159bd9dedSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namespace.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 185b825c3aSIngo Molnar #include <linux/cred.h> 1973cd49ecSMiklos Szeredi #include <linux/idr.h> 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23a07b2000SAl Viro #include <linux/file.h> 24d10577a8SAl Viro #include <linux/uaccess.h> 250bb80f24SDavid Howells #include <linux/proc_ns.h> 2620b4fb48SLinus Torvalds #include <linux/magic.h> 2757c8a661SMike Rapoport #include <linux/memblock.h> 289caccd41SChristian Brauner #include <linux/proc_fs.h> 299ea459e1SAl Viro #include <linux/task_work.h> 309164bb4aSIngo Molnar #include <linux/sched/task.h> 31e262e32dSDavid Howells #include <uapi/linux/mount.h> 329bc61ab1SDavid Howells #include <linux/fs_context.h> 33037f11b4SAl Viro #include <linux/shmem_fs.h> 34bd303368SChristian Brauner #include <linux/mnt_idmapping.h> 359164bb4aSIngo Molnar 3607b20889SRam Pai #include "pnode.h" 37948730b0SAdrian Bunk #include "internal.h" 381da177e4SLinus Torvalds 39d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 40ab171b95SLuis Chamberlain static unsigned int sysctl_mount_max __read_mostly = 100000; 41d2921684SEric W. Biederman 420818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 430818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 440818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 450818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 460818bf27SAl Viro 470818bf27SAl Viro static __initdata unsigned long mhash_entries; 480818bf27SAl Viro static int __init set_mhash_entries(char *str) 490818bf27SAl Viro { 500818bf27SAl Viro if (!str) 510818bf27SAl Viro return 0; 520818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 530818bf27SAl Viro return 1; 540818bf27SAl Viro } 550818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 560818bf27SAl Viro 570818bf27SAl Viro static __initdata unsigned long mphash_entries; 580818bf27SAl Viro static int __init set_mphash_entries(char *str) 590818bf27SAl Viro { 600818bf27SAl Viro if (!str) 610818bf27SAl Viro return 0; 620818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 630818bf27SAl Viro return 1; 640818bf27SAl Viro } 650818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6613f14b4dSEric Dumazet 67c7999c36SAl Viro static u64 event; 6873cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 69719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 705addc5ddSAl Viro 7138129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 720818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 73e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 7459aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 754edbe133SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 764edbe133SAl Viro static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ 771da177e4SLinus Torvalds 782a186721SChristian Brauner struct mount_kattr { 792a186721SChristian Brauner unsigned int attr_set; 802a186721SChristian Brauner unsigned int attr_clr; 812a186721SChristian Brauner unsigned int propagation; 822a186721SChristian Brauner unsigned int lookup_flags; 832a186721SChristian Brauner bool recurse; 849caccd41SChristian Brauner struct user_namespace *mnt_userns; 852a186721SChristian Brauner }; 862a186721SChristian Brauner 87f87fd4c2SMiklos Szeredi /* /sys/fs */ 8800d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 8900d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 90f87fd4c2SMiklos Szeredi 9199b7db7bSNick Piggin /* 9299b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 9399b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 9499b7db7bSNick Piggin * up the tree. 9599b7db7bSNick Piggin * 9699b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 9799b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 9899b7db7bSNick Piggin */ 9948a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 10099b7db7bSNick Piggin 101d033cb67SChristian Brauner static inline void lock_mount_hash(void) 102d033cb67SChristian Brauner { 103d033cb67SChristian Brauner write_seqlock(&mount_lock); 104d033cb67SChristian Brauner } 105d033cb67SChristian Brauner 106d033cb67SChristian Brauner static inline void unlock_mount_hash(void) 107d033cb67SChristian Brauner { 108d033cb67SChristian Brauner write_sequnlock(&mount_lock); 109d033cb67SChristian Brauner } 110d033cb67SChristian Brauner 11138129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 1121da177e4SLinus Torvalds { 1131da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 1141da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 1150818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 1160818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 1170818bf27SAl Viro } 1180818bf27SAl Viro 1190818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 1200818bf27SAl Viro { 1210818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 1220818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 1230818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 1241da177e4SLinus Torvalds } 1251da177e4SLinus Torvalds 126b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 12773cd49ecSMiklos Szeredi { 128169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 12973cd49ecSMiklos Szeredi 130169b480eSMatthew Wilcox if (res < 0) 13173cd49ecSMiklos Szeredi return res; 132169b480eSMatthew Wilcox mnt->mnt_id = res; 133169b480eSMatthew Wilcox return 0; 13473cd49ecSMiklos Szeredi } 13573cd49ecSMiklos Szeredi 136b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 13773cd49ecSMiklos Szeredi { 138169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 13973cd49ecSMiklos Szeredi } 14073cd49ecSMiklos Szeredi 141719f5d7fSMiklos Szeredi /* 142719f5d7fSMiklos Szeredi * Allocate a new peer group ID 143719f5d7fSMiklos Szeredi */ 1444b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 145719f5d7fSMiklos Szeredi { 146169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 147f21f6220SAl Viro 148169b480eSMatthew Wilcox if (res < 0) 149f21f6220SAl Viro return res; 150169b480eSMatthew Wilcox mnt->mnt_group_id = res; 151169b480eSMatthew Wilcox return 0; 152719f5d7fSMiklos Szeredi } 153719f5d7fSMiklos Szeredi 154719f5d7fSMiklos Szeredi /* 155719f5d7fSMiklos Szeredi * Release a peer group ID 156719f5d7fSMiklos Szeredi */ 1574b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 158719f5d7fSMiklos Szeredi { 159169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 16015169fe7SAl Viro mnt->mnt_group_id = 0; 161719f5d7fSMiklos Szeredi } 162719f5d7fSMiklos Szeredi 163b3e19d92SNick Piggin /* 164b3e19d92SNick Piggin * vfsmount lock must be held for read 165b3e19d92SNick Piggin */ 16683adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 167b3e19d92SNick Piggin { 168b3e19d92SNick Piggin #ifdef CONFIG_SMP 16968e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 170b3e19d92SNick Piggin #else 171b3e19d92SNick Piggin preempt_disable(); 17268e8a9feSAl Viro mnt->mnt_count += n; 173b3e19d92SNick Piggin preempt_enable(); 174b3e19d92SNick Piggin #endif 175b3e19d92SNick Piggin } 176b3e19d92SNick Piggin 177b3e19d92SNick Piggin /* 178b3e19d92SNick Piggin * vfsmount lock must be held for write 179b3e19d92SNick Piggin */ 180edf7ddbfSEric Biggers int mnt_get_count(struct mount *mnt) 181b3e19d92SNick Piggin { 182b3e19d92SNick Piggin #ifdef CONFIG_SMP 183edf7ddbfSEric Biggers int count = 0; 184b3e19d92SNick Piggin int cpu; 185b3e19d92SNick Piggin 186b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 18768e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 188b3e19d92SNick Piggin } 189b3e19d92SNick Piggin 190b3e19d92SNick Piggin return count; 191b3e19d92SNick Piggin #else 19268e8a9feSAl Viro return mnt->mnt_count; 193b3e19d92SNick Piggin #endif 194b3e19d92SNick Piggin } 195b3e19d92SNick Piggin 196b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1971da177e4SLinus Torvalds { 198c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 199c63181e6SAl Viro if (mnt) { 20073cd49ecSMiklos Szeredi int err; 20173cd49ecSMiklos Szeredi 202c63181e6SAl Viro err = mnt_alloc_id(mnt); 20388b38782SLi Zefan if (err) 20488b38782SLi Zefan goto out_free_cache; 20588b38782SLi Zefan 20688b38782SLi Zefan if (name) { 20779f6540bSVasily Averin mnt->mnt_devname = kstrdup_const(name, 20879f6540bSVasily Averin GFP_KERNEL_ACCOUNT); 209c63181e6SAl Viro if (!mnt->mnt_devname) 21088b38782SLi Zefan goto out_free_id; 21173cd49ecSMiklos Szeredi } 21273cd49ecSMiklos Szeredi 213b3e19d92SNick Piggin #ifdef CONFIG_SMP 214c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 215c63181e6SAl Viro if (!mnt->mnt_pcp) 216b3e19d92SNick Piggin goto out_free_devname; 217b3e19d92SNick Piggin 218c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 219b3e19d92SNick Piggin #else 220c63181e6SAl Viro mnt->mnt_count = 1; 221c63181e6SAl Viro mnt->mnt_writers = 0; 222b3e19d92SNick Piggin #endif 223b3e19d92SNick Piggin 22438129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 225c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 226c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 227c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 228c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 229c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 230c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 231c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2320a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 23399b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 23456cbb429SAl Viro INIT_HLIST_HEAD(&mnt->mnt_stuck_children); 235a6435940SChristian Brauner mnt->mnt.mnt_userns = &init_user_ns; 2361da177e4SLinus Torvalds } 237c63181e6SAl Viro return mnt; 23888b38782SLi Zefan 239d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 240d3ef3d73Snpiggin@suse.de out_free_devname: 241fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 242d3ef3d73Snpiggin@suse.de #endif 24388b38782SLi Zefan out_free_id: 244c63181e6SAl Viro mnt_free_id(mnt); 24588b38782SLi Zefan out_free_cache: 246c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 24788b38782SLi Zefan return NULL; 2481da177e4SLinus Torvalds } 2491da177e4SLinus Torvalds 2508366025eSDave Hansen /* 2518366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2528366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2538366025eSDave Hansen * We must keep track of when those operations start 2548366025eSDave Hansen * (for permission checks) and when they end, so that 2558366025eSDave Hansen * we can determine when writes are able to occur to 2568366025eSDave Hansen * a filesystem. 2578366025eSDave Hansen */ 2583d733633SDave Hansen /* 2593d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2603d733633SDave Hansen * @mnt: the mount to check for its write status 2613d733633SDave Hansen * 2623d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2633d733633SDave Hansen * It does not guarantee that the filesystem will stay 2643d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2653d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2663d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2673d733633SDave Hansen * r/w. 2683d733633SDave Hansen */ 26943f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 2703d733633SDave Hansen { 27143f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 2723d733633SDave Hansen } 2733d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2743d733633SDave Hansen 27583adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2763d733633SDave Hansen { 277d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 27868e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 279d3ef3d73Snpiggin@suse.de #else 28068e8a9feSAl Viro mnt->mnt_writers++; 281d3ef3d73Snpiggin@suse.de #endif 2823d733633SDave Hansen } 2833d733633SDave Hansen 28483adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2853d733633SDave Hansen { 286d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 28768e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 288d3ef3d73Snpiggin@suse.de #else 28968e8a9feSAl Viro mnt->mnt_writers--; 290d3ef3d73Snpiggin@suse.de #endif 291d3ef3d73Snpiggin@suse.de } 292d3ef3d73Snpiggin@suse.de 29383adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 294d3ef3d73Snpiggin@suse.de { 295d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 296d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2973d733633SDave Hansen int cpu; 2983d733633SDave Hansen 2993d733633SDave Hansen for_each_possible_cpu(cpu) { 30068e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 3013d733633SDave Hansen } 3023d733633SDave Hansen 303d3ef3d73Snpiggin@suse.de return count; 304d3ef3d73Snpiggin@suse.de #else 305d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 306d3ef3d73Snpiggin@suse.de #endif 3073d733633SDave Hansen } 3083d733633SDave Hansen 3094ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3104ed5e82fSMiklos Szeredi { 3114ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3124ed5e82fSMiklos Szeredi return 1; 3134ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3144ed5e82fSMiklos Szeredi smp_rmb(); 3154ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3164ed5e82fSMiklos Szeredi } 3174ed5e82fSMiklos Szeredi 3183d733633SDave Hansen /* 319eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 320eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 321eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 322eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3233d733633SDave Hansen */ 3248366025eSDave Hansen /** 325eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 32683adc753SAl Viro * @m: the mount on which to take a write 3278366025eSDave Hansen * 328eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 329eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 330eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 331eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 332eb04c282SJan Kara * called. This is effectively a refcount. 3338366025eSDave Hansen */ 334eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3358366025eSDave Hansen { 33683adc753SAl Viro struct mount *mnt = real_mount(m); 3373d733633SDave Hansen int ret = 0; 3383d733633SDave Hansen 339d3ef3d73Snpiggin@suse.de preempt_disable(); 340c6653a83SNick Piggin mnt_inc_writers(mnt); 341d3ef3d73Snpiggin@suse.de /* 342c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 343d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 344d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 345d3ef3d73Snpiggin@suse.de */ 346d3ef3d73Snpiggin@suse.de smp_mb(); 3476aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 348d3ef3d73Snpiggin@suse.de cpu_relax(); 349d3ef3d73Snpiggin@suse.de /* 350d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 351d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 352d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 353d3ef3d73Snpiggin@suse.de */ 354d3ef3d73Snpiggin@suse.de smp_rmb(); 3554ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 356c6653a83SNick Piggin mnt_dec_writers(mnt); 3573d733633SDave Hansen ret = -EROFS; 3583d733633SDave Hansen } 359d3ef3d73Snpiggin@suse.de preempt_enable(); 360eb04c282SJan Kara 361eb04c282SJan Kara return ret; 362eb04c282SJan Kara } 363eb04c282SJan Kara 364eb04c282SJan Kara /** 365eb04c282SJan Kara * mnt_want_write - get write access to a mount 366eb04c282SJan Kara * @m: the mount on which to take a write 367eb04c282SJan Kara * 368eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 369eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 370eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 371eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 372eb04c282SJan Kara */ 373eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 374eb04c282SJan Kara { 375eb04c282SJan Kara int ret; 376eb04c282SJan Kara 377eb04c282SJan Kara sb_start_write(m->mnt_sb); 378eb04c282SJan Kara ret = __mnt_want_write(m); 379eb04c282SJan Kara if (ret) 380eb04c282SJan Kara sb_end_write(m->mnt_sb); 3813d733633SDave Hansen return ret; 3828366025eSDave Hansen } 3838366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3848366025eSDave Hansen 3858366025eSDave Hansen /** 386eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 387eb04c282SJan Kara * @file: the file who's mount on which to take a write 388eb04c282SJan Kara * 38914e43bf4SEric Biggers * This is like __mnt_want_write, but if the file is already open for writing it 39014e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 39114e43bf4SEric Biggers * and instead only does the check for emergency r/o remounts. This must be 39214e43bf4SEric Biggers * paired with __mnt_drop_write_file. 393eb04c282SJan Kara */ 394eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 395eb04c282SJan Kara { 39614e43bf4SEric Biggers if (file->f_mode & FMODE_WRITER) { 39714e43bf4SEric Biggers /* 39814e43bf4SEric Biggers * Superblock may have become readonly while there are still 39914e43bf4SEric Biggers * writable fd's, e.g. due to a fs error with errors=remount-ro 40014e43bf4SEric Biggers */ 40114e43bf4SEric Biggers if (__mnt_is_readonly(file->f_path.mnt)) 40214e43bf4SEric Biggers return -EROFS; 40314e43bf4SEric Biggers return 0; 40414e43bf4SEric Biggers } 405eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 406eb04c282SJan Kara } 407eb04c282SJan Kara 408eb04c282SJan Kara /** 4097c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4107c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4117c6893e3SMiklos Szeredi * 41214e43bf4SEric Biggers * This is like mnt_want_write, but if the file is already open for writing it 41314e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 41414e43bf4SEric Biggers * and instead only does the freeze protection and the check for emergency r/o 41514e43bf4SEric Biggers * remounts. This must be paired with mnt_drop_write_file. 4167c6893e3SMiklos Szeredi */ 4177c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4187c6893e3SMiklos Szeredi { 4197c6893e3SMiklos Szeredi int ret; 4207c6893e3SMiklos Szeredi 4217c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4227c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4237c6893e3SMiklos Szeredi if (ret) 4247c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4257c6893e3SMiklos Szeredi return ret; 4267c6893e3SMiklos Szeredi } 42796029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 42896029c4eSnpiggin@suse.de 42996029c4eSnpiggin@suse.de /** 430eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4318366025eSDave Hansen * @mnt: the mount on which to give up write access 4328366025eSDave Hansen * 4338366025eSDave Hansen * Tells the low-level filesystem that we are done 4348366025eSDave Hansen * performing writes to it. Must be matched with 435eb04c282SJan Kara * __mnt_want_write() call above. 4368366025eSDave Hansen */ 437eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4388366025eSDave Hansen { 439d3ef3d73Snpiggin@suse.de preempt_disable(); 44083adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 441d3ef3d73Snpiggin@suse.de preempt_enable(); 4428366025eSDave Hansen } 443eb04c282SJan Kara 444eb04c282SJan Kara /** 445eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 446eb04c282SJan Kara * @mnt: the mount on which to give up write access 447eb04c282SJan Kara * 448eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 449eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 450eb04c282SJan Kara * mnt_want_write() call above. 451eb04c282SJan Kara */ 452eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 453eb04c282SJan Kara { 454eb04c282SJan Kara __mnt_drop_write(mnt); 455eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 456eb04c282SJan Kara } 4578366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4588366025eSDave Hansen 459eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 460eb04c282SJan Kara { 46114e43bf4SEric Biggers if (!(file->f_mode & FMODE_WRITER)) 462eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 463eb04c282SJan Kara } 464eb04c282SJan Kara 4657c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4667c6893e3SMiklos Szeredi { 467a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4687c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4697c6893e3SMiklos Szeredi } 4702a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4712a79f17eSAl Viro 472fbdc2f6cSChristian Brauner static inline int mnt_hold_writers(struct mount *mnt) 4738366025eSDave Hansen { 47483adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 475d3ef3d73Snpiggin@suse.de /* 476d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 477d3ef3d73Snpiggin@suse.de * should be visible before we do. 478d3ef3d73Snpiggin@suse.de */ 479d3ef3d73Snpiggin@suse.de smp_mb(); 480d3ef3d73Snpiggin@suse.de 481d3ef3d73Snpiggin@suse.de /* 482d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 483d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 484d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 485d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 486d3ef3d73Snpiggin@suse.de * 487d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 488d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 489d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 490d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 491d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 492d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 493d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 494d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 495d3ef3d73Snpiggin@suse.de * we're counting up here. 496d3ef3d73Snpiggin@suse.de */ 497c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 498fbdc2f6cSChristian Brauner return -EBUSY; 499fbdc2f6cSChristian Brauner 500fbdc2f6cSChristian Brauner return 0; 501fbdc2f6cSChristian Brauner } 502fbdc2f6cSChristian Brauner 503fbdc2f6cSChristian Brauner static inline void mnt_unhold_writers(struct mount *mnt) 504fbdc2f6cSChristian Brauner { 505d3ef3d73Snpiggin@suse.de /* 506d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 507d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 508d3ef3d73Snpiggin@suse.de */ 509d3ef3d73Snpiggin@suse.de smp_wmb(); 51083adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 511fbdc2f6cSChristian Brauner } 512fbdc2f6cSChristian Brauner 513fbdc2f6cSChristian Brauner static int mnt_make_readonly(struct mount *mnt) 514fbdc2f6cSChristian Brauner { 515fbdc2f6cSChristian Brauner int ret; 516fbdc2f6cSChristian Brauner 517fbdc2f6cSChristian Brauner ret = mnt_hold_writers(mnt); 518fbdc2f6cSChristian Brauner if (!ret) 519fbdc2f6cSChristian Brauner mnt->mnt.mnt_flags |= MNT_READONLY; 520fbdc2f6cSChristian Brauner mnt_unhold_writers(mnt); 5213d733633SDave Hansen return ret; 5223d733633SDave Hansen } 5238366025eSDave Hansen 5244ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5254ed5e82fSMiklos Szeredi { 5264ed5e82fSMiklos Szeredi struct mount *mnt; 5274ed5e82fSMiklos Szeredi int err = 0; 5284ed5e82fSMiklos Szeredi 5298e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5308e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5318e8b8796SMiklos Szeredi return -EBUSY; 5328e8b8796SMiklos Szeredi 533719ea2fbSAl Viro lock_mount_hash(); 5344ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5354ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5364ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5374ed5e82fSMiklos Szeredi smp_mb(); 5384ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5394ed5e82fSMiklos Szeredi err = -EBUSY; 5404ed5e82fSMiklos Szeredi break; 5414ed5e82fSMiklos Szeredi } 5424ed5e82fSMiklos Szeredi } 5434ed5e82fSMiklos Szeredi } 5448e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5458e8b8796SMiklos Szeredi err = -EBUSY; 5468e8b8796SMiklos Szeredi 5474ed5e82fSMiklos Szeredi if (!err) { 5484ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5494ed5e82fSMiklos Szeredi smp_wmb(); 5504ed5e82fSMiklos Szeredi } 5514ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5524ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5534ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5544ed5e82fSMiklos Szeredi } 555719ea2fbSAl Viro unlock_mount_hash(); 5564ed5e82fSMiklos Szeredi 5574ed5e82fSMiklos Szeredi return err; 5584ed5e82fSMiklos Szeredi } 5594ed5e82fSMiklos Szeredi 560b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5611da177e4SLinus Torvalds { 562a6435940SChristian Brauner struct user_namespace *mnt_userns; 563a6435940SChristian Brauner 564a6435940SChristian Brauner mnt_userns = mnt_user_ns(&mnt->mnt); 565bd303368SChristian Brauner if (!initial_idmapping(mnt_userns)) 566a6435940SChristian Brauner put_user_ns(mnt_userns); 567fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 568d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 56968e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 570d3ef3d73Snpiggin@suse.de #endif 571b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5721da177e4SLinus Torvalds } 5731da177e4SLinus Torvalds 5748ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5758ffcb32eSDavid Howells { 5768ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5778ffcb32eSDavid Howells } 5788ffcb32eSDavid Howells 57948a066e7SAl Viro /* call under rcu_read_lock */ 580294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 58148a066e7SAl Viro { 58248a066e7SAl Viro struct mount *mnt; 58348a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 584294d71ffSAl Viro return 1; 58548a066e7SAl Viro if (bastard == NULL) 586294d71ffSAl Viro return 0; 58748a066e7SAl Viro mnt = real_mount(bastard); 58848a066e7SAl Viro mnt_add_count(mnt, 1); 589119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 59048a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 591294d71ffSAl Viro return 0; 59248a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 59348a066e7SAl Viro mnt_add_count(mnt, -1); 594294d71ffSAl Viro return 1; 59548a066e7SAl Viro } 596119e1ef8SAl Viro lock_mount_hash(); 597119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 598119e1ef8SAl Viro mnt_add_count(mnt, -1); 599119e1ef8SAl Viro unlock_mount_hash(); 600119e1ef8SAl Viro return 1; 601119e1ef8SAl Viro } 602119e1ef8SAl Viro unlock_mount_hash(); 603119e1ef8SAl Viro /* caller will mntput() */ 604294d71ffSAl Viro return -1; 605294d71ffSAl Viro } 606294d71ffSAl Viro 607294d71ffSAl Viro /* call under rcu_read_lock */ 608294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 609294d71ffSAl Viro { 610294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 611294d71ffSAl Viro if (likely(!res)) 612294d71ffSAl Viro return true; 613294d71ffSAl Viro if (unlikely(res < 0)) { 61448a066e7SAl Viro rcu_read_unlock(); 61548a066e7SAl Viro mntput(bastard); 61648a066e7SAl Viro rcu_read_lock(); 617294d71ffSAl Viro } 61848a066e7SAl Viro return false; 61948a066e7SAl Viro } 62048a066e7SAl Viro 6211da177e4SLinus Torvalds /* 622474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 62348a066e7SAl Viro * call under rcu_read_lock() 6241da177e4SLinus Torvalds */ 625474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6261da177e4SLinus Torvalds { 62738129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 628474279dcSAl Viro struct mount *p; 6291da177e4SLinus Torvalds 63038129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 631474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 632474279dcSAl Viro return p; 633474279dcSAl Viro return NULL; 6341da177e4SLinus Torvalds } 635474279dcSAl Viro 636474279dcSAl Viro /* 637f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 638f015f126SDavid Howells * 639f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 640f015f126SDavid Howells * following mounts: 641f015f126SDavid Howells * 642f015f126SDavid Howells * mount /dev/sda1 /mnt 643f015f126SDavid Howells * mount /dev/sda2 /mnt 644f015f126SDavid Howells * mount /dev/sda3 /mnt 645f015f126SDavid Howells * 646f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 647f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 648f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 649f015f126SDavid Howells * 650f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 651a05964f3SRam Pai */ 652ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 653a05964f3SRam Pai { 654c7105365SAl Viro struct mount *child_mnt; 65548a066e7SAl Viro struct vfsmount *m; 65648a066e7SAl Viro unsigned seq; 65799b7db7bSNick Piggin 65848a066e7SAl Viro rcu_read_lock(); 65948a066e7SAl Viro do { 66048a066e7SAl Viro seq = read_seqbegin(&mount_lock); 661474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 66248a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 66348a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 66448a066e7SAl Viro rcu_read_unlock(); 66548a066e7SAl Viro return m; 666a05964f3SRam Pai } 667a05964f3SRam Pai 6689f6c61f9SMiklos Szeredi static inline void lock_ns_list(struct mnt_namespace *ns) 6699f6c61f9SMiklos Szeredi { 6709f6c61f9SMiklos Szeredi spin_lock(&ns->ns_lock); 6719f6c61f9SMiklos Szeredi } 6729f6c61f9SMiklos Szeredi 6739f6c61f9SMiklos Szeredi static inline void unlock_ns_list(struct mnt_namespace *ns) 6749f6c61f9SMiklos Szeredi { 6759f6c61f9SMiklos Szeredi spin_unlock(&ns->ns_lock); 6769f6c61f9SMiklos Szeredi } 6779f6c61f9SMiklos Szeredi 6789f6c61f9SMiklos Szeredi static inline bool mnt_is_cursor(struct mount *mnt) 6799f6c61f9SMiklos Szeredi { 6809f6c61f9SMiklos Szeredi return mnt->mnt.mnt_flags & MNT_CURSOR; 6819f6c61f9SMiklos Szeredi } 6829f6c61f9SMiklos Szeredi 6837af1364fSEric W. Biederman /* 6847af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6857af1364fSEric W. Biederman * current mount namespace. 6867af1364fSEric W. Biederman * 6877af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6887af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6897af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6907af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6917af1364fSEric W. Biederman * is a mountpoint. 6927af1364fSEric W. Biederman * 6937af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6947af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6957af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6967af1364fSEric W. Biederman * parent mount. 6977af1364fSEric W. Biederman */ 6987af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6997af1364fSEric W. Biederman { 7007af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7017af1364fSEric W. Biederman struct mount *mnt; 7027af1364fSEric W. Biederman bool is_covered = false; 7037af1364fSEric W. Biederman 7047af1364fSEric W. Biederman down_read(&namespace_sem); 7059f6c61f9SMiklos Szeredi lock_ns_list(ns); 7067af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7079f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 7089f6c61f9SMiklos Szeredi continue; 7097af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7107af1364fSEric W. Biederman if (is_covered) 7117af1364fSEric W. Biederman break; 7127af1364fSEric W. Biederman } 7139f6c61f9SMiklos Szeredi unlock_ns_list(ns); 7147af1364fSEric W. Biederman up_read(&namespace_sem); 7155ad05cc8SNikolay Borisov 7167af1364fSEric W. Biederman return is_covered; 7177af1364fSEric W. Biederman } 7187af1364fSEric W. Biederman 719e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 72084d17192SAl Viro { 7210818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 72284d17192SAl Viro struct mountpoint *mp; 72384d17192SAl Viro 7240818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 72584d17192SAl Viro if (mp->m_dentry == dentry) { 72684d17192SAl Viro mp->m_count++; 72784d17192SAl Viro return mp; 72884d17192SAl Viro } 72984d17192SAl Viro } 730e2dfa935SEric W. Biederman return NULL; 731e2dfa935SEric W. Biederman } 732e2dfa935SEric W. Biederman 7333895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 734e2dfa935SEric W. Biederman { 7353895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 736e2dfa935SEric W. Biederman int ret; 73784d17192SAl Viro 7383895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7391e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7401e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7411e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7423895dbf8SEric W. Biederman mountpoint: 7433895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7443895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7453895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7463895dbf8SEric W. Biederman if (mp) 7473895dbf8SEric W. Biederman goto done; 74884d17192SAl Viro } 749eed81007SMiklos Szeredi 7503895dbf8SEric W. Biederman if (!new) 7513895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7523895dbf8SEric W. Biederman if (!new) 7533895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7543895dbf8SEric W. Biederman 7553895dbf8SEric W. Biederman 7563895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7573895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7583895dbf8SEric W. Biederman 7593895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7603895dbf8SEric W. Biederman if (ret == -EBUSY) 7613895dbf8SEric W. Biederman goto mountpoint; 7623895dbf8SEric W. Biederman 7633895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7643895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7653895dbf8SEric W. Biederman if (ret) 7663895dbf8SEric W. Biederman goto done; 7673895dbf8SEric W. Biederman 7683895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7693895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7704edbe133SAl Viro new->m_dentry = dget(dentry); 7713895dbf8SEric W. Biederman new->m_count = 1; 7723895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7733895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7743895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7753895dbf8SEric W. Biederman 7763895dbf8SEric W. Biederman mp = new; 7773895dbf8SEric W. Biederman new = NULL; 7783895dbf8SEric W. Biederman done: 7793895dbf8SEric W. Biederman kfree(new); 78084d17192SAl Viro return mp; 78184d17192SAl Viro } 78284d17192SAl Viro 7834edbe133SAl Viro /* 7844edbe133SAl Viro * vfsmount lock must be held. Additionally, the caller is responsible 7854edbe133SAl Viro * for serializing calls for given disposal list. 7864edbe133SAl Viro */ 7874edbe133SAl Viro static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) 78884d17192SAl Viro { 78984d17192SAl Viro if (!--mp->m_count) { 79084d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7910a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 79284d17192SAl Viro spin_lock(&dentry->d_lock); 79384d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 79484d17192SAl Viro spin_unlock(&dentry->d_lock); 7954edbe133SAl Viro dput_to_list(dentry, list); 7960818bf27SAl Viro hlist_del(&mp->m_hash); 79784d17192SAl Viro kfree(mp); 79884d17192SAl Viro } 79984d17192SAl Viro } 80084d17192SAl Viro 8014edbe133SAl Viro /* called with namespace_lock and vfsmount lock */ 8024edbe133SAl Viro static void put_mountpoint(struct mountpoint *mp) 8034edbe133SAl Viro { 8044edbe133SAl Viro __put_mountpoint(mp, &ex_mountpoints); 8054edbe133SAl Viro } 8064edbe133SAl Viro 807143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 8081da177e4SLinus Torvalds { 8096b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 8101da177e4SLinus Torvalds } 8111da177e4SLinus Torvalds 81299b7db7bSNick Piggin /* 81399b7db7bSNick Piggin * vfsmount lock must be held for write 81499b7db7bSNick Piggin */ 8156b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 8165addc5ddSAl Viro { 8175addc5ddSAl Viro if (ns) { 8185addc5ddSAl Viro ns->event = ++event; 8195addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8205addc5ddSAl Viro } 8215addc5ddSAl Viro } 8225addc5ddSAl Viro 82399b7db7bSNick Piggin /* 82499b7db7bSNick Piggin * vfsmount lock must be held for write 82599b7db7bSNick Piggin */ 8266b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8275addc5ddSAl Viro { 8285addc5ddSAl Viro if (ns && ns->event != event) { 8295addc5ddSAl Viro ns->event = event; 8305addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8315addc5ddSAl Viro } 8325addc5ddSAl Viro } 8335addc5ddSAl Viro 83499b7db7bSNick Piggin /* 83599b7db7bSNick Piggin * vfsmount lock must be held for write 83699b7db7bSNick Piggin */ 837e4e59906SAl Viro static struct mountpoint *unhash_mnt(struct mount *mnt) 8381da177e4SLinus Torvalds { 839e4e59906SAl Viro struct mountpoint *mp; 8400714a533SAl Viro mnt->mnt_parent = mnt; 841a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8426b41d536SAl Viro list_del_init(&mnt->mnt_child); 84338129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8440a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 845e4e59906SAl Viro mp = mnt->mnt_mp; 84684d17192SAl Viro mnt->mnt_mp = NULL; 847e4e59906SAl Viro return mp; 8487bdb11deSEric W. Biederman } 8497bdb11deSEric W. Biederman 8507bdb11deSEric W. Biederman /* 8517bdb11deSEric W. Biederman * vfsmount lock must be held for write 8527bdb11deSEric W. Biederman */ 8536a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8546a46c573SEric W. Biederman { 855e4e59906SAl Viro put_mountpoint(unhash_mnt(mnt)); 8566a46c573SEric W. Biederman } 8576a46c573SEric W. Biederman 8586a46c573SEric W. Biederman /* 8596a46c573SEric W. Biederman * vfsmount lock must be held for write 8606a46c573SEric W. Biederman */ 86184d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 86284d17192SAl Viro struct mountpoint *mp, 86344d964d6SAl Viro struct mount *child_mnt) 864b90fa9aeSRam Pai { 86584d17192SAl Viro mp->m_count++; 8663a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 8674edbe133SAl Viro child_mnt->mnt_mountpoint = mp->m_dentry; 8683a2393d7SAl Viro child_mnt->mnt_parent = mnt; 86984d17192SAl Viro child_mnt->mnt_mp = mp; 8700a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 871b90fa9aeSRam Pai } 872b90fa9aeSRam Pai 8731064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8741064f874SEric W. Biederman { 8751064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8761064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8771064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8781064f874SEric W. Biederman } 8791064f874SEric W. Biederman 88099b7db7bSNick Piggin /* 88199b7db7bSNick Piggin * vfsmount lock must be held for write 88299b7db7bSNick Piggin */ 88384d17192SAl Viro static void attach_mnt(struct mount *mnt, 88484d17192SAl Viro struct mount *parent, 88584d17192SAl Viro struct mountpoint *mp) 8861da177e4SLinus Torvalds { 88784d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8881064f874SEric W. Biederman __attach_mnt(mnt, parent); 889b90fa9aeSRam Pai } 890b90fa9aeSRam Pai 8911064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 89212a5b529SAl Viro { 8931064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8941064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8951064f874SEric W. Biederman 8961064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8971064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8981064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8991064f874SEric W. Biederman 9001064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 9011064f874SEric W. Biederman 9021064f874SEric W. Biederman put_mountpoint(old_mp); 9031064f874SEric W. Biederman mnt_add_count(old_parent, -1); 90412a5b529SAl Viro } 90512a5b529SAl Viro 906b90fa9aeSRam Pai /* 90799b7db7bSNick Piggin * vfsmount lock must be held for write 908b90fa9aeSRam Pai */ 9091064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 910b90fa9aeSRam Pai { 9110714a533SAl Viro struct mount *parent = mnt->mnt_parent; 91283adc753SAl Viro struct mount *m; 913b90fa9aeSRam Pai LIST_HEAD(head); 914143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 915b90fa9aeSRam Pai 9160714a533SAl Viro BUG_ON(parent == mnt); 917b90fa9aeSRam Pai 9181a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 919f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 920143c8c91SAl Viro m->mnt_ns = n; 921f03c6599SAl Viro 922b90fa9aeSRam Pai list_splice(&head, n->list.prev); 923b90fa9aeSRam Pai 924d2921684SEric W. Biederman n->mounts += n->pending_mounts; 925d2921684SEric W. Biederman n->pending_mounts = 0; 926d2921684SEric W. Biederman 9271064f874SEric W. Biederman __attach_mnt(mnt, parent); 9286b3286edSKirill Korotaev touch_mnt_namespace(n); 9291da177e4SLinus Torvalds } 9301da177e4SLinus Torvalds 931909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9321da177e4SLinus Torvalds { 9336b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9346b41d536SAl Viro if (next == &p->mnt_mounts) { 9351da177e4SLinus Torvalds while (1) { 936909b0a88SAl Viro if (p == root) 9371da177e4SLinus Torvalds return NULL; 9386b41d536SAl Viro next = p->mnt_child.next; 9396b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9401da177e4SLinus Torvalds break; 9410714a533SAl Viro p = p->mnt_parent; 9421da177e4SLinus Torvalds } 9431da177e4SLinus Torvalds } 9446b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9451da177e4SLinus Torvalds } 9461da177e4SLinus Torvalds 947315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9489676f0c6SRam Pai { 9496b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9506b41d536SAl Viro while (prev != &p->mnt_mounts) { 9516b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9526b41d536SAl Viro prev = p->mnt_mounts.prev; 9539676f0c6SRam Pai } 9549676f0c6SRam Pai return p; 9559676f0c6SRam Pai } 9569676f0c6SRam Pai 9578f291889SAl Viro /** 9588f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 9598f291889SAl Viro * @fc: The configuration context with the superblock attached 9608f291889SAl Viro * 9618f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 9628f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 9638f291889SAl Viro * 9648f291889SAl Viro * Note that this does not attach the mount to anything. 9658f291889SAl Viro */ 9668f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 9679d412a43SAl Viro { 968b105e270SAl Viro struct mount *mnt; 969bd303368SChristian Brauner struct user_namespace *fs_userns; 9709d412a43SAl Viro 9718f291889SAl Viro if (!fc->root) 9728f291889SAl Viro return ERR_PTR(-EINVAL); 9739d412a43SAl Viro 9748f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 9759d412a43SAl Viro if (!mnt) 9769d412a43SAl Viro return ERR_PTR(-ENOMEM); 9779d412a43SAl Viro 9788f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 979b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9809d412a43SAl Viro 9818f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 9828f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 9838f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 984a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9850714a533SAl Viro mnt->mnt_parent = mnt; 9868f291889SAl Viro 987bd303368SChristian Brauner fs_userns = mnt->mnt.mnt_sb->s_user_ns; 988bd303368SChristian Brauner if (!initial_idmapping(fs_userns)) 989bd303368SChristian Brauner mnt->mnt.mnt_userns = get_user_ns(fs_userns); 990bd303368SChristian Brauner 991719ea2fbSAl Viro lock_mount_hash(); 9928f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 993719ea2fbSAl Viro unlock_mount_hash(); 994b105e270SAl Viro return &mnt->mnt; 9959d412a43SAl Viro } 9968f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 9978f291889SAl Viro 9988f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 9998f291889SAl Viro { 10008f291889SAl Viro int err = vfs_get_tree(fc); 10018f291889SAl Viro if (!err) { 10028f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 10038f291889SAl Viro return vfs_create_mount(fc); 10048f291889SAl Viro } 10058f291889SAl Viro return ERR_PTR(err); 10068f291889SAl Viro } 10078f291889SAl Viro EXPORT_SYMBOL(fc_mount); 10088f291889SAl Viro 10099bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 10109bc61ab1SDavid Howells int flags, const char *name, 10119bc61ab1SDavid Howells void *data) 10121da177e4SLinus Torvalds { 10139bc61ab1SDavid Howells struct fs_context *fc; 10148f291889SAl Viro struct vfsmount *mnt; 10159bc61ab1SDavid Howells int ret = 0; 10169d412a43SAl Viro 10179d412a43SAl Viro if (!type) 10183e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 10199d412a43SAl Viro 10209bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 10219bc61ab1SDavid Howells if (IS_ERR(fc)) 10229bc61ab1SDavid Howells return ERR_CAST(fc); 10239bc61ab1SDavid Howells 10243e1aeb00SDavid Howells if (name) 10253e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 10263e1aeb00SDavid Howells name, strlen(name)); 10279bc61ab1SDavid Howells if (!ret) 10289bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 10299bc61ab1SDavid Howells if (!ret) 10308f291889SAl Viro mnt = fc_mount(fc); 10318f291889SAl Viro else 10328f291889SAl Viro mnt = ERR_PTR(ret); 10339d412a43SAl Viro 10349bc61ab1SDavid Howells put_fs_context(fc); 10358f291889SAl Viro return mnt; 10369d412a43SAl Viro } 10379d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10389d412a43SAl Viro 103993faccbbSEric W. Biederman struct vfsmount * 104093faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 104193faccbbSEric W. Biederman const char *name, void *data) 104293faccbbSEric W. Biederman { 104393faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 104493faccbbSEric W. Biederman * through from the parent mount to the submount don't support 104593faccbbSEric W. Biederman * unprivileged mounts with submounts. 104693faccbbSEric W. Biederman */ 104793faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 104893faccbbSEric W. Biederman return ERR_PTR(-EPERM); 104993faccbbSEric W. Biederman 1050e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 105193faccbbSEric W. Biederman } 105293faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 105393faccbbSEric W. Biederman 105487129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 105536341f64SRam Pai int flag) 10561da177e4SLinus Torvalds { 105787129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1058be34d1a3SDavid Howells struct mount *mnt; 1059be34d1a3SDavid Howells int err; 10601da177e4SLinus Torvalds 1061be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1062be34d1a3SDavid Howells if (!mnt) 1063be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1064be34d1a3SDavid Howells 10657a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 106615169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1067719f5d7fSMiklos Szeredi else 106815169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1069719f5d7fSMiklos Szeredi 107015169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1071be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1072719f5d7fSMiklos Szeredi if (err) 1073719f5d7fSMiklos Szeredi goto out_free; 1074719f5d7fSMiklos Szeredi } 1075719f5d7fSMiklos Szeredi 107616a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 107716a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 10785ff9d8a6SEric W. Biederman 10791da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1080a6435940SChristian Brauner mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt); 1081bd303368SChristian Brauner if (!initial_idmapping(mnt->mnt.mnt_userns)) 1082a6435940SChristian Brauner mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns); 1083b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1084b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1085a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10860714a533SAl Viro mnt->mnt_parent = mnt; 1087719ea2fbSAl Viro lock_mount_hash(); 108839f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1089719ea2fbSAl Viro unlock_mount_hash(); 1090b90fa9aeSRam Pai 10917a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10927a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10936776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 109432301920SAl Viro mnt->mnt_master = old; 1095fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10968aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1097fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10986776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1099d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 11006776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1101d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 11025235d448SAl Viro } else { 11035235d448SAl Viro CLEAR_MNT_SHARED(mnt); 11045afe0022SRam Pai } 1105b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 11060f0afb1dSAl Viro set_mnt_shared(mnt); 11071da177e4SLinus Torvalds 11081da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 11091da177e4SLinus Torvalds * as the original if that was on one */ 111036341f64SRam Pai if (flag & CL_EXPIRE) { 11116776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 11126776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 11131da177e4SLinus Torvalds } 1114be34d1a3SDavid Howells 1115cb338d06SAl Viro return mnt; 1116719f5d7fSMiklos Szeredi 1117719f5d7fSMiklos Szeredi out_free: 11188ffcb32eSDavid Howells mnt_free_id(mnt); 1119719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1120be34d1a3SDavid Howells return ERR_PTR(err); 11211da177e4SLinus Torvalds } 11221da177e4SLinus Torvalds 11239ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11249ea459e1SAl Viro { 112556cbb429SAl Viro struct hlist_node *p; 112656cbb429SAl Viro struct mount *m; 11279ea459e1SAl Viro /* 112856cbb429SAl Viro * The warning here probably indicates that somebody messed 112956cbb429SAl Viro * up a mnt_want/drop_write() pair. If this happens, the 113056cbb429SAl Viro * filesystem was probably unable to make r/w->r/o transitions. 11319ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11329ea459e1SAl Viro * so mnt_get_writers() below is safe. 11339ea459e1SAl Viro */ 11349ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11359ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11369ea459e1SAl Viro mnt_pin_kill(mnt); 113756cbb429SAl Viro hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { 113856cbb429SAl Viro hlist_del(&m->mnt_umount); 113956cbb429SAl Viro mntput(&m->mnt); 114056cbb429SAl Viro } 11419ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11429ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11439ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11449ea459e1SAl Viro mnt_free_id(mnt); 11459ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11469ea459e1SAl Viro } 11479ea459e1SAl Viro 11489ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11499ea459e1SAl Viro { 11509ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11519ea459e1SAl Viro } 11529ea459e1SAl Viro 11539ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11549ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11559ea459e1SAl Viro { 11569ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 115729785735SByungchul Park struct mount *m, *t; 11589ea459e1SAl Viro 115929785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 116029785735SByungchul Park cleanup_mnt(m); 11619ea459e1SAl Viro } 11629ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11639ea459e1SAl Viro 1164900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11657b7b1aceSAl Viro { 11664edbe133SAl Viro LIST_HEAD(list); 1167edf7ddbfSEric Biggers int count; 11684edbe133SAl Viro 116948a066e7SAl Viro rcu_read_lock(); 11709ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11719ea0a46cSAl Viro /* 11729ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11739ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11749ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11759ea0a46cSAl Viro * be dropped until after an RCU delay done after 11769ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11779ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11789ea0a46cSAl Viro * we are dropping is not the final one. 11799ea0a46cSAl Viro */ 1180aa9c0e07SAl Viro mnt_add_count(mnt, -1); 118148a066e7SAl Viro rcu_read_unlock(); 118299b7db7bSNick Piggin return; 1183b3e19d92SNick Piggin } 1184719ea2fbSAl Viro lock_mount_hash(); 1185119e1ef8SAl Viro /* 1186119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1187119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1188119e1ef8SAl Viro */ 1189119e1ef8SAl Viro smp_mb(); 11909ea0a46cSAl Viro mnt_add_count(mnt, -1); 1191edf7ddbfSEric Biggers count = mnt_get_count(mnt); 1192edf7ddbfSEric Biggers if (count != 0) { 1193edf7ddbfSEric Biggers WARN_ON(count < 0); 119448a066e7SAl Viro rcu_read_unlock(); 1195719ea2fbSAl Viro unlock_mount_hash(); 119699b7db7bSNick Piggin return; 119799b7db7bSNick Piggin } 119848a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 119948a066e7SAl Viro rcu_read_unlock(); 120048a066e7SAl Viro unlock_mount_hash(); 120148a066e7SAl Viro return; 120248a066e7SAl Viro } 120348a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 120448a066e7SAl Viro rcu_read_unlock(); 1205962830dfSAndi Kleen 120639f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1207ce07d891SEric W. Biederman 1208ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1209ce07d891SEric W. Biederman struct mount *p, *tmp; 1210ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 12114edbe133SAl Viro __put_mountpoint(unhash_mnt(p), &list); 121256cbb429SAl Viro hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); 1213ce07d891SEric W. Biederman } 1214ce07d891SEric W. Biederman } 1215719ea2fbSAl Viro unlock_mount_hash(); 12164edbe133SAl Viro shrink_dentry_list(&list); 1217649a795aSAl Viro 12189ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 12199ea459e1SAl Viro struct task_struct *task = current; 12209ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 12219ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 122291989c70SJens Axboe if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) 12239ea459e1SAl Viro return; 12249ea459e1SAl Viro } 12259ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 12269ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 12279ea459e1SAl Viro return; 12289ea459e1SAl Viro } 12299ea459e1SAl Viro cleanup_mnt(mnt); 1230b3e19d92SNick Piggin } 1231b3e19d92SNick Piggin 1232b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1233b3e19d92SNick Piggin { 1234b3e19d92SNick Piggin if (mnt) { 1235863d684fSAl Viro struct mount *m = real_mount(mnt); 1236b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1237863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1238863d684fSAl Viro m->mnt_expiry_mark = 0; 1239863d684fSAl Viro mntput_no_expire(m); 1240b3e19d92SNick Piggin } 1241b3e19d92SNick Piggin } 1242b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1243b3e19d92SNick Piggin 1244b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1245b3e19d92SNick Piggin { 1246b3e19d92SNick Piggin if (mnt) 124783adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1248b3e19d92SNick Piggin return mnt; 1249b3e19d92SNick Piggin } 1250b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1251b3e19d92SNick Piggin 12521f287bc4SRandy Dunlap /** 12531f287bc4SRandy Dunlap * path_is_mountpoint() - Check if path is a mount in the current namespace. 12541f287bc4SRandy Dunlap * @path: path to check 1255c6609c0aSIan Kent * 1256c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1257c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1258c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1259c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1260c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1261c6609c0aSIan Kent * alone. 1262c6609c0aSIan Kent */ 1263c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1264c6609c0aSIan Kent { 1265c6609c0aSIan Kent unsigned seq; 1266c6609c0aSIan Kent bool res; 1267c6609c0aSIan Kent 1268c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1269c6609c0aSIan Kent return false; 1270c6609c0aSIan Kent 1271c6609c0aSIan Kent rcu_read_lock(); 1272c6609c0aSIan Kent do { 1273c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1274c6609c0aSIan Kent res = __path_is_mountpoint(path); 1275c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1276c6609c0aSIan Kent rcu_read_unlock(); 1277c6609c0aSIan Kent 1278c6609c0aSIan Kent return res; 1279c6609c0aSIan Kent } 1280c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1281c6609c0aSIan Kent 1282ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12837b7b1aceSAl Viro { 12843064c356SAl Viro struct mount *p; 12853064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12863064c356SAl Viro if (IS_ERR(p)) 12873064c356SAl Viro return ERR_CAST(p); 12883064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12893064c356SAl Viro return &p->mnt; 12907b7b1aceSAl Viro } 12911da177e4SLinus Torvalds 1292a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12939f6c61f9SMiklos Szeredi static struct mount *mnt_list_next(struct mnt_namespace *ns, 12949f6c61f9SMiklos Szeredi struct list_head *p) 12959f6c61f9SMiklos Szeredi { 12969f6c61f9SMiklos Szeredi struct mount *mnt, *ret = NULL; 12979f6c61f9SMiklos Szeredi 12989f6c61f9SMiklos Szeredi lock_ns_list(ns); 12999f6c61f9SMiklos Szeredi list_for_each_continue(p, &ns->list) { 13009f6c61f9SMiklos Szeredi mnt = list_entry(p, typeof(*mnt), mnt_list); 13019f6c61f9SMiklos Szeredi if (!mnt_is_cursor(mnt)) { 13029f6c61f9SMiklos Szeredi ret = mnt; 13039f6c61f9SMiklos Szeredi break; 13049f6c61f9SMiklos Szeredi } 13059f6c61f9SMiklos Szeredi } 13069f6c61f9SMiklos Szeredi unlock_ns_list(ns); 13079f6c61f9SMiklos Szeredi 13089f6c61f9SMiklos Szeredi return ret; 13099f6c61f9SMiklos Szeredi } 13109f6c61f9SMiklos Szeredi 13110226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 13121da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 13131da177e4SLinus Torvalds { 1314ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13159f6c61f9SMiklos Szeredi struct list_head *prev; 13161da177e4SLinus Torvalds 1317390c6843SRam Pai down_read(&namespace_sem); 13189f6c61f9SMiklos Szeredi if (!*pos) { 13199f6c61f9SMiklos Szeredi prev = &p->ns->list; 13209f6c61f9SMiklos Szeredi } else { 13219f6c61f9SMiklos Szeredi prev = &p->cursor.mnt_list; 13229f6c61f9SMiklos Szeredi 13239f6c61f9SMiklos Szeredi /* Read after we'd reached the end? */ 13249f6c61f9SMiklos Szeredi if (list_empty(prev)) 13259f6c61f9SMiklos Szeredi return NULL; 1326c7999c36SAl Viro } 1327c7999c36SAl Viro 13289f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, prev); 13291da177e4SLinus Torvalds } 13301da177e4SLinus Torvalds 13311da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13321da177e4SLinus Torvalds { 1333ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13349f6c61f9SMiklos Szeredi struct mount *mnt = v; 1335b0765fb8SPavel Emelianov 13369f6c61f9SMiklos Szeredi ++*pos; 13379f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, &mnt->mnt_list); 13381da177e4SLinus Torvalds } 13391da177e4SLinus Torvalds 13401da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13411da177e4SLinus Torvalds { 13429f6c61f9SMiklos Szeredi struct proc_mounts *p = m->private; 13439f6c61f9SMiklos Szeredi struct mount *mnt = v; 13449f6c61f9SMiklos Szeredi 13459f6c61f9SMiklos Szeredi lock_ns_list(p->ns); 13469f6c61f9SMiklos Szeredi if (mnt) 13479f6c61f9SMiklos Szeredi list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); 13489f6c61f9SMiklos Szeredi else 13499f6c61f9SMiklos Szeredi list_del_init(&p->cursor.mnt_list); 13509f6c61f9SMiklos Szeredi unlock_ns_list(p->ns); 1351390c6843SRam Pai up_read(&namespace_sem); 13521da177e4SLinus Torvalds } 13531da177e4SLinus Torvalds 13540226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13559f5596afSAl Viro { 1356ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13579f6c61f9SMiklos Szeredi struct mount *r = v; 13580226f492SAl Viro return p->show(m, &r->mnt); 13591da177e4SLinus Torvalds } 13601da177e4SLinus Torvalds 1361a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13621da177e4SLinus Torvalds .start = m_start, 13631da177e4SLinus Torvalds .next = m_next, 13641da177e4SLinus Torvalds .stop = m_stop, 13650226f492SAl Viro .show = m_show, 1366b4629fe2SChuck Lever }; 13679f6c61f9SMiklos Szeredi 13689f6c61f9SMiklos Szeredi void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) 13699f6c61f9SMiklos Szeredi { 13709f6c61f9SMiklos Szeredi down_read(&namespace_sem); 13719f6c61f9SMiklos Szeredi lock_ns_list(ns); 13729f6c61f9SMiklos Szeredi list_del(&cursor->mnt_list); 13739f6c61f9SMiklos Szeredi unlock_ns_list(ns); 13749f6c61f9SMiklos Szeredi up_read(&namespace_sem); 13759f6c61f9SMiklos Szeredi } 1376a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1377b4629fe2SChuck Lever 13781da177e4SLinus Torvalds /** 13791da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 13801f287bc4SRandy Dunlap * @m: root of mount tree 13811da177e4SLinus Torvalds * 13821da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 13831da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 13841da177e4SLinus Torvalds * busy. 13851da177e4SLinus Torvalds */ 1386909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13871da177e4SLinus Torvalds { 1388909b0a88SAl Viro struct mount *mnt = real_mount(m); 138936341f64SRam Pai int actual_refs = 0; 139036341f64SRam Pai int minimum_refs = 0; 1391315fc83eSAl Viro struct mount *p; 1392909b0a88SAl Viro BUG_ON(!m); 13931da177e4SLinus Torvalds 1394b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1395719ea2fbSAl Viro lock_mount_hash(); 1396909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 139783adc753SAl Viro actual_refs += mnt_get_count(p); 13981da177e4SLinus Torvalds minimum_refs += 2; 13991da177e4SLinus Torvalds } 1400719ea2fbSAl Viro unlock_mount_hash(); 14011da177e4SLinus Torvalds 14021da177e4SLinus Torvalds if (actual_refs > minimum_refs) 14031da177e4SLinus Torvalds return 0; 1404e3474a8eSIan Kent 1405e3474a8eSIan Kent return 1; 14061da177e4SLinus Torvalds } 14071da177e4SLinus Torvalds 14081da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 14091da177e4SLinus Torvalds 14101da177e4SLinus Torvalds /** 14111da177e4SLinus Torvalds * may_umount - check if a mount point is busy 14121da177e4SLinus Torvalds * @mnt: root of mount 14131da177e4SLinus Torvalds * 14141da177e4SLinus Torvalds * This is called to check if a mount point has any 14151da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 14161da177e4SLinus Torvalds * mount has sub mounts this will return busy 14171da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 14181da177e4SLinus Torvalds * 14191da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 14201da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 14211da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 14221da177e4SLinus Torvalds */ 14231da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 14241da177e4SLinus Torvalds { 1425e3474a8eSIan Kent int ret = 1; 14268ad08d8aSAl Viro down_read(&namespace_sem); 1427719ea2fbSAl Viro lock_mount_hash(); 14281ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1429e3474a8eSIan Kent ret = 0; 1430719ea2fbSAl Viro unlock_mount_hash(); 14318ad08d8aSAl Viro up_read(&namespace_sem); 1432a05964f3SRam Pai return ret; 14331da177e4SLinus Torvalds } 14341da177e4SLinus Torvalds 14351da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14361da177e4SLinus Torvalds 143797216be0SAl Viro static void namespace_unlock(void) 14381da177e4SLinus Torvalds { 1439a3b3c562SEric W. Biederman struct hlist_head head; 144056cbb429SAl Viro struct hlist_node *p; 144156cbb429SAl Viro struct mount *m; 14424edbe133SAl Viro LIST_HEAD(list); 144397216be0SAl Viro 1444a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 14454edbe133SAl Viro list_splice_init(&ex_mountpoints, &list); 1446a3b3c562SEric W. Biederman 144797216be0SAl Viro up_write(&namespace_sem); 1448a3b3c562SEric W. Biederman 14494edbe133SAl Viro shrink_dentry_list(&list); 14504edbe133SAl Viro 1451a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 145297216be0SAl Viro return; 145397216be0SAl Viro 145422cb7405SNeilBrown synchronize_rcu_expedited(); 145548a066e7SAl Viro 145656cbb429SAl Viro hlist_for_each_entry_safe(m, p, &head, mnt_umount) { 145756cbb429SAl Viro hlist_del(&m->mnt_umount); 145856cbb429SAl Viro mntput(&m->mnt); 145956cbb429SAl Viro } 146070fbcdf4SRam Pai } 146170fbcdf4SRam Pai 146297216be0SAl Viro static inline void namespace_lock(void) 1463e3197d83SAl Viro { 146497216be0SAl Viro down_write(&namespace_sem); 1465e3197d83SAl Viro } 1466e3197d83SAl Viro 1467e819f152SEric W. Biederman enum umount_tree_flags { 1468e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1469e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1470e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1471e819f152SEric W. Biederman }; 1472f2d0a123SEric W. Biederman 1473f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1474f2d0a123SEric W. Biederman { 1475f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1476f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1477f2d0a123SEric W. Biederman return true; 1478f2d0a123SEric W. Biederman 1479f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1480f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1481f2d0a123SEric W. Biederman return true; 1482f2d0a123SEric W. Biederman 1483f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1484f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1485f2d0a123SEric W. Biederman * connected to mounted mounts. 1486f2d0a123SEric W. Biederman */ 1487f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1488f2d0a123SEric W. Biederman return true; 1489f2d0a123SEric W. Biederman 1490f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1491f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1492f2d0a123SEric W. Biederman return false; 1493f2d0a123SEric W. Biederman 1494f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1495f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1496f2d0a123SEric W. Biederman return false; 1497f2d0a123SEric W. Biederman 1498f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1499f2d0a123SEric W. Biederman return true; 1500f2d0a123SEric W. Biederman } 1501f2d0a123SEric W. Biederman 150299b7db7bSNick Piggin /* 150348a066e7SAl Viro * mount_lock must be held 150499b7db7bSNick Piggin * namespace_sem must be held for write 150599b7db7bSNick Piggin */ 1506e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 150770fbcdf4SRam Pai { 1508c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1509315fc83eSAl Viro struct mount *p; 151070fbcdf4SRam Pai 15115d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 15125d88457eSEric W. Biederman propagate_mount_unlock(mnt); 15135d88457eSEric W. Biederman 1514c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1515590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1516590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1517c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1518590ce4bcSEric W. Biederman } 1519c003b26fSEric W. Biederman 1520411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1521c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1522c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 152338129a13SAl Viro } 152470fbcdf4SRam Pai 1525c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1526e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 15277b8a53fdSAl Viro propagate_umount(&tmp_list); 1528a05964f3SRam Pai 1529c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1530d2921684SEric W. Biederman struct mnt_namespace *ns; 1531ce07d891SEric W. Biederman bool disconnect; 1532c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 15336776db3dSAl Viro list_del_init(&p->mnt_expire); 15341a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1535d2921684SEric W. Biederman ns = p->mnt_ns; 1536d2921684SEric W. Biederman if (ns) { 1537d2921684SEric W. Biederman ns->mounts--; 1538d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1539d2921684SEric W. Biederman } 154063d37a84SAl Viro p->mnt_ns = NULL; 1541e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 154248a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 154387b95ce0SAl Viro 1544f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1545676da58dSAl Viro if (mnt_has_parent(p)) { 154681b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1547ce07d891SEric W. Biederman if (!disconnect) { 1548ce07d891SEric W. Biederman /* Don't forget about p */ 1549ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1550ce07d891SEric W. Biederman } else { 15516a46c573SEric W. Biederman umount_mnt(p); 15527c4b93d8SAl Viro } 1553ce07d891SEric W. Biederman } 15540f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 155519a1c409SAl Viro if (disconnect) 155619a1c409SAl Viro hlist_add_head(&p->mnt_umount, &unmounted); 155738129a13SAl Viro } 15581da177e4SLinus Torvalds } 15591da177e4SLinus Torvalds 1560b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1561c35038beSAl Viro 15628d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 15638d0347f6SDavid Howells { 15648d0347f6SDavid Howells int ret = 0; 15658d0347f6SDavid Howells 15668d0347f6SDavid Howells down_write(&sb->s_umount); 15678d0347f6SDavid Howells if (!sb_rdonly(sb)) { 15688d0347f6SDavid Howells struct fs_context *fc; 15698d0347f6SDavid Howells 15708d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 15718d0347f6SDavid Howells SB_RDONLY); 15728d0347f6SDavid Howells if (IS_ERR(fc)) { 15738d0347f6SDavid Howells ret = PTR_ERR(fc); 15748d0347f6SDavid Howells } else { 15758d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 15768d0347f6SDavid Howells if (!ret) 15778d0347f6SDavid Howells ret = reconfigure_super(fc); 15788d0347f6SDavid Howells put_fs_context(fc); 15798d0347f6SDavid Howells } 15808d0347f6SDavid Howells } 15818d0347f6SDavid Howells up_write(&sb->s_umount); 15828d0347f6SDavid Howells return ret; 15838d0347f6SDavid Howells } 15848d0347f6SDavid Howells 15851ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 15861da177e4SLinus Torvalds { 15871ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 15881da177e4SLinus Torvalds int retval; 15891da177e4SLinus Torvalds 15901ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 15911da177e4SLinus Torvalds if (retval) 15921da177e4SLinus Torvalds return retval; 15931da177e4SLinus Torvalds 15941da177e4SLinus Torvalds /* 15951da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 15961da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 15971da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 15981da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 15991da177e4SLinus Torvalds */ 16001da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 16011ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 16021da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 16031da177e4SLinus Torvalds return -EINVAL; 16041da177e4SLinus Torvalds 1605b3e19d92SNick Piggin /* 1606b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1607b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1608b3e19d92SNick Piggin */ 1609719ea2fbSAl Viro lock_mount_hash(); 161083adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1611719ea2fbSAl Viro unlock_mount_hash(); 16121da177e4SLinus Torvalds return -EBUSY; 1613b3e19d92SNick Piggin } 1614719ea2fbSAl Viro unlock_mount_hash(); 16151da177e4SLinus Torvalds 1616863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 16171da177e4SLinus Torvalds return -EAGAIN; 16181da177e4SLinus Torvalds } 16191da177e4SLinus Torvalds 16201da177e4SLinus Torvalds /* 16211da177e4SLinus Torvalds * If we may have to abort operations to get out of this 16221da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 16231da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 16241da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 16251da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 16261da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 16271da177e4SLinus Torvalds * about for the moment. 16281da177e4SLinus Torvalds */ 16291da177e4SLinus Torvalds 163042faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 163142faad99SAl Viro sb->s_op->umount_begin(sb); 163242faad99SAl Viro } 16331da177e4SLinus Torvalds 16341da177e4SLinus Torvalds /* 16351da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 16361da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 16371da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 16381da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 16391da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 16401da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 16411da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 16421da177e4SLinus Torvalds */ 16431ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 16441da177e4SLinus Torvalds /* 16451da177e4SLinus Torvalds * Special case for "unmounting" root ... 16461da177e4SLinus Torvalds * we just try to remount it readonly. 16471da177e4SLinus Torvalds */ 1648bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1649a1480dccSAndy Lutomirski return -EPERM; 16508d0347f6SDavid Howells return do_umount_root(sb); 16511da177e4SLinus Torvalds } 16521da177e4SLinus Torvalds 165397216be0SAl Viro namespace_lock(); 1654719ea2fbSAl Viro lock_mount_hash(); 16551da177e4SLinus Torvalds 165625d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 165725d202edSEric W. Biederman retval = -EINVAL; 165825d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 165925d202edSEric W. Biederman goto out; 166025d202edSEric W. Biederman 166125d202edSEric W. Biederman event++; 166248a066e7SAl Viro if (flags & MNT_DETACH) { 166348a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1664e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 166548a066e7SAl Viro retval = 0; 166648a066e7SAl Viro } else { 1667b54b9be7SAl Viro shrink_submounts(mnt); 16681da177e4SLinus Torvalds retval = -EBUSY; 166948a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 16701a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1671e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 16721da177e4SLinus Torvalds retval = 0; 16731da177e4SLinus Torvalds } 167448a066e7SAl Viro } 167525d202edSEric W. Biederman out: 1676719ea2fbSAl Viro unlock_mount_hash(); 1677e3197d83SAl Viro namespace_unlock(); 16781da177e4SLinus Torvalds return retval; 16791da177e4SLinus Torvalds } 16801da177e4SLinus Torvalds 16811da177e4SLinus Torvalds /* 168280b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 168380b5dce8SEric W. Biederman * 168480b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 168580b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 168680b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 168780b5dce8SEric W. Biederman * leaking them. 168880b5dce8SEric W. Biederman * 168980b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 169080b5dce8SEric W. Biederman */ 169180b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 169280b5dce8SEric W. Biederman { 169380b5dce8SEric W. Biederman struct mountpoint *mp; 169480b5dce8SEric W. Biederman struct mount *mnt; 169580b5dce8SEric W. Biederman 169680b5dce8SEric W. Biederman namespace_lock(); 16973895dbf8SEric W. Biederman lock_mount_hash(); 169880b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1699adc9b5c0SAl Viro if (!mp) 170080b5dce8SEric W. Biederman goto out_unlock; 170180b5dce8SEric W. Biederman 1702e06b933eSAndrey Ulanov event++; 170380b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 170480b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1705ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1706fe78fcc8SEric W. Biederman umount_mnt(mnt); 170756cbb429SAl Viro hlist_add_head(&mnt->mnt_umount, &unmounted); 1708ce07d891SEric W. Biederman } 1709e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 171080b5dce8SEric W. Biederman } 171180b5dce8SEric W. Biederman put_mountpoint(mp); 171280b5dce8SEric W. Biederman out_unlock: 17133895dbf8SEric W. Biederman unlock_mount_hash(); 171480b5dce8SEric W. Biederman namespace_unlock(); 171580b5dce8SEric W. Biederman } 171680b5dce8SEric W. Biederman 171780b5dce8SEric W. Biederman /* 17189b40bc90SAl Viro * Is the caller allowed to modify his namespace? 17199b40bc90SAl Viro */ 17209b40bc90SAl Viro static inline bool may_mount(void) 17219b40bc90SAl Viro { 17229b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 17239b40bc90SAl Viro } 17249b40bc90SAl Viro 1725f7e33bdbSJeff Layton static void warn_mandlock(void) 17269e8925b6SJeff Layton { 1727f7e33bdbSJeff Layton pr_warn_once("=======================================================\n" 1728f7e33bdbSJeff Layton "WARNING: The mand mount option has been deprecated and\n" 1729f7e33bdbSJeff Layton " and is ignored by this kernel. Remove the mand\n" 1730f7e33bdbSJeff Layton " option from the mount to silence this warning.\n" 1731f7e33bdbSJeff Layton "=======================================================\n"); 17329e8925b6SJeff Layton } 17339e8925b6SJeff Layton 173425ccd24fSChristoph Hellwig static int can_umount(const struct path *path, int flags) 17351da177e4SLinus Torvalds { 173625ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 17371da177e4SLinus Torvalds 17389b40bc90SAl Viro if (!may_mount()) 17399b40bc90SAl Viro return -EPERM; 174041525f56SChristoph Hellwig if (path->dentry != path->mnt->mnt_root) 174125ccd24fSChristoph Hellwig return -EINVAL; 1742143c8c91SAl Viro if (!check_mnt(mnt)) 174325ccd24fSChristoph Hellwig return -EINVAL; 174425d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 174525ccd24fSChristoph Hellwig return -EINVAL; 1746b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 174725ccd24fSChristoph Hellwig return -EPERM; 174825ccd24fSChristoph Hellwig return 0; 174925ccd24fSChristoph Hellwig } 17501da177e4SLinus Torvalds 1751a0a6df9aSAl Viro // caller is responsible for flags being sane 175225ccd24fSChristoph Hellwig int path_umount(struct path *path, int flags) 175325ccd24fSChristoph Hellwig { 175425ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 175525ccd24fSChristoph Hellwig int ret; 175625ccd24fSChristoph Hellwig 175725ccd24fSChristoph Hellwig ret = can_umount(path, flags); 175825ccd24fSChristoph Hellwig if (!ret) 175925ccd24fSChristoph Hellwig ret = do_umount(mnt, flags); 176025ccd24fSChristoph Hellwig 1761429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 176241525f56SChristoph Hellwig dput(path->dentry); 1763900148dcSAl Viro mntput_no_expire(mnt); 176425ccd24fSChristoph Hellwig return ret; 17651da177e4SLinus Torvalds } 17661da177e4SLinus Torvalds 176709267defSChristoph Hellwig static int ksys_umount(char __user *name, int flags) 176841525f56SChristoph Hellwig { 176941525f56SChristoph Hellwig int lookup_flags = LOOKUP_MOUNTPOINT; 177041525f56SChristoph Hellwig struct path path; 177141525f56SChristoph Hellwig int ret; 177241525f56SChristoph Hellwig 1773a0a6df9aSAl Viro // basic validity checks done first 1774a0a6df9aSAl Viro if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1775a0a6df9aSAl Viro return -EINVAL; 1776a0a6df9aSAl Viro 177741525f56SChristoph Hellwig if (!(flags & UMOUNT_NOFOLLOW)) 177841525f56SChristoph Hellwig lookup_flags |= LOOKUP_FOLLOW; 177941525f56SChristoph Hellwig ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); 178041525f56SChristoph Hellwig if (ret) 178141525f56SChristoph Hellwig return ret; 178241525f56SChristoph Hellwig return path_umount(&path, flags); 178341525f56SChristoph Hellwig } 178441525f56SChristoph Hellwig 17853a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 17863a18ef5cSDominik Brodowski { 17873a18ef5cSDominik Brodowski return ksys_umount(name, flags); 17883a18ef5cSDominik Brodowski } 17893a18ef5cSDominik Brodowski 17901da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 17911da177e4SLinus Torvalds 17921da177e4SLinus Torvalds /* 17931da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17941da177e4SLinus Torvalds */ 1795bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17961da177e4SLinus Torvalds { 17973a18ef5cSDominik Brodowski return ksys_umount(name, 0); 17981da177e4SLinus Torvalds } 17991da177e4SLinus Torvalds 18001da177e4SLinus Torvalds #endif 18011da177e4SLinus Torvalds 18024ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 18038823c079SEric W. Biederman { 18044ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1805e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1806e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 18074ce5d2b1SEric W. Biederman } 18084ce5d2b1SEric W. Biederman 1809213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 181058be2825SAl Viro { 181158be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 181258be2825SAl Viro } 181358be2825SAl Viro 1814303cc571SChristian Brauner struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) 1815303cc571SChristian Brauner { 1816303cc571SChristian Brauner return &mnt->ns; 1817303cc571SChristian Brauner } 1818303cc571SChristian Brauner 18194ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 18204ce5d2b1SEric W. Biederman { 18214ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 18224ce5d2b1SEric W. Biederman * mount namespace loop? 18234ce5d2b1SEric W. Biederman */ 18244ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 18254ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 18264ce5d2b1SEric W. Biederman return false; 18274ce5d2b1SEric W. Biederman 1828f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 18298823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 18308823c079SEric W. Biederman } 18318823c079SEric W. Biederman 183287129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 183336341f64SRam Pai int flag) 18341da177e4SLinus Torvalds { 183584d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 18361da177e4SLinus Torvalds 18374ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 18384ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 18394ce5d2b1SEric W. Biederman 18404ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1841be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 18429676f0c6SRam Pai 184336341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1844be34d1a3SDavid Howells if (IS_ERR(q)) 1845be34d1a3SDavid Howells return q; 1846be34d1a3SDavid Howells 1847a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 18481da177e4SLinus Torvalds 18491da177e4SLinus Torvalds p = mnt; 18506b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1851315fc83eSAl Viro struct mount *s; 18527ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 18531da177e4SLinus Torvalds continue; 18541da177e4SLinus Torvalds 1855909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 18564ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 18574ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1858df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1859df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1860df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1861df7342b2SEric W. Biederman goto out; 1862df7342b2SEric W. Biederman } else { 18634ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 18644ce5d2b1SEric W. Biederman continue; 18654ce5d2b1SEric W. Biederman } 1866df7342b2SEric W. Biederman } 18674ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 18684ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 18699676f0c6SRam Pai s = skip_mnt_tree(s); 18709676f0c6SRam Pai continue; 18719676f0c6SRam Pai } 18720714a533SAl Viro while (p != s->mnt_parent) { 18730714a533SAl Viro p = p->mnt_parent; 18740714a533SAl Viro q = q->mnt_parent; 18751da177e4SLinus Torvalds } 187687129cc0SAl Viro p = s; 187784d17192SAl Viro parent = q; 187887129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1879be34d1a3SDavid Howells if (IS_ERR(q)) 1880be34d1a3SDavid Howells goto out; 1881719ea2fbSAl Viro lock_mount_hash(); 18821a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 18831064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1884719ea2fbSAl Viro unlock_mount_hash(); 18851da177e4SLinus Torvalds } 18861da177e4SLinus Torvalds } 18871da177e4SLinus Torvalds return res; 1888be34d1a3SDavid Howells out: 18891da177e4SLinus Torvalds if (res) { 1890719ea2fbSAl Viro lock_mount_hash(); 1891e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1892719ea2fbSAl Viro unlock_mount_hash(); 18931da177e4SLinus Torvalds } 1894be34d1a3SDavid Howells return q; 18951da177e4SLinus Torvalds } 18961da177e4SLinus Torvalds 1897be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1898be34d1a3SDavid Howells 1899ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 19008aec0809SAl Viro { 1901cb338d06SAl Viro struct mount *tree; 190297216be0SAl Viro namespace_lock(); 1903cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1904cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1905cd4a4017SEric W. Biederman else 190687129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 190787129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1908328e6d90SAl Viro namespace_unlock(); 1909be34d1a3SDavid Howells if (IS_ERR(tree)) 191052e220d3SDan Carpenter return ERR_CAST(tree); 1911be34d1a3SDavid Howells return &tree->mnt; 19128aec0809SAl Viro } 19138aec0809SAl Viro 1914a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1915a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1916a07b2000SAl Viro 1917a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1918a07b2000SAl Viro { 1919a07b2000SAl Viro struct mnt_namespace *ns; 1920a07b2000SAl Viro namespace_lock(); 1921a07b2000SAl Viro lock_mount_hash(); 1922a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 192344dfd84aSDavid Howells if (ns) { 192444dfd84aSDavid Howells if (is_anon_ns(ns)) 1925a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 192644dfd84aSDavid Howells else 192744dfd84aSDavid Howells ns = NULL; 192844dfd84aSDavid Howells } 1929a07b2000SAl Viro unlock_mount_hash(); 1930a07b2000SAl Viro namespace_unlock(); 193144dfd84aSDavid Howells if (ns) 1932a07b2000SAl Viro free_mnt_ns(ns); 1933a07b2000SAl Viro } 1934a07b2000SAl Viro 19358aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 19368aec0809SAl Viro { 193797216be0SAl Viro namespace_lock(); 1938719ea2fbSAl Viro lock_mount_hash(); 19399c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1940719ea2fbSAl Viro unlock_mount_hash(); 19413ab6abeeSAl Viro namespace_unlock(); 19428aec0809SAl Viro } 19438aec0809SAl Viro 1944427215d8SMiklos Szeredi static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 1945427215d8SMiklos Szeredi { 1946427215d8SMiklos Szeredi struct mount *child; 1947427215d8SMiklos Szeredi 1948427215d8SMiklos Szeredi list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 1949427215d8SMiklos Szeredi if (!is_subdir(child->mnt_mountpoint, dentry)) 1950427215d8SMiklos Szeredi continue; 1951427215d8SMiklos Szeredi 1952427215d8SMiklos Szeredi if (child->mnt.mnt_flags & MNT_LOCKED) 1953427215d8SMiklos Szeredi return true; 1954427215d8SMiklos Szeredi } 1955427215d8SMiklos Szeredi return false; 1956427215d8SMiklos Szeredi } 1957427215d8SMiklos Szeredi 1958c771d683SMiklos Szeredi /** 1959c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 19601f287bc4SRandy Dunlap * @path: path to clone 1961c771d683SMiklos Szeredi * 19621f287bc4SRandy Dunlap * This creates a new vfsmount, which will be the clone of @path. The new mount 19631f287bc4SRandy Dunlap * will not be attached anywhere in the namespace and will be private (i.e. 19641f287bc4SRandy Dunlap * changes to the originating mount won't be propagated into this). 1965c771d683SMiklos Szeredi * 1966c771d683SMiklos Szeredi * Release with mntput(). 1967c771d683SMiklos Szeredi */ 1968ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1969c771d683SMiklos Szeredi { 1970c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1971c771d683SMiklos Szeredi struct mount *new_mnt; 1972c771d683SMiklos Szeredi 1973427215d8SMiklos Szeredi down_read(&namespace_sem); 1974c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1975427215d8SMiklos Szeredi goto invalid; 1976427215d8SMiklos Szeredi 1977427215d8SMiklos Szeredi if (!check_mnt(old_mnt)) 1978427215d8SMiklos Szeredi goto invalid; 1979427215d8SMiklos Szeredi 1980427215d8SMiklos Szeredi if (has_locked_children(old_mnt, path->dentry)) 1981427215d8SMiklos Szeredi goto invalid; 1982c771d683SMiklos Szeredi 1983c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1984427215d8SMiklos Szeredi up_read(&namespace_sem); 1985427215d8SMiklos Szeredi 1986c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1987c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1988c771d683SMiklos Szeredi 1989df820f8dSMiklos Szeredi /* Longterm mount to be removed by kern_unmount*() */ 1990df820f8dSMiklos Szeredi new_mnt->mnt_ns = MNT_NS_INTERNAL; 1991df820f8dSMiklos Szeredi 1992c771d683SMiklos Szeredi return &new_mnt->mnt; 1993427215d8SMiklos Szeredi 1994427215d8SMiklos Szeredi invalid: 1995427215d8SMiklos Szeredi up_read(&namespace_sem); 1996427215d8SMiklos Szeredi return ERR_PTR(-EINVAL); 1997c771d683SMiklos Szeredi } 1998c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1999c771d683SMiklos Szeredi 20001f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 20011f707137SAl Viro struct vfsmount *root) 20021f707137SAl Viro { 20031a4eeaf2SAl Viro struct mount *mnt; 20041f707137SAl Viro int res = f(root, arg); 20051f707137SAl Viro if (res) 20061f707137SAl Viro return res; 20071a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 20081a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 20091f707137SAl Viro if (res) 20101f707137SAl Viro return res; 20111f707137SAl Viro } 20121f707137SAl Viro return 0; 20131f707137SAl Viro } 20141f707137SAl Viro 20153bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 20163bd045ccSAl Viro { 20173bd045ccSAl Viro struct mount *p; 20183bd045ccSAl Viro 20193bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 20203bd045ccSAl Viro int flags = p->mnt.mnt_flags; 20213bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 20223bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 20233bd045ccSAl Viro 20243bd045ccSAl Viro if (flags & MNT_READONLY) 20253bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 20263bd045ccSAl Viro 20273bd045ccSAl Viro if (flags & MNT_NODEV) 20283bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 20293bd045ccSAl Viro 20303bd045ccSAl Viro if (flags & MNT_NOSUID) 20313bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 20323bd045ccSAl Viro 20333bd045ccSAl Viro if (flags & MNT_NOEXEC) 20343bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 20353bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 20363bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 20373bd045ccSAl Viro flags |= MNT_LOCKED; 20383bd045ccSAl Viro p->mnt.mnt_flags = flags; 20393bd045ccSAl Viro } 20403bd045ccSAl Viro } 20413bd045ccSAl Viro 20424b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 2043719f5d7fSMiklos Szeredi { 2044315fc83eSAl Viro struct mount *p; 2045719f5d7fSMiklos Szeredi 2046909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 2047fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 20484b8b21f4SAl Viro mnt_release_group_id(p); 2049719f5d7fSMiklos Szeredi } 2050719f5d7fSMiklos Szeredi } 2051719f5d7fSMiklos Szeredi 20524b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 2053719f5d7fSMiklos Szeredi { 2054315fc83eSAl Viro struct mount *p; 2055719f5d7fSMiklos Szeredi 2056909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 2057fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 20584b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 2059719f5d7fSMiklos Szeredi if (err) { 20604b8b21f4SAl Viro cleanup_group_ids(mnt, p); 2061719f5d7fSMiklos Szeredi return err; 2062719f5d7fSMiklos Szeredi } 2063719f5d7fSMiklos Szeredi } 2064719f5d7fSMiklos Szeredi } 2065719f5d7fSMiklos Szeredi 2066719f5d7fSMiklos Szeredi return 0; 2067719f5d7fSMiklos Szeredi } 2068719f5d7fSMiklos Szeredi 2069d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 2070d2921684SEric W. Biederman { 2071d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 2072d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 2073d2921684SEric W. Biederman struct mount *p; 2074d2921684SEric W. Biederman 2075d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 2076d2921684SEric W. Biederman mounts++; 2077d2921684SEric W. Biederman 2078d2921684SEric W. Biederman old = ns->mounts; 2079d2921684SEric W. Biederman pending = ns->pending_mounts; 2080d2921684SEric W. Biederman sum = old + pending; 2081d2921684SEric W. Biederman if ((old > sum) || 2082d2921684SEric W. Biederman (pending > sum) || 2083d2921684SEric W. Biederman (max < sum) || 2084d2921684SEric W. Biederman (mounts > (max - sum))) 2085d2921684SEric W. Biederman return -ENOSPC; 2086d2921684SEric W. Biederman 2087d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 2088d2921684SEric W. Biederman return 0; 2089d2921684SEric W. Biederman } 2090d2921684SEric W. Biederman 2091b90fa9aeSRam Pai /* 2092b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 2093b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 209421444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 209521444403SRam Pai * store the parent mount and mountpoint dentry. 209621444403SRam Pai * (done when source_mnt is moved) 2097b90fa9aeSRam Pai * 2098b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 2099b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 21009676f0c6SRam Pai * --------------------------------------------------------------------------- 2101b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 21029676f0c6SRam Pai * |************************************************************************** 21039676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21049676f0c6SRam Pai * | dest | | | | | 21059676f0c6SRam Pai * | | | | | | | 21069676f0c6SRam Pai * | v | | | | | 21079676f0c6SRam Pai * |************************************************************************** 21089676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 21095afe0022SRam Pai * | | | | | | 21109676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 21119676f0c6SRam Pai * *************************************************************************** 2112b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 2113b90fa9aeSRam Pai * destination mount. 2114b90fa9aeSRam Pai * 2115b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2116b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2117b90fa9aeSRam Pai * the peer group of the source mount. 2118b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2119b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2120b90fa9aeSRam Pai * mount. 21215afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 21225afe0022SRam Pai * of the destination mount and the cloned mount is made slave 21235afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 21245afe0022SRam Pai * is marked as 'shared and slave'. 21255afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 21265afe0022SRam Pai * source mount. 21275afe0022SRam Pai * 21289676f0c6SRam Pai * --------------------------------------------------------------------------- 212921444403SRam Pai * | MOVE MOUNT OPERATION | 21309676f0c6SRam Pai * |************************************************************************** 21319676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21329676f0c6SRam Pai * | dest | | | | | 21339676f0c6SRam Pai * | | | | | | | 21349676f0c6SRam Pai * | v | | | | | 21359676f0c6SRam Pai * |************************************************************************** 21369676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 21375afe0022SRam Pai * | | | | | | 21389676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 21399676f0c6SRam Pai * *************************************************************************** 21405afe0022SRam Pai * 21415afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 21425afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 214321444403SRam Pai * (+*) the mount is moved to the destination. 21445afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 21455afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 21465afe0022SRam Pai * the mount is marked as 'shared and slave'. 21475afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2148b90fa9aeSRam Pai * 2149b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2150b90fa9aeSRam Pai * applied to each mount in the tree. 2151b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2152b90fa9aeSRam Pai * in allocations. 2153b90fa9aeSRam Pai */ 21540fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 215584d17192SAl Viro struct mount *dest_mnt, 215684d17192SAl Viro struct mountpoint *dest_mp, 21572763d119SAl Viro bool moving) 2158b90fa9aeSRam Pai { 21593bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 216038129a13SAl Viro HLIST_HEAD(tree_list); 2161d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 21621064f874SEric W. Biederman struct mountpoint *smp; 2163315fc83eSAl Viro struct mount *child, *p; 216438129a13SAl Viro struct hlist_node *n; 2165719f5d7fSMiklos Szeredi int err; 2166b90fa9aeSRam Pai 21671064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 21681064f874SEric W. Biederman * to be tucked under other mounts. 21691064f874SEric W. Biederman */ 21701064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 21711064f874SEric W. Biederman if (IS_ERR(smp)) 21721064f874SEric W. Biederman return PTR_ERR(smp); 21731064f874SEric W. Biederman 2174d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 21752763d119SAl Viro if (!moving) { 2176d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2177d2921684SEric W. Biederman if (err) 2178d2921684SEric W. Biederman goto out; 2179d2921684SEric W. Biederman } 2180d2921684SEric W. Biederman 2181fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 21820fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2183719f5d7fSMiklos Szeredi if (err) 2184719f5d7fSMiklos Szeredi goto out; 218584d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2186f2ebb3a9SAl Viro lock_mount_hash(); 2187719f5d7fSMiklos Szeredi if (err) 2188719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2189909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 21900f0afb1dSAl Viro set_mnt_shared(p); 21910b1b901bSAl Viro } else { 21920b1b901bSAl Viro lock_mount_hash(); 2193b90fa9aeSRam Pai } 21942763d119SAl Viro if (moving) { 21952763d119SAl Viro unhash_mnt(source_mnt); 219684d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2197143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 219821444403SRam Pai } else { 219944dfd84aSDavid Howells if (source_mnt->mnt_ns) { 220044dfd84aSDavid Howells /* move from anon - the caller will destroy */ 220144dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 220244dfd84aSDavid Howells } 220384d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 22041064f874SEric W. Biederman commit_tree(source_mnt); 220521444403SRam Pai } 2206b90fa9aeSRam Pai 220738129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 22081d6a32acSAl Viro struct mount *q; 220938129a13SAl Viro hlist_del_init(&child->mnt_hash); 22101064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 22111d6a32acSAl Viro child->mnt_mountpoint); 22121064f874SEric W. Biederman if (q) 22131064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 22143bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 22153bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 22163bd045ccSAl Viro lock_mnt_tree(child); 2217d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 22181064f874SEric W. Biederman commit_tree(child); 2219b90fa9aeSRam Pai } 22201064f874SEric W. Biederman put_mountpoint(smp); 2221719ea2fbSAl Viro unlock_mount_hash(); 222299b7db7bSNick Piggin 2223b90fa9aeSRam Pai return 0; 2224719f5d7fSMiklos Szeredi 2225719f5d7fSMiklos Szeredi out_cleanup_ids: 2226f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2227f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2228d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2229e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2230f2ebb3a9SAl Viro } 2231f2ebb3a9SAl Viro unlock_mount_hash(); 22320fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2233719f5d7fSMiklos Szeredi out: 2234d2921684SEric W. Biederman ns->pending_mounts = 0; 22351064f874SEric W. Biederman 22361064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 22371064f874SEric W. Biederman put_mountpoint(smp); 22381064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 22391064f874SEric W. Biederman 2240719f5d7fSMiklos Szeredi return err; 2241b90fa9aeSRam Pai } 2242b90fa9aeSRam Pai 224384d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2244b12cea91SAl Viro { 2245b12cea91SAl Viro struct vfsmount *mnt; 224684d17192SAl Viro struct dentry *dentry = path->dentry; 2247b12cea91SAl Viro retry: 22485955102cSAl Viro inode_lock(dentry->d_inode); 224984d17192SAl Viro if (unlikely(cant_mount(dentry))) { 22505955102cSAl Viro inode_unlock(dentry->d_inode); 225184d17192SAl Viro return ERR_PTR(-ENOENT); 2252b12cea91SAl Viro } 225397216be0SAl Viro namespace_lock(); 2254b12cea91SAl Viro mnt = lookup_mnt(path); 225584d17192SAl Viro if (likely(!mnt)) { 22563895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 225784d17192SAl Viro if (IS_ERR(mp)) { 225897216be0SAl Viro namespace_unlock(); 22595955102cSAl Viro inode_unlock(dentry->d_inode); 226084d17192SAl Viro return mp; 226184d17192SAl Viro } 226284d17192SAl Viro return mp; 226384d17192SAl Viro } 226497216be0SAl Viro namespace_unlock(); 22655955102cSAl Viro inode_unlock(path->dentry->d_inode); 2266b12cea91SAl Viro path_put(path); 2267b12cea91SAl Viro path->mnt = mnt; 226884d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2269b12cea91SAl Viro goto retry; 2270b12cea91SAl Viro } 2271b12cea91SAl Viro 227284d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2273b12cea91SAl Viro { 227484d17192SAl Viro struct dentry *dentry = where->m_dentry; 22753895dbf8SEric W. Biederman 22763895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 227784d17192SAl Viro put_mountpoint(where); 22783895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 22793895dbf8SEric W. Biederman 2280328e6d90SAl Viro namespace_unlock(); 22815955102cSAl Viro inode_unlock(dentry->d_inode); 2282b12cea91SAl Viro } 2283b12cea91SAl Viro 228484d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 22851da177e4SLinus Torvalds { 2286e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 22871da177e4SLinus Torvalds return -EINVAL; 22881da177e4SLinus Torvalds 2289e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2290e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 22911da177e4SLinus Torvalds return -ENOTDIR; 22921da177e4SLinus Torvalds 22932763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 22941da177e4SLinus Torvalds } 22951da177e4SLinus Torvalds 22961da177e4SLinus Torvalds /* 22977a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 22987a2e8a8fSValerie Aurora */ 22997a2e8a8fSValerie Aurora 2300e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 23017a2e8a8fSValerie Aurora { 2302e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 23037a2e8a8fSValerie Aurora 23047a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 23057a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 23067a2e8a8fSValerie Aurora return 0; 23077a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 23087a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 23097a2e8a8fSValerie Aurora return 0; 23107a2e8a8fSValerie Aurora return type; 23117a2e8a8fSValerie Aurora } 23127a2e8a8fSValerie Aurora 23137a2e8a8fSValerie Aurora /* 231407b20889SRam Pai * recursively change the type of the mountpoint. 231507b20889SRam Pai */ 2316e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 231707b20889SRam Pai { 2318315fc83eSAl Viro struct mount *m; 23194b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2320e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 23217a2e8a8fSValerie Aurora int type; 2322719f5d7fSMiklos Szeredi int err = 0; 232307b20889SRam Pai 23242d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 232507b20889SRam Pai return -EINVAL; 232607b20889SRam Pai 2327e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 23287a2e8a8fSValerie Aurora if (!type) 23297a2e8a8fSValerie Aurora return -EINVAL; 23307a2e8a8fSValerie Aurora 233197216be0SAl Viro namespace_lock(); 2332719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2333719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2334719f5d7fSMiklos Szeredi if (err) 2335719f5d7fSMiklos Szeredi goto out_unlock; 2336719f5d7fSMiklos Szeredi } 2337719f5d7fSMiklos Szeredi 2338719ea2fbSAl Viro lock_mount_hash(); 2339909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 23400f0afb1dSAl Viro change_mnt_propagation(m, type); 2341719ea2fbSAl Viro unlock_mount_hash(); 2342719f5d7fSMiklos Szeredi 2343719f5d7fSMiklos Szeredi out_unlock: 234497216be0SAl Viro namespace_unlock(); 2345719f5d7fSMiklos Szeredi return err; 234607b20889SRam Pai } 234707b20889SRam Pai 2348a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2349a07b2000SAl Viro { 2350a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2351a07b2000SAl Viro 2352a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2353a07b2000SAl Viro return mnt; 2354a07b2000SAl Viro 2355a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2356a07b2000SAl Viro return mnt; 2357a07b2000SAl Viro 2358a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2359a07b2000SAl Viro return mnt; 2360a07b2000SAl Viro 2361a07b2000SAl Viro if (recurse) 2362a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2363a07b2000SAl Viro else 2364a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2365a07b2000SAl Viro 2366a07b2000SAl Viro if (!IS_ERR(mnt)) 2367a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2368a07b2000SAl Viro 2369a07b2000SAl Viro return mnt; 2370a07b2000SAl Viro } 2371a07b2000SAl Viro 237207b20889SRam Pai /* 23731da177e4SLinus Torvalds * do loopback mount. 23741da177e4SLinus Torvalds */ 2375808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 23762dafe1c4SEric Sandeen int recurse) 23771da177e4SLinus Torvalds { 23782d92ab3cSAl Viro struct path old_path; 2379a07b2000SAl Viro struct mount *mnt = NULL, *parent; 238084d17192SAl Viro struct mountpoint *mp; 238157eccb83SAl Viro int err; 23821da177e4SLinus Torvalds if (!old_name || !*old_name) 23831da177e4SLinus Torvalds return -EINVAL; 2384815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 23851da177e4SLinus Torvalds if (err) 23861da177e4SLinus Torvalds return err; 23871da177e4SLinus Torvalds 23888823c079SEric W. Biederman err = -EINVAL; 23894ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 23908823c079SEric W. Biederman goto out; 23918823c079SEric W. Biederman 239284d17192SAl Viro mp = lock_mount(path); 2393a07b2000SAl Viro if (IS_ERR(mp)) { 239484d17192SAl Viro err = PTR_ERR(mp); 23959676f0c6SRam Pai goto out; 2396a07b2000SAl Viro } 23979676f0c6SRam Pai 239884d17192SAl Viro parent = real_mount(path->mnt); 2399e149ed2bSAl Viro if (!check_mnt(parent)) 2400e149ed2bSAl Viro goto out2; 2401e149ed2bSAl Viro 2402a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2403be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2404be34d1a3SDavid Howells err = PTR_ERR(mnt); 2405e9c5d8a5SAndrey Vagin goto out2; 2406be34d1a3SDavid Howells } 2407ccd48bc7SAl Viro 240884d17192SAl Viro err = graft_tree(mnt, parent, mp); 24091da177e4SLinus Torvalds if (err) { 2410719ea2fbSAl Viro lock_mount_hash(); 2411e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2412719ea2fbSAl Viro unlock_mount_hash(); 24135b83d2c5SRam Pai } 2414b12cea91SAl Viro out2: 241584d17192SAl Viro unlock_mount(mp); 2416ccd48bc7SAl Viro out: 24172d92ab3cSAl Viro path_put(&old_path); 24181da177e4SLinus Torvalds return err; 24191da177e4SLinus Torvalds } 24201da177e4SLinus Torvalds 2421a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2422a07b2000SAl Viro { 2423a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2424a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2425a07b2000SAl Viro struct mount *mnt, *p; 2426a07b2000SAl Viro struct file *file; 2427a07b2000SAl Viro 2428a07b2000SAl Viro if (IS_ERR(ns)) 2429a07b2000SAl Viro return ERR_CAST(ns); 2430a07b2000SAl Viro 2431a07b2000SAl Viro namespace_lock(); 2432a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2433a07b2000SAl Viro if (IS_ERR(mnt)) { 2434a07b2000SAl Viro namespace_unlock(); 2435a07b2000SAl Viro free_mnt_ns(ns); 2436a07b2000SAl Viro return ERR_CAST(mnt); 2437a07b2000SAl Viro } 2438a07b2000SAl Viro 2439a07b2000SAl Viro lock_mount_hash(); 2440a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2441a07b2000SAl Viro p->mnt_ns = ns; 2442a07b2000SAl Viro ns->mounts++; 2443a07b2000SAl Viro } 2444a07b2000SAl Viro ns->root = mnt; 2445a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2446a07b2000SAl Viro mntget(&mnt->mnt); 2447a07b2000SAl Viro unlock_mount_hash(); 2448a07b2000SAl Viro namespace_unlock(); 2449a07b2000SAl Viro 2450a07b2000SAl Viro mntput(path->mnt); 2451a07b2000SAl Viro path->mnt = &mnt->mnt; 2452a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2453a07b2000SAl Viro if (IS_ERR(file)) 2454a07b2000SAl Viro dissolve_on_fput(path->mnt); 2455a07b2000SAl Viro else 2456a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2457a07b2000SAl Viro return file; 2458a07b2000SAl Viro } 2459a07b2000SAl Viro 24602658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2461a07b2000SAl Viro { 2462a07b2000SAl Viro struct file *file; 2463a07b2000SAl Viro struct path path; 2464a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2465a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2466a07b2000SAl Viro int error; 2467a07b2000SAl Viro int fd; 2468a07b2000SAl Viro 2469a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2470a07b2000SAl Viro 2471a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2472a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2473a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2474a07b2000SAl Viro return -EINVAL; 2475a07b2000SAl Viro 2476a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2477a07b2000SAl Viro return -EINVAL; 2478a07b2000SAl Viro 2479a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2480a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2481a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2482a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2483a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2484a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2485a07b2000SAl Viro 2486a07b2000SAl Viro if (detached && !may_mount()) 2487a07b2000SAl Viro return -EPERM; 2488a07b2000SAl Viro 2489a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2490a07b2000SAl Viro if (fd < 0) 2491a07b2000SAl Viro return fd; 2492a07b2000SAl Viro 2493a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2494a07b2000SAl Viro if (unlikely(error)) { 2495a07b2000SAl Viro file = ERR_PTR(error); 2496a07b2000SAl Viro } else { 2497a07b2000SAl Viro if (detached) 2498a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2499a07b2000SAl Viro else 2500a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2501a07b2000SAl Viro path_put(&path); 2502a07b2000SAl Viro } 2503a07b2000SAl Viro if (IS_ERR(file)) { 2504a07b2000SAl Viro put_unused_fd(fd); 2505a07b2000SAl Viro return PTR_ERR(file); 2506a07b2000SAl Viro } 2507a07b2000SAl Viro fd_install(fd, file); 2508a07b2000SAl Viro return fd; 2509a07b2000SAl Viro } 2510a07b2000SAl Viro 251143f5e655SDavid Howells /* 251243f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 251343f5e655SDavid Howells * 251443f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 251543f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 251643f5e655SDavid Howells */ 251743f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 25182e4b7fcdSDave Hansen { 251943f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 25202e4b7fcdSDave Hansen 252143f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 252243f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 252343f5e655SDavid Howells return false; 252443f5e655SDavid Howells 252543f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 252643f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 252743f5e655SDavid Howells return false; 252843f5e655SDavid Howells 252943f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 253043f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 253143f5e655SDavid Howells return false; 253243f5e655SDavid Howells 253343f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 253443f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 253543f5e655SDavid Howells return false; 253643f5e655SDavid Howells 253743f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 253843f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 253943f5e655SDavid Howells return false; 254043f5e655SDavid Howells 254143f5e655SDavid Howells return true; 254243f5e655SDavid Howells } 254343f5e655SDavid Howells 254443f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 254543f5e655SDavid Howells { 254643f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 254743f5e655SDavid Howells 254843f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 25492e4b7fcdSDave Hansen return 0; 25502e4b7fcdSDave Hansen 25512e4b7fcdSDave Hansen if (readonly_request) 255243f5e655SDavid Howells return mnt_make_readonly(mnt); 255343f5e655SDavid Howells 255468847c94SChristian Brauner mnt->mnt.mnt_flags &= ~MNT_READONLY; 255568847c94SChristian Brauner return 0; 255643f5e655SDavid Howells } 255743f5e655SDavid Howells 255843f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 255943f5e655SDavid Howells { 256043f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 256143f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 256243f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 256343f5e655SDavid Howells } 256443f5e655SDavid Howells 2565f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2566f8b92ba6SDeepa Dinamani { 2567f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2568f8b92ba6SDeepa Dinamani 2569f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2570f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2571f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2572f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2573f8b92ba6SDeepa Dinamani struct tm tm; 2574f8b92ba6SDeepa Dinamani 2575f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2576f8b92ba6SDeepa Dinamani 25770ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 25780ecee669SEric Biggers sb->s_type->name, 25790ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 25800ecee669SEric Biggers mntpath, 2581f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2582f8b92ba6SDeepa Dinamani 2583f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2584f8b92ba6SDeepa Dinamani } 2585f8b92ba6SDeepa Dinamani } 2586f8b92ba6SDeepa Dinamani 258743f5e655SDavid Howells /* 258843f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 258943f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 259043f5e655SDavid Howells * to mount(2). 259143f5e655SDavid Howells */ 259243f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 259343f5e655SDavid Howells { 259443f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 259543f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 259643f5e655SDavid Howells int ret; 259743f5e655SDavid Howells 259843f5e655SDavid Howells if (!check_mnt(mnt)) 259943f5e655SDavid Howells return -EINVAL; 260043f5e655SDavid Howells 260143f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 260243f5e655SDavid Howells return -EINVAL; 260343f5e655SDavid Howells 260443f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 260543f5e655SDavid Howells return -EPERM; 260643f5e655SDavid Howells 2607e58ace1aSChristian Brauner /* 2608e58ace1aSChristian Brauner * We're only checking whether the superblock is read-only not 2609e58ace1aSChristian Brauner * changing it, so only take down_read(&sb->s_umount). 2610e58ace1aSChristian Brauner */ 2611e58ace1aSChristian Brauner down_read(&sb->s_umount); 261268847c94SChristian Brauner lock_mount_hash(); 261343f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 261443f5e655SDavid Howells if (ret == 0) 261543f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 261668847c94SChristian Brauner unlock_mount_hash(); 2617e58ace1aSChristian Brauner up_read(&sb->s_umount); 2618f8b92ba6SDeepa Dinamani 2619f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2620f8b92ba6SDeepa Dinamani 262143f5e655SDavid Howells return ret; 26222e4b7fcdSDave Hansen } 26232e4b7fcdSDave Hansen 26241da177e4SLinus Torvalds /* 26251da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 26261da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 26271da177e4SLinus Torvalds * on it - tough luck. 26281da177e4SLinus Torvalds */ 2629e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2630e462ec50SDavid Howells int mnt_flags, void *data) 26311da177e4SLinus Torvalds { 26321da177e4SLinus Torvalds int err; 26332d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2634143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 26358d0347f6SDavid Howells struct fs_context *fc; 26361da177e4SLinus Torvalds 2637143c8c91SAl Viro if (!check_mnt(mnt)) 26381da177e4SLinus Torvalds return -EINVAL; 26391da177e4SLinus Torvalds 26402d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 26411da177e4SLinus Torvalds return -EINVAL; 26421da177e4SLinus Torvalds 264343f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 264407b64558SEric W. Biederman return -EPERM; 26459566d674SEric W. Biederman 26468d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 26478d0347f6SDavid Howells if (IS_ERR(fc)) 26488d0347f6SDavid Howells return PTR_ERR(fc); 2649ff36fe2cSEric Paris 2650b330966fSMiklos Szeredi fc->oldapi = true; 26518d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 26528d0347f6SDavid Howells if (!err) { 26531da177e4SLinus Torvalds down_write(&sb->s_umount); 265457eccb83SAl Viro err = -EPERM; 265543f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 26568d0347f6SDavid Howells err = reconfigure_super(fc); 265768847c94SChristian Brauner if (!err) { 265868847c94SChristian Brauner lock_mount_hash(); 265943f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 266068847c94SChristian Brauner unlock_mount_hash(); 266168847c94SChristian Brauner } 26620e55a7ccSDan Williams } 26636339dab8SAl Viro up_write(&sb->s_umount); 26648d0347f6SDavid Howells } 2665f8b92ba6SDeepa Dinamani 2666f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2667f8b92ba6SDeepa Dinamani 26688d0347f6SDavid Howells put_fs_context(fc); 26691da177e4SLinus Torvalds return err; 26701da177e4SLinus Torvalds } 26711da177e4SLinus Torvalds 2672cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 26739676f0c6SRam Pai { 2674315fc83eSAl Viro struct mount *p; 2675909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2676fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 26779676f0c6SRam Pai return 1; 26789676f0c6SRam Pai } 26799676f0c6SRam Pai return 0; 26809676f0c6SRam Pai } 26819676f0c6SRam Pai 268244dfd84aSDavid Howells /* 268344dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 268444dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 268544dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 268644dfd84aSDavid Howells * cycles to be made. 268744dfd84aSDavid Howells */ 268844dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 268944dfd84aSDavid Howells { 269044dfd84aSDavid Howells struct mount *p; 269144dfd84aSDavid Howells bool ret = false; 269244dfd84aSDavid Howells 269344dfd84aSDavid Howells lock_mount_hash(); 269444dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 269544dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 269644dfd84aSDavid Howells goto out; 269744dfd84aSDavid Howells 269844dfd84aSDavid Howells ret = true; 269944dfd84aSDavid Howells out: 270044dfd84aSDavid Howells unlock_mount_hash(); 270144dfd84aSDavid Howells return ret; 270244dfd84aSDavid Howells } 270344dfd84aSDavid Howells 27049ffb14efSPavel Tikhomirov static int do_set_group(struct path *from_path, struct path *to_path) 27059ffb14efSPavel Tikhomirov { 27069ffb14efSPavel Tikhomirov struct mount *from, *to; 27079ffb14efSPavel Tikhomirov int err; 27089ffb14efSPavel Tikhomirov 27099ffb14efSPavel Tikhomirov from = real_mount(from_path->mnt); 27109ffb14efSPavel Tikhomirov to = real_mount(to_path->mnt); 27119ffb14efSPavel Tikhomirov 27129ffb14efSPavel Tikhomirov namespace_lock(); 27139ffb14efSPavel Tikhomirov 27149ffb14efSPavel Tikhomirov err = -EINVAL; 27159ffb14efSPavel Tikhomirov /* To and From must be mounted */ 27169ffb14efSPavel Tikhomirov if (!is_mounted(&from->mnt)) 27179ffb14efSPavel Tikhomirov goto out; 27189ffb14efSPavel Tikhomirov if (!is_mounted(&to->mnt)) 27199ffb14efSPavel Tikhomirov goto out; 27209ffb14efSPavel Tikhomirov 27219ffb14efSPavel Tikhomirov err = -EPERM; 27229ffb14efSPavel Tikhomirov /* We should be allowed to modify mount namespaces of both mounts */ 27239ffb14efSPavel Tikhomirov if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27249ffb14efSPavel Tikhomirov goto out; 27259ffb14efSPavel Tikhomirov if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27269ffb14efSPavel Tikhomirov goto out; 27279ffb14efSPavel Tikhomirov 27289ffb14efSPavel Tikhomirov err = -EINVAL; 27299ffb14efSPavel Tikhomirov /* To and From paths should be mount roots */ 27309ffb14efSPavel Tikhomirov if (from_path->dentry != from_path->mnt->mnt_root) 27319ffb14efSPavel Tikhomirov goto out; 27329ffb14efSPavel Tikhomirov if (to_path->dentry != to_path->mnt->mnt_root) 27339ffb14efSPavel Tikhomirov goto out; 27349ffb14efSPavel Tikhomirov 27359ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed across same superblock */ 27369ffb14efSPavel Tikhomirov if (from->mnt.mnt_sb != to->mnt.mnt_sb) 27379ffb14efSPavel Tikhomirov goto out; 27389ffb14efSPavel Tikhomirov 27399ffb14efSPavel Tikhomirov /* From mount root should be wider than To mount root */ 27409ffb14efSPavel Tikhomirov if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) 27419ffb14efSPavel Tikhomirov goto out; 27429ffb14efSPavel Tikhomirov 27439ffb14efSPavel Tikhomirov /* From mount should not have locked children in place of To's root */ 27449ffb14efSPavel Tikhomirov if (has_locked_children(from, to->mnt.mnt_root)) 27459ffb14efSPavel Tikhomirov goto out; 27469ffb14efSPavel Tikhomirov 27479ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed on private mounts */ 27489ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) 27499ffb14efSPavel Tikhomirov goto out; 27509ffb14efSPavel Tikhomirov 27519ffb14efSPavel Tikhomirov /* From should not be private */ 27529ffb14efSPavel Tikhomirov if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) 27539ffb14efSPavel Tikhomirov goto out; 27549ffb14efSPavel Tikhomirov 27559ffb14efSPavel Tikhomirov if (IS_MNT_SLAVE(from)) { 27569ffb14efSPavel Tikhomirov struct mount *m = from->mnt_master; 27579ffb14efSPavel Tikhomirov 27589ffb14efSPavel Tikhomirov list_add(&to->mnt_slave, &m->mnt_slave_list); 27599ffb14efSPavel Tikhomirov to->mnt_master = m; 27609ffb14efSPavel Tikhomirov } 27619ffb14efSPavel Tikhomirov 27629ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(from)) { 27639ffb14efSPavel Tikhomirov to->mnt_group_id = from->mnt_group_id; 27649ffb14efSPavel Tikhomirov list_add(&to->mnt_share, &from->mnt_share); 27659ffb14efSPavel Tikhomirov lock_mount_hash(); 27669ffb14efSPavel Tikhomirov set_mnt_shared(to); 27679ffb14efSPavel Tikhomirov unlock_mount_hash(); 27689ffb14efSPavel Tikhomirov } 27699ffb14efSPavel Tikhomirov 27709ffb14efSPavel Tikhomirov err = 0; 27719ffb14efSPavel Tikhomirov out: 27729ffb14efSPavel Tikhomirov namespace_unlock(); 27739ffb14efSPavel Tikhomirov return err; 27749ffb14efSPavel Tikhomirov } 27759ffb14efSPavel Tikhomirov 27762db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 27771da177e4SLinus Torvalds { 277844dfd84aSDavid Howells struct mnt_namespace *ns; 2779676da58dSAl Viro struct mount *p; 27800fb54e50SAl Viro struct mount *old; 27812763d119SAl Viro struct mount *parent; 27822763d119SAl Viro struct mountpoint *mp, *old_mp; 278357eccb83SAl Viro int err; 278444dfd84aSDavid Howells bool attached; 27851da177e4SLinus Torvalds 27862db154b3SDavid Howells mp = lock_mount(new_path); 278784d17192SAl Viro if (IS_ERR(mp)) 27882db154b3SDavid Howells return PTR_ERR(mp); 2789cc53ce53SDavid Howells 27902db154b3SDavid Howells old = real_mount(old_path->mnt); 27912db154b3SDavid Howells p = real_mount(new_path->mnt); 27922763d119SAl Viro parent = old->mnt_parent; 279344dfd84aSDavid Howells attached = mnt_has_parent(old); 27942763d119SAl Viro old_mp = old->mnt_mp; 279544dfd84aSDavid Howells ns = old->mnt_ns; 2796143c8c91SAl Viro 27971da177e4SLinus Torvalds err = -EINVAL; 279844dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 279944dfd84aSDavid Howells if (!check_mnt(p)) 28002db154b3SDavid Howells goto out; 28011da177e4SLinus Torvalds 2802570d7a98SEric Biggers /* The thing moved must be mounted... */ 2803570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 280444dfd84aSDavid Howells goto out; 280544dfd84aSDavid Howells 2806570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2807570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 28082db154b3SDavid Howells goto out; 28091da177e4SLinus Torvalds 28102db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 28112db154b3SDavid Howells goto out; 28122db154b3SDavid Howells 28132db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 28142db154b3SDavid Howells goto out; 28152db154b3SDavid Howells 28162db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 28172db154b3SDavid Howells d_is_dir(old_path->dentry)) 28182db154b3SDavid Howells goto out; 281921444403SRam Pai /* 282021444403SRam Pai * Don't move a mount residing in a shared parent. 282121444403SRam Pai */ 28222763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 28232db154b3SDavid Howells goto out; 28249676f0c6SRam Pai /* 28259676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 28269676f0c6SRam Pai * mount which is shared. 28279676f0c6SRam Pai */ 2828fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 28292db154b3SDavid Howells goto out; 28301da177e4SLinus Torvalds err = -ELOOP; 283144dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 283244dfd84aSDavid Howells goto out; 2833fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2834676da58dSAl Viro if (p == old) 28352db154b3SDavid Howells goto out; 28361da177e4SLinus Torvalds 28372db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 28382763d119SAl Viro attached); 28394ac91378SJan Blunck if (err) 28402db154b3SDavid Howells goto out; 28411da177e4SLinus Torvalds 28421da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 28431da177e4SLinus Torvalds * automatically */ 28446776db3dSAl Viro list_del_init(&old->mnt_expire); 28452763d119SAl Viro if (attached) 28462763d119SAl Viro put_mountpoint(old_mp); 28471da177e4SLinus Torvalds out: 28482db154b3SDavid Howells unlock_mount(mp); 284944dfd84aSDavid Howells if (!err) { 28502763d119SAl Viro if (attached) 28512763d119SAl Viro mntput_no_expire(parent); 28522763d119SAl Viro else 285344dfd84aSDavid Howells free_mnt_ns(ns); 285444dfd84aSDavid Howells } 28552db154b3SDavid Howells return err; 28562db154b3SDavid Howells } 28572db154b3SDavid Howells 28582db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 28592db154b3SDavid Howells { 28602db154b3SDavid Howells struct path old_path; 28612db154b3SDavid Howells int err; 28622db154b3SDavid Howells 28632db154b3SDavid Howells if (!old_name || !*old_name) 28642db154b3SDavid Howells return -EINVAL; 28652db154b3SDavid Howells 28662db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 28672db154b3SDavid Howells if (err) 28682db154b3SDavid Howells return err; 28692db154b3SDavid Howells 28702db154b3SDavid Howells err = do_move_mount(&old_path, path); 28712d92ab3cSAl Viro path_put(&old_path); 28721da177e4SLinus Torvalds return err; 28731da177e4SLinus Torvalds } 28741da177e4SLinus Torvalds 28759d412a43SAl Viro /* 28769d412a43SAl Viro * add a mount into a namespace's mount tree 28779d412a43SAl Viro */ 28788f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 2879*1e2d8464SAl Viro const struct path *path, int mnt_flags) 28809d412a43SAl Viro { 28818f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 28829d412a43SAl Viro 2883f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 28849d412a43SAl Viro 288584d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2886156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2887156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 28888f11538eSAl Viro return -EINVAL; 2889156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 289084d17192SAl Viro if (!parent->mnt_ns) 28918f11538eSAl Viro return -EINVAL; 2892156cacb1SAl Viro } 28939d412a43SAl Viro 28949d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 289595bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 28969d412a43SAl Viro path->mnt->mnt_root == path->dentry) 28978f11538eSAl Viro return -EBUSY; 28989d412a43SAl Viro 2899e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 29008f11538eSAl Viro return -EINVAL; 29019d412a43SAl Viro 290295bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 29038f11538eSAl Viro return graft_tree(newmnt, parent, mp); 29049d412a43SAl Viro } 2905b1e75df4SAl Viro 2906132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2907132e4608SDavid Howells 2908132e4608SDavid Howells /* 2909132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2910132e4608SDavid Howells * be added to the namespace tree. 2911132e4608SDavid Howells */ 2912132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2913132e4608SDavid Howells unsigned int mnt_flags) 2914132e4608SDavid Howells { 2915132e4608SDavid Howells struct vfsmount *mnt; 29168f11538eSAl Viro struct mountpoint *mp; 2917132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2918132e4608SDavid Howells int error; 2919132e4608SDavid Howells 2920c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2921c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2922c9ce29edSAl Viro error = -EPERM; 2923c9ce29edSAl Viro 2924c9ce29edSAl Viro if (unlikely(error)) { 2925c9ce29edSAl Viro fc_drop_locked(fc); 2926c9ce29edSAl Viro return error; 2927132e4608SDavid Howells } 2928132e4608SDavid Howells 2929132e4608SDavid Howells up_write(&sb->s_umount); 2930132e4608SDavid Howells 2931132e4608SDavid Howells mnt = vfs_create_mount(fc); 2932132e4608SDavid Howells if (IS_ERR(mnt)) 2933132e4608SDavid Howells return PTR_ERR(mnt); 2934132e4608SDavid Howells 2935f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 2936f8b92ba6SDeepa Dinamani 29378f11538eSAl Viro mp = lock_mount(mountpoint); 29388f11538eSAl Viro if (IS_ERR(mp)) { 29398f11538eSAl Viro mntput(mnt); 29408f11538eSAl Viro return PTR_ERR(mp); 29418f11538eSAl Viro } 29428f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 29438f11538eSAl Viro unlock_mount(mp); 29440ecee669SEric Biggers if (error < 0) 29450ecee669SEric Biggers mntput(mnt); 2946f8b92ba6SDeepa Dinamani return error; 2947f8b92ba6SDeepa Dinamani } 2948f8b92ba6SDeepa Dinamani 29491da177e4SLinus Torvalds /* 29501da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 29511da177e4SLinus Torvalds * namespace's tree 29521da177e4SLinus Torvalds */ 2953e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2954808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 29551da177e4SLinus Torvalds { 29560c55cfc4SEric W. Biederman struct file_system_type *type; 2957a0c9a8b8SAl Viro struct fs_context *fc; 2958a0c9a8b8SAl Viro const char *subtype = NULL; 2959a0c9a8b8SAl Viro int err = 0; 29601da177e4SLinus Torvalds 29610c55cfc4SEric W. Biederman if (!fstype) 29621da177e4SLinus Torvalds return -EINVAL; 29631da177e4SLinus Torvalds 29640c55cfc4SEric W. Biederman type = get_fs_type(fstype); 29650c55cfc4SEric W. Biederman if (!type) 29660c55cfc4SEric W. Biederman return -ENODEV; 29670c55cfc4SEric W. Biederman 2968a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 2969a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 2970a0c9a8b8SAl Viro if (subtype) { 2971a0c9a8b8SAl Viro subtype++; 2972a0c9a8b8SAl Viro if (!*subtype) { 29730c55cfc4SEric W. Biederman put_filesystem(type); 2974a0c9a8b8SAl Viro return -EINVAL; 2975a0c9a8b8SAl Viro } 2976a0c9a8b8SAl Viro } 29778654df4eSEric W. Biederman } 29788654df4eSEric W. Biederman 2979a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 2980a0c9a8b8SAl Viro put_filesystem(type); 2981a0c9a8b8SAl Viro if (IS_ERR(fc)) 2982a0c9a8b8SAl Viro return PTR_ERR(fc); 2983a0c9a8b8SAl Viro 29843e1aeb00SDavid Howells if (subtype) 29853e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 29863e1aeb00SDavid Howells subtype, strlen(subtype)); 29873e1aeb00SDavid Howells if (!err && name) 29883e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 2989a0c9a8b8SAl Viro if (!err) 2990a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 2991c3aabf07SAl Viro if (!err && !mount_capable(fc)) 2992c3aabf07SAl Viro err = -EPERM; 2993a0c9a8b8SAl Viro if (!err) 2994a0c9a8b8SAl Viro err = vfs_get_tree(fc); 2995132e4608SDavid Howells if (!err) 2996132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 2997a0c9a8b8SAl Viro 2998a0c9a8b8SAl Viro put_fs_context(fc); 299915f9a3f3SAl Viro return err; 30001da177e4SLinus Torvalds } 30011da177e4SLinus Torvalds 3002*1e2d8464SAl Viro int finish_automount(struct vfsmount *m, const struct path *path) 300319a167afSAl Viro { 300426df6034SAl Viro struct dentry *dentry = path->dentry; 30058f11538eSAl Viro struct mountpoint *mp; 300625e195aaSAl Viro struct mount *mnt; 300719a167afSAl Viro int err; 300825e195aaSAl Viro 300925e195aaSAl Viro if (!m) 301025e195aaSAl Viro return 0; 301125e195aaSAl Viro if (IS_ERR(m)) 301225e195aaSAl Viro return PTR_ERR(m); 301325e195aaSAl Viro 301425e195aaSAl Viro mnt = real_mount(m); 301519a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 301619a167afSAl Viro * expired before we get a chance to add it 301719a167afSAl Viro */ 30186776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 301919a167afSAl Viro 302019a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 302126df6034SAl Viro m->mnt_root == dentry) { 3022b1e75df4SAl Viro err = -ELOOP; 302326df6034SAl Viro goto discard; 302419a167afSAl Viro } 302519a167afSAl Viro 302626df6034SAl Viro /* 302726df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 302826df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 302926df6034SAl Viro * got", not "try to mount it on top". 303026df6034SAl Viro */ 303126df6034SAl Viro inode_lock(dentry->d_inode); 303226df6034SAl Viro namespace_lock(); 303326df6034SAl Viro if (unlikely(cant_mount(dentry))) { 303426df6034SAl Viro err = -ENOENT; 303526df6034SAl Viro goto discard_locked; 303626df6034SAl Viro } 303726df6034SAl Viro rcu_read_lock(); 303826df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 303926df6034SAl Viro rcu_read_unlock(); 304026df6034SAl Viro err = 0; 304126df6034SAl Viro goto discard_locked; 304226df6034SAl Viro } 304326df6034SAl Viro rcu_read_unlock(); 304426df6034SAl Viro mp = get_mountpoint(dentry); 30458f11538eSAl Viro if (IS_ERR(mp)) { 30468f11538eSAl Viro err = PTR_ERR(mp); 304726df6034SAl Viro goto discard_locked; 30488f11538eSAl Viro } 304926df6034SAl Viro 30508f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 30518f11538eSAl Viro unlock_mount(mp); 305226df6034SAl Viro if (unlikely(err)) 305326df6034SAl Viro goto discard; 305426df6034SAl Viro mntput(m); 3055b1e75df4SAl Viro return 0; 305626df6034SAl Viro 305726df6034SAl Viro discard_locked: 305826df6034SAl Viro namespace_unlock(); 305926df6034SAl Viro inode_unlock(dentry->d_inode); 306026df6034SAl Viro discard: 3061b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 30626776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 306397216be0SAl Viro namespace_lock(); 30646776db3dSAl Viro list_del_init(&mnt->mnt_expire); 306597216be0SAl Viro namespace_unlock(); 306619a167afSAl Viro } 3067b1e75df4SAl Viro mntput(m); 3068b1e75df4SAl Viro mntput(m); 306919a167afSAl Viro return err; 307019a167afSAl Viro } 307119a167afSAl Viro 3072ea5b778aSDavid Howells /** 3073ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 3074ea5b778aSDavid Howells * @mnt: The mount to list. 3075ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 3076ea5b778aSDavid Howells */ 3077ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 3078ea5b778aSDavid Howells { 307997216be0SAl Viro namespace_lock(); 3080ea5b778aSDavid Howells 30816776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 3082ea5b778aSDavid Howells 308397216be0SAl Viro namespace_unlock(); 3084ea5b778aSDavid Howells } 3085ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 3086ea5b778aSDavid Howells 3087ea5b778aSDavid Howells /* 30881da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 30891da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 30901da177e4SLinus Torvalds * here 30911da177e4SLinus Torvalds */ 30921da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 30931da177e4SLinus Torvalds { 3094761d5c38SAl Viro struct mount *mnt, *next; 30951da177e4SLinus Torvalds LIST_HEAD(graveyard); 30961da177e4SLinus Torvalds 30971da177e4SLinus Torvalds if (list_empty(mounts)) 30981da177e4SLinus Torvalds return; 30991da177e4SLinus Torvalds 310097216be0SAl Viro namespace_lock(); 3101719ea2fbSAl Viro lock_mount_hash(); 31021da177e4SLinus Torvalds 31031da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 31041da177e4SLinus Torvalds * following criteria: 31051da177e4SLinus Torvalds * - only referenced by its parent vfsmount 31061da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 31071da177e4SLinus Torvalds * cleared by mntput()) 31081da177e4SLinus Torvalds */ 31096776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 3110863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 31111ab59738SAl Viro propagate_mount_busy(mnt, 1)) 31121da177e4SLinus Torvalds continue; 31136776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 31141da177e4SLinus Torvalds } 3115bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 31166776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 3117143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 3118e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3119bcc5c7d2SAl Viro } 3120719ea2fbSAl Viro unlock_mount_hash(); 31213ab6abeeSAl Viro namespace_unlock(); 31221da177e4SLinus Torvalds } 31231da177e4SLinus Torvalds 31241da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 31251da177e4SLinus Torvalds 31261da177e4SLinus Torvalds /* 31275528f911STrond Myklebust * Ripoff of 'select_parent()' 31285528f911STrond Myklebust * 31295528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 31305528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 31315528f911STrond Myklebust */ 3132692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 31335528f911STrond Myklebust { 3134692afc31SAl Viro struct mount *this_parent = parent; 31355528f911STrond Myklebust struct list_head *next; 31365528f911STrond Myklebust int found = 0; 31375528f911STrond Myklebust 31385528f911STrond Myklebust repeat: 31396b41d536SAl Viro next = this_parent->mnt_mounts.next; 31405528f911STrond Myklebust resume: 31416b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 31425528f911STrond Myklebust struct list_head *tmp = next; 31436b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 31445528f911STrond Myklebust 31455528f911STrond Myklebust next = tmp->next; 3146692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 31475528f911STrond Myklebust continue; 31485528f911STrond Myklebust /* 31495528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 31505528f911STrond Myklebust */ 31516b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 31525528f911STrond Myklebust this_parent = mnt; 31535528f911STrond Myklebust goto repeat; 31545528f911STrond Myklebust } 31555528f911STrond Myklebust 31561ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 31576776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 31585528f911STrond Myklebust found++; 31595528f911STrond Myklebust } 31605528f911STrond Myklebust } 31615528f911STrond Myklebust /* 31625528f911STrond Myklebust * All done at this level ... ascend and resume the search 31635528f911STrond Myklebust */ 31645528f911STrond Myklebust if (this_parent != parent) { 31656b41d536SAl Viro next = this_parent->mnt_child.next; 31660714a533SAl Viro this_parent = this_parent->mnt_parent; 31675528f911STrond Myklebust goto resume; 31685528f911STrond Myklebust } 31695528f911STrond Myklebust return found; 31705528f911STrond Myklebust } 31715528f911STrond Myklebust 31725528f911STrond Myklebust /* 31735528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 31745528f911STrond Myklebust * submounts of a specific parent mountpoint 317599b7db7bSNick Piggin * 317648a066e7SAl Viro * mount_lock must be held for write 31775528f911STrond Myklebust */ 3178b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 31795528f911STrond Myklebust { 31805528f911STrond Myklebust LIST_HEAD(graveyard); 3181761d5c38SAl Viro struct mount *m; 31825528f911STrond Myklebust 31835528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 3184c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3185bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3186761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 31876776db3dSAl Viro mnt_expire); 3188143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3189e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3190bcc5c7d2SAl Viro } 3191bcc5c7d2SAl Viro } 31925528f911STrond Myklebust } 31935528f911STrond Myklebust 3194028abd92SChristoph Hellwig static void *copy_mount_options(const void __user * data) 31951da177e4SLinus Torvalds { 3196b40ef869SAl Viro char *copy; 3197d563d678SCatalin Marinas unsigned left, offset; 31981da177e4SLinus Torvalds 31991da177e4SLinus Torvalds if (!data) 3200b40ef869SAl Viro return NULL; 32011da177e4SLinus Torvalds 3202b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3203b40ef869SAl Viro if (!copy) 3204b40ef869SAl Viro return ERR_PTR(-ENOMEM); 32051da177e4SLinus Torvalds 3206d563d678SCatalin Marinas left = copy_from_user(copy, data, PAGE_SIZE); 32071da177e4SLinus Torvalds 3208d563d678SCatalin Marinas /* 3209d563d678SCatalin Marinas * Not all architectures have an exact copy_from_user(). Resort to 3210d563d678SCatalin Marinas * byte at a time. 3211d563d678SCatalin Marinas */ 3212d563d678SCatalin Marinas offset = PAGE_SIZE - left; 3213d563d678SCatalin Marinas while (left) { 3214d563d678SCatalin Marinas char c; 3215d563d678SCatalin Marinas if (get_user(c, (const char __user *)data + offset)) 3216d563d678SCatalin Marinas break; 3217d563d678SCatalin Marinas copy[offset] = c; 3218d563d678SCatalin Marinas left--; 3219d563d678SCatalin Marinas offset++; 3220d563d678SCatalin Marinas } 3221d563d678SCatalin Marinas 3222d563d678SCatalin Marinas if (left == PAGE_SIZE) { 3223b40ef869SAl Viro kfree(copy); 3224b40ef869SAl Viro return ERR_PTR(-EFAULT); 32251da177e4SLinus Torvalds } 3226d563d678SCatalin Marinas 3227b40ef869SAl Viro return copy; 32281da177e4SLinus Torvalds } 32291da177e4SLinus Torvalds 3230028abd92SChristoph Hellwig static char *copy_mount_string(const void __user *data) 3231eca6f534SVegard Nossum { 3232fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3233eca6f534SVegard Nossum } 3234eca6f534SVegard Nossum 32351da177e4SLinus Torvalds /* 32361da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 32371da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 32381da177e4SLinus Torvalds * 32391da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 32401da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 32411da177e4SLinus Torvalds * information (or be NULL). 32421da177e4SLinus Torvalds * 32431da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 32441da177e4SLinus Torvalds * When the flags word was introduced its top half was required 32451da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 32461da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 32471da177e4SLinus Torvalds * and must be discarded. 32481da177e4SLinus Torvalds */ 3249c60166f0SChristoph Hellwig int path_mount(const char *dev_name, struct path *path, 3250808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 32511da177e4SLinus Torvalds { 3252e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 3253a1e6aaa3SChristoph Hellwig int ret; 32541da177e4SLinus Torvalds 32551da177e4SLinus Torvalds /* Discard magic */ 32561da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 32571da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 32581da177e4SLinus Torvalds 32591da177e4SLinus Torvalds /* Basic sanity checks */ 32601da177e4SLinus Torvalds if (data_page) 32611da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 32621da177e4SLinus Torvalds 3263e462ec50SDavid Howells if (flags & MS_NOUSER) 3264e462ec50SDavid Howells return -EINVAL; 3265e462ec50SDavid Howells 3266a1e6aaa3SChristoph Hellwig ret = security_sb_mount(dev_name, path, type_page, flags, data_page); 3267a1e6aaa3SChristoph Hellwig if (ret) 3268a1e6aaa3SChristoph Hellwig return ret; 3269a1e6aaa3SChristoph Hellwig if (!may_mount()) 3270a1e6aaa3SChristoph Hellwig return -EPERM; 3271f7e33bdbSJeff Layton if (flags & SB_MANDLOCK) 3272f7e33bdbSJeff Layton warn_mandlock(); 3273a27ab9f2STetsuo Handa 3274613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3275613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 32760a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 32770a1c01c9SMatthew Garrett 32781da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 32791da177e4SLinus Torvalds if (flags & MS_NOSUID) 32801da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 32811da177e4SLinus Torvalds if (flags & MS_NODEV) 32821da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 32831da177e4SLinus Torvalds if (flags & MS_NOEXEC) 32841da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3285fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3286fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3287fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3288fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3289d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3290d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3291a9e5b732SDavid Howells if (flags & MS_RDONLY) 32922e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3293dab741e0SMattias Nissler if (flags & MS_NOSYMFOLLOW) 3294dab741e0SMattias Nissler mnt_flags |= MNT_NOSYMFOLLOW; 3295fc33a7bbSChristoph Hellwig 3296ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3297ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3298ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3299ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3300ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3301a1e6aaa3SChristoph Hellwig mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; 3302ffbc6f0eSEric W. Biederman } 3303ffbc6f0eSEric W. Biederman 3304e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3305e462ec50SDavid Howells SB_SYNCHRONOUS | 3306e462ec50SDavid Howells SB_MANDLOCK | 3307e462ec50SDavid Howells SB_DIRSYNC | 3308e462ec50SDavid Howells SB_SILENT | 3309917086ffSMimi Zohar SB_POSIXACL | 3310d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3311917086ffSMimi Zohar SB_I_VERSION); 33121da177e4SLinus Torvalds 331343f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 3314a1e6aaa3SChristoph Hellwig return do_reconfigure_mnt(path, mnt_flags); 3315a1e6aaa3SChristoph Hellwig if (flags & MS_REMOUNT) 3316a1e6aaa3SChristoph Hellwig return do_remount(path, flags, sb_flags, mnt_flags, data_page); 3317a1e6aaa3SChristoph Hellwig if (flags & MS_BIND) 3318a1e6aaa3SChristoph Hellwig return do_loopback(path, dev_name, flags & MS_REC); 3319a1e6aaa3SChristoph Hellwig if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 3320a1e6aaa3SChristoph Hellwig return do_change_type(path, flags); 3321a1e6aaa3SChristoph Hellwig if (flags & MS_MOVE) 3322a1e6aaa3SChristoph Hellwig return do_move_mount_old(path, dev_name); 3323a1e6aaa3SChristoph Hellwig 3324a1e6aaa3SChristoph Hellwig return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, 33251da177e4SLinus Torvalds data_page); 3326a1e6aaa3SChristoph Hellwig } 3327a1e6aaa3SChristoph Hellwig 3328a1e6aaa3SChristoph Hellwig long do_mount(const char *dev_name, const char __user *dir_name, 3329a1e6aaa3SChristoph Hellwig const char *type_page, unsigned long flags, void *data_page) 3330a1e6aaa3SChristoph Hellwig { 3331a1e6aaa3SChristoph Hellwig struct path path; 3332a1e6aaa3SChristoph Hellwig int ret; 3333a1e6aaa3SChristoph Hellwig 3334a1e6aaa3SChristoph Hellwig ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3335a1e6aaa3SChristoph Hellwig if (ret) 3336a1e6aaa3SChristoph Hellwig return ret; 3337a1e6aaa3SChristoph Hellwig ret = path_mount(dev_name, &path, type_page, flags, data_page); 33382d92ab3cSAl Viro path_put(&path); 3339a1e6aaa3SChristoph Hellwig return ret; 33401da177e4SLinus Torvalds } 33411da177e4SLinus Torvalds 3342537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3343537f7ccbSEric W. Biederman { 3344537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3345537f7ccbSEric W. Biederman } 3346537f7ccbSEric W. Biederman 3347537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3348537f7ccbSEric W. Biederman { 3349537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3350537f7ccbSEric W. Biederman } 3351537f7ccbSEric W. Biederman 3352771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3353771b1371SEric W. Biederman { 335474e83122SAl Viro if (!is_anon_ns(ns)) 33556344c433SAl Viro ns_free_inum(&ns->ns); 3356537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3357771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3358771b1371SEric W. Biederman kfree(ns); 3359771b1371SEric W. Biederman } 3360771b1371SEric W. Biederman 33618823c079SEric W. Biederman /* 33628823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 33638823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 33648823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 33658823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 33668823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 33678823c079SEric W. Biederman */ 33688823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 33698823c079SEric W. Biederman 337074e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3371cf8d2c11STrond Myklebust { 3372cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3373537f7ccbSEric W. Biederman struct ucounts *ucounts; 337498f842e6SEric W. Biederman int ret; 3375cf8d2c11STrond Myklebust 3376537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3377537f7ccbSEric W. Biederman if (!ucounts) 3378df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3379537f7ccbSEric W. Biederman 338030acd0bdSVasily Averin new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); 3381537f7ccbSEric W. Biederman if (!new_ns) { 3382537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3383cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3384537f7ccbSEric W. Biederman } 338574e83122SAl Viro if (!anon) { 33866344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 338798f842e6SEric W. Biederman if (ret) { 338898f842e6SEric W. Biederman kfree(new_ns); 3389537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 339098f842e6SEric W. Biederman return ERR_PTR(ret); 339198f842e6SEric W. Biederman } 339274e83122SAl Viro } 339333c42940SAl Viro new_ns->ns.ops = &mntns_operations; 339474e83122SAl Viro if (!anon) 33958823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 33961a7b8969SKirill Tkhai refcount_set(&new_ns->ns.count, 1); 3397cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3398cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 33999f6c61f9SMiklos Szeredi spin_lock_init(&new_ns->ns_lock); 3400771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3401537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3402cf8d2c11STrond Myklebust return new_ns; 3403cf8d2c11STrond Myklebust } 3404cf8d2c11STrond Myklebust 34050766f788SEmese Revfy __latent_entropy 34069559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 34079559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 34081da177e4SLinus Torvalds { 34096b3286edSKirill Korotaev struct mnt_namespace *new_ns; 34107f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3411315fc83eSAl Viro struct mount *p, *q; 34129559f689SAl Viro struct mount *old; 3413cb338d06SAl Viro struct mount *new; 34147a472ef4SEric W. Biederman int copy_flags; 34151da177e4SLinus Torvalds 34169559f689SAl Viro BUG_ON(!ns); 34179559f689SAl Viro 34189559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 34199559f689SAl Viro get_mnt_ns(ns); 34209559f689SAl Viro return ns; 34219559f689SAl Viro } 34229559f689SAl Viro 34239559f689SAl Viro old = ns->root; 34249559f689SAl Viro 342574e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3426cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3427cf8d2c11STrond Myklebust return new_ns; 34281da177e4SLinus Torvalds 342997216be0SAl Viro namespace_lock(); 34301da177e4SLinus Torvalds /* First pass: copy the tree topology */ 34314ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 34329559f689SAl Viro if (user_ns != ns->user_ns) 34333bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 34347a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3435be34d1a3SDavid Howells if (IS_ERR(new)) { 3436328e6d90SAl Viro namespace_unlock(); 3437771b1371SEric W. Biederman free_mnt_ns(new_ns); 3438be34d1a3SDavid Howells return ERR_CAST(new); 34391da177e4SLinus Torvalds } 34403bd045ccSAl Viro if (user_ns != ns->user_ns) { 34413bd045ccSAl Viro lock_mount_hash(); 34423bd045ccSAl Viro lock_mnt_tree(new); 34433bd045ccSAl Viro unlock_mount_hash(); 34443bd045ccSAl Viro } 3445be08d6d2SAl Viro new_ns->root = new; 34461a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 34471da177e4SLinus Torvalds 34481da177e4SLinus Torvalds /* 34491da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 34501da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 34511da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 34521da177e4SLinus Torvalds */ 3453909b0a88SAl Viro p = old; 3454cb338d06SAl Viro q = new; 34551da177e4SLinus Torvalds while (p) { 3456143c8c91SAl Viro q->mnt_ns = new_ns; 3457d2921684SEric W. Biederman new_ns->mounts++; 34589559f689SAl Viro if (new_fs) { 34599559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 34609559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3461315fc83eSAl Viro rootmnt = &p->mnt; 34621da177e4SLinus Torvalds } 34639559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 34649559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3465315fc83eSAl Viro pwdmnt = &p->mnt; 34661da177e4SLinus Torvalds } 34671da177e4SLinus Torvalds } 3468909b0a88SAl Viro p = next_mnt(p, old); 3469909b0a88SAl Viro q = next_mnt(q, new); 34704ce5d2b1SEric W. Biederman if (!q) 34714ce5d2b1SEric W. Biederman break; 34724ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 34734ce5d2b1SEric W. Biederman p = next_mnt(p, old); 34741da177e4SLinus Torvalds } 3475328e6d90SAl Viro namespace_unlock(); 34761da177e4SLinus Torvalds 34771da177e4SLinus Torvalds if (rootmnt) 3478f03c6599SAl Viro mntput(rootmnt); 34791da177e4SLinus Torvalds if (pwdmnt) 3480f03c6599SAl Viro mntput(pwdmnt); 34811da177e4SLinus Torvalds 3482741a2951SJANAK DESAI return new_ns; 3483741a2951SJANAK DESAI } 3484741a2951SJANAK DESAI 348574e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3486cf8d2c11STrond Myklebust { 34871a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3488ea441d11SAl Viro struct mnt_namespace *ns; 3489d31da0f0SAl Viro struct super_block *s; 3490ea441d11SAl Viro struct path path; 3491ea441d11SAl Viro int err; 3492ea441d11SAl Viro 349374e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 349474e83122SAl Viro if (IS_ERR(ns)) { 349574e83122SAl Viro mntput(m); 3496ea441d11SAl Viro return ERR_CAST(ns); 349774e83122SAl Viro } 349874e83122SAl Viro mnt->mnt_ns = ns; 349974e83122SAl Viro ns->root = mnt; 350074e83122SAl Viro ns->mounts++; 350174e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3502ea441d11SAl Viro 350374e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3504ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3505ea441d11SAl Viro 3506ea441d11SAl Viro put_mnt_ns(ns); 3507ea441d11SAl Viro 3508ea441d11SAl Viro if (err) 3509ea441d11SAl Viro return ERR_PTR(err); 3510ea441d11SAl Viro 3511ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3512d31da0f0SAl Viro s = path.mnt->mnt_sb; 3513d31da0f0SAl Viro atomic_inc(&s->s_active); 3514ea441d11SAl Viro mntput(path.mnt); 3515ea441d11SAl Viro /* lock the sucker */ 3516d31da0f0SAl Viro down_write(&s->s_umount); 3517ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3518ea441d11SAl Viro return path.dentry; 3519ea441d11SAl Viro } 3520ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3521ea441d11SAl Viro 3522cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3523cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 35241da177e4SLinus Torvalds { 3525eca6f534SVegard Nossum int ret; 3526eca6f534SVegard Nossum char *kernel_type; 3527eca6f534SVegard Nossum char *kernel_dev; 3528b40ef869SAl Viro void *options; 35291da177e4SLinus Torvalds 3530b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3531b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3532b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3533eca6f534SVegard Nossum goto out_type; 35341da177e4SLinus Torvalds 3535b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3536b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3537b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3538eca6f534SVegard Nossum goto out_dev; 35391da177e4SLinus Torvalds 3540b40ef869SAl Viro options = copy_mount_options(data); 3541b40ef869SAl Viro ret = PTR_ERR(options); 3542b40ef869SAl Viro if (IS_ERR(options)) 3543eca6f534SVegard Nossum goto out_data; 35441da177e4SLinus Torvalds 3545b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3546eca6f534SVegard Nossum 3547b40ef869SAl Viro kfree(options); 3548eca6f534SVegard Nossum out_data: 3549eca6f534SVegard Nossum kfree(kernel_dev); 3550eca6f534SVegard Nossum out_dev: 3551eca6f534SVegard Nossum kfree(kernel_type); 3552eca6f534SVegard Nossum out_type: 3553eca6f534SVegard Nossum return ret; 35541da177e4SLinus Torvalds } 35551da177e4SLinus Torvalds 35565b490500SChristian Brauner #define FSMOUNT_VALID_FLAGS \ 35575b490500SChristian Brauner (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ 3558dd8b477fSChristian Brauner MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ 3559dd8b477fSChristian Brauner MOUNT_ATTR_NOSYMFOLLOW) 35605b490500SChristian Brauner 35619caccd41SChristian Brauner #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) 35622a186721SChristian Brauner 35632a186721SChristian Brauner #define MOUNT_SETATTR_PROPAGATION_FLAGS \ 35642a186721SChristian Brauner (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) 35652a186721SChristian Brauner 35665b490500SChristian Brauner static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) 35675b490500SChristian Brauner { 35685b490500SChristian Brauner unsigned int mnt_flags = 0; 35695b490500SChristian Brauner 35705b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_RDONLY) 35715b490500SChristian Brauner mnt_flags |= MNT_READONLY; 35725b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOSUID) 35735b490500SChristian Brauner mnt_flags |= MNT_NOSUID; 35745b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODEV) 35755b490500SChristian Brauner mnt_flags |= MNT_NODEV; 35765b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOEXEC) 35775b490500SChristian Brauner mnt_flags |= MNT_NOEXEC; 35785b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODIRATIME) 35795b490500SChristian Brauner mnt_flags |= MNT_NODIRATIME; 3580dd8b477fSChristian Brauner if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) 3581dd8b477fSChristian Brauner mnt_flags |= MNT_NOSYMFOLLOW; 35825b490500SChristian Brauner 35835b490500SChristian Brauner return mnt_flags; 35845b490500SChristian Brauner } 35855b490500SChristian Brauner 35861da177e4SLinus Torvalds /* 358793766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 358893766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 358993766fbdSDavid Howells */ 359093766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 359193766fbdSDavid Howells unsigned int, attr_flags) 359293766fbdSDavid Howells { 359393766fbdSDavid Howells struct mnt_namespace *ns; 359493766fbdSDavid Howells struct fs_context *fc; 359593766fbdSDavid Howells struct file *file; 359693766fbdSDavid Howells struct path newmount; 359793766fbdSDavid Howells struct mount *mnt; 359893766fbdSDavid Howells struct fd f; 359993766fbdSDavid Howells unsigned int mnt_flags = 0; 360093766fbdSDavid Howells long ret; 360193766fbdSDavid Howells 360293766fbdSDavid Howells if (!may_mount()) 360393766fbdSDavid Howells return -EPERM; 360493766fbdSDavid Howells 360593766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 360693766fbdSDavid Howells return -EINVAL; 360793766fbdSDavid Howells 36085b490500SChristian Brauner if (attr_flags & ~FSMOUNT_VALID_FLAGS) 360993766fbdSDavid Howells return -EINVAL; 361093766fbdSDavid Howells 36115b490500SChristian Brauner mnt_flags = attr_flags_to_mnt_flags(attr_flags); 361293766fbdSDavid Howells 361393766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 361493766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 361593766fbdSDavid Howells break; 361693766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 361793766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 361893766fbdSDavid Howells break; 361993766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 362093766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 362193766fbdSDavid Howells break; 362293766fbdSDavid Howells default: 362393766fbdSDavid Howells return -EINVAL; 362493766fbdSDavid Howells } 362593766fbdSDavid Howells 362693766fbdSDavid Howells f = fdget(fs_fd); 362793766fbdSDavid Howells if (!f.file) 362893766fbdSDavid Howells return -EBADF; 362993766fbdSDavid Howells 363093766fbdSDavid Howells ret = -EINVAL; 363193766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 363293766fbdSDavid Howells goto err_fsfd; 363393766fbdSDavid Howells 363493766fbdSDavid Howells fc = f.file->private_data; 363593766fbdSDavid Howells 363693766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 363793766fbdSDavid Howells if (ret < 0) 363893766fbdSDavid Howells goto err_fsfd; 363993766fbdSDavid Howells 364093766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 364193766fbdSDavid Howells ret = -EINVAL; 364293766fbdSDavid Howells if (!fc->root) 364393766fbdSDavid Howells goto err_unlock; 364493766fbdSDavid Howells 364593766fbdSDavid Howells ret = -EPERM; 364693766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 364793766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 364893766fbdSDavid Howells goto err_unlock; 364993766fbdSDavid Howells } 365093766fbdSDavid Howells 365193766fbdSDavid Howells ret = -EBUSY; 365293766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 365393766fbdSDavid Howells goto err_unlock; 365493766fbdSDavid Howells 3655f7e33bdbSJeff Layton if (fc->sb_flags & SB_MANDLOCK) 3656f7e33bdbSJeff Layton warn_mandlock(); 365793766fbdSDavid Howells 365893766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 365993766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 366093766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 366193766fbdSDavid Howells goto err_unlock; 366293766fbdSDavid Howells } 366393766fbdSDavid Howells newmount.dentry = dget(fc->root); 366493766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 366593766fbdSDavid Howells 366693766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 366793766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 366893766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 366993766fbdSDavid Howells * don't want to have to handle any errors incurred. 367093766fbdSDavid Howells */ 367193766fbdSDavid Howells vfs_clean_context(fc); 367293766fbdSDavid Howells 367393766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 367493766fbdSDavid Howells if (IS_ERR(ns)) { 367593766fbdSDavid Howells ret = PTR_ERR(ns); 367693766fbdSDavid Howells goto err_path; 367793766fbdSDavid Howells } 367893766fbdSDavid Howells mnt = real_mount(newmount.mnt); 367993766fbdSDavid Howells mnt->mnt_ns = ns; 368093766fbdSDavid Howells ns->root = mnt; 368193766fbdSDavid Howells ns->mounts = 1; 368293766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 36831b0b9cc8SEric Biggers mntget(newmount.mnt); 368493766fbdSDavid Howells 368593766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 368693766fbdSDavid Howells * it, not just simply put it. 368793766fbdSDavid Howells */ 368893766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 368993766fbdSDavid Howells if (IS_ERR(file)) { 369093766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 369193766fbdSDavid Howells ret = PTR_ERR(file); 369293766fbdSDavid Howells goto err_path; 369393766fbdSDavid Howells } 369493766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 369593766fbdSDavid Howells 369693766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 369793766fbdSDavid Howells if (ret >= 0) 369893766fbdSDavid Howells fd_install(ret, file); 369993766fbdSDavid Howells else 370093766fbdSDavid Howells fput(file); 370193766fbdSDavid Howells 370293766fbdSDavid Howells err_path: 370393766fbdSDavid Howells path_put(&newmount); 370493766fbdSDavid Howells err_unlock: 370593766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 370693766fbdSDavid Howells err_fsfd: 370793766fbdSDavid Howells fdput(f); 370893766fbdSDavid Howells return ret; 370993766fbdSDavid Howells } 371093766fbdSDavid Howells 371193766fbdSDavid Howells /* 371293766fbdSDavid Howells * Move a mount from one place to another. In combination with 371393766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 371493766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 371593766fbdSDavid Howells * a mount subtree. 37162db154b3SDavid Howells * 37172db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 37182db154b3SDavid Howells */ 37192db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 37202658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 37212658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 37222db154b3SDavid Howells unsigned int, flags) 37232db154b3SDavid Howells { 37242db154b3SDavid Howells struct path from_path, to_path; 37252db154b3SDavid Howells unsigned int lflags; 37262db154b3SDavid Howells int ret = 0; 37272db154b3SDavid Howells 37282db154b3SDavid Howells if (!may_mount()) 37292db154b3SDavid Howells return -EPERM; 37302db154b3SDavid Howells 37312db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 37322db154b3SDavid Howells return -EINVAL; 37332db154b3SDavid Howells 37342db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 37352db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 37362db154b3SDavid Howells * to clear it afterwards. 37372db154b3SDavid Howells */ 37382db154b3SDavid Howells lflags = 0; 37392db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37402db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37412db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 37422db154b3SDavid Howells 37432db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 37442db154b3SDavid Howells if (ret < 0) 37452db154b3SDavid Howells return ret; 37462db154b3SDavid Howells 37472db154b3SDavid Howells lflags = 0; 37482db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37492db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37502db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 37512db154b3SDavid Howells 37522db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 37532db154b3SDavid Howells if (ret < 0) 37542db154b3SDavid Howells goto out_from; 37552db154b3SDavid Howells 37562db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 37572db154b3SDavid Howells if (ret < 0) 37582db154b3SDavid Howells goto out_to; 37592db154b3SDavid Howells 37609ffb14efSPavel Tikhomirov if (flags & MOVE_MOUNT_SET_GROUP) 37619ffb14efSPavel Tikhomirov ret = do_set_group(&from_path, &to_path); 37629ffb14efSPavel Tikhomirov else 37632db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 37642db154b3SDavid Howells 37652db154b3SDavid Howells out_to: 37662db154b3SDavid Howells path_put(&to_path); 37672db154b3SDavid Howells out_from: 37682db154b3SDavid Howells path_put(&from_path); 37692db154b3SDavid Howells return ret; 37702db154b3SDavid Howells } 37712db154b3SDavid Howells 37722db154b3SDavid Howells /* 3773afac7cbaSAl Viro * Return true if path is reachable from root 3774afac7cbaSAl Viro * 377548a066e7SAl Viro * namespace_sem or mount_lock is held 3776afac7cbaSAl Viro */ 3777643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3778afac7cbaSAl Viro const struct path *root) 3779afac7cbaSAl Viro { 3780643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3781a73324daSAl Viro dentry = mnt->mnt_mountpoint; 37820714a533SAl Viro mnt = mnt->mnt_parent; 3783afac7cbaSAl Viro } 3784643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3785afac7cbaSAl Viro } 3786afac7cbaSAl Viro 3787640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3788afac7cbaSAl Viro { 378925ab4c9bSYaowei Bai bool res; 379048a066e7SAl Viro read_seqlock_excl(&mount_lock); 3791643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 379248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3793afac7cbaSAl Viro return res; 3794afac7cbaSAl Viro } 3795afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3796afac7cbaSAl Viro 3797afac7cbaSAl Viro /* 37981da177e4SLinus Torvalds * pivot_root Semantics: 37991da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 38001da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 38011da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 38021da177e4SLinus Torvalds * 38031da177e4SLinus Torvalds * Restrictions: 38041da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 38051da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 38061da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 38071da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 38081da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 38091da177e4SLinus Torvalds * 38104a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 38110c1bc6b8SMauro Carvalho Chehab * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives 38124a0d11faSNeil Brown * in this situation. 38134a0d11faSNeil Brown * 38141da177e4SLinus Torvalds * Notes: 38151da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 38161da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 38171da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 38181da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 38191da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 38201da177e4SLinus Torvalds * first. 38211da177e4SLinus Torvalds */ 38223480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 38233480b257SHeiko Carstens const char __user *, put_old) 38241da177e4SLinus Torvalds { 38252763d119SAl Viro struct path new, old, root; 38262763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 382784d17192SAl Viro struct mountpoint *old_mp, *root_mp; 38281da177e4SLinus Torvalds int error; 38291da177e4SLinus Torvalds 38309b40bc90SAl Viro if (!may_mount()) 38311da177e4SLinus Torvalds return -EPERM; 38321da177e4SLinus Torvalds 3833ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3834ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 38351da177e4SLinus Torvalds if (error) 38361da177e4SLinus Torvalds goto out0; 38371da177e4SLinus Torvalds 3838ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 3839ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 38401da177e4SLinus Torvalds if (error) 38411da177e4SLinus Torvalds goto out1; 38421da177e4SLinus Torvalds 38432d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3844b12cea91SAl Viro if (error) 3845b12cea91SAl Viro goto out2; 38461da177e4SLinus Torvalds 3847f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 384884d17192SAl Viro old_mp = lock_mount(&old); 384984d17192SAl Viro error = PTR_ERR(old_mp); 385084d17192SAl Viro if (IS_ERR(old_mp)) 3851b12cea91SAl Viro goto out3; 3852b12cea91SAl Viro 38531da177e4SLinus Torvalds error = -EINVAL; 3854419148daSAl Viro new_mnt = real_mount(new.mnt); 3855419148daSAl Viro root_mnt = real_mount(root.mnt); 385684d17192SAl Viro old_mnt = real_mount(old.mnt); 38572763d119SAl Viro ex_parent = new_mnt->mnt_parent; 38582763d119SAl Viro root_parent = root_mnt->mnt_parent; 385984d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 38602763d119SAl Viro IS_MNT_SHARED(ex_parent) || 38612763d119SAl Viro IS_MNT_SHARED(root_parent)) 3862b12cea91SAl Viro goto out4; 3863143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3864b12cea91SAl Viro goto out4; 38655ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 38665ff9d8a6SEric W. Biederman goto out4; 38671da177e4SLinus Torvalds error = -ENOENT; 3868f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3869b12cea91SAl Viro goto out4; 38701da177e4SLinus Torvalds error = -EBUSY; 387184d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3872b12cea91SAl Viro goto out4; /* loop, on the same file system */ 38731da177e4SLinus Torvalds error = -EINVAL; 38748c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3875b12cea91SAl Viro goto out4; /* not a mountpoint */ 3876676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3877b12cea91SAl Viro goto out4; /* not attached */ 38782d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3879b12cea91SAl Viro goto out4; /* not a mountpoint */ 3880676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3881b12cea91SAl Viro goto out4; /* not attached */ 38824ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 388384d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3884b12cea91SAl Viro goto out4; 38850d082601SEric W. Biederman /* make certain new is below the root */ 38860d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 38870d082601SEric W. Biederman goto out4; 3888719ea2fbSAl Viro lock_mount_hash(); 38892763d119SAl Viro umount_mnt(new_mnt); 38902763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 38915ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 38925ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 38935ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 38945ff9d8a6SEric W. Biederman } 38954ac91378SJan Blunck /* mount old root on put_old */ 389684d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 38974ac91378SJan Blunck /* mount new_root on / */ 38982763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 38992763d119SAl Viro mnt_add_count(root_parent, -1); 39006b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 39014fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 39024fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 39033895dbf8SEric W. Biederman put_mountpoint(root_mp); 3904719ea2fbSAl Viro unlock_mount_hash(); 39052d8f3038SAl Viro chroot_fs_refs(&root, &new); 39061da177e4SLinus Torvalds error = 0; 3907b12cea91SAl Viro out4: 390884d17192SAl Viro unlock_mount(old_mp); 39092763d119SAl Viro if (!error) 39102763d119SAl Viro mntput_no_expire(ex_parent); 3911b12cea91SAl Viro out3: 39128c3ee42eSAl Viro path_put(&root); 3913b12cea91SAl Viro out2: 39142d8f3038SAl Viro path_put(&old); 39151da177e4SLinus Torvalds out1: 39162d8f3038SAl Viro path_put(&new); 39171da177e4SLinus Torvalds out0: 39181da177e4SLinus Torvalds return error; 39191da177e4SLinus Torvalds } 39201da177e4SLinus Torvalds 39212a186721SChristian Brauner static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) 39222a186721SChristian Brauner { 39232a186721SChristian Brauner unsigned int flags = mnt->mnt.mnt_flags; 39242a186721SChristian Brauner 39252a186721SChristian Brauner /* flags to clear */ 39262a186721SChristian Brauner flags &= ~kattr->attr_clr; 39272a186721SChristian Brauner /* flags to raise */ 39282a186721SChristian Brauner flags |= kattr->attr_set; 39292a186721SChristian Brauner 39302a186721SChristian Brauner return flags; 39312a186721SChristian Brauner } 39322a186721SChristian Brauner 39339caccd41SChristian Brauner static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 39349caccd41SChristian Brauner { 39359caccd41SChristian Brauner struct vfsmount *m = &mnt->mnt; 3936bd303368SChristian Brauner struct user_namespace *fs_userns = m->mnt_sb->s_user_ns; 39379caccd41SChristian Brauner 39389caccd41SChristian Brauner if (!kattr->mnt_userns) 39399caccd41SChristian Brauner return 0; 39409caccd41SChristian Brauner 39419caccd41SChristian Brauner /* 3942bd303368SChristian Brauner * Creating an idmapped mount with the filesystem wide idmapping 3943bd303368SChristian Brauner * doesn't make sense so block that. We don't allow mushy semantics. 3944bd303368SChristian Brauner */ 3945bd303368SChristian Brauner if (kattr->mnt_userns == fs_userns) 3946bd303368SChristian Brauner return -EINVAL; 3947bd303368SChristian Brauner 3948bd303368SChristian Brauner /* 39499caccd41SChristian Brauner * Once a mount has been idmapped we don't allow it to change its 39509caccd41SChristian Brauner * mapping. It makes things simpler and callers can just create 39519caccd41SChristian Brauner * another bind-mount they can idmap if they want to. 39529caccd41SChristian Brauner */ 3953bb49e9e7SChristian Brauner if (is_idmapped_mnt(m)) 39549caccd41SChristian Brauner return -EPERM; 39559caccd41SChristian Brauner 39569caccd41SChristian Brauner /* The underlying filesystem doesn't support idmapped mounts yet. */ 39579caccd41SChristian Brauner if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) 39589caccd41SChristian Brauner return -EINVAL; 39599caccd41SChristian Brauner 39609caccd41SChristian Brauner /* We're not controlling the superblock. */ 3961bd303368SChristian Brauner if (!ns_capable(fs_userns, CAP_SYS_ADMIN)) 39629caccd41SChristian Brauner return -EPERM; 39639caccd41SChristian Brauner 39649caccd41SChristian Brauner /* Mount has already been visible in the filesystem hierarchy. */ 39659caccd41SChristian Brauner if (!is_anon_ns(mnt->mnt_ns)) 39669caccd41SChristian Brauner return -EINVAL; 39679caccd41SChristian Brauner 39689caccd41SChristian Brauner return 0; 39699caccd41SChristian Brauner } 39709caccd41SChristian Brauner 39712a186721SChristian Brauner static struct mount *mount_setattr_prepare(struct mount_kattr *kattr, 39722a186721SChristian Brauner struct mount *mnt, int *err) 39732a186721SChristian Brauner { 39742a186721SChristian Brauner struct mount *m = mnt, *last = NULL; 39752a186721SChristian Brauner 39762a186721SChristian Brauner if (!is_mounted(&m->mnt)) { 39772a186721SChristian Brauner *err = -EINVAL; 39782a186721SChristian Brauner goto out; 39792a186721SChristian Brauner } 39802a186721SChristian Brauner 39812a186721SChristian Brauner if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) { 39822a186721SChristian Brauner *err = -EINVAL; 39832a186721SChristian Brauner goto out; 39842a186721SChristian Brauner } 39852a186721SChristian Brauner 39862a186721SChristian Brauner do { 39872a186721SChristian Brauner unsigned int flags; 39882a186721SChristian Brauner 39892a186721SChristian Brauner flags = recalc_flags(kattr, m); 39902a186721SChristian Brauner if (!can_change_locked_flags(m, flags)) { 39912a186721SChristian Brauner *err = -EPERM; 39922a186721SChristian Brauner goto out; 39932a186721SChristian Brauner } 39942a186721SChristian Brauner 39959caccd41SChristian Brauner *err = can_idmap_mount(kattr, m); 39969caccd41SChristian Brauner if (*err) 39979caccd41SChristian Brauner goto out; 39989caccd41SChristian Brauner 39992a186721SChristian Brauner last = m; 40002a186721SChristian Brauner 40012a186721SChristian Brauner if ((kattr->attr_set & MNT_READONLY) && 40022a186721SChristian Brauner !(m->mnt.mnt_flags & MNT_READONLY)) { 40032a186721SChristian Brauner *err = mnt_hold_writers(m); 40042a186721SChristian Brauner if (*err) 40052a186721SChristian Brauner goto out; 40062a186721SChristian Brauner } 40072a186721SChristian Brauner } while (kattr->recurse && (m = next_mnt(m, mnt))); 40082a186721SChristian Brauner 40092a186721SChristian Brauner out: 40102a186721SChristian Brauner return last; 40112a186721SChristian Brauner } 40122a186721SChristian Brauner 40139caccd41SChristian Brauner static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 40149caccd41SChristian Brauner { 4015bd303368SChristian Brauner struct user_namespace *mnt_userns, *old_mnt_userns; 40169caccd41SChristian Brauner 40179caccd41SChristian Brauner if (!kattr->mnt_userns) 40189caccd41SChristian Brauner return; 40199caccd41SChristian Brauner 4020bd303368SChristian Brauner /* 4021bd303368SChristian Brauner * We're the only ones able to change the mount's idmapping. So 4022bd303368SChristian Brauner * mnt->mnt.mnt_userns is stable and we can retrieve it directly. 4023bd303368SChristian Brauner */ 4024bd303368SChristian Brauner old_mnt_userns = mnt->mnt.mnt_userns; 4025bd303368SChristian Brauner 40269caccd41SChristian Brauner mnt_userns = get_user_ns(kattr->mnt_userns); 40279caccd41SChristian Brauner /* Pairs with smp_load_acquire() in mnt_user_ns(). */ 40289caccd41SChristian Brauner smp_store_release(&mnt->mnt.mnt_userns, mnt_userns); 4029bd303368SChristian Brauner 4030bd303368SChristian Brauner /* 4031bd303368SChristian Brauner * If this is an idmapped filesystem drop the reference we've taken 4032bd303368SChristian Brauner * in vfs_create_mount() before. 4033bd303368SChristian Brauner */ 4034bd303368SChristian Brauner if (!initial_idmapping(old_mnt_userns)) 4035bd303368SChristian Brauner put_user_ns(old_mnt_userns); 40369caccd41SChristian Brauner } 40379caccd41SChristian Brauner 40382a186721SChristian Brauner static void mount_setattr_commit(struct mount_kattr *kattr, 40392a186721SChristian Brauner struct mount *mnt, struct mount *last, 40402a186721SChristian Brauner int err) 40412a186721SChristian Brauner { 40422a186721SChristian Brauner struct mount *m = mnt; 40432a186721SChristian Brauner 40442a186721SChristian Brauner do { 40452a186721SChristian Brauner if (!err) { 40462a186721SChristian Brauner unsigned int flags; 40472a186721SChristian Brauner 40489caccd41SChristian Brauner do_idmap_mount(kattr, m); 40492a186721SChristian Brauner flags = recalc_flags(kattr, m); 40502a186721SChristian Brauner WRITE_ONCE(m->mnt.mnt_flags, flags); 40512a186721SChristian Brauner } 40522a186721SChristian Brauner 40532a186721SChristian Brauner /* 40542a186721SChristian Brauner * We either set MNT_READONLY above so make it visible 40552a186721SChristian Brauner * before ~MNT_WRITE_HOLD or we failed to recursively 40562a186721SChristian Brauner * apply mount options. 40572a186721SChristian Brauner */ 40582a186721SChristian Brauner if ((kattr->attr_set & MNT_READONLY) && 40592a186721SChristian Brauner (m->mnt.mnt_flags & MNT_WRITE_HOLD)) 40602a186721SChristian Brauner mnt_unhold_writers(m); 40612a186721SChristian Brauner 40622a186721SChristian Brauner if (!err && kattr->propagation) 40632a186721SChristian Brauner change_mnt_propagation(m, kattr->propagation); 40642a186721SChristian Brauner 40652a186721SChristian Brauner /* 40662a186721SChristian Brauner * On failure, only cleanup until we found the first mount 40672a186721SChristian Brauner * we failed to handle. 40682a186721SChristian Brauner */ 40692a186721SChristian Brauner if (err && m == last) 40702a186721SChristian Brauner break; 40712a186721SChristian Brauner } while (kattr->recurse && (m = next_mnt(m, mnt))); 40722a186721SChristian Brauner 40732a186721SChristian Brauner if (!err) 40742a186721SChristian Brauner touch_mnt_namespace(mnt->mnt_ns); 40752a186721SChristian Brauner } 40762a186721SChristian Brauner 40772a186721SChristian Brauner static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) 40782a186721SChristian Brauner { 40792a186721SChristian Brauner struct mount *mnt = real_mount(path->mnt), *last = NULL; 40802a186721SChristian Brauner int err = 0; 40812a186721SChristian Brauner 40822a186721SChristian Brauner if (path->dentry != mnt->mnt.mnt_root) 40832a186721SChristian Brauner return -EINVAL; 40842a186721SChristian Brauner 40852a186721SChristian Brauner if (kattr->propagation) { 40862a186721SChristian Brauner /* 40872a186721SChristian Brauner * Only take namespace_lock() if we're actually changing 40882a186721SChristian Brauner * propagation. 40892a186721SChristian Brauner */ 40902a186721SChristian Brauner namespace_lock(); 40912a186721SChristian Brauner if (kattr->propagation == MS_SHARED) { 40922a186721SChristian Brauner err = invent_group_ids(mnt, kattr->recurse); 40932a186721SChristian Brauner if (err) { 40942a186721SChristian Brauner namespace_unlock(); 40952a186721SChristian Brauner return err; 40962a186721SChristian Brauner } 40972a186721SChristian Brauner } 40982a186721SChristian Brauner } 40992a186721SChristian Brauner 41002a186721SChristian Brauner lock_mount_hash(); 41012a186721SChristian Brauner 41022a186721SChristian Brauner /* 41032a186721SChristian Brauner * Get the mount tree in a shape where we can change mount 41042a186721SChristian Brauner * properties without failure. 41052a186721SChristian Brauner */ 41062a186721SChristian Brauner last = mount_setattr_prepare(kattr, mnt, &err); 41072a186721SChristian Brauner if (last) /* Commit all changes or revert to the old state. */ 41082a186721SChristian Brauner mount_setattr_commit(kattr, mnt, last, err); 41092a186721SChristian Brauner 41102a186721SChristian Brauner unlock_mount_hash(); 41112a186721SChristian Brauner 41122a186721SChristian Brauner if (kattr->propagation) { 41132a186721SChristian Brauner namespace_unlock(); 41142a186721SChristian Brauner if (err) 41152a186721SChristian Brauner cleanup_group_ids(mnt, NULL); 41162a186721SChristian Brauner } 41172a186721SChristian Brauner 41182a186721SChristian Brauner return err; 41192a186721SChristian Brauner } 41202a186721SChristian Brauner 41219caccd41SChristian Brauner static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, 41229caccd41SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 41239caccd41SChristian Brauner { 41249caccd41SChristian Brauner int err = 0; 41259caccd41SChristian Brauner struct ns_common *ns; 41269caccd41SChristian Brauner struct user_namespace *mnt_userns; 41279caccd41SChristian Brauner struct file *file; 41289caccd41SChristian Brauner 41299caccd41SChristian Brauner if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) 41309caccd41SChristian Brauner return 0; 41319caccd41SChristian Brauner 41329caccd41SChristian Brauner /* 41339caccd41SChristian Brauner * We currently do not support clearing an idmapped mount. If this ever 41349caccd41SChristian Brauner * is a use-case we can revisit this but for now let's keep it simple 41359caccd41SChristian Brauner * and not allow it. 41369caccd41SChristian Brauner */ 41379caccd41SChristian Brauner if (attr->attr_clr & MOUNT_ATTR_IDMAP) 41389caccd41SChristian Brauner return -EINVAL; 41399caccd41SChristian Brauner 41409caccd41SChristian Brauner if (attr->userns_fd > INT_MAX) 41419caccd41SChristian Brauner return -EINVAL; 41429caccd41SChristian Brauner 41439caccd41SChristian Brauner file = fget(attr->userns_fd); 41449caccd41SChristian Brauner if (!file) 41459caccd41SChristian Brauner return -EBADF; 41469caccd41SChristian Brauner 41479caccd41SChristian Brauner if (!proc_ns_file(file)) { 41489caccd41SChristian Brauner err = -EINVAL; 41499caccd41SChristian Brauner goto out_fput; 41509caccd41SChristian Brauner } 41519caccd41SChristian Brauner 41529caccd41SChristian Brauner ns = get_proc_ns(file_inode(file)); 41539caccd41SChristian Brauner if (ns->ops->type != CLONE_NEWUSER) { 41549caccd41SChristian Brauner err = -EINVAL; 41559caccd41SChristian Brauner goto out_fput; 41569caccd41SChristian Brauner } 41579caccd41SChristian Brauner 41589caccd41SChristian Brauner /* 4159bd303368SChristian Brauner * The initial idmapping cannot be used to create an idmapped 4160bd303368SChristian Brauner * mount. We use the initial idmapping as an indicator of a mount 4161bd303368SChristian Brauner * that is not idmapped. It can simply be passed into helpers that 4162bd303368SChristian Brauner * are aware of idmapped mounts as a convenient shortcut. A user 4163bd303368SChristian Brauner * can just create a dedicated identity mapping to achieve the same 4164bd303368SChristian Brauner * result. 41659caccd41SChristian Brauner */ 41669caccd41SChristian Brauner mnt_userns = container_of(ns, struct user_namespace, ns); 4167bd303368SChristian Brauner if (initial_idmapping(mnt_userns)) { 41689caccd41SChristian Brauner err = -EPERM; 41699caccd41SChristian Brauner goto out_fput; 41709caccd41SChristian Brauner } 41719caccd41SChristian Brauner kattr->mnt_userns = get_user_ns(mnt_userns); 41729caccd41SChristian Brauner 41739caccd41SChristian Brauner out_fput: 41749caccd41SChristian Brauner fput(file); 41759caccd41SChristian Brauner return err; 41769caccd41SChristian Brauner } 41779caccd41SChristian Brauner 41789caccd41SChristian Brauner static int build_mount_kattr(const struct mount_attr *attr, size_t usize, 41792a186721SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 41802a186721SChristian Brauner { 41812a186721SChristian Brauner unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 41822a186721SChristian Brauner 41832a186721SChristian Brauner if (flags & AT_NO_AUTOMOUNT) 41842a186721SChristian Brauner lookup_flags &= ~LOOKUP_AUTOMOUNT; 41852a186721SChristian Brauner if (flags & AT_SYMLINK_NOFOLLOW) 41862a186721SChristian Brauner lookup_flags &= ~LOOKUP_FOLLOW; 41872a186721SChristian Brauner if (flags & AT_EMPTY_PATH) 41882a186721SChristian Brauner lookup_flags |= LOOKUP_EMPTY; 41892a186721SChristian Brauner 41902a186721SChristian Brauner *kattr = (struct mount_kattr) { 41912a186721SChristian Brauner .lookup_flags = lookup_flags, 41922a186721SChristian Brauner .recurse = !!(flags & AT_RECURSIVE), 41932a186721SChristian Brauner }; 41942a186721SChristian Brauner 41952a186721SChristian Brauner if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) 41962a186721SChristian Brauner return -EINVAL; 41972a186721SChristian Brauner if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) 41982a186721SChristian Brauner return -EINVAL; 41992a186721SChristian Brauner kattr->propagation = attr->propagation; 42002a186721SChristian Brauner 42012a186721SChristian Brauner if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) 42022a186721SChristian Brauner return -EINVAL; 42032a186721SChristian Brauner 42042a186721SChristian Brauner kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); 42052a186721SChristian Brauner kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); 42062a186721SChristian Brauner 42072a186721SChristian Brauner /* 42082a186721SChristian Brauner * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, 42092a186721SChristian Brauner * users wanting to transition to a different atime setting cannot 42102a186721SChristian Brauner * simply specify the atime setting in @attr_set, but must also 42112a186721SChristian Brauner * specify MOUNT_ATTR__ATIME in the @attr_clr field. 42122a186721SChristian Brauner * So ensure that MOUNT_ATTR__ATIME can't be partially set in 42132a186721SChristian Brauner * @attr_clr and that @attr_set can't have any atime bits set if 42142a186721SChristian Brauner * MOUNT_ATTR__ATIME isn't set in @attr_clr. 42152a186721SChristian Brauner */ 42162a186721SChristian Brauner if (attr->attr_clr & MOUNT_ATTR__ATIME) { 42172a186721SChristian Brauner if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) 42182a186721SChristian Brauner return -EINVAL; 42192a186721SChristian Brauner 42202a186721SChristian Brauner /* 42212a186721SChristian Brauner * Clear all previous time settings as they are mutually 42222a186721SChristian Brauner * exclusive. 42232a186721SChristian Brauner */ 42242a186721SChristian Brauner kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; 42252a186721SChristian Brauner switch (attr->attr_set & MOUNT_ATTR__ATIME) { 42262a186721SChristian Brauner case MOUNT_ATTR_RELATIME: 42272a186721SChristian Brauner kattr->attr_set |= MNT_RELATIME; 42282a186721SChristian Brauner break; 42292a186721SChristian Brauner case MOUNT_ATTR_NOATIME: 42302a186721SChristian Brauner kattr->attr_set |= MNT_NOATIME; 42312a186721SChristian Brauner break; 42322a186721SChristian Brauner case MOUNT_ATTR_STRICTATIME: 42332a186721SChristian Brauner break; 42342a186721SChristian Brauner default: 42352a186721SChristian Brauner return -EINVAL; 42362a186721SChristian Brauner } 42372a186721SChristian Brauner } else { 42382a186721SChristian Brauner if (attr->attr_set & MOUNT_ATTR__ATIME) 42392a186721SChristian Brauner return -EINVAL; 42402a186721SChristian Brauner } 42412a186721SChristian Brauner 42429caccd41SChristian Brauner return build_mount_idmapped(attr, usize, kattr, flags); 42439caccd41SChristian Brauner } 42449caccd41SChristian Brauner 42459caccd41SChristian Brauner static void finish_mount_kattr(struct mount_kattr *kattr) 42469caccd41SChristian Brauner { 42479caccd41SChristian Brauner put_user_ns(kattr->mnt_userns); 42489caccd41SChristian Brauner kattr->mnt_userns = NULL; 42492a186721SChristian Brauner } 42502a186721SChristian Brauner 42512a186721SChristian Brauner SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, 42522a186721SChristian Brauner unsigned int, flags, struct mount_attr __user *, uattr, 42532a186721SChristian Brauner size_t, usize) 42542a186721SChristian Brauner { 42552a186721SChristian Brauner int err; 42562a186721SChristian Brauner struct path target; 42572a186721SChristian Brauner struct mount_attr attr; 42582a186721SChristian Brauner struct mount_kattr kattr; 42592a186721SChristian Brauner 42602a186721SChristian Brauner BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); 42612a186721SChristian Brauner 42622a186721SChristian Brauner if (flags & ~(AT_EMPTY_PATH | 42632a186721SChristian Brauner AT_RECURSIVE | 42642a186721SChristian Brauner AT_SYMLINK_NOFOLLOW | 42652a186721SChristian Brauner AT_NO_AUTOMOUNT)) 42662a186721SChristian Brauner return -EINVAL; 42672a186721SChristian Brauner 42682a186721SChristian Brauner if (unlikely(usize > PAGE_SIZE)) 42692a186721SChristian Brauner return -E2BIG; 42702a186721SChristian Brauner if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) 42712a186721SChristian Brauner return -EINVAL; 42722a186721SChristian Brauner 42732a186721SChristian Brauner if (!may_mount()) 42742a186721SChristian Brauner return -EPERM; 42752a186721SChristian Brauner 42762a186721SChristian Brauner err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); 42772a186721SChristian Brauner if (err) 42782a186721SChristian Brauner return err; 42792a186721SChristian Brauner 42802a186721SChristian Brauner /* Don't bother walking through the mounts if this is a nop. */ 42812a186721SChristian Brauner if (attr.attr_set == 0 && 42822a186721SChristian Brauner attr.attr_clr == 0 && 42832a186721SChristian Brauner attr.propagation == 0) 42842a186721SChristian Brauner return 0; 42852a186721SChristian Brauner 42869caccd41SChristian Brauner err = build_mount_kattr(&attr, usize, &kattr, flags); 42872a186721SChristian Brauner if (err) 42882a186721SChristian Brauner return err; 42892a186721SChristian Brauner 42902a186721SChristian Brauner err = user_path_at(dfd, path, kattr.lookup_flags, &target); 4291012e3322SChristian Brauner if (!err) { 42922a186721SChristian Brauner err = do_mount_setattr(&target, &kattr); 42932a186721SChristian Brauner path_put(&target); 4294012e3322SChristian Brauner } 4295012e3322SChristian Brauner finish_mount_kattr(&kattr); 42962a186721SChristian Brauner return err; 42972a186721SChristian Brauner } 42982a186721SChristian Brauner 42991da177e4SLinus Torvalds static void __init init_mount_tree(void) 43001da177e4SLinus Torvalds { 43011da177e4SLinus Torvalds struct vfsmount *mnt; 430274e83122SAl Viro struct mount *m; 43036b3286edSKirill Korotaev struct mnt_namespace *ns; 4304ac748a09SJan Blunck struct path root; 43051da177e4SLinus Torvalds 4306fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 43071da177e4SLinus Torvalds if (IS_ERR(mnt)) 43081da177e4SLinus Torvalds panic("Can't create rootfs"); 4309b3e19d92SNick Piggin 431074e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 43113b22edc5STrond Myklebust if (IS_ERR(ns)) 43121da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 431374e83122SAl Viro m = real_mount(mnt); 431474e83122SAl Viro m->mnt_ns = ns; 431574e83122SAl Viro ns->root = m; 431674e83122SAl Viro ns->mounts = 1; 431774e83122SAl Viro list_add(&m->mnt_list, &ns->list); 43186b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 43196b3286edSKirill Korotaev get_mnt_ns(ns); 43201da177e4SLinus Torvalds 4321be08d6d2SAl Viro root.mnt = mnt; 4322be08d6d2SAl Viro root.dentry = mnt->mnt_root; 4323da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 4324ac748a09SJan Blunck 4325ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 4326ac748a09SJan Blunck set_fs_root(current->fs, &root); 43271da177e4SLinus Torvalds } 43281da177e4SLinus Torvalds 432974bf17cfSDenis Cheng void __init mnt_init(void) 43301da177e4SLinus Torvalds { 433115a67dd8SRandy Dunlap int err; 43321da177e4SLinus Torvalds 43337d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 433479f6540bSVasily Averin 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); 43351da177e4SLinus Torvalds 43360818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 433738129a13SAl Viro sizeof(struct hlist_head), 43380818bf27SAl Viro mhash_entries, 19, 43393d375d78SPavel Tatashin HASH_ZERO, 43400818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 43410818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 43420818bf27SAl Viro sizeof(struct hlist_head), 43430818bf27SAl Viro mphash_entries, 19, 43443d375d78SPavel Tatashin HASH_ZERO, 43450818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 43461da177e4SLinus Torvalds 434784d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 43481da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 43491da177e4SLinus Torvalds 43504b93dc9bSTejun Heo kernfs_init(); 43514b93dc9bSTejun Heo 435215a67dd8SRandy Dunlap err = sysfs_init(); 435315a67dd8SRandy Dunlap if (err) 435415a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 43558e24eea7SHarvey Harrison __func__, err); 435600d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 435700d26666SGreg Kroah-Hartman if (!fs_kobj) 43588e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 4359037f11b4SAl Viro shmem_init(); 43601da177e4SLinus Torvalds init_rootfs(); 43611da177e4SLinus Torvalds init_mount_tree(); 43621da177e4SLinus Torvalds } 43631da177e4SLinus Torvalds 4364616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 43651da177e4SLinus Torvalds { 43661a7b8969SKirill Tkhai if (!refcount_dec_and_test(&ns->ns.count)) 4367616511d0STrond Myklebust return; 43687b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 4369771b1371SEric W. Biederman free_mnt_ns(ns); 43701da177e4SLinus Torvalds } 43719d412a43SAl Viro 4372d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 43739d412a43SAl Viro { 4374423e0ab0STim Chen struct vfsmount *mnt; 4375d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 4376423e0ab0STim Chen if (!IS_ERR(mnt)) { 4377423e0ab0STim Chen /* 4378423e0ab0STim Chen * it is a longterm mount, don't release mnt until 4379423e0ab0STim Chen * we unmount before file sys is unregistered 4380423e0ab0STim Chen */ 4381f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 4382423e0ab0STim Chen } 4383423e0ab0STim Chen return mnt; 43849d412a43SAl Viro } 4385d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 4386423e0ab0STim Chen 4387423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 4388423e0ab0STim Chen { 4389423e0ab0STim Chen /* release long term mount so mount point can be released */ 4390423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 4391f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 439248a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 4393423e0ab0STim Chen mntput(mnt); 4394423e0ab0STim Chen } 4395423e0ab0STim Chen } 4396423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 439702125a82SAl Viro 4398df820f8dSMiklos Szeredi void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) 4399df820f8dSMiklos Szeredi { 4400df820f8dSMiklos Szeredi unsigned int i; 4401df820f8dSMiklos Szeredi 4402df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4403df820f8dSMiklos Szeredi if (mnt[i]) 4404df820f8dSMiklos Szeredi real_mount(mnt[i])->mnt_ns = NULL; 4405df820f8dSMiklos Szeredi synchronize_rcu_expedited(); 4406df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4407df820f8dSMiklos Szeredi mntput(mnt[i]); 4408df820f8dSMiklos Szeredi } 4409df820f8dSMiklos Szeredi EXPORT_SYMBOL(kern_unmount_array); 4410df820f8dSMiklos Szeredi 441102125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 441202125a82SAl Viro { 4413143c8c91SAl Viro return check_mnt(real_mount(mnt)); 441402125a82SAl Viro } 44158823c079SEric W. Biederman 44163151527eSEric W. Biederman bool current_chrooted(void) 44173151527eSEric W. Biederman { 44183151527eSEric W. Biederman /* Does the current process have a non-standard root */ 44193151527eSEric W. Biederman struct path ns_root; 44203151527eSEric W. Biederman struct path fs_root; 44213151527eSEric W. Biederman bool chrooted; 44223151527eSEric W. Biederman 44233151527eSEric W. Biederman /* Find the namespace root */ 44243151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 44253151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 44263151527eSEric W. Biederman path_get(&ns_root); 44273151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 44283151527eSEric W. Biederman ; 44293151527eSEric W. Biederman 44303151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 44313151527eSEric W. Biederman 44323151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 44333151527eSEric W. Biederman 44343151527eSEric W. Biederman path_put(&fs_root); 44353151527eSEric W. Biederman path_put(&ns_root); 44363151527eSEric W. Biederman 44373151527eSEric W. Biederman return chrooted; 44383151527eSEric W. Biederman } 44393151527eSEric W. Biederman 4440132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 4441132e4608SDavid Howells const struct super_block *sb, 44428654df4eSEric W. Biederman int *new_mnt_flags) 444387a8ebd6SEric W. Biederman { 44448c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 444587a8ebd6SEric W. Biederman struct mount *mnt; 4446e51db735SEric W. Biederman bool visible = false; 444787a8ebd6SEric W. Biederman 444844bb4385SAl Viro down_read(&namespace_sem); 44499f6c61f9SMiklos Szeredi lock_ns_list(ns); 445087a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 4451e51db735SEric W. Biederman struct mount *child; 445277b1a97dSEric W. Biederman int mnt_flags; 445377b1a97dSEric W. Biederman 44549f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 44559f6c61f9SMiklos Szeredi continue; 44569f6c61f9SMiklos Szeredi 4457132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 4458e51db735SEric W. Biederman continue; 4459e51db735SEric W. Biederman 44607e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 44617e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 44627e96c1b0SEric W. Biederman */ 44637e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 44647e96c1b0SEric W. Biederman continue; 44657e96c1b0SEric W. Biederman 4466a1935c17SEric W. Biederman /* A local view of the mount flags */ 446777b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 446877b1a97dSEric W. Biederman 4469695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 4470bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 4471695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 4472695e9df0SEric W. Biederman 44738c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 44748c6cf9ccSEric W. Biederman * than the proposed new mount. 44758c6cf9ccSEric W. Biederman */ 447677b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 44778c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 44788c6cf9ccSEric W. Biederman continue; 447977b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 448077b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 44818c6cf9ccSEric W. Biederman continue; 44828c6cf9ccSEric W. Biederman 4483ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 4484ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 4485ceeb0e5dSEric W. Biederman * empty directories. 4486e51db735SEric W. Biederman */ 4487e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 4488e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 4489ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 4490d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 4491ceeb0e5dSEric W. Biederman continue; 44927236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 44937236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 4494e51db735SEric W. Biederman goto next; 449587a8ebd6SEric W. Biederman } 44968c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 449777b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 44988c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 4499e51db735SEric W. Biederman visible = true; 4500e51db735SEric W. Biederman goto found; 4501e51db735SEric W. Biederman next: ; 450287a8ebd6SEric W. Biederman } 4503e51db735SEric W. Biederman found: 45049f6c61f9SMiklos Szeredi unlock_ns_list(ns); 450544bb4385SAl Viro up_read(&namespace_sem); 4506e51db735SEric W. Biederman return visible; 450787a8ebd6SEric W. Biederman } 450887a8ebd6SEric W. Biederman 4509132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 45108654df4eSEric W. Biederman { 4511a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 45128654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 45138654df4eSEric W. Biederman unsigned long s_iflags; 45148654df4eSEric W. Biederman 45158654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 45168654df4eSEric W. Biederman return false; 45178654df4eSEric W. Biederman 45188654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 4519132e4608SDavid Howells s_iflags = sb->s_iflags; 45208654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 45218654df4eSEric W. Biederman return false; 45228654df4eSEric W. Biederman 4523a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 4524a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 4525a1935c17SEric W. Biederman required_iflags); 4526a1935c17SEric W. Biederman return true; 4527a1935c17SEric W. Biederman } 4528a1935c17SEric W. Biederman 4529132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 45308654df4eSEric W. Biederman } 45318654df4eSEric W. Biederman 4532380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 4533380cf5baSAndy Lutomirski { 4534380cf5baSAndy Lutomirski /* 4535380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 4536380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 4537380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 4538380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 4539380cf5baSAndy Lutomirski * in other namespaces. 4540380cf5baSAndy Lutomirski */ 4541380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 4542380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 4543380cf5baSAndy Lutomirski } 4544380cf5baSAndy Lutomirski 454564964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 45468823c079SEric W. Biederman { 454758be2825SAl Viro struct ns_common *ns = NULL; 45488823c079SEric W. Biederman struct nsproxy *nsproxy; 45498823c079SEric W. Biederman 4550728dba3aSEric W. Biederman task_lock(task); 4551728dba3aSEric W. Biederman nsproxy = task->nsproxy; 45528823c079SEric W. Biederman if (nsproxy) { 455358be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 455458be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 45558823c079SEric W. Biederman } 4556728dba3aSEric W. Biederman task_unlock(task); 45578823c079SEric W. Biederman 45588823c079SEric W. Biederman return ns; 45598823c079SEric W. Biederman } 45608823c079SEric W. Biederman 456164964528SAl Viro static void mntns_put(struct ns_common *ns) 45628823c079SEric W. Biederman { 456358be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 45648823c079SEric W. Biederman } 45658823c079SEric W. Biederman 4566f2a8d52eSChristian Brauner static int mntns_install(struct nsset *nsset, struct ns_common *ns) 45678823c079SEric W. Biederman { 4568f2a8d52eSChristian Brauner struct nsproxy *nsproxy = nsset->nsproxy; 4569f2a8d52eSChristian Brauner struct fs_struct *fs = nsset->fs; 45704f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 4571f2a8d52eSChristian Brauner struct user_namespace *user_ns = nsset->cred->user_ns; 45728823c079SEric W. Biederman struct path root; 45734f757f3cSAl Viro int err; 45748823c079SEric W. Biederman 45750c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 4576f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_CHROOT) || 4577f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_ADMIN)) 4578ae11e0f1SZhao Hongjiang return -EPERM; 45798823c079SEric W. Biederman 458074e83122SAl Viro if (is_anon_ns(mnt_ns)) 458174e83122SAl Viro return -EINVAL; 458274e83122SAl Viro 45838823c079SEric W. Biederman if (fs->users != 1) 45848823c079SEric W. Biederman return -EINVAL; 45858823c079SEric W. Biederman 45868823c079SEric W. Biederman get_mnt_ns(mnt_ns); 45874f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 45888823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 45898823c079SEric W. Biederman 45908823c079SEric W. Biederman /* Find the root */ 45914f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 45924f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 45934f757f3cSAl Viro if (err) { 45944f757f3cSAl Viro /* revert to old namespace */ 45954f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 45964f757f3cSAl Viro put_mnt_ns(mnt_ns); 45974f757f3cSAl Viro return err; 45984f757f3cSAl Viro } 45998823c079SEric W. Biederman 46004068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 46014068367cSAndrei Vagin 46028823c079SEric W. Biederman /* Update the pwd and root */ 46038823c079SEric W. Biederman set_fs_pwd(fs, &root); 46048823c079SEric W. Biederman set_fs_root(fs, &root); 46058823c079SEric W. Biederman 46068823c079SEric W. Biederman path_put(&root); 46078823c079SEric W. Biederman return 0; 46088823c079SEric W. Biederman } 46098823c079SEric W. Biederman 4610bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 4611bcac25a5SAndrey Vagin { 4612bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 4613bcac25a5SAndrey Vagin } 4614bcac25a5SAndrey Vagin 46158823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 46168823c079SEric W. Biederman .name = "mnt", 46178823c079SEric W. Biederman .type = CLONE_NEWNS, 46188823c079SEric W. Biederman .get = mntns_get, 46198823c079SEric W. Biederman .put = mntns_put, 46208823c079SEric W. Biederman .install = mntns_install, 4621bcac25a5SAndrey Vagin .owner = mntns_owner, 46228823c079SEric W. Biederman }; 4623ab171b95SLuis Chamberlain 4624ab171b95SLuis Chamberlain #ifdef CONFIG_SYSCTL 4625ab171b95SLuis Chamberlain static struct ctl_table fs_namespace_sysctls[] = { 4626ab171b95SLuis Chamberlain { 4627ab171b95SLuis Chamberlain .procname = "mount-max", 4628ab171b95SLuis Chamberlain .data = &sysctl_mount_max, 4629ab171b95SLuis Chamberlain .maxlen = sizeof(unsigned int), 4630ab171b95SLuis Chamberlain .mode = 0644, 4631ab171b95SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 4632ab171b95SLuis Chamberlain .extra1 = SYSCTL_ONE, 4633ab171b95SLuis Chamberlain }, 4634ab171b95SLuis Chamberlain { } 4635ab171b95SLuis Chamberlain }; 4636ab171b95SLuis Chamberlain 4637ab171b95SLuis Chamberlain static int __init init_fs_namespace_sysctls(void) 4638ab171b95SLuis Chamberlain { 4639ab171b95SLuis Chamberlain register_sysctl_init("fs", fs_namespace_sysctls); 4640ab171b95SLuis Chamberlain return 0; 4641ab171b95SLuis Chamberlain } 4642ab171b95SLuis Chamberlain fs_initcall(init_fs_namespace_sysctls); 4643ab171b95SLuis Chamberlain 4644ab171b95SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 4645