177e4SLinus Torvalds /* 16771da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 16781da177e4SLinus Torvalds */ 1679bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 16801da177e4SLinus Torvalds { 16813a18ef5cSDominik Brodowski return ksys_umount(name, 0); 16821da177e4SLinus Torvalds } 16831da177e4SLinus Torvalds 16841da177e4SLinus Torvalds #endif 16851da177e4SLinus Torvalds 16864ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 16878823c079SEric W. Biederman { 16884ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1689e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1690e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 16914ce5d2b1SEric W. Biederman } 16924ce5d2b1SEric W. Biederman 169358be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 169458be2825SAl Viro { 169558be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 169658be2825SAl Viro } 169758be2825SAl Viro 16984ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 16994ce5d2b1SEric W. Biederman { 17004ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 17014ce5d2b1SEric W. Biederman * mount namespace loop? 17024ce5d2b1SEric W. Biederman */ 17034ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 17044ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 17054ce5d2b1SEric W. Biederman return false; 17064ce5d2b1SEric W. Biederman 1707f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 17088823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 17098823c079SEric W. Biederman } 17108823c079SEric W. Biederman 171187129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 171236341f64SRam Pai int flag) 17131da177e4SLinus Torvalds { 171484d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 17151da177e4SLinus Torvalds 17164ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 17174ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 17184ce5d2b1SEric W. Biederman 17194ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1720be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 17219676f0c6SRam Pai 172236341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1723be34d1a3SDavid Howells if (IS_ERR(q)) 1724be34d1a3SDavid Howells return q; 1725be34d1a3SDavid Howells 1726a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17271da177e4SLinus Torvalds 17281da177e4SLinus Torvalds p = mnt; 17296b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1730315fc83eSAl Viro struct mount *s; 17317ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17321da177e4SLinus Torvalds continue; 17331da177e4SLinus Torvalds 1734909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 17354ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17364ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1737df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1738df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1739df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1740df7342b2SEric W. Biederman goto out; 1741df7342b2SEric W. Biederman } else { 17424ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17434ce5d2b1SEric W. Biederman continue; 17444ce5d2b1SEric W. Biederman } 1745df7342b2SEric W. Biederman } 17464ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17474ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17489676f0c6SRam Pai s = skip_mnt_tree(s); 17499676f0c6SRam Pai continue; 17509676f0c6SRam Pai } 17510714a533SAl Viro while (p != s->mnt_parent) { 17520714a533SAl Viro p = p->mnt_parent; 17530714a533SAl Viro q = q->mnt_parent; 17541da177e4SLinus Torvalds } 175587129cc0SAl Viro p = s; 175684d17192SAl Viro parent = q; 175787129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1758be34d1a3SDavid Howells if (IS_ERR(q)) 1759be34d1a3SDavid Howells goto out; 1760719ea2fbSAl Viro lock_mount_hash(); 17611a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 17621064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1763719ea2fbSAl Viro unlock_mount_hash(); 17641da177e4SLinus Torvalds } 17651da177e4SLinus Torvalds } 17661da177e4SLinus Torvalds return res; 1767be34d1a3SDavid Howells out: 17681da177e4SLinus Torvalds if (res) { 1769719ea2fbSAl Viro lock_mount_hash(); 1770e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1771719ea2fbSAl Viro unlock_mount_hash(); 17721da177e4SLinus Torvalds } 1773be34d1a3SDavid Howells return q; 17741da177e4SLinus Torvalds } 17751da177e4SLinus Torvalds 1776be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1777be34d1a3SDavid Howells 1778ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 17798aec0809SAl Viro { 1780cb338d06SAl Viro struct mount *tree; 178197216be0SAl Viro namespace_lock(); 1782cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1783cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1784cd4a4017SEric W. Biederman else 178587129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 178687129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1787328e6d90SAl Viro namespace_unlock(); 1788be34d1a3SDavid Howells if (IS_ERR(tree)) 178952e220d3SDan Carpenter return ERR_CAST(tree); 1790be34d1a3SDavid Howells return &tree->mnt; 17918aec0809SAl Viro } 17928aec0809SAl Viro 17938aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 17948aec0809SAl Viro { 179597216be0SAl Viro namespace_lock(); 1796719ea2fbSAl Viro lock_mount_hash(); 17979c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1798719ea2fbSAl Viro unlock_mount_hash(); 17993ab6abeeSAl Viro namespace_unlock(); 18008aec0809SAl Viro } 18018aec0809SAl Viro 1802c771d683SMiklos Szeredi /** 1803c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1804c771d683SMiklos Szeredi * 1805c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1806c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1807c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1808c771d683SMiklos Szeredi * 1809c771d683SMiklos Szeredi * Release with mntput(). 1810c771d683SMiklos Szeredi */ 1811ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1812c771d683SMiklos Szeredi { 1813c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1814c771d683SMiklos Szeredi struct mount *new_mnt; 1815c771d683SMiklos Szeredi 1816c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1817c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1818c771d683SMiklos Szeredi 1819c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1820c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1821c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1822c771d683SMiklos Szeredi 1823c771d683SMiklos Szeredi return &new_mnt->mnt; 1824c771d683SMiklos Szeredi } 1825c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1826c771d683SMiklos Szeredi 18271f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18281f707137SAl Viro struct vfsmount *root) 18291f707137SAl Viro { 18301a4eeaf2SAl Viro struct mount *mnt; 18311f707137SAl Viro int res = f(root, arg); 18321f707137SAl Viro if (res) 18331f707137SAl Viro return res; 18341a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18351a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18361f707137SAl Viro if (res) 18371f707137SAl Viro return res; 18381f707137SAl Viro } 18391f707137SAl Viro return 0; 18401f707137SAl Viro } 18411f707137SAl Viro 18424b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1843719f5d7fSMiklos Szeredi { 1844315fc83eSAl Viro struct mount *p; 1845719f5d7fSMiklos Szeredi 1846909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1847fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 18484b8b21f4SAl Viro mnt_release_group_id(p); 1849719f5d7fSMiklos Szeredi } 1850719f5d7fSMiklos Szeredi } 1851719f5d7fSMiklos Szeredi 18524b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1853719f5d7fSMiklos Szeredi { 1854315fc83eSAl Viro struct mount *p; 1855719f5d7fSMiklos Szeredi 1856909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1857fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 18584b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1859719f5d7fSMiklos Szeredi if (err) { 18604b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1861719f5d7fSMiklos Szeredi return err; 1862719f5d7fSMiklos Szeredi } 1863719f5d7fSMiklos Szeredi } 1864719f5d7fSMiklos Szeredi } 1865719f5d7fSMiklos Szeredi 1866719f5d7fSMiklos Szeredi return 0; 1867719f5d7fSMiklos Szeredi } 1868719f5d7fSMiklos Szeredi 1869d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1870d2921684SEric W. Biederman { 1871d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1872d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1873d2921684SEric W. Biederman struct mount *p; 1874d2921684SEric W. Biederman 1875d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1876d2921684SEric W. Biederman mounts++; 1877d2921684SEric W. Biederman 1878d2921684SEric W. Biederman old = ns->mounts; 1879d2921684SEric W. Biederman pending = ns->pending_mounts; 1880d2921684SEric W. Biederman sum = old + pending; 1881d2921684SEric W. Biederman if ((old > sum) || 1882d2921684SEric W. Biederman (pending > sum) || 1883d2921684SEric W. Biederman (max < sum) || 1884d2921684SEric W. Biederman (mounts > (max - sum))) 1885d2921684SEric W. Biederman return -ENOSPC; 1886d2921684SEric W. Biederman 1887d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1888d2921684SEric W. Biederman return 0; 1889d2921684SEric W. Biederman } 1890d2921684SEric W. Biederman 1891b90fa9aeSRam Pai /* 1892b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1893b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 189421444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 189521444403SRam Pai * store the parent mount and mountpoint dentry. 189621444403SRam Pai * (done when source_mnt is moved) 1897b90fa9aeSRam Pai * 1898b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1899b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 19009676f0c6SRam Pai * --------------------------------------------------------------------------- 1901b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 19029676f0c6SRam Pai * |************************************************************************** 19039676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19049676f0c6SRam Pai * | dest | | | | | 19059676f0c6SRam Pai * | | | | | | | 19069676f0c6SRam Pai * | v | | | | | 19079676f0c6SRam Pai * |************************************************************************** 19089676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 19095afe0022SRam Pai * | | | | | | 19109676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 19119676f0c6SRam Pai * *************************************************************************** 1912b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1913b90fa9aeSRam Pai * destination mount. 1914b90fa9aeSRam Pai * 1915b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1916b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1917b90fa9aeSRam Pai * the peer group of the source mount. 1918b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1919b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1920b90fa9aeSRam Pai * mount. 19215afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 19225afe0022SRam Pai * of the destination mount and the cloned mount is made slave 19235afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 19245afe0022SRam Pai * is marked as 'shared and slave'. 19255afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 19265afe0022SRam Pai * source mount. 19275afe0022SRam Pai * 19289676f0c6SRam Pai * --------------------------------------------------------------------------- 192921444403SRam Pai * | MOVE MOUNT OPERATION | 19309676f0c6SRam Pai * |************************************************************************** 19319676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19329676f0c6SRam Pai * | dest | | | | | 19339676f0c6SRam Pai * | | | | | | | 19349676f0c6SRam Pai * | v | | | | | 19359676f0c6SRam Pai * |************************************************************************** 19369676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 19375afe0022SRam Pai * | | | | | | 19389676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 19399676f0c6SRam Pai * *************************************************************************** 19405afe0022SRam Pai * 19415afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 19425afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 194321444403SRam Pai * (+*) the mount is moved to the destination. 19445afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 19455afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 19465afe0022SRam Pai * the mount is marked as 'shared and slave'. 19475afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1948b90fa9aeSRam Pai * 1949b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1950b90fa9aeSRam Pai * applied to each mount in the tree. 1951b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1952b90fa9aeSRam Pai * in allocations. 1953b90fa9aeSRam Pai */ 19540fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 195584d17192SAl Viro struct mount *dest_mnt, 195684d17192SAl Viro struct mountpoint *dest_mp, 195784d17192SAl Viro struct path *parent_path) 1958b90fa9aeSRam Pai { 195938129a13SAl Viro HLIST_HEAD(tree_list); 1960d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 19611064f874SEric W. Biederman struct mountpoint *smp; 1962315fc83eSAl Viro struct mount *child, *p; 196338129a13SAl Viro struct hlist_node *n; 1964719f5d7fSMiklos Szeredi int err; 1965b90fa9aeSRam Pai 19661064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 19671064f874SEric W. Biederman * to be tucked under other mounts. 19681064f874SEric W. Biederman */ 19691064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 19701064f874SEric W. Biederman if (IS_ERR(smp)) 19711064f874SEric W. Biederman return PTR_ERR(smp); 19721064f874SEric W. Biederman 1973d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 1974d2921684SEric W. Biederman if (!parent_path) { 1975d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 1976d2921684SEric W. Biederman if (err) 1977d2921684SEric W. Biederman goto out; 1978d2921684SEric W. Biederman } 1979d2921684SEric W. Biederman 1980fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 19810fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1982719f5d7fSMiklos Szeredi if (err) 1983719f5d7fSMiklos Szeredi goto out; 198484d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1985f2ebb3a9SAl Viro lock_mount_hash(); 1986719f5d7fSMiklos Szeredi if (err) 1987719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1988909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 19890f0afb1dSAl Viro set_mnt_shared(p); 19900b1b901bSAl Viro } else { 19910b1b901bSAl Viro lock_mount_hash(); 1992b90fa9aeSRam Pai } 19931a390689SAl Viro if (parent_path) { 19940fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 199584d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1996143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 199721444403SRam Pai } else { 199884d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 19991064f874SEric W. Biederman commit_tree(source_mnt); 200021444403SRam Pai } 2001b90fa9aeSRam Pai 200238129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 20031d6a32acSAl Viro struct mount *q; 200438129a13SAl Viro hlist_del_init(&child->mnt_hash); 20051064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 20061d6a32acSAl Viro child->mnt_mountpoint); 20071064f874SEric W. Biederman if (q) 20081064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 20091064f874SEric W. Biederman commit_tree(child); 2010b90fa9aeSRam Pai } 20111064f874SEric W. Biederman put_mountpoint(smp); 2012719ea2fbSAl Viro unlock_mount_hash(); 201399b7db7bSNick Piggin 2014b90fa9aeSRam Pai return 0; 2015719f5d7fSMiklos Szeredi 2016719f5d7fSMiklos Szeredi out_cleanup_ids: 2017f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2018f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2019d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2020e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2021f2ebb3a9SAl Viro } 2022f2ebb3a9SAl Viro unlock_mount_hash(); 20230fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2024719f5d7fSMiklos Szeredi out: 2025d2921684SEric W. Biederman ns->pending_mounts = 0; 20261064f874SEric W. Biederman 20271064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 20281064f874SEric W. Biederman put_mountpoint(smp); 20291064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 20301064f874SEric W. Biederman 2031719f5d7fSMiklos Szeredi return err; 2032b90fa9aeSRam Pai } 2033b90fa9aeSRam Pai 203484d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2035b12cea91SAl Viro { 2036b12cea91SAl Viro struct vfsmount *mnt; 203784d17192SAl Viro struct dentry *dentry = path->dentry; 2038b12cea91SAl Viro retry: 20395955102cSAl Viro inode_lock(dentry->d_inode); 204084d17192SAl Viro if (unlikely(cant_mount(dentry))) { 20415955102cSAl Viro inode_unlock(dentry->d_inode); 204284d17192SAl Viro return ERR_PTR(-ENOENT); 2043b12cea91SAl Viro } 204497216be0SAl Viro namespace_lock(); 2045b12cea91SAl Viro mnt = lookup_mnt(path); 204684d17192SAl Viro if (likely(!mnt)) { 20473895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 204884d17192SAl Viro if (IS_ERR(mp)) { 204997216be0SAl Viro namespace_unlock(); 20505955102cSAl Viro inode_unlock(dentry->d_inode); 205184d17192SAl Viro return mp; 205284d17192SAl Viro } 205384d17192SAl Viro return mp; 205484d17192SAl Viro } 205597216be0SAl Viro namespace_unlock(); 20565955102cSAl Viro inode_unlock(path->dentry->d_inode); 2057b12cea91SAl Viro path_put(path); 2058b12cea91SAl Viro path->mnt = mnt; 205984d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2060b12cea91SAl Viro goto retry; 2061b12cea91SAl Viro } 2062b12cea91SAl Viro 206384d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2064b12cea91SAl Viro { 206584d17192SAl Viro struct dentry *dentry = where->m_dentry; 20663895dbf8SEric W. Biederman 20673895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 206884d17192SAl Viro put_mountpoint(where); 20693895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 20703895dbf8SEric W. Biederman 2071328e6d90SAl Viro namespace_unlock(); 20725955102cSAl Viro inode_unlock(dentry->d_inode); 2073b12cea91SAl Viro } 2074b12cea91SAl Viro 207584d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 20761da177e4SLinus Torvalds { 2077e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 20781da177e4SLinus Torvalds return -EINVAL; 20791da177e4SLinus Torvalds 2080e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2081e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 20821da177e4SLinus Torvalds return -ENOTDIR; 20831da177e4SLinus Torvalds 208484d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 20851da177e4SLinus Torvalds } 20861da177e4SLinus Torvalds 20871da177e4SLinus Torvalds /* 20887a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 20897a2e8a8fSValerie Aurora */ 20907a2e8a8fSValerie Aurora 2091e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 20927a2e8a8fSValerie Aurora { 2093e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 20947a2e8a8fSValerie Aurora 20957a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 20967a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 20977a2e8a8fSValerie Aurora return 0; 20987a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 20997a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 21007a2e8a8fSValerie Aurora return 0; 21017a2e8a8fSValerie Aurora return type; 21027a2e8a8fSValerie Aurora } 21037a2e8a8fSValerie Aurora 21047a2e8a8fSValerie Aurora /* 210507b20889SRam Pai * recursively change the type of the mountpoint. 210607b20889SRam Pai */ 2107e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 210807b20889SRam Pai { 2109315fc83eSAl Viro struct mount *m; 21104b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2111e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 21127a2e8a8fSValerie Aurora int type; 2113719f5d7fSMiklos Szeredi int err = 0; 211407b20889SRam Pai 21152d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 211607b20889SRam Pai return -EINVAL; 211707b20889SRam Pai 2118e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 21197a2e8a8fSValerie Aurora if (!type) 21207a2e8a8fSValerie Aurora return -EINVAL; 21217a2e8a8fSValerie Aurora 212297216be0SAl Viro namespace_lock(); 2123719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2124719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2125719f5d7fSMiklos Szeredi if (err) 2126719f5d7fSMiklos Szeredi goto out_unlock; 2127719f5d7fSMiklos Szeredi } 2128719f5d7fSMiklos Szeredi 2129719ea2fbSAl Viro lock_mount_hash(); 2130909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 21310f0afb1dSAl Viro change_mnt_propagation(m, type); 2132719ea2fbSAl Viro unlock_mount_hash(); 2133719f5d7fSMiklos Szeredi 2134719f5d7fSMiklos Szeredi out_unlock: 213597216be0SAl Viro namespace_unlock(); 2136719f5d7fSMiklos Szeredi return err; 213707b20889SRam Pai } 213807b20889SRam Pai 21395ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 21405ff9d8a6SEric W. Biederman { 21415ff9d8a6SEric W. Biederman struct mount *child; 21425ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 21435ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 21445ff9d8a6SEric W. Biederman continue; 21455ff9d8a6SEric W. Biederman 21465ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 21475ff9d8a6SEric W. Biederman return true; 21485ff9d8a6SEric W. Biederman } 21495ff9d8a6SEric W. Biederman return false; 21505ff9d8a6SEric W. Biederman } 21515ff9d8a6SEric W. Biederman 215207b20889SRam Pai /* 21531da177e4SLinus Torvalds * do loopback mount. 21541da177e4SLinus Torvalds */ 2155808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 21562dafe1c4SEric Sandeen int recurse) 21571da177e4SLinus Torvalds { 21582d92ab3cSAl Viro struct path old_path; 215984d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 216084d17192SAl Viro struct mountpoint *mp; 216157eccb83SAl Viro int err; 21621da177e4SLinus Torvalds if (!old_name || !*old_name) 21631da177e4SLinus Torvalds return -EINVAL; 2164815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 21651da177e4SLinus Torvalds if (err) 21661da177e4SLinus Torvalds return err; 21671da177e4SLinus Torvalds 21688823c079SEric W. Biederman err = -EINVAL; 21694ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 21708823c079SEric W. Biederman goto out; 21718823c079SEric W. Biederman 217284d17192SAl Viro mp = lock_mount(path); 217384d17192SAl Viro err = PTR_ERR(mp); 217484d17192SAl Viro if (IS_ERR(mp)) 21759676f0c6SRam Pai goto out; 21769676f0c6SRam Pai 217787129cc0SAl Viro old = real_mount(old_path.mnt); 217884d17192SAl Viro parent = real_mount(path->mnt); 217987129cc0SAl Viro 2180b12cea91SAl Viro err = -EINVAL; 2181fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2182b12cea91SAl Viro goto out2; 2183b12cea91SAl Viro 2184e149ed2bSAl Viro if (!check_mnt(parent)) 2185e149ed2bSAl Viro goto out2; 2186e149ed2bSAl Viro 2187e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2188b12cea91SAl Viro goto out2; 2189ccd48bc7SAl Viro 21905ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 21915ff9d8a6SEric W. Biederman goto out2; 21925ff9d8a6SEric W. Biederman 21931da177e4SLinus Torvalds if (recurse) 21944ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 21951da177e4SLinus Torvalds else 219687129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 21971da177e4SLinus Torvalds 2198be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2199be34d1a3SDavid Howells err = PTR_ERR(mnt); 2200e9c5d8a5SAndrey Vagin goto out2; 2201be34d1a3SDavid Howells } 2202ccd48bc7SAl Viro 22035ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 22045ff9d8a6SEric W. Biederman 220584d17192SAl Viro err = graft_tree(mnt, parent, mp); 22061da177e4SLinus Torvalds if (err) { 2207719ea2fbSAl Viro lock_mount_hash(); 2208e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2209719ea2fbSAl Viro unlock_mount_hash(); 22105b83d2c5SRam Pai } 2211b12cea91SAl Viro out2: 221284d17192SAl Viro unlock_mount(mp); 2213ccd48bc7SAl Viro out: 22142d92ab3cSAl Viro path_put(&old_path); 22151da177e4SLinus Torvalds return err; 22161da177e4SLinus Torvalds } 22171da177e4SLinus Torvalds 22182e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 22192e4b7fcdSDave Hansen { 22202e4b7fcdSDave Hansen int error = 0; 22212e4b7fcdSDave Hansen int readonly_request = 0; 22222e4b7fcdSDave Hansen 22232e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 22242e4b7fcdSDave Hansen readonly_request = 1; 22252e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 22262e4b7fcdSDave Hansen return 0; 22272e4b7fcdSDave Hansen 22282e4b7fcdSDave Hansen if (readonly_request) 222983adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 22302e4b7fcdSDave Hansen else 223183adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 22322e4b7fcdSDave Hansen return error; 22332e4b7fcdSDave Hansen } 22342e4b7fcdSDave Hansen 22351da177e4SLinus Torvalds /* 22361da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 22371da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 22381da177e4SLinus Torvalds * on it - tough luck. 22391da177e4SLinus Torvalds */ 2240e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2241e462ec50SDavid Howells int mnt_flags, void *data) 22421da177e4SLinus Torvalds { 22431da177e4SLinus Torvalds int err; 22442d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2245143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 22461da177e4SLinus Torvalds 2247143c8c91SAl Viro if (!check_mnt(mnt)) 22481da177e4SLinus Torvalds return -EINVAL; 22491da177e4SLinus Torvalds 22502d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 22511da177e4SLinus Torvalds return -EINVAL; 22521da177e4SLinus Torvalds 225307b64558SEric W. Biederman /* Don't allow changing of locked mnt flags. 225407b64558SEric W. Biederman * 225507b64558SEric W. Biederman * No locks need to be held here while testing the various 225607b64558SEric W. Biederman * MNT_LOCK flags because those flags can never be cleared 225707b64558SEric W. Biederman * once they are set. 225807b64558SEric W. Biederman */ 225907b64558SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 226007b64558SEric W. Biederman !(mnt_flags & MNT_READONLY)) { 226107b64558SEric W. Biederman return -EPERM; 226207b64558SEric W. Biederman } 22639566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 22649566d674SEric W. Biederman !(mnt_flags & MNT_NODEV)) { 22659566d674SEric W. Biederman return -EPERM; 22669566d674SEric W. Biederman } 22679566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && 22689566d674SEric W. Biederman !(mnt_flags & MNT_NOSUID)) { 22699566d674SEric W. Biederman return -EPERM; 22709566d674SEric W. Biederman } 22719566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && 22729566d674SEric W. Biederman !(mnt_flags & MNT_NOEXEC)) { 22739566d674SEric W. Biederman return -EPERM; 22749566d674SEric W. Biederman } 22759566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 22769566d674SEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { 22779566d674SEric W. Biederman return -EPERM; 22789566d674SEric W. Biederman } 22799566d674SEric W. Biederman 2280ff36fe2cSEric Paris err = security_sb_remount(sb, data); 2281ff36fe2cSEric Paris if (err) 2282ff36fe2cSEric Paris return err; 2283ff36fe2cSEric Paris 22841da177e4SLinus Torvalds down_write(&sb->s_umount); 2285e462ec50SDavid Howells if (ms_flags & MS_BIND) 2286e462ec50SDavid Howells err = change_mount_flags(path->mnt, ms_flags); 2287bc6155d1SEric W. Biederman else if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 228857eccb83SAl Viro err = -EPERM; 22894aa98cf7SAl Viro else 2290e462ec50SDavid Howells err = do_remount_sb(sb, sb_flags, data, 0); 22917b43a79fSAl Viro if (!err) { 2292719ea2fbSAl Viro lock_mount_hash(); 2293a6138db8SEric W. Biederman mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 2294143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 2295143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2296719ea2fbSAl Viro unlock_mount_hash(); 22970e55a7ccSDan Williams } 22986339dab8SAl Viro up_write(&sb->s_umount); 22991da177e4SLinus Torvalds return err; 23001da177e4SLinus Torvalds } 23011da177e4SLinus Torvalds 2302cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 23039676f0c6SRam Pai { 2304315fc83eSAl Viro struct mount *p; 2305909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2306fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 23079676f0c6SRam Pai return 1; 23089676f0c6SRam Pai } 23099676f0c6SRam Pai return 0; 23109676f0c6SRam Pai } 23119676f0c6SRam Pai 2312808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 23131da177e4SLinus Torvalds { 23142d92ab3cSAl Viro struct path old_path, parent_path; 2315676da58dSAl Viro struct mount *p; 23160fb54e50SAl Viro struct mount *old; 231784d17192SAl Viro struct mountpoint *mp; 231857eccb83SAl Viro int err; 23191da177e4SLinus Torvalds if (!old_name || !*old_name) 23201da177e4SLinus Torvalds return -EINVAL; 23212d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 23221da177e4SLinus Torvalds if (err) 23231da177e4SLinus Torvalds return err; 23241da177e4SLinus Torvalds 232584d17192SAl Viro mp = lock_mount(path); 232684d17192SAl Viro err = PTR_ERR(mp); 232784d17192SAl Viro if (IS_ERR(mp)) 2328cc53ce53SDavid Howells goto out; 2329cc53ce53SDavid Howells 2330143c8c91SAl Viro old = real_mount(old_path.mnt); 2331fc7be130SAl Viro p = real_mount(path->mnt); 2332143c8c91SAl Viro 23331da177e4SLinus Torvalds err = -EINVAL; 2334fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 23351da177e4SLinus Torvalds goto out1; 23361da177e4SLinus Torvalds 23375ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 23385ff9d8a6SEric W. Biederman goto out1; 23395ff9d8a6SEric W. Biederman 23401da177e4SLinus Torvalds err = -EINVAL; 23412d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 234221444403SRam Pai goto out1; 23431da177e4SLinus Torvalds 2344676da58dSAl Viro if (!mnt_has_parent(old)) 234521444403SRam Pai goto out1; 23461da177e4SLinus Torvalds 2347e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2348e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 234921444403SRam Pai goto out1; 235021444403SRam Pai /* 235121444403SRam Pai * Don't move a mount residing in a shared parent. 235221444403SRam Pai */ 2353fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 235421444403SRam Pai goto out1; 23559676f0c6SRam Pai /* 23569676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 23579676f0c6SRam Pai * mount which is shared. 23589676f0c6SRam Pai */ 2359fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 23609676f0c6SRam Pai goto out1; 23611da177e4SLinus Torvalds err = -ELOOP; 2362fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2363676da58dSAl Viro if (p == old) 236421444403SRam Pai goto out1; 23651da177e4SLinus Torvalds 236684d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 23674ac91378SJan Blunck if (err) 236821444403SRam Pai goto out1; 23691da177e4SLinus Torvalds 23701da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 23711da177e4SLinus Torvalds * automatically */ 23726776db3dSAl Viro list_del_init(&old->mnt_expire); 23731da177e4SLinus Torvalds out1: 237484d17192SAl Viro unlock_mount(mp); 23751da177e4SLinus Torvalds out: 23761da177e4SLinus Torvalds if (!err) 23771a390689SAl Viro path_put(&parent_path); 23782d92ab3cSAl Viro path_put(&old_path); 23791da177e4SLinus Torvalds return err; 23801da177e4SLinus Torvalds } 23811da177e4SLinus Torvalds 23829d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 23839d412a43SAl Viro { 23849d412a43SAl Viro int err; 23859d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 23869d412a43SAl Viro if (subtype) { 23879d412a43SAl Viro subtype++; 23889d412a43SAl Viro err = -EINVAL; 23899d412a43SAl Viro if (!subtype[0]) 23909d412a43SAl Viro goto err; 23919d412a43SAl Viro } else 23929d412a43SAl Viro subtype = ""; 23939d412a43SAl Viro 23949d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 23959d412a43SAl Viro err = -ENOMEM; 23969d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 23979d412a43SAl Viro goto err; 23989d412a43SAl Viro return mnt; 23999d412a43SAl Viro 24009d412a43SAl Viro err: 24019d412a43SAl Viro mntput(mnt); 24029d412a43SAl Viro return ERR_PTR(err); 24039d412a43SAl Viro } 24049d412a43SAl Viro 24059d412a43SAl Viro /* 24069d412a43SAl Viro * add a mount into a namespace's mount tree 24079d412a43SAl Viro */ 240895bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 24099d412a43SAl Viro { 241084d17192SAl Viro struct mountpoint *mp; 241184d17192SAl Viro struct mount *parent; 24129d412a43SAl Viro int err; 24139d412a43SAl Viro 2414f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 24159d412a43SAl Viro 241684d17192SAl Viro mp = lock_mount(path); 241784d17192SAl Viro if (IS_ERR(mp)) 241884d17192SAl Viro return PTR_ERR(mp); 24199d412a43SAl Viro 242084d17192SAl Viro parent = real_mount(path->mnt); 24219d412a43SAl Viro err = -EINVAL; 242284d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2423156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2424156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 24259d412a43SAl Viro goto unlock; 2426156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 242784d17192SAl Viro if (!parent->mnt_ns) 2428156cacb1SAl Viro goto unlock; 2429156cacb1SAl Viro } 24309d412a43SAl Viro 24319d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 24329d412a43SAl Viro err = -EBUSY; 243395bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 24349d412a43SAl Viro path->mnt->mnt_root == path->dentry) 24359d412a43SAl Viro goto unlock; 24369d412a43SAl Viro 24379d412a43SAl Viro err = -EINVAL; 2438e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 24399d412a43SAl Viro goto unlock; 24409d412a43SAl Viro 244195bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 244284d17192SAl Viro err = graft_tree(newmnt, parent, mp); 24439d412a43SAl Viro 24449d412a43SAl Viro unlock: 244584d17192SAl Viro unlock_mount(mp); 24469d412a43SAl Viro return err; 24479d412a43SAl Viro } 2448b1e75df4SAl Viro 24498654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); 24501b852bceSEric W. Biederman 24511da177e4SLinus Torvalds /* 24521da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 24531da177e4SLinus Torvalds * namespace's tree 24541da177e4SLinus Torvalds */ 2455e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2456808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 24571da177e4SLinus Torvalds { 24580c55cfc4SEric W. Biederman struct file_system_type *type; 24591da177e4SLinus Torvalds struct vfsmount *mnt; 246015f9a3f3SAl Viro int err; 24611da177e4SLinus Torvalds 24620c55cfc4SEric W. Biederman if (!fstype) 24631da177e4SLinus Torvalds return -EINVAL; 24641da177e4SLinus Torvalds 24650c55cfc4SEric W. Biederman type = get_fs_type(fstype); 24660c55cfc4SEric W. Biederman if (!type) 24670c55cfc4SEric W. Biederman return -ENODEV; 24680c55cfc4SEric W. Biederman 2469e462ec50SDavid Howells mnt = vfs_kern_mount(type, sb_flags, name, data); 24700c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 24710c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 24720c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 24730c55cfc4SEric W. Biederman 24740c55cfc4SEric W. Biederman put_filesystem(type); 24751da177e4SLinus Torvalds if (IS_ERR(mnt)) 24761da177e4SLinus Torvalds return PTR_ERR(mnt); 24771da177e4SLinus Torvalds 24788654df4eSEric W. Biederman if (mount_too_revealing(mnt, &mnt_flags)) { 24798654df4eSEric W. Biederman mntput(mnt); 24808654df4eSEric W. Biederman return -EPERM; 24818654df4eSEric W. Biederman } 24828654df4eSEric W. Biederman 248395bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 248415f9a3f3SAl Viro if (err) 248515f9a3f3SAl Viro mntput(mnt); 248615f9a3f3SAl Viro return err; 24871da177e4SLinus Torvalds } 24881da177e4SLinus Torvalds 248919a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 249019a167afSAl Viro { 24916776db3dSAl Viro struct mount *mnt = real_mount(m); 249219a167afSAl Viro int err; 249319a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 249419a167afSAl Viro * expired before we get a chance to add it 249519a167afSAl Viro */ 24966776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 249719a167afSAl Viro 249819a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 249919a167afSAl Viro m->mnt_root == path->dentry) { 2500b1e75df4SAl Viro err = -ELOOP; 2501b1e75df4SAl Viro goto fail; 250219a167afSAl Viro } 250319a167afSAl Viro 250495bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2505b1e75df4SAl Viro if (!err) 2506b1e75df4SAl Viro return 0; 2507b1e75df4SAl Viro fail: 2508b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 25096776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 251097216be0SAl Viro namespace_lock(); 25116776db3dSAl Viro list_del_init(&mnt->mnt_expire); 251297216be0SAl Viro namespace_unlock(); 251319a167afSAl Viro } 2514b1e75df4SAl Viro mntput(m); 2515b1e75df4SAl Viro mntput(m); 251619a167afSAl Viro return err; 251719a167afSAl Viro } 251819a167afSAl Viro 2519ea5b778aSDavid Howells /** 2520ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2521ea5b778aSDavid Howells * @mnt: The mount to list. 2522ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2523ea5b778aSDavid Howells */ 2524ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2525ea5b778aSDavid Howells { 252697216be0SAl Viro namespace_lock(); 2527ea5b778aSDavid Howells 25286776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2529ea5b778aSDavid Howells 253097216be0SAl Viro namespace_unlock(); 2531ea5b778aSDavid Howells } 2532ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2533ea5b778aSDavid Howells 2534ea5b778aSDavid Howells /* 25351da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 25361da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 25371da177e4SLinus Torvalds * here 25381da177e4SLinus Torvalds */ 25391da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 25401da177e4SLinus Torvalds { 2541761d5c38SAl Viro struct mount *mnt, *next; 25421da177e4SLinus Torvalds LIST_HEAD(graveyard); 25431da177e4SLinus Torvalds 25441da177e4SLinus Torvalds if (list_empty(mounts)) 25451da177e4SLinus Torvalds return; 25461da177e4SLinus Torvalds 254797216be0SAl Viro namespace_lock(); 2548719ea2fbSAl Viro lock_mount_hash(); 25491da177e4SLinus Torvalds 25501da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 25511da177e4SLinus Torvalds * following criteria: 25521da177e4SLinus Torvalds * - only referenced by its parent vfsmount 25531da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 25541da177e4SLinus Torvalds * cleared by mntput()) 25551da177e4SLinus Torvalds */ 25566776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2557863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 25581ab59738SAl Viro propagate_mount_busy(mnt, 1)) 25591da177e4SLinus Torvalds continue; 25606776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 25611da177e4SLinus Torvalds } 2562bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 25636776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2564143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2565e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2566bcc5c7d2SAl Viro } 2567719ea2fbSAl Viro unlock_mount_hash(); 25683ab6abeeSAl Viro namespace_unlock(); 25691da177e4SLinus Torvalds } 25701da177e4SLinus Torvalds 25711da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 25721da177e4SLinus Torvalds 25731da177e4SLinus Torvalds /* 25745528f911STrond Myklebust * Ripoff of 'select_parent()' 25755528f911STrond Myklebust * 25765528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 25775528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 25785528f911STrond Myklebust */ 2579692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 25805528f911STrond Myklebust { 2581692afc31SAl Viro struct mount *this_parent = parent; 25825528f911STrond Myklebust struct list_head *next; 25835528f911STrond Myklebust int found = 0; 25845528f911STrond Myklebust 25855528f911STrond Myklebust repeat: 25866b41d536SAl Viro next = this_parent->mnt_mounts.next; 25875528f911STrond Myklebust resume: 25886b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 25895528f911STrond Myklebust struct list_head *tmp = next; 25906b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 25915528f911STrond Myklebust 25925528f911STrond Myklebust next = tmp->next; 2593692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 25945528f911STrond Myklebust continue; 25955528f911STrond Myklebust /* 25965528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 25975528f911STrond Myklebust */ 25986b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 25995528f911STrond Myklebust this_parent = mnt; 26005528f911STrond Myklebust goto repeat; 26015528f911STrond Myklebust } 26025528f911STrond Myklebust 26031ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 26046776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 26055528f911STrond Myklebust found++; 26065528f911STrond Myklebust } 26075528f911STrond Myklebust } 26085528f911STrond Myklebust /* 26095528f911STrond Myklebust * All done at this level ... ascend and resume the search 26105528f911STrond Myklebust */ 26115528f911STrond Myklebust if (this_parent != parent) { 26126b41d536SAl Viro next = this_parent->mnt_child.next; 26130714a533SAl Viro this_parent = this_parent->mnt_parent; 26145528f911STrond Myklebust goto resume; 26155528f911STrond Myklebust } 26165528f911STrond Myklebust return found; 26175528f911STrond Myklebust } 26185528f911STrond Myklebust 26195528f911STrond Myklebust /* 26205528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 26215528f911STrond Myklebust * submounts of a specific parent mountpoint 262299b7db7bSNick Piggin * 262348a066e7SAl Viro * mount_lock must be held for write 26245528f911STrond Myklebust */ 2625b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 26265528f911STrond Myklebust { 26275528f911STrond Myklebust LIST_HEAD(graveyard); 2628761d5c38SAl Viro struct mount *m; 26295528f911STrond Myklebust 26305528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2631c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2632bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2633761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 26346776db3dSAl Viro mnt_expire); 2635143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2636e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2637bcc5c7d2SAl Viro } 2638bcc5c7d2SAl Viro } 26395528f911STrond Myklebust } 26405528f911STrond Myklebust 26415528f911STrond Myklebust /* 26421da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 26431da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 26441da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 26451da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 26461da177e4SLinus Torvalds */ 2647b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2648b58fed8bSRam Pai unsigned long n) 26491da177e4SLinus Torvalds { 26501da177e4SLinus Torvalds char *t = to; 26511da177e4SLinus Torvalds const char __user *f = from; 26521da177e4SLinus Torvalds char c; 26531da177e4SLinus Torvalds 26541da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 26551da177e4SLinus Torvalds return n; 26561da177e4SLinus Torvalds 26579da3f2b7SJann Horn current->kernel_uaccess_faults_ok++; 26581da177e4SLinus Torvalds while (n) { 26591da177e4SLinus Torvalds if (__get_user(c, f)) { 26601da177e4SLinus Torvalds memset(t, 0, n); 26611da177e4SLinus Torvalds break; 26621da177e4SLinus Torvalds } 26631da177e4SLinus Torvalds *t++ = c; 26641da177e4SLinus Torvalds f++; 26651da177e4SLinus Torvalds n--; 26661da177e4SLinus Torvalds } 26679da3f2b7SJann Horn current->kernel_uaccess_faults_ok--; 26681da177e4SLinus Torvalds return n; 26691da177e4SLinus Torvalds } 26701da177e4SLinus Torvalds 2671b40ef869SAl Viro void *copy_mount_options(const void __user * data) 26721da177e4SLinus Torvalds { 26731da177e4SLinus Torvalds int i; 26741da177e4SLinus Torvalds unsigned long size; 2675b40ef869SAl Viro char *copy; 26761da177e4SLinus Torvalds 26771da177e4SLinus Torvalds if (!data) 2678b40ef869SAl Viro return NULL; 26791da177e4SLinus Torvalds 2680b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2681b40ef869SAl Viro if (!copy) 2682b40ef869SAl Viro return ERR_PTR(-ENOMEM); 26831da177e4SLinus Torvalds 26841da177e4SLinus Torvalds /* We only care that *some* data at the address the user 26851da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 26861da177e4SLinus Torvalds * the remainder of the page. 26871da177e4SLinus Torvalds */ 26881da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 26891da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 26901da177e4SLinus Torvalds if (size > PAGE_SIZE) 26911da177e4SLinus Torvalds size = PAGE_SIZE; 26921da177e4SLinus Torvalds 2693b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 26941da177e4SLinus Torvalds if (!i) { 2695b40ef869SAl Viro kfree(copy); 2696b40ef869SAl Viro return ERR_PTR(-EFAULT); 26971da177e4SLinus Torvalds } 26981da177e4SLinus Torvalds if (i != PAGE_SIZE) 2699b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2700b40ef869SAl Viro return copy; 27011da177e4SLinus Torvalds } 27021da177e4SLinus Torvalds 2703b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2704eca6f534SVegard Nossum { 2705b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2706eca6f534SVegard Nossum } 2707eca6f534SVegard Nossum 27081da177e4SLinus Torvalds /* 27091da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 27101da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 27111da177e4SLinus Torvalds * 27121da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 27131da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 27141da177e4SLinus Torvalds * information (or be NULL). 27151da177e4SLinus Torvalds * 27161da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 27171da177e4SLinus Torvalds * When the flags word was introduced its top half was required 27181da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 27191da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 27201da177e4SLinus Torvalds * and must be discarded. 27211da177e4SLinus Torvalds */ 27225e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2723808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 27241da177e4SLinus Torvalds { 27252d92ab3cSAl Viro struct path path; 2726e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 27271da177e4SLinus Torvalds int retval = 0; 27281da177e4SLinus Torvalds 27291da177e4SLinus Torvalds /* Discard magic */ 27301da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 27311da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 27321da177e4SLinus Torvalds 27331da177e4SLinus Torvalds /* Basic sanity checks */ 27341da177e4SLinus Torvalds if (data_page) 27351da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 27361da177e4SLinus Torvalds 2737e462ec50SDavid Howells if (flags & MS_NOUSER) 2738e462ec50SDavid Howells return -EINVAL; 2739e462ec50SDavid Howells 2740a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 27415e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2742a27ab9f2STetsuo Handa if (retval) 2743a27ab9f2STetsuo Handa return retval; 2744a27ab9f2STetsuo Handa 2745a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2746a27ab9f2STetsuo Handa type_page, flags, data_page); 27470d5cadb8SAl Viro if (!retval && !may_mount()) 27480d5cadb8SAl Viro retval = -EPERM; 2749e462ec50SDavid Howells if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) 27509e8925b6SJeff Layton retval = -EPERM; 2751a27ab9f2STetsuo Handa if (retval) 2752a27ab9f2STetsuo Handa goto dput_out; 2753a27ab9f2STetsuo Handa 2754613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2755613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 27560a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 27570a1c01c9SMatthew Garrett 27581da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 27591da177e4SLinus Torvalds if (flags & MS_NOSUID) 27601da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 27611da177e4SLinus Torvalds if (flags & MS_NODEV) 27621da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 27631da177e4SLinus Torvalds if (flags & MS_NOEXEC) 27641da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2765fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2766fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2767fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2768fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2769d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2770d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 2771a9e5b732SDavid Howells if (flags & MS_RDONLY) 27722e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2773fc33a7bbSChristoph Hellwig 2774ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2775ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2776ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2777ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2778ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2779ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2780ffbc6f0eSEric W. Biederman } 2781ffbc6f0eSEric W. Biederman 2782e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 2783e462ec50SDavid Howells SB_SYNCHRONOUS | 2784e462ec50SDavid Howells SB_MANDLOCK | 2785e462ec50SDavid Howells SB_DIRSYNC | 2786e462ec50SDavid Howells SB_SILENT | 2787917086ffSMimi Zohar SB_POSIXACL | 2788d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 2789917086ffSMimi Zohar SB_I_VERSION); 27901da177e4SLinus Torvalds 27911da177e4SLinus Torvalds if (flags & MS_REMOUNT) 2792e462ec50SDavid Howells retval = do_remount(&path, flags, sb_flags, mnt_flags, 27931da177e4SLinus Torvalds data_page); 27941da177e4SLinus Torvalds else if (flags & MS_BIND) 27952d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 27969676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 27972d92ab3cSAl Viro retval = do_change_type(&path, flags); 27981da177e4SLinus Torvalds else if (flags & MS_MOVE) 27992d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 28001da177e4SLinus Torvalds else 2801e462ec50SDavid Howells retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, 28021da177e4SLinus Torvalds dev_name, data_page); 28031da177e4SLinus Torvalds dput_out: 28042d92ab3cSAl Viro path_put(&path); 28051da177e4SLinus Torvalds return retval; 28061da177e4SLinus Torvalds } 28071da177e4SLinus Torvalds 2808537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 2809537f7ccbSEric W. Biederman { 2810537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 2811537f7ccbSEric W. Biederman } 2812537f7ccbSEric W. Biederman 2813537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 2814537f7ccbSEric W. Biederman { 2815537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 2816537f7ccbSEric W. Biederman } 2817537f7ccbSEric W. Biederman 2818771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2819771b1371SEric W. Biederman { 28206344c433SAl Viro ns_free_inum(&ns->ns); 2821537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 2822771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2823771b1371SEric W. Biederman kfree(ns); 2824771b1371SEric W. Biederman } 2825771b1371SEric W. Biederman 28268823c079SEric W. Biederman /* 28278823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 28288823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 28298823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 28308823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 28318823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 28328823c079SEric W. Biederman */ 28338823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 28348823c079SEric W. Biederman 2835771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2836cf8d2c11STrond Myklebust { 2837cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2838537f7ccbSEric W. Biederman struct ucounts *ucounts; 283998f842e6SEric W. Biederman int ret; 2840cf8d2c11STrond Myklebust 2841537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 2842537f7ccbSEric W. Biederman if (!ucounts) 2843df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 2844537f7ccbSEric W. Biederman 2845cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2846537f7ccbSEric W. Biederman if (!new_ns) { 2847537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 2848cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2849537f7ccbSEric W. Biederman } 28506344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 285198f842e6SEric W. Biederman if (ret) { 285298f842e6SEric W. Biederman kfree(new_ns); 2853537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 285498f842e6SEric W. Biederman return ERR_PTR(ret); 285598f842e6SEric W. Biederman } 285633c42940SAl Viro new_ns->ns.ops = &mntns_operations; 28578823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2858cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2859cf8d2c11STrond Myklebust new_ns->root = NULL; 2860cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2861cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2862cf8d2c11STrond Myklebust new_ns->event = 0; 2863771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2864537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 2865d2921684SEric W. Biederman new_ns->mounts = 0; 2866d2921684SEric W. Biederman new_ns->pending_mounts = 0; 2867cf8d2c11STrond Myklebust return new_ns; 2868cf8d2c11STrond Myklebust } 2869cf8d2c11STrond Myklebust 28700766f788SEmese Revfy __latent_entropy 28719559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 28729559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 28731da177e4SLinus Torvalds { 28746b3286edSKirill Korotaev struct mnt_namespace *new_ns; 28757f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2876315fc83eSAl Viro struct mount *p, *q; 28779559f689SAl Viro struct mount *old; 2878cb338d06SAl Viro struct mount *new; 28797a472ef4SEric W. Biederman int copy_flags; 28801da177e4SLinus Torvalds 28819559f689SAl Viro BUG_ON(!ns); 28829559f689SAl Viro 28839559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 28849559f689SAl Viro get_mnt_ns(ns); 28859559f689SAl Viro return ns; 28869559f689SAl Viro } 28879559f689SAl Viro 28889559f689SAl Viro old = ns->root; 28899559f689SAl Viro 2890771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2891cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2892cf8d2c11STrond Myklebust return new_ns; 28931da177e4SLinus Torvalds 289497216be0SAl Viro namespace_lock(); 28951da177e4SLinus Torvalds /* First pass: copy the tree topology */ 28964ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 28979559f689SAl Viro if (user_ns != ns->user_ns) 2898132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 28997a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2900be34d1a3SDavid Howells if (IS_ERR(new)) { 2901328e6d90SAl Viro namespace_unlock(); 2902771b1371SEric W. Biederman free_mnt_ns(new_ns); 2903be34d1a3SDavid Howells return ERR_CAST(new); 29041da177e4SLinus Torvalds } 2905be08d6d2SAl Viro new_ns->root = new; 29061a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 29071da177e4SLinus Torvalds 29081da177e4SLinus Torvalds /* 29091da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 29101da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 29111da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 29121da177e4SLinus Torvalds */ 2913909b0a88SAl Viro p = old; 2914cb338d06SAl Viro q = new; 29151da177e4SLinus Torvalds while (p) { 2916143c8c91SAl Viro q->mnt_ns = new_ns; 2917d2921684SEric W. Biederman new_ns->mounts++; 29189559f689SAl Viro if (new_fs) { 29199559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 29209559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2921315fc83eSAl Viro rootmnt = &p->mnt; 29221da177e4SLinus Torvalds } 29239559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 29249559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2925315fc83eSAl Viro pwdmnt = &p->mnt; 29261da177e4SLinus Torvalds } 29271da177e4SLinus Torvalds } 2928909b0a88SAl Viro p = next_mnt(p, old); 2929909b0a88SAl Viro q = next_mnt(q, new); 29304ce5d2b1SEric W. Biederman if (!q) 29314ce5d2b1SEric W. Biederman break; 29324ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 29334ce5d2b1SEric W. Biederman p = next_mnt(p, old); 29341da177e4SLinus Torvalds } 2935328e6d90SAl Viro namespace_unlock(); 29361da177e4SLinus Torvalds 29371da177e4SLinus Torvalds if (rootmnt) 2938f03c6599SAl Viro mntput(rootmnt); 29391da177e4SLinus Torvalds if (pwdmnt) 2940f03c6599SAl Viro mntput(pwdmnt); 29411da177e4SLinus Torvalds 2942741a2951SJANAK DESAI return new_ns; 2943741a2951SJANAK DESAI } 2944741a2951SJANAK DESAI 2945cf8d2c11STrond Myklebust /** 2946cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2947cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2948cf8d2c11STrond Myklebust */ 29491a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2950cf8d2c11STrond Myklebust { 2951771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2952cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 29531a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 29541a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2955be08d6d2SAl Viro new_ns->root = mnt; 2956d2921684SEric W. Biederman new_ns->mounts++; 2957b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2958c1334495SAl Viro } else { 29591a4eeaf2SAl Viro mntput(m); 2960cf8d2c11STrond Myklebust } 2961cf8d2c11STrond Myklebust return new_ns; 2962cf8d2c11STrond Myklebust } 2963cf8d2c11STrond Myklebust 2964ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2965ea441d11SAl Viro { 2966ea441d11SAl Viro struct mnt_namespace *ns; 2967d31da0f0SAl Viro struct super_block *s; 2968ea441d11SAl Viro struct path path; 2969ea441d11SAl Viro int err; 2970ea441d11SAl Viro 2971ea441d11SAl Viro ns = create_mnt_ns(mnt); 2972ea441d11SAl Viro if (IS_ERR(ns)) 2973ea441d11SAl Viro return ERR_CAST(ns); 2974ea441d11SAl Viro 2975ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2976ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2977ea441d11SAl Viro 2978ea441d11SAl Viro put_mnt_ns(ns); 2979ea441d11SAl Viro 2980ea441d11SAl Viro if (err) 2981ea441d11SAl Viro return ERR_PTR(err); 2982ea441d11SAl Viro 2983ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2984d31da0f0SAl Viro s = path.mnt->mnt_sb; 2985d31da0f0SAl Viro atomic_inc(&s->s_active); 2986ea441d11SAl Viro mntput(path.mnt); 2987ea441d11SAl Viro /* lock the sucker */ 2988d31da0f0SAl Viro down_write(&s->s_umount); 2989ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2990ea441d11SAl Viro return path.dentry; 2991ea441d11SAl Viro } 2992ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2993ea441d11SAl Viro 2994312db1aaSDominik Brodowski int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, 2995312db1aaSDominik Brodowski unsigned long flags, void __user *data) 29961da177e4SLinus Torvalds { 2997eca6f534SVegard Nossum int ret; 2998eca6f534SVegard Nossum char *kernel_type; 2999eca6f534SVegard Nossum char *kernel_dev; 3000b40ef869SAl Viro void *options; 30011da177e4SLinus Torvalds 3002b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3003b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3004b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3005eca6f534SVegard Nossum goto out_type; 30061da177e4SLinus Torvalds 3007b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3008b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3009b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3010eca6f534SVegard Nossum goto out_dev; 30111da177e4SLinus Torvalds 3012b40ef869SAl Viro options = copy_mount_options(data); 3013b40ef869SAl Viro ret = PTR_ERR(options); 3014b40ef869SAl Viro if (IS_ERR(options)) 3015eca6f534SVegard Nossum goto out_data; 30161da177e4SLinus Torvalds 3017b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3018eca6f534SVegard Nossum 3019b40ef869SAl Viro kfree(options); 3020eca6f534SVegard Nossum out_data: 3021eca6f534SVegard Nossum kfree(kernel_dev); 3022eca6f534SVegard Nossum out_dev: 3023eca6f534SVegard Nossum kfree(kernel_type); 3024eca6f534SVegard Nossum out_type: 3025eca6f534SVegard Nossum return ret; 30261da177e4SLinus Torvalds } 30271da177e4SLinus Torvalds 3028312db1aaSDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3029312db1aaSDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 3030312db1aaSDominik Brodowski { 3031312db1aaSDominik Brodowski return ksys_mount(dev_name, dir_name, type, flags, data); 3032312db1aaSDominik Brodowski } 3033312db1aaSDominik Brodowski 30341da177e4SLinus Torvalds /* 3035afac7cbaSAl Viro * Return true if path is reachable from root 3036afac7cbaSAl Viro * 303748a066e7SAl Viro * namespace_sem or mount_lock is held 3038afac7cbaSAl Viro */ 3039643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3040afac7cbaSAl Viro const struct path *root) 3041afac7cbaSAl Viro { 3042643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3043a73324daSAl Viro dentry = mnt->mnt_mountpoint; 30440714a533SAl Viro mnt = mnt->mnt_parent; 3045afac7cbaSAl Viro } 3046643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3047afac7cbaSAl Viro } 3048afac7cbaSAl Viro 3049640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3050afac7cbaSAl Viro { 305125ab4c9bSYaowei Bai bool res; 305248a066e7SAl Viro read_seqlock_excl(&mount_lock); 3053643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 305448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3055afac7cbaSAl Viro return res; 3056afac7cbaSAl Viro } 3057afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3058afac7cbaSAl Viro 3059afac7cbaSAl Viro /* 30601da177e4SLinus Torvalds * pivot_root Semantics: 30611da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 30621da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 30631da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 30641da177e4SLinus Torvalds * 30651da177e4SLinus Torvalds * Restrictions: 30661da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 30671da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 30681da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 30691da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 30701da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 30711da177e4SLinus Torvalds * 30724a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 30734a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 30744a0d11faSNeil Brown * in this situation. 30754a0d11faSNeil Brown * 30761da177e4SLinus Torvalds * Notes: 30771da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 30781da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 30791da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 30801da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 30811da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 30821da177e4SLinus Torvalds * first. 30831da177e4SLinus Torvalds */ 30843480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 30853480b257SHeiko Carstens const char __user *, put_old) 30861da177e4SLinus Torvalds { 30872d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 308884d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 308984d17192SAl Viro struct mountpoint *old_mp, *root_mp; 30901da177e4SLinus Torvalds int error; 30911da177e4SLinus Torvalds 30929b40bc90SAl Viro if (!may_mount()) 30931da177e4SLinus Torvalds return -EPERM; 30941da177e4SLinus Torvalds 30952d8f3038SAl Viro error = user_path_dir(new_root, &new); 30961da177e4SLinus Torvalds if (error) 30971da177e4SLinus Torvalds goto out0; 30981da177e4SLinus Torvalds 30992d8f3038SAl Viro error = user_path_dir(put_old, &old); 31001da177e4SLinus Torvalds if (error) 31011da177e4SLinus Torvalds goto out1; 31021da177e4SLinus Torvalds 31032d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3104b12cea91SAl Viro if (error) 3105b12cea91SAl Viro goto out2; 31061da177e4SLinus Torvalds 3107f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 310884d17192SAl Viro old_mp = lock_mount(&old); 310984d17192SAl Viro error = PTR_ERR(old_mp); 311084d17192SAl Viro if (IS_ERR(old_mp)) 3111b12cea91SAl Viro goto out3; 3112b12cea91SAl Viro 31131da177e4SLinus Torvalds error = -EINVAL; 3114419148daSAl Viro new_mnt = real_mount(new.mnt); 3115419148daSAl Viro root_mnt = real_mount(root.mnt); 311684d17192SAl Viro old_mnt = real_mount(old.mnt); 311784d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3118fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3119fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3120b12cea91SAl Viro goto out4; 3121143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3122b12cea91SAl Viro goto out4; 31235ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 31245ff9d8a6SEric W. Biederman goto out4; 31251da177e4SLinus Torvalds error = -ENOENT; 3126f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3127b12cea91SAl Viro goto out4; 31281da177e4SLinus Torvalds error = -EBUSY; 312984d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3130b12cea91SAl Viro goto out4; /* loop, on the same file system */ 31311da177e4SLinus Torvalds error = -EINVAL; 31328c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3133b12cea91SAl Viro goto out4; /* not a mountpoint */ 3134676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3135b12cea91SAl Viro goto out4; /* not attached */ 313684d17192SAl Viro root_mp = root_mnt->mnt_mp; 31372d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3138b12cea91SAl Viro goto out4; /* not a mountpoint */ 3139676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3140b12cea91SAl Viro goto out4; /* not attached */ 31414ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 314284d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3143b12cea91SAl Viro goto out4; 31440d082601SEric W. Biederman /* make certain new is below the root */ 31450d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 31460d082601SEric W. Biederman goto out4; 314784d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3148719ea2fbSAl Viro lock_mount_hash(); 3149419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3150419148daSAl Viro detach_mnt(root_mnt, &root_parent); 31515ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 31525ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 31535ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 31545ff9d8a6SEric W. Biederman } 31554ac91378SJan Blunck /* mount old root on put_old */ 315684d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 31574ac91378SJan Blunck /* mount new_root on / */ 315884d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 31596b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 31604fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 31614fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 31623895dbf8SEric W. Biederman put_mountpoint(root_mp); 3163719ea2fbSAl Viro unlock_mount_hash(); 31642d8f3038SAl Viro chroot_fs_refs(&root, &new); 31651da177e4SLinus Torvalds error = 0; 3166b12cea91SAl Viro out4: 316784d17192SAl Viro unlock_mount(old_mp); 3168b12cea91SAl Viro if (!error) { 31691a390689SAl Viro path_put(&root_parent); 31701a390689SAl Viro path_put(&parent_path); 3171b12cea91SAl Viro } 3172b12cea91SAl Viro out3: 31738c3ee42eSAl Viro path_put(&root); 3174b12cea91SAl Viro out2: 31752d8f3038SAl Viro path_put(&old); 31761da177e4SLinus Torvalds out1: 31772d8f3038SAl Viro path_put(&new); 31781da177e4SLinus Torvalds out0: 31791da177e4SLinus Torvalds return error; 31801da177e4SLinus Torvalds } 31811da177e4SLinus Torvalds 31821da177e4SLinus Torvalds static void __init init_mount_tree(void) 31831da177e4SLinus Torvalds { 31841da177e4SLinus Torvalds struct vfsmount *mnt; 31856b3286edSKirill Korotaev struct mnt_namespace *ns; 3186ac748a09SJan Blunck struct path root; 31870c55cfc4SEric W. Biederman struct file_system_type *type; 31881da177e4SLinus Torvalds 31890c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 31900c55cfc4SEric W. Biederman if (!type) 31910c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 31920c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 31930c55cfc4SEric W. Biederman put_filesystem(type); 31941da177e4SLinus Torvalds if (IS_ERR(mnt)) 31951da177e4SLinus Torvalds panic("Can't create rootfs"); 3196b3e19d92SNick Piggin 31973b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 31983b22edc5STrond Myklebust if (IS_ERR(ns)) 31991da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 32001da177e4SLinus Torvalds 32016b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 32026b3286edSKirill Korotaev get_mnt_ns(ns); 32031da177e4SLinus Torvalds 3204be08d6d2SAl Viro root.mnt = mnt; 3205be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3206da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3207ac748a09SJan Blunck 3208ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3209ac748a09SJan Blunck set_fs_root(current->fs, &root); 32101da177e4SLinus Torvalds } 32111da177e4SLinus Torvalds 321274bf17cfSDenis Cheng void __init mnt_init(void) 32131da177e4SLinus Torvalds { 321415a67dd8SRandy Dunlap int err; 32151da177e4SLinus Torvalds 32167d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 321720c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 32181da177e4SLinus Torvalds 32190818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 322038129a13SAl Viro sizeof(struct hlist_head), 32210818bf27SAl Viro mhash_entries, 19, 32223d375d78SPavel Tatashin HASH_ZERO, 32230818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 32240818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 32250818bf27SAl Viro sizeof(struct hlist_head), 32260818bf27SAl Viro mphash_entries, 19, 32273d375d78SPavel Tatashin HASH_ZERO, 32280818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 32291da177e4SLinus Torvalds 323084d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 32311da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 32321da177e4SLinus Torvalds 32334b93dc9bSTejun Heo kernfs_init(); 32344b93dc9bSTejun Heo 323515a67dd8SRandy Dunlap err = sysfs_init(); 323615a67dd8SRandy Dunlap if (err) 323715a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 32388e24eea7SHarvey Harrison __func__, err); 323900d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 324000d26666SGreg Kroah-Hartman if (!fs_kobj) 32418e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 32421da177e4SLinus Torvalds init_rootfs(); 32431da177e4SLinus Torvalds init_mount_tree(); 32441da177e4SLinus Torvalds } 32451da177e4SLinus Torvalds 3246616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 32471da177e4SLinus Torvalds { 3248d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3249616511d0STrond Myklebust return; 32507b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3251771b1371SEric W. Biederman free_mnt_ns(ns); 32521da177e4SLinus Torvalds } 32539d412a43SAl Viro 32549d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 32559d412a43SAl Viro { 3256423e0ab0STim Chen struct vfsmount *mnt; 3257e462ec50SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); 3258423e0ab0STim Chen if (!IS_ERR(mnt)) { 3259423e0ab0STim Chen /* 3260423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3261423e0ab0STim Chen * we unmount before file sys is unregistered 3262423e0ab0STim Chen */ 3263f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3264423e0ab0STim Chen } 3265423e0ab0STim Chen return mnt; 32669d412a43SAl Viro } 32679d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3268423e0ab0STim Chen 3269423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3270423e0ab0STim Chen { 3271423e0ab0STim Chen /* release long term mount so mount point can be released */ 3272423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3273f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 327448a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3275423e0ab0STim Chen mntput(mnt); 3276423e0ab0STim Chen } 3277423e0ab0STim Chen } 3278423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 327902125a82SAl Viro 328002125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 328102125a82SAl Viro { 3282143c8c91SAl Viro return check_mnt(real_mount(mnt)); 328302125a82SAl Viro } 32848823c079SEric W. Biederman 32853151527eSEric W. Biederman bool current_chrooted(void) 32863151527eSEric W. Biederman { 32873151527eSEric W. Biederman /* Does the current process have a non-standard root */ 32883151527eSEric W. Biederman struct path ns_root; 32893151527eSEric W. Biederman struct path fs_root; 32903151527eSEric W. Biederman bool chrooted; 32913151527eSEric W. Biederman 32923151527eSEric W. Biederman /* Find the namespace root */ 32933151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 32943151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 32953151527eSEric W. Biederman path_get(&ns_root); 32963151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 32973151527eSEric W. Biederman ; 32983151527eSEric W. Biederman 32993151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 33003151527eSEric W. Biederman 33013151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 33023151527eSEric W. Biederman 33033151527eSEric W. Biederman path_put(&fs_root); 33043151527eSEric W. Biederman path_put(&ns_root); 33053151527eSEric W. Biederman 33063151527eSEric W. Biederman return chrooted; 33073151527eSEric W. Biederman } 33083151527eSEric W. Biederman 33098654df4eSEric W. Biederman static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, 33108654df4eSEric W. Biederman int *new_mnt_flags) 331187a8ebd6SEric W. Biederman { 33128c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 331387a8ebd6SEric W. Biederman struct mount *mnt; 3314e51db735SEric W. Biederman bool visible = false; 331587a8ebd6SEric W. Biederman 331644bb4385SAl Viro down_read(&namespace_sem); 331787a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3318e51db735SEric W. Biederman struct mount *child; 331977b1a97dSEric W. Biederman int mnt_flags; 332077b1a97dSEric W. Biederman 33218654df4eSEric W. Biederman if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) 3322e51db735SEric W. Biederman continue; 3323e51db735SEric W. Biederman 33247e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 33257e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 33267e96c1b0SEric W. Biederman */ 33277e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 33287e96c1b0SEric W. Biederman continue; 33297e96c1b0SEric W. Biederman 3330a1935c17SEric W. Biederman /* A local view of the mount flags */ 333177b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 333277b1a97dSEric W. Biederman 3333695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3334bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 3335695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3336695e9df0SEric W. Biederman 33378c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 33388c6cf9ccSEric W. Biederman * than the proposed new mount. 33398c6cf9ccSEric W. Biederman */ 334077b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 33418c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 33428c6cf9ccSEric W. Biederman continue; 334377b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 334477b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 33458c6cf9ccSEric W. Biederman continue; 33468c6cf9ccSEric W. Biederman 3347ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3348ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3349ceeb0e5dSEric W. Biederman * empty directories. 3350e51db735SEric W. Biederman */ 3351e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3352e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3353ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3354d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3355ceeb0e5dSEric W. Biederman continue; 33567236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 33577236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3358e51db735SEric W. Biederman goto next; 335987a8ebd6SEric W. Biederman } 33608c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 336177b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 33628c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3363e51db735SEric W. Biederman visible = true; 3364e51db735SEric W. Biederman goto found; 3365e51db735SEric W. Biederman next: ; 336687a8ebd6SEric W. Biederman } 3367e51db735SEric W. Biederman found: 336844bb4385SAl Viro up_read(&namespace_sem); 3369e51db735SEric W. Biederman return visible; 337087a8ebd6SEric W. Biederman } 337187a8ebd6SEric W. Biederman 33728654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) 33738654df4eSEric W. Biederman { 3374a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 33758654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 33768654df4eSEric W. Biederman unsigned long s_iflags; 33778654df4eSEric W. Biederman 33788654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 33798654df4eSEric W. Biederman return false; 33808654df4eSEric W. Biederman 33818654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 33828654df4eSEric W. Biederman s_iflags = mnt->mnt_sb->s_iflags; 33838654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 33848654df4eSEric W. Biederman return false; 33858654df4eSEric W. Biederman 3386a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3387a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3388a1935c17SEric W. Biederman required_iflags); 3389a1935c17SEric W. Biederman return true; 3390a1935c17SEric W. Biederman } 3391a1935c17SEric W. Biederman 33928654df4eSEric W. Biederman return !mnt_already_visible(ns, mnt, new_mnt_flags); 33938654df4eSEric W. Biederman } 33948654df4eSEric W. Biederman 3395380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3396380cf5baSAndy Lutomirski { 3397380cf5baSAndy Lutomirski /* 3398380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3399380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3400380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3401380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3402380cf5baSAndy Lutomirski * in other namespaces. 3403380cf5baSAndy Lutomirski */ 3404380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3405380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3406380cf5baSAndy Lutomirski } 3407380cf5baSAndy Lutomirski 340864964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 34098823c079SEric W. Biederman { 341058be2825SAl Viro struct ns_common *ns = NULL; 34118823c079SEric W. Biederman struct nsproxy *nsproxy; 34128823c079SEric W. Biederman 3413728dba3aSEric W. Biederman task_lock(task); 3414728dba3aSEric W. Biederman nsproxy = task->nsproxy; 34158823c079SEric W. Biederman if (nsproxy) { 341658be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 341758be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 34188823c079SEric W. Biederman } 3419728dba3aSEric W. Biederman task_unlock(task); 34208823c079SEric W. Biederman 34218823c079SEric W. Biederman return ns; 34228823c079SEric W. Biederman } 34238823c079SEric W. Biederman 342464964528SAl Viro static void mntns_put(struct ns_common *ns) 34258823c079SEric W. Biederman { 342658be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 34278823c079SEric W. Biederman } 34288823c079SEric W. Biederman 342964964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 34308823c079SEric W. Biederman { 34318823c079SEric W. Biederman struct fs_struct *fs = current->fs; 34324f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 34338823c079SEric W. Biederman struct path root; 34344f757f3cSAl Viro int err; 34358823c079SEric W. Biederman 34360c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3437c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3438c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3439ae11e0f1SZhao Hongjiang return -EPERM; 34408823c079SEric W. Biederman 34418823c079SEric W. Biederman if (fs->users != 1) 34428823c079SEric W. Biederman return -EINVAL; 34438823c079SEric W. Biederman 34448823c079SEric W. Biederman get_mnt_ns(mnt_ns); 34454f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 34468823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 34478823c079SEric W. Biederman 34488823c079SEric W. Biederman /* Find the root */ 34494f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 34504f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 34514f757f3cSAl Viro if (err) { 34524f757f3cSAl Viro /* revert to old namespace */ 34534f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 34544f757f3cSAl Viro put_mnt_ns(mnt_ns); 34554f757f3cSAl Viro return err; 34564f757f3cSAl Viro } 34578823c079SEric W. Biederman 34584068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 34594068367cSAndrei Vagin 34608823c079SEric W. Biederman /* Update the pwd and root */ 34618823c079SEric W. Biederman set_fs_pwd(fs, &root); 34628823c079SEric W. Biederman set_fs_root(fs, &root); 34638823c079SEric W. Biederman 34648823c079SEric W. Biederman path_put(&root); 34658823c079SEric W. Biederman return 0; 34668823c079SEric W. Biederman } 34678823c079SEric W. Biederman 3468bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3469bcac25a5SAndrey Vagin { 3470bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3471bcac25a5SAndrey Vagin } 3472bcac25a5SAndrey Vagin 34738823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 34748823c079SEric W. Biederman .name = "mnt", 34758823c079SEric W. Biederman .type = CLONE_NEWNS, 34768823c079SEric W. Biederman .get = mntns_get, 34778823c079SEric W. Biederman .put = mntns_put, 34788823c079SEric W. Biederman .install = mntns_install, 3479bcac25a5SAndrey Vagin .owner = mntns_owner, 34808823c079SEric W. Biederman }; 3481