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 472538f4f02SChristian Brauner /** 473538f4f02SChristian Brauner * mnt_hold_writers - prevent write access to the given mount 474538f4f02SChristian Brauner * @mnt: mnt to prevent write access to 475538f4f02SChristian Brauner * 476538f4f02SChristian Brauner * Prevents write access to @mnt if there are no active writers for @mnt. 477538f4f02SChristian Brauner * This function needs to be called and return successfully before changing 478538f4f02SChristian Brauner * properties of @mnt that need to remain stable for callers with write access 479538f4f02SChristian Brauner * to @mnt. 480538f4f02SChristian Brauner * 481538f4f02SChristian Brauner * After this functions has been called successfully callers must pair it with 482538f4f02SChristian Brauner * a call to mnt_unhold_writers() in order to stop preventing write access to 483538f4f02SChristian Brauner * @mnt. 484538f4f02SChristian Brauner * 485538f4f02SChristian Brauner * Context: This function expects lock_mount_hash() to be held serializing 486538f4f02SChristian Brauner * setting MNT_WRITE_HOLD. 487538f4f02SChristian Brauner * Return: On success 0 is returned. 488538f4f02SChristian Brauner * On error, -EBUSY is returned. 489538f4f02SChristian Brauner */ 490fbdc2f6cSChristian Brauner static inline int mnt_hold_writers(struct mount *mnt) 4918366025eSDave Hansen { 49283adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 493d3ef3d73Snpiggin@suse.de /* 494d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 495d3ef3d73Snpiggin@suse.de * should be visible before we do. 496d3ef3d73Snpiggin@suse.de */ 497d3ef3d73Snpiggin@suse.de smp_mb(); 498d3ef3d73Snpiggin@suse.de 499d3ef3d73Snpiggin@suse.de /* 500d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 501d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 502d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 503d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 504d3ef3d73Snpiggin@suse.de * 505d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 506d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 507d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 508d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 509d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 510d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 511d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 512d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 513d3ef3d73Snpiggin@suse.de * we're counting up here. 514d3ef3d73Snpiggin@suse.de */ 515c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 516fbdc2f6cSChristian Brauner return -EBUSY; 517fbdc2f6cSChristian Brauner 518fbdc2f6cSChristian Brauner return 0; 519fbdc2f6cSChristian Brauner } 520fbdc2f6cSChristian Brauner 521538f4f02SChristian Brauner /** 522538f4f02SChristian Brauner * mnt_unhold_writers - stop preventing write access to the given mount 523538f4f02SChristian Brauner * @mnt: mnt to stop preventing write access to 524538f4f02SChristian Brauner * 525538f4f02SChristian Brauner * Stop preventing write access to @mnt allowing callers to gain write access 526538f4f02SChristian Brauner * to @mnt again. 527538f4f02SChristian Brauner * 528538f4f02SChristian Brauner * This function can only be called after a successful call to 529538f4f02SChristian Brauner * mnt_hold_writers(). 530538f4f02SChristian Brauner * 531538f4f02SChristian Brauner * Context: This function expects lock_mount_hash() to be held. 532538f4f02SChristian Brauner */ 533fbdc2f6cSChristian Brauner static inline void mnt_unhold_writers(struct mount *mnt) 534fbdc2f6cSChristian Brauner { 535d3ef3d73Snpiggin@suse.de /* 536d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 537d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 538d3ef3d73Snpiggin@suse.de */ 539d3ef3d73Snpiggin@suse.de smp_wmb(); 54083adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 541fbdc2f6cSChristian Brauner } 542fbdc2f6cSChristian Brauner 543fbdc2f6cSChristian Brauner static int mnt_make_readonly(struct mount *mnt) 544fbdc2f6cSChristian Brauner { 545fbdc2f6cSChristian Brauner int ret; 546fbdc2f6cSChristian Brauner 547fbdc2f6cSChristian Brauner ret = mnt_hold_writers(mnt); 548fbdc2f6cSChristian Brauner if (!ret) 549fbdc2f6cSChristian Brauner mnt->mnt.mnt_flags |= MNT_READONLY; 550fbdc2f6cSChristian Brauner mnt_unhold_writers(mnt); 5513d733633SDave Hansen return ret; 5523d733633SDave Hansen } 5538366025eSDave Hansen 5544ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5554ed5e82fSMiklos Szeredi { 5564ed5e82fSMiklos Szeredi struct mount *mnt; 5574ed5e82fSMiklos Szeredi int err = 0; 5584ed5e82fSMiklos Szeredi 5598e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5608e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5618e8b8796SMiklos Szeredi return -EBUSY; 5628e8b8796SMiklos Szeredi 563719ea2fbSAl Viro lock_mount_hash(); 5644ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5654ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5664ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5674ed5e82fSMiklos Szeredi smp_mb(); 5684ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5694ed5e82fSMiklos Szeredi err = -EBUSY; 5704ed5e82fSMiklos Szeredi break; 5714ed5e82fSMiklos Szeredi } 5724ed5e82fSMiklos Szeredi } 5734ed5e82fSMiklos Szeredi } 5748e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5758e8b8796SMiklos Szeredi err = -EBUSY; 5768e8b8796SMiklos Szeredi 5774ed5e82fSMiklos Szeredi if (!err) { 5784ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5794ed5e82fSMiklos Szeredi smp_wmb(); 5804ed5e82fSMiklos Szeredi } 5814ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5824ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5834ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5844ed5e82fSMiklos Szeredi } 585719ea2fbSAl Viro unlock_mount_hash(); 5864ed5e82fSMiklos Szeredi 5874ed5e82fSMiklos Szeredi return err; 5884ed5e82fSMiklos Szeredi } 5894ed5e82fSMiklos Szeredi 590b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5911da177e4SLinus Torvalds { 592a6435940SChristian Brauner struct user_namespace *mnt_userns; 593a6435940SChristian Brauner 594a6435940SChristian Brauner mnt_userns = mnt_user_ns(&mnt->mnt); 595bd303368SChristian Brauner if (!initial_idmapping(mnt_userns)) 596a6435940SChristian Brauner put_user_ns(mnt_userns); 597fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 598d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 59968e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 600d3ef3d73Snpiggin@suse.de #endif 601b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 6021da177e4SLinus Torvalds } 6031da177e4SLinus Torvalds 6048ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 6058ffcb32eSDavid Howells { 6068ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 6078ffcb32eSDavid Howells } 6088ffcb32eSDavid Howells 60948a066e7SAl Viro /* call under rcu_read_lock */ 610294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 61148a066e7SAl Viro { 61248a066e7SAl Viro struct mount *mnt; 61348a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 614294d71ffSAl Viro return 1; 61548a066e7SAl Viro if (bastard == NULL) 616294d71ffSAl Viro return 0; 61748a066e7SAl Viro mnt = real_mount(bastard); 61848a066e7SAl Viro mnt_add_count(mnt, 1); 619119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 62048a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 621294d71ffSAl Viro return 0; 62248a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 62348a066e7SAl Viro mnt_add_count(mnt, -1); 624294d71ffSAl Viro return 1; 62548a066e7SAl Viro } 626119e1ef8SAl Viro lock_mount_hash(); 627119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 628119e1ef8SAl Viro mnt_add_count(mnt, -1); 629119e1ef8SAl Viro unlock_mount_hash(); 630119e1ef8SAl Viro return 1; 631119e1ef8SAl Viro } 632119e1ef8SAl Viro unlock_mount_hash(); 633119e1ef8SAl Viro /* caller will mntput() */ 634294d71ffSAl Viro return -1; 635294d71ffSAl Viro } 636294d71ffSAl Viro 637294d71ffSAl Viro /* call under rcu_read_lock */ 638294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 639294d71ffSAl Viro { 640294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 641294d71ffSAl Viro if (likely(!res)) 642294d71ffSAl Viro return true; 643294d71ffSAl Viro if (unlikely(res < 0)) { 64448a066e7SAl Viro rcu_read_unlock(); 64548a066e7SAl Viro mntput(bastard); 64648a066e7SAl Viro rcu_read_lock(); 647294d71ffSAl Viro } 64848a066e7SAl Viro return false; 64948a066e7SAl Viro } 65048a066e7SAl Viro 6511da177e4SLinus Torvalds /* 652474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 65348a066e7SAl Viro * call under rcu_read_lock() 6541da177e4SLinus Torvalds */ 655474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6561da177e4SLinus Torvalds { 65738129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 658474279dcSAl Viro struct mount *p; 6591da177e4SLinus Torvalds 66038129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 661474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 662474279dcSAl Viro return p; 663474279dcSAl Viro return NULL; 6641da177e4SLinus Torvalds } 665474279dcSAl Viro 666474279dcSAl Viro /* 667f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 668f015f126SDavid Howells * 669f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 670f015f126SDavid Howells * following mounts: 671f015f126SDavid Howells * 672f015f126SDavid Howells * mount /dev/sda1 /mnt 673f015f126SDavid Howells * mount /dev/sda2 /mnt 674f015f126SDavid Howells * mount /dev/sda3 /mnt 675f015f126SDavid Howells * 676f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 677f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 678f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 679f015f126SDavid Howells * 680f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 681a05964f3SRam Pai */ 682ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 683a05964f3SRam Pai { 684c7105365SAl Viro struct mount *child_mnt; 68548a066e7SAl Viro struct vfsmount *m; 68648a066e7SAl Viro unsigned seq; 68799b7db7bSNick Piggin 68848a066e7SAl Viro rcu_read_lock(); 68948a066e7SAl Viro do { 69048a066e7SAl Viro seq = read_seqbegin(&mount_lock); 691474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 69248a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 69348a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 69448a066e7SAl Viro rcu_read_unlock(); 69548a066e7SAl Viro return m; 696a05964f3SRam Pai } 697a05964f3SRam Pai 6989f6c61f9SMiklos Szeredi static inline void lock_ns_list(struct mnt_namespace *ns) 6999f6c61f9SMiklos Szeredi { 7009f6c61f9SMiklos Szeredi spin_lock(&ns->ns_lock); 7019f6c61f9SMiklos Szeredi } 7029f6c61f9SMiklos Szeredi 7039f6c61f9SMiklos Szeredi static inline void unlock_ns_list(struct mnt_namespace *ns) 7049f6c61f9SMiklos Szeredi { 7059f6c61f9SMiklos Szeredi spin_unlock(&ns->ns_lock); 7069f6c61f9SMiklos Szeredi } 7079f6c61f9SMiklos Szeredi 7089f6c61f9SMiklos Szeredi static inline bool mnt_is_cursor(struct mount *mnt) 7099f6c61f9SMiklos Szeredi { 7109f6c61f9SMiklos Szeredi return mnt->mnt.mnt_flags & MNT_CURSOR; 7119f6c61f9SMiklos Szeredi } 7129f6c61f9SMiklos Szeredi 7137af1364fSEric W. Biederman /* 7147af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 7157af1364fSEric W. Biederman * current mount namespace. 7167af1364fSEric W. Biederman * 7177af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 7187af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 7197af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 7207af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 7217af1364fSEric W. Biederman * is a mountpoint. 7227af1364fSEric W. Biederman * 7237af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 7247af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 7257af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 7267af1364fSEric W. Biederman * parent mount. 7277af1364fSEric W. Biederman */ 7287af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 7297af1364fSEric W. Biederman { 7307af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7317af1364fSEric W. Biederman struct mount *mnt; 7327af1364fSEric W. Biederman bool is_covered = false; 7337af1364fSEric W. Biederman 7347af1364fSEric W. Biederman down_read(&namespace_sem); 7359f6c61f9SMiklos Szeredi lock_ns_list(ns); 7367af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7379f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 7389f6c61f9SMiklos Szeredi continue; 7397af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7407af1364fSEric W. Biederman if (is_covered) 7417af1364fSEric W. Biederman break; 7427af1364fSEric W. Biederman } 7439f6c61f9SMiklos Szeredi unlock_ns_list(ns); 7447af1364fSEric W. Biederman up_read(&namespace_sem); 7455ad05cc8SNikolay Borisov 7467af1364fSEric W. Biederman return is_covered; 7477af1364fSEric W. Biederman } 7487af1364fSEric W. Biederman 749e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 75084d17192SAl Viro { 7510818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 75284d17192SAl Viro struct mountpoint *mp; 75384d17192SAl Viro 7540818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 75584d17192SAl Viro if (mp->m_dentry == dentry) { 75684d17192SAl Viro mp->m_count++; 75784d17192SAl Viro return mp; 75884d17192SAl Viro } 75984d17192SAl Viro } 760e2dfa935SEric W. Biederman return NULL; 761e2dfa935SEric W. Biederman } 762e2dfa935SEric W. Biederman 7633895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 764e2dfa935SEric W. Biederman { 7653895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 766e2dfa935SEric W. Biederman int ret; 76784d17192SAl Viro 7683895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7691e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7701e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7711e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7723895dbf8SEric W. Biederman mountpoint: 7733895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7743895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7753895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7763895dbf8SEric W. Biederman if (mp) 7773895dbf8SEric W. Biederman goto done; 77884d17192SAl Viro } 779eed81007SMiklos Szeredi 7803895dbf8SEric W. Biederman if (!new) 7813895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7823895dbf8SEric W. Biederman if (!new) 7833895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7843895dbf8SEric W. Biederman 7853895dbf8SEric W. Biederman 7863895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7873895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7883895dbf8SEric W. Biederman 7893895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7903895dbf8SEric W. Biederman if (ret == -EBUSY) 7913895dbf8SEric W. Biederman goto mountpoint; 7923895dbf8SEric W. Biederman 7933895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7943895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7953895dbf8SEric W. Biederman if (ret) 7963895dbf8SEric W. Biederman goto done; 7973895dbf8SEric W. Biederman 7983895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7993895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 8004edbe133SAl Viro new->m_dentry = dget(dentry); 8013895dbf8SEric W. Biederman new->m_count = 1; 8023895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 8033895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 8043895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 8053895dbf8SEric W. Biederman 8063895dbf8SEric W. Biederman mp = new; 8073895dbf8SEric W. Biederman new = NULL; 8083895dbf8SEric W. Biederman done: 8093895dbf8SEric W. Biederman kfree(new); 81084d17192SAl Viro return mp; 81184d17192SAl Viro } 81284d17192SAl Viro 8134edbe133SAl Viro /* 8144edbe133SAl Viro * vfsmount lock must be held. Additionally, the caller is responsible 8154edbe133SAl Viro * for serializing calls for given disposal list. 8164edbe133SAl Viro */ 8174edbe133SAl Viro static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) 81884d17192SAl Viro { 81984d17192SAl Viro if (!--mp->m_count) { 82084d17192SAl Viro struct dentry *dentry = mp->m_dentry; 8210a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 82284d17192SAl Viro spin_lock(&dentry->d_lock); 82384d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 82484d17192SAl Viro spin_unlock(&dentry->d_lock); 8254edbe133SAl Viro dput_to_list(dentry, list); 8260818bf27SAl Viro hlist_del(&mp->m_hash); 82784d17192SAl Viro kfree(mp); 82884d17192SAl Viro } 82984d17192SAl Viro } 83084d17192SAl Viro 8314edbe133SAl Viro /* called with namespace_lock and vfsmount lock */ 8324edbe133SAl Viro static void put_mountpoint(struct mountpoint *mp) 8334edbe133SAl Viro { 8344edbe133SAl Viro __put_mountpoint(mp, &ex_mountpoints); 8354edbe133SAl Viro } 8364edbe133SAl Viro 837143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 8381da177e4SLinus Torvalds { 8396b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds 84299b7db7bSNick Piggin /* 84399b7db7bSNick Piggin * vfsmount lock must be held for write 84499b7db7bSNick Piggin */ 8456b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 8465addc5ddSAl Viro { 8475addc5ddSAl Viro if (ns) { 8485addc5ddSAl Viro ns->event = ++event; 8495addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8505addc5ddSAl Viro } 8515addc5ddSAl Viro } 8525addc5ddSAl Viro 85399b7db7bSNick Piggin /* 85499b7db7bSNick Piggin * vfsmount lock must be held for write 85599b7db7bSNick Piggin */ 8566b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8575addc5ddSAl Viro { 8585addc5ddSAl Viro if (ns && ns->event != event) { 8595addc5ddSAl Viro ns->event = event; 8605addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8615addc5ddSAl Viro } 8625addc5ddSAl Viro } 8635addc5ddSAl Viro 86499b7db7bSNick Piggin /* 86599b7db7bSNick Piggin * vfsmount lock must be held for write 86699b7db7bSNick Piggin */ 867e4e59906SAl Viro static struct mountpoint *unhash_mnt(struct mount *mnt) 8681da177e4SLinus Torvalds { 869e4e59906SAl Viro struct mountpoint *mp; 8700714a533SAl Viro mnt->mnt_parent = mnt; 871a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8726b41d536SAl Viro list_del_init(&mnt->mnt_child); 87338129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8740a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 875e4e59906SAl Viro mp = mnt->mnt_mp; 87684d17192SAl Viro mnt->mnt_mp = NULL; 877e4e59906SAl Viro return mp; 8787bdb11deSEric W. Biederman } 8797bdb11deSEric W. Biederman 8807bdb11deSEric W. Biederman /* 8817bdb11deSEric W. Biederman * vfsmount lock must be held for write 8827bdb11deSEric W. Biederman */ 8836a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8846a46c573SEric W. Biederman { 885e4e59906SAl Viro put_mountpoint(unhash_mnt(mnt)); 8866a46c573SEric W. Biederman } 8876a46c573SEric W. Biederman 8886a46c573SEric W. Biederman /* 8896a46c573SEric W. Biederman * vfsmount lock must be held for write 8906a46c573SEric W. Biederman */ 89184d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 89284d17192SAl Viro struct mountpoint *mp, 89344d964d6SAl Viro struct mount *child_mnt) 894b90fa9aeSRam Pai { 89584d17192SAl Viro mp->m_count++; 8963a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 8974edbe133SAl Viro child_mnt->mnt_mountpoint = mp->m_dentry; 8983a2393d7SAl Viro child_mnt->mnt_parent = mnt; 89984d17192SAl Viro child_mnt->mnt_mp = mp; 9000a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 901b90fa9aeSRam Pai } 902b90fa9aeSRam Pai 9031064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 9041064f874SEric W. Biederman { 9051064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 9061064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 9071064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 9081064f874SEric W. Biederman } 9091064f874SEric W. Biederman 91099b7db7bSNick Piggin /* 91199b7db7bSNick Piggin * vfsmount lock must be held for write 91299b7db7bSNick Piggin */ 91384d17192SAl Viro static void attach_mnt(struct mount *mnt, 91484d17192SAl Viro struct mount *parent, 91584d17192SAl Viro struct mountpoint *mp) 9161da177e4SLinus Torvalds { 91784d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 9181064f874SEric W. Biederman __attach_mnt(mnt, parent); 919b90fa9aeSRam Pai } 920b90fa9aeSRam Pai 9211064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 92212a5b529SAl Viro { 9231064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 9241064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 9251064f874SEric W. Biederman 9261064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 9271064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 9281064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 9291064f874SEric W. Biederman 9301064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 9311064f874SEric W. Biederman 9321064f874SEric W. Biederman put_mountpoint(old_mp); 9331064f874SEric W. Biederman mnt_add_count(old_parent, -1); 93412a5b529SAl Viro } 93512a5b529SAl Viro 936b90fa9aeSRam Pai /* 93799b7db7bSNick Piggin * vfsmount lock must be held for write 938b90fa9aeSRam Pai */ 9391064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 940b90fa9aeSRam Pai { 9410714a533SAl Viro struct mount *parent = mnt->mnt_parent; 94283adc753SAl Viro struct mount *m; 943b90fa9aeSRam Pai LIST_HEAD(head); 944143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 945b90fa9aeSRam Pai 9460714a533SAl Viro BUG_ON(parent == mnt); 947b90fa9aeSRam Pai 9481a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 949f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 950143c8c91SAl Viro m->mnt_ns = n; 951f03c6599SAl Viro 952b90fa9aeSRam Pai list_splice(&head, n->list.prev); 953b90fa9aeSRam Pai 954d2921684SEric W. Biederman n->mounts += n->pending_mounts; 955d2921684SEric W. Biederman n->pending_mounts = 0; 956d2921684SEric W. Biederman 9571064f874SEric W. Biederman __attach_mnt(mnt, parent); 9586b3286edSKirill Korotaev touch_mnt_namespace(n); 9591da177e4SLinus Torvalds } 9601da177e4SLinus Torvalds 961909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9621da177e4SLinus Torvalds { 9636b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9646b41d536SAl Viro if (next == &p->mnt_mounts) { 9651da177e4SLinus Torvalds while (1) { 966909b0a88SAl Viro if (p == root) 9671da177e4SLinus Torvalds return NULL; 9686b41d536SAl Viro next = p->mnt_child.next; 9696b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9701da177e4SLinus Torvalds break; 9710714a533SAl Viro p = p->mnt_parent; 9721da177e4SLinus Torvalds } 9731da177e4SLinus Torvalds } 9746b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9751da177e4SLinus Torvalds } 9761da177e4SLinus Torvalds 977315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9789676f0c6SRam Pai { 9796b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9806b41d536SAl Viro while (prev != &p->mnt_mounts) { 9816b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9826b41d536SAl Viro prev = p->mnt_mounts.prev; 9839676f0c6SRam Pai } 9849676f0c6SRam Pai return p; 9859676f0c6SRam Pai } 9869676f0c6SRam Pai 9878f291889SAl Viro /** 9888f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 9898f291889SAl Viro * @fc: The configuration context with the superblock attached 9908f291889SAl Viro * 9918f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 9928f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 9938f291889SAl Viro * 9948f291889SAl Viro * Note that this does not attach the mount to anything. 9958f291889SAl Viro */ 9968f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 9979d412a43SAl Viro { 998b105e270SAl Viro struct mount *mnt; 999bd303368SChristian Brauner struct user_namespace *fs_userns; 10009d412a43SAl Viro 10018f291889SAl Viro if (!fc->root) 10028f291889SAl Viro return ERR_PTR(-EINVAL); 10039d412a43SAl Viro 10048f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 10059d412a43SAl Viro if (!mnt) 10069d412a43SAl Viro return ERR_PTR(-ENOMEM); 10079d412a43SAl Viro 10088f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 1009b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 10109d412a43SAl Viro 10118f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 10128f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 10138f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 1014a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10150714a533SAl Viro mnt->mnt_parent = mnt; 10168f291889SAl Viro 1017bd303368SChristian Brauner fs_userns = mnt->mnt.mnt_sb->s_user_ns; 1018bd303368SChristian Brauner if (!initial_idmapping(fs_userns)) 1019bd303368SChristian Brauner mnt->mnt.mnt_userns = get_user_ns(fs_userns); 1020bd303368SChristian Brauner 1021719ea2fbSAl Viro lock_mount_hash(); 10228f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 1023719ea2fbSAl Viro unlock_mount_hash(); 1024b105e270SAl Viro return &mnt->mnt; 10259d412a43SAl Viro } 10268f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 10278f291889SAl Viro 10288f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 10298f291889SAl Viro { 10308f291889SAl Viro int err = vfs_get_tree(fc); 10318f291889SAl Viro if (!err) { 10328f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 10338f291889SAl Viro return vfs_create_mount(fc); 10348f291889SAl Viro } 10358f291889SAl Viro return ERR_PTR(err); 10368f291889SAl Viro } 10378f291889SAl Viro EXPORT_SYMBOL(fc_mount); 10388f291889SAl Viro 10399bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 10409bc61ab1SDavid Howells int flags, const char *name, 10419bc61ab1SDavid Howells void *data) 10421da177e4SLinus Torvalds { 10439bc61ab1SDavid Howells struct fs_context *fc; 10448f291889SAl Viro struct vfsmount *mnt; 10459bc61ab1SDavid Howells int ret = 0; 10469d412a43SAl Viro 10479d412a43SAl Viro if (!type) 10483e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 10499d412a43SAl Viro 10509bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 10519bc61ab1SDavid Howells if (IS_ERR(fc)) 10529bc61ab1SDavid Howells return ERR_CAST(fc); 10539bc61ab1SDavid Howells 10543e1aeb00SDavid Howells if (name) 10553e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 10563e1aeb00SDavid Howells name, strlen(name)); 10579bc61ab1SDavid Howells if (!ret) 10589bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 10599bc61ab1SDavid Howells if (!ret) 10608f291889SAl Viro mnt = fc_mount(fc); 10618f291889SAl Viro else 10628f291889SAl Viro mnt = ERR_PTR(ret); 10639d412a43SAl Viro 10649bc61ab1SDavid Howells put_fs_context(fc); 10658f291889SAl Viro return mnt; 10669d412a43SAl Viro } 10679d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10689d412a43SAl Viro 106993faccbbSEric W. Biederman struct vfsmount * 107093faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 107193faccbbSEric W. Biederman const char *name, void *data) 107293faccbbSEric W. Biederman { 107393faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 107493faccbbSEric W. Biederman * through from the parent mount to the submount don't support 107593faccbbSEric W. Biederman * unprivileged mounts with submounts. 107693faccbbSEric W. Biederman */ 107793faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 107893faccbbSEric W. Biederman return ERR_PTR(-EPERM); 107993faccbbSEric W. Biederman 1080e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 108193faccbbSEric W. Biederman } 108293faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 108393faccbbSEric W. Biederman 108487129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 108536341f64SRam Pai int flag) 10861da177e4SLinus Torvalds { 108787129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1088be34d1a3SDavid Howells struct mount *mnt; 1089be34d1a3SDavid Howells int err; 10901da177e4SLinus Torvalds 1091be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1092be34d1a3SDavid Howells if (!mnt) 1093be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1094be34d1a3SDavid Howells 10957a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 109615169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1097719f5d7fSMiklos Szeredi else 109815169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1099719f5d7fSMiklos Szeredi 110015169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1101be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1102719f5d7fSMiklos Szeredi if (err) 1103719f5d7fSMiklos Szeredi goto out_free; 1104719f5d7fSMiklos Szeredi } 1105719f5d7fSMiklos Szeredi 110616a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 110716a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 11085ff9d8a6SEric W. Biederman 11091da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1110a6435940SChristian Brauner mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt); 1111bd303368SChristian Brauner if (!initial_idmapping(mnt->mnt.mnt_userns)) 1112a6435940SChristian Brauner mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns); 1113b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1114b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1115a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 11160714a533SAl Viro mnt->mnt_parent = mnt; 1117719ea2fbSAl Viro lock_mount_hash(); 111839f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1119719ea2fbSAl Viro unlock_mount_hash(); 1120b90fa9aeSRam Pai 11217a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 11227a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 11236776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 112432301920SAl Viro mnt->mnt_master = old; 1125fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 11268aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1127fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 11286776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1129d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 11306776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1131d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 11325235d448SAl Viro } else { 11335235d448SAl Viro CLEAR_MNT_SHARED(mnt); 11345afe0022SRam Pai } 1135b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 11360f0afb1dSAl Viro set_mnt_shared(mnt); 11371da177e4SLinus Torvalds 11381da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 11391da177e4SLinus Torvalds * as the original if that was on one */ 114036341f64SRam Pai if (flag & CL_EXPIRE) { 11416776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 11426776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 11431da177e4SLinus Torvalds } 1144be34d1a3SDavid Howells 1145cb338d06SAl Viro return mnt; 1146719f5d7fSMiklos Szeredi 1147719f5d7fSMiklos Szeredi out_free: 11488ffcb32eSDavid Howells mnt_free_id(mnt); 1149719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1150be34d1a3SDavid Howells return ERR_PTR(err); 11511da177e4SLinus Torvalds } 11521da177e4SLinus Torvalds 11539ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11549ea459e1SAl Viro { 115556cbb429SAl Viro struct hlist_node *p; 115656cbb429SAl Viro struct mount *m; 11579ea459e1SAl Viro /* 115856cbb429SAl Viro * The warning here probably indicates that somebody messed 115956cbb429SAl Viro * up a mnt_want/drop_write() pair. If this happens, the 116056cbb429SAl Viro * filesystem was probably unable to make r/w->r/o transitions. 11619ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11629ea459e1SAl Viro * so mnt_get_writers() below is safe. 11639ea459e1SAl Viro */ 11649ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11659ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11669ea459e1SAl Viro mnt_pin_kill(mnt); 116756cbb429SAl Viro hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { 116856cbb429SAl Viro hlist_del(&m->mnt_umount); 116956cbb429SAl Viro mntput(&m->mnt); 117056cbb429SAl Viro } 11719ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11729ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11739ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11749ea459e1SAl Viro mnt_free_id(mnt); 11759ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11769ea459e1SAl Viro } 11779ea459e1SAl Viro 11789ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11799ea459e1SAl Viro { 11809ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11819ea459e1SAl Viro } 11829ea459e1SAl Viro 11839ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11849ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11859ea459e1SAl Viro { 11869ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 118729785735SByungchul Park struct mount *m, *t; 11889ea459e1SAl Viro 118929785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 119029785735SByungchul Park cleanup_mnt(m); 11919ea459e1SAl Viro } 11929ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11939ea459e1SAl Viro 1194900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11957b7b1aceSAl Viro { 11964edbe133SAl Viro LIST_HEAD(list); 1197edf7ddbfSEric Biggers int count; 11984edbe133SAl Viro 119948a066e7SAl Viro rcu_read_lock(); 12009ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 12019ea0a46cSAl Viro /* 12029ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 12039ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 12049ea0a46cSAl Viro * non-NULL, then there's a reference that won't 12059ea0a46cSAl Viro * be dropped until after an RCU delay done after 12069ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 12079ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 12089ea0a46cSAl Viro * we are dropping is not the final one. 12099ea0a46cSAl Viro */ 1210aa9c0e07SAl Viro mnt_add_count(mnt, -1); 121148a066e7SAl Viro rcu_read_unlock(); 121299b7db7bSNick Piggin return; 1213b3e19d92SNick Piggin } 1214719ea2fbSAl Viro lock_mount_hash(); 1215119e1ef8SAl Viro /* 1216119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1217119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1218119e1ef8SAl Viro */ 1219119e1ef8SAl Viro smp_mb(); 12209ea0a46cSAl Viro mnt_add_count(mnt, -1); 1221edf7ddbfSEric Biggers count = mnt_get_count(mnt); 1222edf7ddbfSEric Biggers if (count != 0) { 1223edf7ddbfSEric Biggers WARN_ON(count < 0); 122448a066e7SAl Viro rcu_read_unlock(); 1225719ea2fbSAl Viro unlock_mount_hash(); 122699b7db7bSNick Piggin return; 122799b7db7bSNick Piggin } 122848a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 122948a066e7SAl Viro rcu_read_unlock(); 123048a066e7SAl Viro unlock_mount_hash(); 123148a066e7SAl Viro return; 123248a066e7SAl Viro } 123348a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 123448a066e7SAl Viro rcu_read_unlock(); 1235962830dfSAndi Kleen 123639f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1237ce07d891SEric W. Biederman 1238ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1239ce07d891SEric W. Biederman struct mount *p, *tmp; 1240ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 12414edbe133SAl Viro __put_mountpoint(unhash_mnt(p), &list); 124256cbb429SAl Viro hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); 1243ce07d891SEric W. Biederman } 1244ce07d891SEric W. Biederman } 1245719ea2fbSAl Viro unlock_mount_hash(); 12464edbe133SAl Viro shrink_dentry_list(&list); 1247649a795aSAl Viro 12489ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 12499ea459e1SAl Viro struct task_struct *task = current; 12509ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 12519ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 125291989c70SJens Axboe if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) 12539ea459e1SAl Viro return; 12549ea459e1SAl Viro } 12559ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 12569ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 12579ea459e1SAl Viro return; 12589ea459e1SAl Viro } 12599ea459e1SAl Viro cleanup_mnt(mnt); 1260b3e19d92SNick Piggin } 1261b3e19d92SNick Piggin 1262b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1263b3e19d92SNick Piggin { 1264b3e19d92SNick Piggin if (mnt) { 1265863d684fSAl Viro struct mount *m = real_mount(mnt); 1266b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1267863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1268863d684fSAl Viro m->mnt_expiry_mark = 0; 1269863d684fSAl Viro mntput_no_expire(m); 1270b3e19d92SNick Piggin } 1271b3e19d92SNick Piggin } 1272b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1273b3e19d92SNick Piggin 1274b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1275b3e19d92SNick Piggin { 1276b3e19d92SNick Piggin if (mnt) 127783adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1278b3e19d92SNick Piggin return mnt; 1279b3e19d92SNick Piggin } 1280b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1281b3e19d92SNick Piggin 12821f287bc4SRandy Dunlap /** 12831f287bc4SRandy Dunlap * path_is_mountpoint() - Check if path is a mount in the current namespace. 12841f287bc4SRandy Dunlap * @path: path to check 1285c6609c0aSIan Kent * 1286c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1287c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1288c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1289c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1290c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1291c6609c0aSIan Kent * alone. 1292c6609c0aSIan Kent */ 1293c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1294c6609c0aSIan Kent { 1295c6609c0aSIan Kent unsigned seq; 1296c6609c0aSIan Kent bool res; 1297c6609c0aSIan Kent 1298c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1299c6609c0aSIan Kent return false; 1300c6609c0aSIan Kent 1301c6609c0aSIan Kent rcu_read_lock(); 1302c6609c0aSIan Kent do { 1303c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1304c6609c0aSIan Kent res = __path_is_mountpoint(path); 1305c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1306c6609c0aSIan Kent rcu_read_unlock(); 1307c6609c0aSIan Kent 1308c6609c0aSIan Kent return res; 1309c6609c0aSIan Kent } 1310c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1311c6609c0aSIan Kent 1312ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 13137b7b1aceSAl Viro { 13143064c356SAl Viro struct mount *p; 13153064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 13163064c356SAl Viro if (IS_ERR(p)) 13173064c356SAl Viro return ERR_CAST(p); 13183064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 13193064c356SAl Viro return &p->mnt; 13207b7b1aceSAl Viro } 13211da177e4SLinus Torvalds 1322a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 13239f6c61f9SMiklos Szeredi static struct mount *mnt_list_next(struct mnt_namespace *ns, 13249f6c61f9SMiklos Szeredi struct list_head *p) 13259f6c61f9SMiklos Szeredi { 13269f6c61f9SMiklos Szeredi struct mount *mnt, *ret = NULL; 13279f6c61f9SMiklos Szeredi 13289f6c61f9SMiklos Szeredi lock_ns_list(ns); 13299f6c61f9SMiklos Szeredi list_for_each_continue(p, &ns->list) { 13309f6c61f9SMiklos Szeredi mnt = list_entry(p, typeof(*mnt), mnt_list); 13319f6c61f9SMiklos Szeredi if (!mnt_is_cursor(mnt)) { 13329f6c61f9SMiklos Szeredi ret = mnt; 13339f6c61f9SMiklos Szeredi break; 13349f6c61f9SMiklos Szeredi } 13359f6c61f9SMiklos Szeredi } 13369f6c61f9SMiklos Szeredi unlock_ns_list(ns); 13379f6c61f9SMiklos Szeredi 13389f6c61f9SMiklos Szeredi return ret; 13399f6c61f9SMiklos Szeredi } 13409f6c61f9SMiklos Szeredi 13410226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 13421da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 13431da177e4SLinus Torvalds { 1344ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13459f6c61f9SMiklos Szeredi struct list_head *prev; 13461da177e4SLinus Torvalds 1347390c6843SRam Pai down_read(&namespace_sem); 13489f6c61f9SMiklos Szeredi if (!*pos) { 13499f6c61f9SMiklos Szeredi prev = &p->ns->list; 13509f6c61f9SMiklos Szeredi } else { 13519f6c61f9SMiklos Szeredi prev = &p->cursor.mnt_list; 13529f6c61f9SMiklos Szeredi 13539f6c61f9SMiklos Szeredi /* Read after we'd reached the end? */ 13549f6c61f9SMiklos Szeredi if (list_empty(prev)) 13559f6c61f9SMiklos Szeredi return NULL; 1356c7999c36SAl Viro } 1357c7999c36SAl Viro 13589f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, prev); 13591da177e4SLinus Torvalds } 13601da177e4SLinus Torvalds 13611da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13621da177e4SLinus Torvalds { 1363ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13649f6c61f9SMiklos Szeredi struct mount *mnt = v; 1365b0765fb8SPavel Emelianov 13669f6c61f9SMiklos Szeredi ++*pos; 13679f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, &mnt->mnt_list); 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13711da177e4SLinus Torvalds { 13729f6c61f9SMiklos Szeredi struct proc_mounts *p = m->private; 13739f6c61f9SMiklos Szeredi struct mount *mnt = v; 13749f6c61f9SMiklos Szeredi 13759f6c61f9SMiklos Szeredi lock_ns_list(p->ns); 13769f6c61f9SMiklos Szeredi if (mnt) 13779f6c61f9SMiklos Szeredi list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); 13789f6c61f9SMiklos Szeredi else 13799f6c61f9SMiklos Szeredi list_del_init(&p->cursor.mnt_list); 13809f6c61f9SMiklos Szeredi unlock_ns_list(p->ns); 1381390c6843SRam Pai up_read(&namespace_sem); 13821da177e4SLinus Torvalds } 13831da177e4SLinus Torvalds 13840226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13859f5596afSAl Viro { 1386ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13879f6c61f9SMiklos Szeredi struct mount *r = v; 13880226f492SAl Viro return p->show(m, &r->mnt); 13891da177e4SLinus Torvalds } 13901da177e4SLinus Torvalds 1391a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13921da177e4SLinus Torvalds .start = m_start, 13931da177e4SLinus Torvalds .next = m_next, 13941da177e4SLinus Torvalds .stop = m_stop, 13950226f492SAl Viro .show = m_show, 1396b4629fe2SChuck Lever }; 13979f6c61f9SMiklos Szeredi 13989f6c61f9SMiklos Szeredi void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) 13999f6c61f9SMiklos Szeredi { 14009f6c61f9SMiklos Szeredi down_read(&namespace_sem); 14019f6c61f9SMiklos Szeredi lock_ns_list(ns); 14029f6c61f9SMiklos Szeredi list_del(&cursor->mnt_list); 14039f6c61f9SMiklos Szeredi unlock_ns_list(ns); 14049f6c61f9SMiklos Szeredi up_read(&namespace_sem); 14059f6c61f9SMiklos Szeredi } 1406a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1407b4629fe2SChuck Lever 14081da177e4SLinus Torvalds /** 14091da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 14101f287bc4SRandy Dunlap * @m: root of mount tree 14111da177e4SLinus Torvalds * 14121da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 14131da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 14141da177e4SLinus Torvalds * busy. 14151da177e4SLinus Torvalds */ 1416909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 14171da177e4SLinus Torvalds { 1418909b0a88SAl Viro struct mount *mnt = real_mount(m); 141936341f64SRam Pai int actual_refs = 0; 142036341f64SRam Pai int minimum_refs = 0; 1421315fc83eSAl Viro struct mount *p; 1422909b0a88SAl Viro BUG_ON(!m); 14231da177e4SLinus Torvalds 1424b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1425719ea2fbSAl Viro lock_mount_hash(); 1426909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 142783adc753SAl Viro actual_refs += mnt_get_count(p); 14281da177e4SLinus Torvalds minimum_refs += 2; 14291da177e4SLinus Torvalds } 1430719ea2fbSAl Viro unlock_mount_hash(); 14311da177e4SLinus Torvalds 14321da177e4SLinus Torvalds if (actual_refs > minimum_refs) 14331da177e4SLinus Torvalds return 0; 1434e3474a8eSIan Kent 1435e3474a8eSIan Kent return 1; 14361da177e4SLinus Torvalds } 14371da177e4SLinus Torvalds 14381da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 14391da177e4SLinus Torvalds 14401da177e4SLinus Torvalds /** 14411da177e4SLinus Torvalds * may_umount - check if a mount point is busy 14421da177e4SLinus Torvalds * @mnt: root of mount 14431da177e4SLinus Torvalds * 14441da177e4SLinus Torvalds * This is called to check if a mount point has any 14451da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 14461da177e4SLinus Torvalds * mount has sub mounts this will return busy 14471da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 14481da177e4SLinus Torvalds * 14491da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 14501da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 14511da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 14521da177e4SLinus Torvalds */ 14531da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 14541da177e4SLinus Torvalds { 1455e3474a8eSIan Kent int ret = 1; 14568ad08d8aSAl Viro down_read(&namespace_sem); 1457719ea2fbSAl Viro lock_mount_hash(); 14581ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1459e3474a8eSIan Kent ret = 0; 1460719ea2fbSAl Viro unlock_mount_hash(); 14618ad08d8aSAl Viro up_read(&namespace_sem); 1462a05964f3SRam Pai return ret; 14631da177e4SLinus Torvalds } 14641da177e4SLinus Torvalds 14651da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14661da177e4SLinus Torvalds 146797216be0SAl Viro static void namespace_unlock(void) 14681da177e4SLinus Torvalds { 1469a3b3c562SEric W. Biederman struct hlist_head head; 147056cbb429SAl Viro struct hlist_node *p; 147156cbb429SAl Viro struct mount *m; 14724edbe133SAl Viro LIST_HEAD(list); 147397216be0SAl Viro 1474a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 14754edbe133SAl Viro list_splice_init(&ex_mountpoints, &list); 1476a3b3c562SEric W. Biederman 147797216be0SAl Viro up_write(&namespace_sem); 1478a3b3c562SEric W. Biederman 14794edbe133SAl Viro shrink_dentry_list(&list); 14804edbe133SAl Viro 1481a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 148297216be0SAl Viro return; 148397216be0SAl Viro 148422cb7405SNeilBrown synchronize_rcu_expedited(); 148548a066e7SAl Viro 148656cbb429SAl Viro hlist_for_each_entry_safe(m, p, &head, mnt_umount) { 148756cbb429SAl Viro hlist_del(&m->mnt_umount); 148856cbb429SAl Viro mntput(&m->mnt); 148956cbb429SAl Viro } 149070fbcdf4SRam Pai } 149170fbcdf4SRam Pai 149297216be0SAl Viro static inline void namespace_lock(void) 1493e3197d83SAl Viro { 149497216be0SAl Viro down_write(&namespace_sem); 1495e3197d83SAl Viro } 1496e3197d83SAl Viro 1497e819f152SEric W. Biederman enum umount_tree_flags { 1498e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1499e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1500e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1501e819f152SEric W. Biederman }; 1502f2d0a123SEric W. Biederman 1503f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1504f2d0a123SEric W. Biederman { 1505f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1506f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1507f2d0a123SEric W. Biederman return true; 1508f2d0a123SEric W. Biederman 1509f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1510f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1511f2d0a123SEric W. Biederman return true; 1512f2d0a123SEric W. Biederman 1513f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1514f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1515f2d0a123SEric W. Biederman * connected to mounted mounts. 1516f2d0a123SEric W. Biederman */ 1517f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1518f2d0a123SEric W. Biederman return true; 1519f2d0a123SEric W. Biederman 1520f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1521f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1522f2d0a123SEric W. Biederman return false; 1523f2d0a123SEric W. Biederman 1524f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1525f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1526f2d0a123SEric W. Biederman return false; 1527f2d0a123SEric W. Biederman 1528f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1529f2d0a123SEric W. Biederman return true; 1530f2d0a123SEric W. Biederman } 1531f2d0a123SEric W. Biederman 153299b7db7bSNick Piggin /* 153348a066e7SAl Viro * mount_lock must be held 153499b7db7bSNick Piggin * namespace_sem must be held for write 153599b7db7bSNick Piggin */ 1536e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 153770fbcdf4SRam Pai { 1538c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1539315fc83eSAl Viro struct mount *p; 154070fbcdf4SRam Pai 15415d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 15425d88457eSEric W. Biederman propagate_mount_unlock(mnt); 15435d88457eSEric W. Biederman 1544c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1545590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1546590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1547c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1548590ce4bcSEric W. Biederman } 1549c003b26fSEric W. Biederman 1550411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1551c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1552c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 155338129a13SAl Viro } 155470fbcdf4SRam Pai 1555c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1556e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 15577b8a53fdSAl Viro propagate_umount(&tmp_list); 1558a05964f3SRam Pai 1559c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1560d2921684SEric W. Biederman struct mnt_namespace *ns; 1561ce07d891SEric W. Biederman bool disconnect; 1562c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 15636776db3dSAl Viro list_del_init(&p->mnt_expire); 15641a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1565d2921684SEric W. Biederman ns = p->mnt_ns; 1566d2921684SEric W. Biederman if (ns) { 1567d2921684SEric W. Biederman ns->mounts--; 1568d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1569d2921684SEric W. Biederman } 157063d37a84SAl Viro p->mnt_ns = NULL; 1571e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 157248a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 157387b95ce0SAl Viro 1574f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1575676da58dSAl Viro if (mnt_has_parent(p)) { 157681b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1577ce07d891SEric W. Biederman if (!disconnect) { 1578ce07d891SEric W. Biederman /* Don't forget about p */ 1579ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1580ce07d891SEric W. Biederman } else { 15816a46c573SEric W. Biederman umount_mnt(p); 15827c4b93d8SAl Viro } 1583ce07d891SEric W. Biederman } 15840f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 158519a1c409SAl Viro if (disconnect) 158619a1c409SAl Viro hlist_add_head(&p->mnt_umount, &unmounted); 158738129a13SAl Viro } 15881da177e4SLinus Torvalds } 15891da177e4SLinus Torvalds 1590b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1591c35038beSAl Viro 15928d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 15938d0347f6SDavid Howells { 15948d0347f6SDavid Howells int ret = 0; 15958d0347f6SDavid Howells 15968d0347f6SDavid Howells down_write(&sb->s_umount); 15978d0347f6SDavid Howells if (!sb_rdonly(sb)) { 15988d0347f6SDavid Howells struct fs_context *fc; 15998d0347f6SDavid Howells 16008d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 16018d0347f6SDavid Howells SB_RDONLY); 16028d0347f6SDavid Howells if (IS_ERR(fc)) { 16038d0347f6SDavid Howells ret = PTR_ERR(fc); 16048d0347f6SDavid Howells } else { 16058d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 16068d0347f6SDavid Howells if (!ret) 16078d0347f6SDavid Howells ret = reconfigure_super(fc); 16088d0347f6SDavid Howells put_fs_context(fc); 16098d0347f6SDavid Howells } 16108d0347f6SDavid Howells } 16118d0347f6SDavid Howells up_write(&sb->s_umount); 16128d0347f6SDavid Howells return ret; 16138d0347f6SDavid Howells } 16148d0347f6SDavid Howells 16151ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 16161da177e4SLinus Torvalds { 16171ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 16181da177e4SLinus Torvalds int retval; 16191da177e4SLinus Torvalds 16201ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 16211da177e4SLinus Torvalds if (retval) 16221da177e4SLinus Torvalds return retval; 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds /* 16251da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 16261da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 16271da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 16281da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 16291da177e4SLinus Torvalds */ 16301da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 16311ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 16321da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 16331da177e4SLinus Torvalds return -EINVAL; 16341da177e4SLinus Torvalds 1635b3e19d92SNick Piggin /* 1636b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1637b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1638b3e19d92SNick Piggin */ 1639719ea2fbSAl Viro lock_mount_hash(); 164083adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1641719ea2fbSAl Viro unlock_mount_hash(); 16421da177e4SLinus Torvalds return -EBUSY; 1643b3e19d92SNick Piggin } 1644719ea2fbSAl Viro unlock_mount_hash(); 16451da177e4SLinus Torvalds 1646863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 16471da177e4SLinus Torvalds return -EAGAIN; 16481da177e4SLinus Torvalds } 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds /* 16511da177e4SLinus Torvalds * If we may have to abort operations to get out of this 16521da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 16531da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 16541da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 16551da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 16561da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 16571da177e4SLinus Torvalds * about for the moment. 16581da177e4SLinus Torvalds */ 16591da177e4SLinus Torvalds 166042faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 166142faad99SAl Viro sb->s_op->umount_begin(sb); 166242faad99SAl Viro } 16631da177e4SLinus Torvalds 16641da177e4SLinus Torvalds /* 16651da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 16661da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 16671da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 16681da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 16691da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 16701da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 16711da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 16721da177e4SLinus Torvalds */ 16731ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 16741da177e4SLinus Torvalds /* 16751da177e4SLinus Torvalds * Special case for "unmounting" root ... 16761da177e4SLinus Torvalds * we just try to remount it readonly. 16771da177e4SLinus Torvalds */ 1678bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1679a1480dccSAndy Lutomirski return -EPERM; 16808d0347f6SDavid Howells return do_umount_root(sb); 16811da177e4SLinus Torvalds } 16821da177e4SLinus Torvalds 168397216be0SAl Viro namespace_lock(); 1684719ea2fbSAl Viro lock_mount_hash(); 16851da177e4SLinus Torvalds 168625d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 168725d202edSEric W. Biederman retval = -EINVAL; 168825d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 168925d202edSEric W. Biederman goto out; 169025d202edSEric W. Biederman 169125d202edSEric W. Biederman event++; 169248a066e7SAl Viro if (flags & MNT_DETACH) { 169348a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1694e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 169548a066e7SAl Viro retval = 0; 169648a066e7SAl Viro } else { 1697b54b9be7SAl Viro shrink_submounts(mnt); 16981da177e4SLinus Torvalds retval = -EBUSY; 169948a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 17001a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1701e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 17021da177e4SLinus Torvalds retval = 0; 17031da177e4SLinus Torvalds } 170448a066e7SAl Viro } 170525d202edSEric W. Biederman out: 1706719ea2fbSAl Viro unlock_mount_hash(); 1707e3197d83SAl Viro namespace_unlock(); 17081da177e4SLinus Torvalds return retval; 17091da177e4SLinus Torvalds } 17101da177e4SLinus Torvalds 17111da177e4SLinus Torvalds /* 171280b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 171380b5dce8SEric W. Biederman * 171480b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 171580b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 171680b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 171780b5dce8SEric W. Biederman * leaking them. 171880b5dce8SEric W. Biederman * 171980b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 172080b5dce8SEric W. Biederman */ 172180b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 172280b5dce8SEric W. Biederman { 172380b5dce8SEric W. Biederman struct mountpoint *mp; 172480b5dce8SEric W. Biederman struct mount *mnt; 172580b5dce8SEric W. Biederman 172680b5dce8SEric W. Biederman namespace_lock(); 17273895dbf8SEric W. Biederman lock_mount_hash(); 172880b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1729adc9b5c0SAl Viro if (!mp) 173080b5dce8SEric W. Biederman goto out_unlock; 173180b5dce8SEric W. Biederman 1732e06b933eSAndrey Ulanov event++; 173380b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 173480b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1735ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1736fe78fcc8SEric W. Biederman umount_mnt(mnt); 173756cbb429SAl Viro hlist_add_head(&mnt->mnt_umount, &unmounted); 1738ce07d891SEric W. Biederman } 1739e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 174080b5dce8SEric W. Biederman } 174180b5dce8SEric W. Biederman put_mountpoint(mp); 174280b5dce8SEric W. Biederman out_unlock: 17433895dbf8SEric W. Biederman unlock_mount_hash(); 174480b5dce8SEric W. Biederman namespace_unlock(); 174580b5dce8SEric W. Biederman } 174680b5dce8SEric W. Biederman 174780b5dce8SEric W. Biederman /* 17489b40bc90SAl Viro * Is the caller allowed to modify his namespace? 17499b40bc90SAl Viro */ 17509b40bc90SAl Viro static inline bool may_mount(void) 17519b40bc90SAl Viro { 17529b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 17539b40bc90SAl Viro } 17549b40bc90SAl Viro 1755f7e33bdbSJeff Layton static void warn_mandlock(void) 17569e8925b6SJeff Layton { 1757f7e33bdbSJeff Layton pr_warn_once("=======================================================\n" 1758f7e33bdbSJeff Layton "WARNING: The mand mount option has been deprecated and\n" 1759f7e33bdbSJeff Layton " and is ignored by this kernel. Remove the mand\n" 1760f7e33bdbSJeff Layton " option from the mount to silence this warning.\n" 1761f7e33bdbSJeff Layton "=======================================================\n"); 17629e8925b6SJeff Layton } 17639e8925b6SJeff Layton 176425ccd24fSChristoph Hellwig static int can_umount(const struct path *path, int flags) 17651da177e4SLinus Torvalds { 176625ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 17671da177e4SLinus Torvalds 17689b40bc90SAl Viro if (!may_mount()) 17699b40bc90SAl Viro return -EPERM; 177041525f56SChristoph Hellwig if (path->dentry != path->mnt->mnt_root) 177125ccd24fSChristoph Hellwig return -EINVAL; 1772143c8c91SAl Viro if (!check_mnt(mnt)) 177325ccd24fSChristoph Hellwig return -EINVAL; 177425d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 177525ccd24fSChristoph Hellwig return -EINVAL; 1776b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 177725ccd24fSChristoph Hellwig return -EPERM; 177825ccd24fSChristoph Hellwig return 0; 177925ccd24fSChristoph Hellwig } 17801da177e4SLinus Torvalds 1781a0a6df9aSAl Viro // caller is responsible for flags being sane 178225ccd24fSChristoph Hellwig int path_umount(struct path *path, int flags) 178325ccd24fSChristoph Hellwig { 178425ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 178525ccd24fSChristoph Hellwig int ret; 178625ccd24fSChristoph Hellwig 178725ccd24fSChristoph Hellwig ret = can_umount(path, flags); 178825ccd24fSChristoph Hellwig if (!ret) 178925ccd24fSChristoph Hellwig ret = do_umount(mnt, flags); 179025ccd24fSChristoph Hellwig 1791429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 179241525f56SChristoph Hellwig dput(path->dentry); 1793900148dcSAl Viro mntput_no_expire(mnt); 179425ccd24fSChristoph Hellwig return ret; 17951da177e4SLinus Torvalds } 17961da177e4SLinus Torvalds 179709267defSChristoph Hellwig static int ksys_umount(char __user *name, int flags) 179841525f56SChristoph Hellwig { 179941525f56SChristoph Hellwig int lookup_flags = LOOKUP_MOUNTPOINT; 180041525f56SChristoph Hellwig struct path path; 180141525f56SChristoph Hellwig int ret; 180241525f56SChristoph Hellwig 1803a0a6df9aSAl Viro // basic validity checks done first 1804a0a6df9aSAl Viro if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1805a0a6df9aSAl Viro return -EINVAL; 1806a0a6df9aSAl Viro 180741525f56SChristoph Hellwig if (!(flags & UMOUNT_NOFOLLOW)) 180841525f56SChristoph Hellwig lookup_flags |= LOOKUP_FOLLOW; 180941525f56SChristoph Hellwig ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); 181041525f56SChristoph Hellwig if (ret) 181141525f56SChristoph Hellwig return ret; 181241525f56SChristoph Hellwig return path_umount(&path, flags); 181341525f56SChristoph Hellwig } 181441525f56SChristoph Hellwig 18153a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 18163a18ef5cSDominik Brodowski { 18173a18ef5cSDominik Brodowski return ksys_umount(name, flags); 18183a18ef5cSDominik Brodowski } 18193a18ef5cSDominik Brodowski 18201da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds /* 18231da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 18241da177e4SLinus Torvalds */ 1825bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 18261da177e4SLinus Torvalds { 18273a18ef5cSDominik Brodowski return ksys_umount(name, 0); 18281da177e4SLinus Torvalds } 18291da177e4SLinus Torvalds 18301da177e4SLinus Torvalds #endif 18311da177e4SLinus Torvalds 18324ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 18338823c079SEric W. Biederman { 18344ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1835e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1836e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 18374ce5d2b1SEric W. Biederman } 18384ce5d2b1SEric W. Biederman 1839213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 184058be2825SAl Viro { 184158be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 184258be2825SAl Viro } 184358be2825SAl Viro 1844303cc571SChristian Brauner struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) 1845303cc571SChristian Brauner { 1846303cc571SChristian Brauner return &mnt->ns; 1847303cc571SChristian Brauner } 1848303cc571SChristian Brauner 18494ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 18504ce5d2b1SEric W. Biederman { 18514ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 18524ce5d2b1SEric W. Biederman * mount namespace loop? 18534ce5d2b1SEric W. Biederman */ 18544ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 18554ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 18564ce5d2b1SEric W. Biederman return false; 18574ce5d2b1SEric W. Biederman 1858f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 18598823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 18608823c079SEric W. Biederman } 18618823c079SEric W. Biederman 186287129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 186336341f64SRam Pai int flag) 18641da177e4SLinus Torvalds { 186584d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 18661da177e4SLinus Torvalds 18674ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 18684ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 18694ce5d2b1SEric W. Biederman 18704ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1871be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 18729676f0c6SRam Pai 187336341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1874be34d1a3SDavid Howells if (IS_ERR(q)) 1875be34d1a3SDavid Howells return q; 1876be34d1a3SDavid Howells 1877a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 18781da177e4SLinus Torvalds 18791da177e4SLinus Torvalds p = mnt; 18806b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1881315fc83eSAl Viro struct mount *s; 18827ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 18831da177e4SLinus Torvalds continue; 18841da177e4SLinus Torvalds 1885909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 18864ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 18874ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1888df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1889df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1890df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1891df7342b2SEric W. Biederman goto out; 1892df7342b2SEric W. Biederman } else { 18934ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 18944ce5d2b1SEric W. Biederman continue; 18954ce5d2b1SEric W. Biederman } 1896df7342b2SEric W. Biederman } 18974ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 18984ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 18999676f0c6SRam Pai s = skip_mnt_tree(s); 19009676f0c6SRam Pai continue; 19019676f0c6SRam Pai } 19020714a533SAl Viro while (p != s->mnt_parent) { 19030714a533SAl Viro p = p->mnt_parent; 19040714a533SAl Viro q = q->mnt_parent; 19051da177e4SLinus Torvalds } 190687129cc0SAl Viro p = s; 190784d17192SAl Viro parent = q; 190887129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1909be34d1a3SDavid Howells if (IS_ERR(q)) 1910be34d1a3SDavid Howells goto out; 1911719ea2fbSAl Viro lock_mount_hash(); 19121a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 19131064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1914719ea2fbSAl Viro unlock_mount_hash(); 19151da177e4SLinus Torvalds } 19161da177e4SLinus Torvalds } 19171da177e4SLinus Torvalds return res; 1918be34d1a3SDavid Howells out: 19191da177e4SLinus Torvalds if (res) { 1920719ea2fbSAl Viro lock_mount_hash(); 1921e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1922719ea2fbSAl Viro unlock_mount_hash(); 19231da177e4SLinus Torvalds } 1924be34d1a3SDavid Howells return q; 19251da177e4SLinus Torvalds } 19261da177e4SLinus Torvalds 1927be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1928be34d1a3SDavid Howells 1929ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 19308aec0809SAl Viro { 1931cb338d06SAl Viro struct mount *tree; 193297216be0SAl Viro namespace_lock(); 1933cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1934cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1935cd4a4017SEric W. Biederman else 193687129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 193787129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1938328e6d90SAl Viro namespace_unlock(); 1939be34d1a3SDavid Howells if (IS_ERR(tree)) 194052e220d3SDan Carpenter return ERR_CAST(tree); 1941be34d1a3SDavid Howells return &tree->mnt; 19428aec0809SAl Viro } 19438aec0809SAl Viro 1944a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1945a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1946a07b2000SAl Viro 1947a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1948a07b2000SAl Viro { 1949a07b2000SAl Viro struct mnt_namespace *ns; 1950a07b2000SAl Viro namespace_lock(); 1951a07b2000SAl Viro lock_mount_hash(); 1952a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 195344dfd84aSDavid Howells if (ns) { 195444dfd84aSDavid Howells if (is_anon_ns(ns)) 1955a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 195644dfd84aSDavid Howells else 195744dfd84aSDavid Howells ns = NULL; 195844dfd84aSDavid Howells } 1959a07b2000SAl Viro unlock_mount_hash(); 1960a07b2000SAl Viro namespace_unlock(); 196144dfd84aSDavid Howells if (ns) 1962a07b2000SAl Viro free_mnt_ns(ns); 1963a07b2000SAl Viro } 1964a07b2000SAl Viro 19658aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 19668aec0809SAl Viro { 196797216be0SAl Viro namespace_lock(); 1968719ea2fbSAl Viro lock_mount_hash(); 19699c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1970719ea2fbSAl Viro unlock_mount_hash(); 19713ab6abeeSAl Viro namespace_unlock(); 19728aec0809SAl Viro } 19738aec0809SAl Viro 1974427215d8SMiklos Szeredi static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 1975427215d8SMiklos Szeredi { 1976427215d8SMiklos Szeredi struct mount *child; 1977427215d8SMiklos Szeredi 1978427215d8SMiklos Szeredi list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 1979427215d8SMiklos Szeredi if (!is_subdir(child->mnt_mountpoint, dentry)) 1980427215d8SMiklos Szeredi continue; 1981427215d8SMiklos Szeredi 1982427215d8SMiklos Szeredi if (child->mnt.mnt_flags & MNT_LOCKED) 1983427215d8SMiklos Szeredi return true; 1984427215d8SMiklos Szeredi } 1985427215d8SMiklos Szeredi return false; 1986427215d8SMiklos Szeredi } 1987427215d8SMiklos Szeredi 1988c771d683SMiklos Szeredi /** 1989c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 19901f287bc4SRandy Dunlap * @path: path to clone 1991c771d683SMiklos Szeredi * 19921f287bc4SRandy Dunlap * This creates a new vfsmount, which will be the clone of @path. The new mount 19931f287bc4SRandy Dunlap * will not be attached anywhere in the namespace and will be private (i.e. 19941f287bc4SRandy Dunlap * changes to the originating mount won't be propagated into this). 1995c771d683SMiklos Szeredi * 1996c771d683SMiklos Szeredi * Release with mntput(). 1997c771d683SMiklos Szeredi */ 1998ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1999c771d683SMiklos Szeredi { 2000c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 2001c771d683SMiklos Szeredi struct mount *new_mnt; 2002c771d683SMiklos Szeredi 2003427215d8SMiklos Szeredi down_read(&namespace_sem); 2004c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 2005427215d8SMiklos Szeredi goto invalid; 2006427215d8SMiklos Szeredi 2007427215d8SMiklos Szeredi if (!check_mnt(old_mnt)) 2008427215d8SMiklos Szeredi goto invalid; 2009427215d8SMiklos Szeredi 2010427215d8SMiklos Szeredi if (has_locked_children(old_mnt, path->dentry)) 2011427215d8SMiklos Szeredi goto invalid; 2012c771d683SMiklos Szeredi 2013c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 2014427215d8SMiklos Szeredi up_read(&namespace_sem); 2015427215d8SMiklos Szeredi 2016c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 2017c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 2018c771d683SMiklos Szeredi 2019df820f8dSMiklos Szeredi /* Longterm mount to be removed by kern_unmount*() */ 2020df820f8dSMiklos Szeredi new_mnt->mnt_ns = MNT_NS_INTERNAL; 2021df820f8dSMiklos Szeredi 2022c771d683SMiklos Szeredi return &new_mnt->mnt; 2023427215d8SMiklos Szeredi 2024427215d8SMiklos Szeredi invalid: 2025427215d8SMiklos Szeredi up_read(&namespace_sem); 2026427215d8SMiklos Szeredi return ERR_PTR(-EINVAL); 2027c771d683SMiklos Szeredi } 2028c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 2029c771d683SMiklos Szeredi 20301f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 20311f707137SAl Viro struct vfsmount *root) 20321f707137SAl Viro { 20331a4eeaf2SAl Viro struct mount *mnt; 20341f707137SAl Viro int res = f(root, arg); 20351f707137SAl Viro if (res) 20361f707137SAl Viro return res; 20371a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 20381a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 20391f707137SAl Viro if (res) 20401f707137SAl Viro return res; 20411f707137SAl Viro } 20421f707137SAl Viro return 0; 20431f707137SAl Viro } 20441f707137SAl Viro 20453bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 20463bd045ccSAl Viro { 20473bd045ccSAl Viro struct mount *p; 20483bd045ccSAl Viro 20493bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 20503bd045ccSAl Viro int flags = p->mnt.mnt_flags; 20513bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 20523bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 20533bd045ccSAl Viro 20543bd045ccSAl Viro if (flags & MNT_READONLY) 20553bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 20563bd045ccSAl Viro 20573bd045ccSAl Viro if (flags & MNT_NODEV) 20583bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 20593bd045ccSAl Viro 20603bd045ccSAl Viro if (flags & MNT_NOSUID) 20613bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 20623bd045ccSAl Viro 20633bd045ccSAl Viro if (flags & MNT_NOEXEC) 20643bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 20653bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 20663bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 20673bd045ccSAl Viro flags |= MNT_LOCKED; 20683bd045ccSAl Viro p->mnt.mnt_flags = flags; 20693bd045ccSAl Viro } 20703bd045ccSAl Viro } 20713bd045ccSAl Viro 20724b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 2073719f5d7fSMiklos Szeredi { 2074315fc83eSAl Viro struct mount *p; 2075719f5d7fSMiklos Szeredi 2076909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 2077fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 20784b8b21f4SAl Viro mnt_release_group_id(p); 2079719f5d7fSMiklos Szeredi } 2080719f5d7fSMiklos Szeredi } 2081719f5d7fSMiklos Szeredi 20824b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 2083719f5d7fSMiklos Szeredi { 2084315fc83eSAl Viro struct mount *p; 2085719f5d7fSMiklos Szeredi 2086909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 2087fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 20884b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 2089719f5d7fSMiklos Szeredi if (err) { 20904b8b21f4SAl Viro cleanup_group_ids(mnt, p); 2091719f5d7fSMiklos Szeredi return err; 2092719f5d7fSMiklos Szeredi } 2093719f5d7fSMiklos Szeredi } 2094719f5d7fSMiklos Szeredi } 2095719f5d7fSMiklos Szeredi 2096719f5d7fSMiklos Szeredi return 0; 2097719f5d7fSMiklos Szeredi } 2098719f5d7fSMiklos Szeredi 2099d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 2100d2921684SEric W. Biederman { 2101d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 2102d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 2103d2921684SEric W. Biederman struct mount *p; 2104d2921684SEric W. Biederman 2105d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 2106d2921684SEric W. Biederman mounts++; 2107d2921684SEric W. Biederman 2108d2921684SEric W. Biederman old = ns->mounts; 2109d2921684SEric W. Biederman pending = ns->pending_mounts; 2110d2921684SEric W. Biederman sum = old + pending; 2111d2921684SEric W. Biederman if ((old > sum) || 2112d2921684SEric W. Biederman (pending > sum) || 2113d2921684SEric W. Biederman (max < sum) || 2114d2921684SEric W. Biederman (mounts > (max - sum))) 2115d2921684SEric W. Biederman return -ENOSPC; 2116d2921684SEric W. Biederman 2117d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 2118d2921684SEric W. Biederman return 0; 2119d2921684SEric W. Biederman } 2120d2921684SEric W. Biederman 2121b90fa9aeSRam Pai /* 2122b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 2123b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 212421444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 212521444403SRam Pai * store the parent mount and mountpoint dentry. 212621444403SRam Pai * (done when source_mnt is moved) 2127b90fa9aeSRam Pai * 2128b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 2129b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 21309676f0c6SRam Pai * --------------------------------------------------------------------------- 2131b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 21329676f0c6SRam Pai * |************************************************************************** 21339676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21349676f0c6SRam Pai * | dest | | | | | 21359676f0c6SRam Pai * | | | | | | | 21369676f0c6SRam Pai * | v | | | | | 21379676f0c6SRam Pai * |************************************************************************** 21389676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 21395afe0022SRam Pai * | | | | | | 21409676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 21419676f0c6SRam Pai * *************************************************************************** 2142b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 2143b90fa9aeSRam Pai * destination mount. 2144b90fa9aeSRam Pai * 2145b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2146b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2147b90fa9aeSRam Pai * the peer group of the source mount. 2148b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2149b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2150b90fa9aeSRam Pai * mount. 21515afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 21525afe0022SRam Pai * of the destination mount and the cloned mount is made slave 21535afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 21545afe0022SRam Pai * is marked as 'shared and slave'. 21555afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 21565afe0022SRam Pai * source mount. 21575afe0022SRam Pai * 21589676f0c6SRam Pai * --------------------------------------------------------------------------- 215921444403SRam Pai * | MOVE MOUNT OPERATION | 21609676f0c6SRam Pai * |************************************************************************** 21619676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21629676f0c6SRam Pai * | dest | | | | | 21639676f0c6SRam Pai * | | | | | | | 21649676f0c6SRam Pai * | v | | | | | 21659676f0c6SRam Pai * |************************************************************************** 21669676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 21675afe0022SRam Pai * | | | | | | 21689676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 21699676f0c6SRam Pai * *************************************************************************** 21705afe0022SRam Pai * 21715afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 21725afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 217321444403SRam Pai * (+*) the mount is moved to the destination. 21745afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 21755afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 21765afe0022SRam Pai * the mount is marked as 'shared and slave'. 21775afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2178b90fa9aeSRam Pai * 2179b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2180b90fa9aeSRam Pai * applied to each mount in the tree. 2181b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2182b90fa9aeSRam Pai * in allocations. 2183b90fa9aeSRam Pai */ 21840fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 218584d17192SAl Viro struct mount *dest_mnt, 218684d17192SAl Viro struct mountpoint *dest_mp, 21872763d119SAl Viro bool moving) 2188b90fa9aeSRam Pai { 21893bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 219038129a13SAl Viro HLIST_HEAD(tree_list); 2191d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 21921064f874SEric W. Biederman struct mountpoint *smp; 2193315fc83eSAl Viro struct mount *child, *p; 219438129a13SAl Viro struct hlist_node *n; 2195719f5d7fSMiklos Szeredi int err; 2196b90fa9aeSRam Pai 21971064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 21981064f874SEric W. Biederman * to be tucked under other mounts. 21991064f874SEric W. Biederman */ 22001064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 22011064f874SEric W. Biederman if (IS_ERR(smp)) 22021064f874SEric W. Biederman return PTR_ERR(smp); 22031064f874SEric W. Biederman 2204d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 22052763d119SAl Viro if (!moving) { 2206d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2207d2921684SEric W. Biederman if (err) 2208d2921684SEric W. Biederman goto out; 2209d2921684SEric W. Biederman } 2210d2921684SEric W. Biederman 2211fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 22120fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2213719f5d7fSMiklos Szeredi if (err) 2214719f5d7fSMiklos Szeredi goto out; 221584d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2216f2ebb3a9SAl Viro lock_mount_hash(); 2217719f5d7fSMiklos Szeredi if (err) 2218719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2219909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 22200f0afb1dSAl Viro set_mnt_shared(p); 22210b1b901bSAl Viro } else { 22220b1b901bSAl Viro lock_mount_hash(); 2223b90fa9aeSRam Pai } 22242763d119SAl Viro if (moving) { 22252763d119SAl Viro unhash_mnt(source_mnt); 222684d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2227143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 222821444403SRam Pai } else { 222944dfd84aSDavid Howells if (source_mnt->mnt_ns) { 223044dfd84aSDavid Howells /* move from anon - the caller will destroy */ 223144dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 223244dfd84aSDavid Howells } 223384d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 22341064f874SEric W. Biederman commit_tree(source_mnt); 223521444403SRam Pai } 2236b90fa9aeSRam Pai 223738129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 22381d6a32acSAl Viro struct mount *q; 223938129a13SAl Viro hlist_del_init(&child->mnt_hash); 22401064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 22411d6a32acSAl Viro child->mnt_mountpoint); 22421064f874SEric W. Biederman if (q) 22431064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 22443bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 22453bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 22463bd045ccSAl Viro lock_mnt_tree(child); 2247d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 22481064f874SEric W. Biederman commit_tree(child); 2249b90fa9aeSRam Pai } 22501064f874SEric W. Biederman put_mountpoint(smp); 2251719ea2fbSAl Viro unlock_mount_hash(); 225299b7db7bSNick Piggin 2253b90fa9aeSRam Pai return 0; 2254719f5d7fSMiklos Szeredi 2255719f5d7fSMiklos Szeredi out_cleanup_ids: 2256f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2257f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2258d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2259e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2260f2ebb3a9SAl Viro } 2261f2ebb3a9SAl Viro unlock_mount_hash(); 22620fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2263719f5d7fSMiklos Szeredi out: 2264d2921684SEric W. Biederman ns->pending_mounts = 0; 22651064f874SEric W. Biederman 22661064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 22671064f874SEric W. Biederman put_mountpoint(smp); 22681064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 22691064f874SEric W. Biederman 2270719f5d7fSMiklos Szeredi return err; 2271b90fa9aeSRam Pai } 2272b90fa9aeSRam Pai 227384d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2274b12cea91SAl Viro { 2275b12cea91SAl Viro struct vfsmount *mnt; 227684d17192SAl Viro struct dentry *dentry = path->dentry; 2277b12cea91SAl Viro retry: 22785955102cSAl Viro inode_lock(dentry->d_inode); 227984d17192SAl Viro if (unlikely(cant_mount(dentry))) { 22805955102cSAl Viro inode_unlock(dentry->d_inode); 228184d17192SAl Viro return ERR_PTR(-ENOENT); 2282b12cea91SAl Viro } 228397216be0SAl Viro namespace_lock(); 2284b12cea91SAl Viro mnt = lookup_mnt(path); 228584d17192SAl Viro if (likely(!mnt)) { 22863895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 228784d17192SAl Viro if (IS_ERR(mp)) { 228897216be0SAl Viro namespace_unlock(); 22895955102cSAl Viro inode_unlock(dentry->d_inode); 229084d17192SAl Viro return mp; 229184d17192SAl Viro } 229284d17192SAl Viro return mp; 229384d17192SAl Viro } 229497216be0SAl Viro namespace_unlock(); 22955955102cSAl Viro inode_unlock(path->dentry->d_inode); 2296b12cea91SAl Viro path_put(path); 2297b12cea91SAl Viro path->mnt = mnt; 229884d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2299b12cea91SAl Viro goto retry; 2300b12cea91SAl Viro } 2301b12cea91SAl Viro 230284d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2303b12cea91SAl Viro { 230484d17192SAl Viro struct dentry *dentry = where->m_dentry; 23053895dbf8SEric W. Biederman 23063895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 230784d17192SAl Viro put_mountpoint(where); 23083895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 23093895dbf8SEric W. Biederman 2310328e6d90SAl Viro namespace_unlock(); 23115955102cSAl Viro inode_unlock(dentry->d_inode); 2312b12cea91SAl Viro } 2313b12cea91SAl Viro 231484d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 23151da177e4SLinus Torvalds { 2316e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 23171da177e4SLinus Torvalds return -EINVAL; 23181da177e4SLinus Torvalds 2319e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2320e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 23211da177e4SLinus Torvalds return -ENOTDIR; 23221da177e4SLinus Torvalds 23232763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 23241da177e4SLinus Torvalds } 23251da177e4SLinus Torvalds 23261da177e4SLinus Torvalds /* 23277a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 23287a2e8a8fSValerie Aurora */ 23297a2e8a8fSValerie Aurora 2330e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 23317a2e8a8fSValerie Aurora { 2332e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 23337a2e8a8fSValerie Aurora 23347a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 23357a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 23367a2e8a8fSValerie Aurora return 0; 23377a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 23387a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 23397a2e8a8fSValerie Aurora return 0; 23407a2e8a8fSValerie Aurora return type; 23417a2e8a8fSValerie Aurora } 23427a2e8a8fSValerie Aurora 23437a2e8a8fSValerie Aurora /* 234407b20889SRam Pai * recursively change the type of the mountpoint. 234507b20889SRam Pai */ 2346e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 234707b20889SRam Pai { 2348315fc83eSAl Viro struct mount *m; 23494b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2350e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 23517a2e8a8fSValerie Aurora int type; 2352719f5d7fSMiklos Szeredi int err = 0; 235307b20889SRam Pai 23542d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 235507b20889SRam Pai return -EINVAL; 235607b20889SRam Pai 2357e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 23587a2e8a8fSValerie Aurora if (!type) 23597a2e8a8fSValerie Aurora return -EINVAL; 23607a2e8a8fSValerie Aurora 236197216be0SAl Viro namespace_lock(); 2362719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2363719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2364719f5d7fSMiklos Szeredi if (err) 2365719f5d7fSMiklos Szeredi goto out_unlock; 2366719f5d7fSMiklos Szeredi } 2367719f5d7fSMiklos Szeredi 2368719ea2fbSAl Viro lock_mount_hash(); 2369909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 23700f0afb1dSAl Viro change_mnt_propagation(m, type); 2371719ea2fbSAl Viro unlock_mount_hash(); 2372719f5d7fSMiklos Szeredi 2373719f5d7fSMiklos Szeredi out_unlock: 237497216be0SAl Viro namespace_unlock(); 2375719f5d7fSMiklos Szeredi return err; 237607b20889SRam Pai } 237707b20889SRam Pai 2378a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2379a07b2000SAl Viro { 2380a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2381a07b2000SAl Viro 2382a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2383a07b2000SAl Viro return mnt; 2384a07b2000SAl Viro 2385a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2386a07b2000SAl Viro return mnt; 2387a07b2000SAl Viro 2388a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2389a07b2000SAl Viro return mnt; 2390a07b2000SAl Viro 2391a07b2000SAl Viro if (recurse) 2392a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2393a07b2000SAl Viro else 2394a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2395a07b2000SAl Viro 2396a07b2000SAl Viro if (!IS_ERR(mnt)) 2397a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2398a07b2000SAl Viro 2399a07b2000SAl Viro return mnt; 2400a07b2000SAl Viro } 2401a07b2000SAl Viro 240207b20889SRam Pai /* 24031da177e4SLinus Torvalds * do loopback mount. 24041da177e4SLinus Torvalds */ 2405808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 24062dafe1c4SEric Sandeen int recurse) 24071da177e4SLinus Torvalds { 24082d92ab3cSAl Viro struct path old_path; 2409a07b2000SAl Viro struct mount *mnt = NULL, *parent; 241084d17192SAl Viro struct mountpoint *mp; 241157eccb83SAl Viro int err; 24121da177e4SLinus Torvalds if (!old_name || !*old_name) 24131da177e4SLinus Torvalds return -EINVAL; 2414815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 24151da177e4SLinus Torvalds if (err) 24161da177e4SLinus Torvalds return err; 24171da177e4SLinus Torvalds 24188823c079SEric W. Biederman err = -EINVAL; 24194ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 24208823c079SEric W. Biederman goto out; 24218823c079SEric W. Biederman 242284d17192SAl Viro mp = lock_mount(path); 2423a07b2000SAl Viro if (IS_ERR(mp)) { 242484d17192SAl Viro err = PTR_ERR(mp); 24259676f0c6SRam Pai goto out; 2426a07b2000SAl Viro } 24279676f0c6SRam Pai 242884d17192SAl Viro parent = real_mount(path->mnt); 2429e149ed2bSAl Viro if (!check_mnt(parent)) 2430e149ed2bSAl Viro goto out2; 2431e149ed2bSAl Viro 2432a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2433be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2434be34d1a3SDavid Howells err = PTR_ERR(mnt); 2435e9c5d8a5SAndrey Vagin goto out2; 2436be34d1a3SDavid Howells } 2437ccd48bc7SAl Viro 243884d17192SAl Viro err = graft_tree(mnt, parent, mp); 24391da177e4SLinus Torvalds if (err) { 2440719ea2fbSAl Viro lock_mount_hash(); 2441e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2442719ea2fbSAl Viro unlock_mount_hash(); 24435b83d2c5SRam Pai } 2444b12cea91SAl Viro out2: 244584d17192SAl Viro unlock_mount(mp); 2446ccd48bc7SAl Viro out: 24472d92ab3cSAl Viro path_put(&old_path); 24481da177e4SLinus Torvalds return err; 24491da177e4SLinus Torvalds } 24501da177e4SLinus Torvalds 2451a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2452a07b2000SAl Viro { 2453a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2454a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2455a07b2000SAl Viro struct mount *mnt, *p; 2456a07b2000SAl Viro struct file *file; 2457a07b2000SAl Viro 2458a07b2000SAl Viro if (IS_ERR(ns)) 2459a07b2000SAl Viro return ERR_CAST(ns); 2460a07b2000SAl Viro 2461a07b2000SAl Viro namespace_lock(); 2462a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2463a07b2000SAl Viro if (IS_ERR(mnt)) { 2464a07b2000SAl Viro namespace_unlock(); 2465a07b2000SAl Viro free_mnt_ns(ns); 2466a07b2000SAl Viro return ERR_CAST(mnt); 2467a07b2000SAl Viro } 2468a07b2000SAl Viro 2469a07b2000SAl Viro lock_mount_hash(); 2470a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2471a07b2000SAl Viro p->mnt_ns = ns; 2472a07b2000SAl Viro ns->mounts++; 2473a07b2000SAl Viro } 2474a07b2000SAl Viro ns->root = mnt; 2475a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2476a07b2000SAl Viro mntget(&mnt->mnt); 2477a07b2000SAl Viro unlock_mount_hash(); 2478a07b2000SAl Viro namespace_unlock(); 2479a07b2000SAl Viro 2480a07b2000SAl Viro mntput(path->mnt); 2481a07b2000SAl Viro path->mnt = &mnt->mnt; 2482a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2483a07b2000SAl Viro if (IS_ERR(file)) 2484a07b2000SAl Viro dissolve_on_fput(path->mnt); 2485a07b2000SAl Viro else 2486a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2487a07b2000SAl Viro return file; 2488a07b2000SAl Viro } 2489a07b2000SAl Viro 24902658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2491a07b2000SAl Viro { 2492a07b2000SAl Viro struct file *file; 2493a07b2000SAl Viro struct path path; 2494a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2495a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2496a07b2000SAl Viro int error; 2497a07b2000SAl Viro int fd; 2498a07b2000SAl Viro 2499a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2500a07b2000SAl Viro 2501a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2502a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2503a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2504a07b2000SAl Viro return -EINVAL; 2505a07b2000SAl Viro 2506a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2507a07b2000SAl Viro return -EINVAL; 2508a07b2000SAl Viro 2509a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2510a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2511a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2512a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2513a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2514a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2515a07b2000SAl Viro 2516a07b2000SAl Viro if (detached && !may_mount()) 2517a07b2000SAl Viro return -EPERM; 2518a07b2000SAl Viro 2519a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2520a07b2000SAl Viro if (fd < 0) 2521a07b2000SAl Viro return fd; 2522a07b2000SAl Viro 2523a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2524a07b2000SAl Viro if (unlikely(error)) { 2525a07b2000SAl Viro file = ERR_PTR(error); 2526a07b2000SAl Viro } else { 2527a07b2000SAl Viro if (detached) 2528a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2529a07b2000SAl Viro else 2530a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2531a07b2000SAl Viro path_put(&path); 2532a07b2000SAl Viro } 2533a07b2000SAl Viro if (IS_ERR(file)) { 2534a07b2000SAl Viro put_unused_fd(fd); 2535a07b2000SAl Viro return PTR_ERR(file); 2536a07b2000SAl Viro } 2537a07b2000SAl Viro fd_install(fd, file); 2538a07b2000SAl Viro return fd; 2539a07b2000SAl Viro } 2540a07b2000SAl Viro 254143f5e655SDavid Howells /* 254243f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 254343f5e655SDavid Howells * 254443f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 254543f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 254643f5e655SDavid Howells */ 254743f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 25482e4b7fcdSDave Hansen { 254943f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 25502e4b7fcdSDave Hansen 255143f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 255243f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 255343f5e655SDavid Howells return false; 255443f5e655SDavid Howells 255543f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 255643f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 255743f5e655SDavid Howells return false; 255843f5e655SDavid Howells 255943f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 256043f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 256143f5e655SDavid Howells return false; 256243f5e655SDavid Howells 256343f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 256443f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 256543f5e655SDavid Howells return false; 256643f5e655SDavid Howells 256743f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 256843f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 256943f5e655SDavid Howells return false; 257043f5e655SDavid Howells 257143f5e655SDavid Howells return true; 257243f5e655SDavid Howells } 257343f5e655SDavid Howells 257443f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 257543f5e655SDavid Howells { 257643f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 257743f5e655SDavid Howells 257843f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 25792e4b7fcdSDave Hansen return 0; 25802e4b7fcdSDave Hansen 25812e4b7fcdSDave Hansen if (readonly_request) 258243f5e655SDavid Howells return mnt_make_readonly(mnt); 258343f5e655SDavid Howells 258468847c94SChristian Brauner mnt->mnt.mnt_flags &= ~MNT_READONLY; 258568847c94SChristian Brauner return 0; 258643f5e655SDavid Howells } 258743f5e655SDavid Howells 258843f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 258943f5e655SDavid Howells { 259043f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 259143f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 259243f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 259343f5e655SDavid Howells } 259443f5e655SDavid Howells 2595f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2596f8b92ba6SDeepa Dinamani { 2597f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2598f8b92ba6SDeepa Dinamani 2599f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2600*a128b054SAnthony Iliopoulos (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) && 2601f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2602f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2603f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2604f8b92ba6SDeepa Dinamani struct tm tm; 2605f8b92ba6SDeepa Dinamani 2606f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2607f8b92ba6SDeepa Dinamani 26080ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 26090ecee669SEric Biggers sb->s_type->name, 26100ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 26110ecee669SEric Biggers mntpath, 2612f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2613f8b92ba6SDeepa Dinamani 2614f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2615*a128b054SAnthony Iliopoulos sb->s_iflags |= SB_I_TS_EXPIRY_WARNED; 2616f8b92ba6SDeepa Dinamani } 2617f8b92ba6SDeepa Dinamani } 2618f8b92ba6SDeepa Dinamani 261943f5e655SDavid Howells /* 262043f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 262143f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 262243f5e655SDavid Howells * to mount(2). 262343f5e655SDavid Howells */ 262443f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 262543f5e655SDavid Howells { 262643f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 262743f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 262843f5e655SDavid Howells int ret; 262943f5e655SDavid Howells 263043f5e655SDavid Howells if (!check_mnt(mnt)) 263143f5e655SDavid Howells return -EINVAL; 263243f5e655SDavid Howells 263343f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 263443f5e655SDavid Howells return -EINVAL; 263543f5e655SDavid Howells 263643f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 263743f5e655SDavid Howells return -EPERM; 263843f5e655SDavid Howells 2639e58ace1aSChristian Brauner /* 2640e58ace1aSChristian Brauner * We're only checking whether the superblock is read-only not 2641e58ace1aSChristian Brauner * changing it, so only take down_read(&sb->s_umount). 2642e58ace1aSChristian Brauner */ 2643e58ace1aSChristian Brauner down_read(&sb->s_umount); 264468847c94SChristian Brauner lock_mount_hash(); 264543f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 264643f5e655SDavid Howells if (ret == 0) 264743f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 264868847c94SChristian Brauner unlock_mount_hash(); 2649e58ace1aSChristian Brauner up_read(&sb->s_umount); 2650f8b92ba6SDeepa Dinamani 2651f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2652f8b92ba6SDeepa Dinamani 265343f5e655SDavid Howells return ret; 26542e4b7fcdSDave Hansen } 26552e4b7fcdSDave Hansen 26561da177e4SLinus Torvalds /* 26571da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 26581da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 26591da177e4SLinus Torvalds * on it - tough luck. 26601da177e4SLinus Torvalds */ 2661e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2662e462ec50SDavid Howells int mnt_flags, void *data) 26631da177e4SLinus Torvalds { 26641da177e4SLinus Torvalds int err; 26652d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2666143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 26678d0347f6SDavid Howells struct fs_context *fc; 26681da177e4SLinus Torvalds 2669143c8c91SAl Viro if (!check_mnt(mnt)) 26701da177e4SLinus Torvalds return -EINVAL; 26711da177e4SLinus Torvalds 26722d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 26731da177e4SLinus Torvalds return -EINVAL; 26741da177e4SLinus Torvalds 267543f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 267607b64558SEric W. Biederman return -EPERM; 26779566d674SEric W. Biederman 26788d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 26798d0347f6SDavid Howells if (IS_ERR(fc)) 26808d0347f6SDavid Howells return PTR_ERR(fc); 2681ff36fe2cSEric Paris 2682b330966fSMiklos Szeredi fc->oldapi = true; 26838d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 26848d0347f6SDavid Howells if (!err) { 26851da177e4SLinus Torvalds down_write(&sb->s_umount); 268657eccb83SAl Viro err = -EPERM; 268743f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 26888d0347f6SDavid Howells err = reconfigure_super(fc); 268968847c94SChristian Brauner if (!err) { 269068847c94SChristian Brauner lock_mount_hash(); 269143f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 269268847c94SChristian Brauner unlock_mount_hash(); 269368847c94SChristian Brauner } 26940e55a7ccSDan Williams } 26956339dab8SAl Viro up_write(&sb->s_umount); 26968d0347f6SDavid Howells } 2697f8b92ba6SDeepa Dinamani 2698f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2699f8b92ba6SDeepa Dinamani 27008d0347f6SDavid Howells put_fs_context(fc); 27011da177e4SLinus Torvalds return err; 27021da177e4SLinus Torvalds } 27031da177e4SLinus Torvalds 2704cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 27059676f0c6SRam Pai { 2706315fc83eSAl Viro struct mount *p; 2707909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2708fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 27099676f0c6SRam Pai return 1; 27109676f0c6SRam Pai } 27119676f0c6SRam Pai return 0; 27129676f0c6SRam Pai } 27139676f0c6SRam Pai 271444dfd84aSDavid Howells /* 271544dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 271644dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 271744dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 271844dfd84aSDavid Howells * cycles to be made. 271944dfd84aSDavid Howells */ 272044dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 272144dfd84aSDavid Howells { 272244dfd84aSDavid Howells struct mount *p; 272344dfd84aSDavid Howells bool ret = false; 272444dfd84aSDavid Howells 272544dfd84aSDavid Howells lock_mount_hash(); 272644dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 272744dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 272844dfd84aSDavid Howells goto out; 272944dfd84aSDavid Howells 273044dfd84aSDavid Howells ret = true; 273144dfd84aSDavid Howells out: 273244dfd84aSDavid Howells unlock_mount_hash(); 273344dfd84aSDavid Howells return ret; 273444dfd84aSDavid Howells } 273544dfd84aSDavid Howells 27369ffb14efSPavel Tikhomirov static int do_set_group(struct path *from_path, struct path *to_path) 27379ffb14efSPavel Tikhomirov { 27389ffb14efSPavel Tikhomirov struct mount *from, *to; 27399ffb14efSPavel Tikhomirov int err; 27409ffb14efSPavel Tikhomirov 27419ffb14efSPavel Tikhomirov from = real_mount(from_path->mnt); 27429ffb14efSPavel Tikhomirov to = real_mount(to_path->mnt); 27439ffb14efSPavel Tikhomirov 27449ffb14efSPavel Tikhomirov namespace_lock(); 27459ffb14efSPavel Tikhomirov 27469ffb14efSPavel Tikhomirov err = -EINVAL; 27479ffb14efSPavel Tikhomirov /* To and From must be mounted */ 27489ffb14efSPavel Tikhomirov if (!is_mounted(&from->mnt)) 27499ffb14efSPavel Tikhomirov goto out; 27509ffb14efSPavel Tikhomirov if (!is_mounted(&to->mnt)) 27519ffb14efSPavel Tikhomirov goto out; 27529ffb14efSPavel Tikhomirov 27539ffb14efSPavel Tikhomirov err = -EPERM; 27549ffb14efSPavel Tikhomirov /* We should be allowed to modify mount namespaces of both mounts */ 27559ffb14efSPavel Tikhomirov if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27569ffb14efSPavel Tikhomirov goto out; 27579ffb14efSPavel Tikhomirov if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) 27589ffb14efSPavel Tikhomirov goto out; 27599ffb14efSPavel Tikhomirov 27609ffb14efSPavel Tikhomirov err = -EINVAL; 27619ffb14efSPavel Tikhomirov /* To and From paths should be mount roots */ 27629ffb14efSPavel Tikhomirov if (from_path->dentry != from_path->mnt->mnt_root) 27639ffb14efSPavel Tikhomirov goto out; 27649ffb14efSPavel Tikhomirov if (to_path->dentry != to_path->mnt->mnt_root) 27659ffb14efSPavel Tikhomirov goto out; 27669ffb14efSPavel Tikhomirov 27679ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed across same superblock */ 27689ffb14efSPavel Tikhomirov if (from->mnt.mnt_sb != to->mnt.mnt_sb) 27699ffb14efSPavel Tikhomirov goto out; 27709ffb14efSPavel Tikhomirov 27719ffb14efSPavel Tikhomirov /* From mount root should be wider than To mount root */ 27729ffb14efSPavel Tikhomirov if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) 27739ffb14efSPavel Tikhomirov goto out; 27749ffb14efSPavel Tikhomirov 27759ffb14efSPavel Tikhomirov /* From mount should not have locked children in place of To's root */ 27769ffb14efSPavel Tikhomirov if (has_locked_children(from, to->mnt.mnt_root)) 27779ffb14efSPavel Tikhomirov goto out; 27789ffb14efSPavel Tikhomirov 27799ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed on private mounts */ 27809ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) 27819ffb14efSPavel Tikhomirov goto out; 27829ffb14efSPavel Tikhomirov 27839ffb14efSPavel Tikhomirov /* From should not be private */ 27849ffb14efSPavel Tikhomirov if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) 27859ffb14efSPavel Tikhomirov goto out; 27869ffb14efSPavel Tikhomirov 27879ffb14efSPavel Tikhomirov if (IS_MNT_SLAVE(from)) { 27889ffb14efSPavel Tikhomirov struct mount *m = from->mnt_master; 27899ffb14efSPavel Tikhomirov 27909ffb14efSPavel Tikhomirov list_add(&to->mnt_slave, &m->mnt_slave_list); 27919ffb14efSPavel Tikhomirov to->mnt_master = m; 27929ffb14efSPavel Tikhomirov } 27939ffb14efSPavel Tikhomirov 27949ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(from)) { 27959ffb14efSPavel Tikhomirov to->mnt_group_id = from->mnt_group_id; 27969ffb14efSPavel Tikhomirov list_add(&to->mnt_share, &from->mnt_share); 27979ffb14efSPavel Tikhomirov lock_mount_hash(); 27989ffb14efSPavel Tikhomirov set_mnt_shared(to); 27999ffb14efSPavel Tikhomirov unlock_mount_hash(); 28009ffb14efSPavel Tikhomirov } 28019ffb14efSPavel Tikhomirov 28029ffb14efSPavel Tikhomirov err = 0; 28039ffb14efSPavel Tikhomirov out: 28049ffb14efSPavel Tikhomirov namespace_unlock(); 28059ffb14efSPavel Tikhomirov return err; 28069ffb14efSPavel Tikhomirov } 28079ffb14efSPavel Tikhomirov 28082db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 28091da177e4SLinus Torvalds { 281044dfd84aSDavid Howells struct mnt_namespace *ns; 2811676da58dSAl Viro struct mount *p; 28120fb54e50SAl Viro struct mount *old; 28132763d119SAl Viro struct mount *parent; 28142763d119SAl Viro struct mountpoint *mp, *old_mp; 281557eccb83SAl Viro int err; 281644dfd84aSDavid Howells bool attached; 28171da177e4SLinus Torvalds 28182db154b3SDavid Howells mp = lock_mount(new_path); 281984d17192SAl Viro if (IS_ERR(mp)) 28202db154b3SDavid Howells return PTR_ERR(mp); 2821cc53ce53SDavid Howells 28222db154b3SDavid Howells old = real_mount(old_path->mnt); 28232db154b3SDavid Howells p = real_mount(new_path->mnt); 28242763d119SAl Viro parent = old->mnt_parent; 282544dfd84aSDavid Howells attached = mnt_has_parent(old); 28262763d119SAl Viro old_mp = old->mnt_mp; 282744dfd84aSDavid Howells ns = old->mnt_ns; 2828143c8c91SAl Viro 28291da177e4SLinus Torvalds err = -EINVAL; 283044dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 283144dfd84aSDavid Howells if (!check_mnt(p)) 28322db154b3SDavid Howells goto out; 28331da177e4SLinus Torvalds 2834570d7a98SEric Biggers /* The thing moved must be mounted... */ 2835570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 283644dfd84aSDavid Howells goto out; 283744dfd84aSDavid Howells 2838570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2839570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 28402db154b3SDavid Howells goto out; 28411da177e4SLinus Torvalds 28422db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 28432db154b3SDavid Howells goto out; 28442db154b3SDavid Howells 28452db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 28462db154b3SDavid Howells goto out; 28472db154b3SDavid Howells 28482db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 28492db154b3SDavid Howells d_is_dir(old_path->dentry)) 28502db154b3SDavid Howells goto out; 285121444403SRam Pai /* 285221444403SRam Pai * Don't move a mount residing in a shared parent. 285321444403SRam Pai */ 28542763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 28552db154b3SDavid Howells goto out; 28569676f0c6SRam Pai /* 28579676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 28589676f0c6SRam Pai * mount which is shared. 28599676f0c6SRam Pai */ 2860fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 28612db154b3SDavid Howells goto out; 28621da177e4SLinus Torvalds err = -ELOOP; 286344dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 286444dfd84aSDavid Howells goto out; 2865fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2866676da58dSAl Viro if (p == old) 28672db154b3SDavid Howells goto out; 28681da177e4SLinus Torvalds 28692db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 28702763d119SAl Viro attached); 28714ac91378SJan Blunck if (err) 28722db154b3SDavid Howells goto out; 28731da177e4SLinus Torvalds 28741da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 28751da177e4SLinus Torvalds * automatically */ 28766776db3dSAl Viro list_del_init(&old->mnt_expire); 28772763d119SAl Viro if (attached) 28782763d119SAl Viro put_mountpoint(old_mp); 28791da177e4SLinus Torvalds out: 28802db154b3SDavid Howells unlock_mount(mp); 288144dfd84aSDavid Howells if (!err) { 28822763d119SAl Viro if (attached) 28832763d119SAl Viro mntput_no_expire(parent); 28842763d119SAl Viro else 288544dfd84aSDavid Howells free_mnt_ns(ns); 288644dfd84aSDavid Howells } 28872db154b3SDavid Howells return err; 28882db154b3SDavid Howells } 28892db154b3SDavid Howells 28902db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 28912db154b3SDavid Howells { 28922db154b3SDavid Howells struct path old_path; 28932db154b3SDavid Howells int err; 28942db154b3SDavid Howells 28952db154b3SDavid Howells if (!old_name || !*old_name) 28962db154b3SDavid Howells return -EINVAL; 28972db154b3SDavid Howells 28982db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 28992db154b3SDavid Howells if (err) 29002db154b3SDavid Howells return err; 29012db154b3SDavid Howells 29022db154b3SDavid Howells err = do_move_mount(&old_path, path); 29032d92ab3cSAl Viro path_put(&old_path); 29041da177e4SLinus Torvalds return err; 29051da177e4SLinus Torvalds } 29061da177e4SLinus Torvalds 29079d412a43SAl Viro /* 29089d412a43SAl Viro * add a mount into a namespace's mount tree 29099d412a43SAl Viro */ 29108f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 29118f11538eSAl Viro struct path *path, int mnt_flags) 29129d412a43SAl Viro { 29138f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 29149d412a43SAl Viro 2915f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 29169d412a43SAl Viro 291784d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2918156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2919156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 29208f11538eSAl Viro return -EINVAL; 2921156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 292284d17192SAl Viro if (!parent->mnt_ns) 29238f11538eSAl Viro return -EINVAL; 2924156cacb1SAl Viro } 29259d412a43SAl Viro 29269d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 292795bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 29289d412a43SAl Viro path->mnt->mnt_root == path->dentry) 29298f11538eSAl Viro return -EBUSY; 29309d412a43SAl Viro 2931e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 29328f11538eSAl Viro return -EINVAL; 29339d412a43SAl Viro 293495bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 29358f11538eSAl Viro return graft_tree(newmnt, parent, mp); 29369d412a43SAl Viro } 2937b1e75df4SAl Viro 2938132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2939132e4608SDavid Howells 2940132e4608SDavid Howells /* 2941132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2942132e4608SDavid Howells * be added to the namespace tree. 2943132e4608SDavid Howells */ 2944132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2945132e4608SDavid Howells unsigned int mnt_flags) 2946132e4608SDavid Howells { 2947132e4608SDavid Howells struct vfsmount *mnt; 29488f11538eSAl Viro struct mountpoint *mp; 2949132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2950132e4608SDavid Howells int error; 2951132e4608SDavid Howells 2952c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2953c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2954c9ce29edSAl Viro error = -EPERM; 2955c9ce29edSAl Viro 2956c9ce29edSAl Viro if (unlikely(error)) { 2957c9ce29edSAl Viro fc_drop_locked(fc); 2958c9ce29edSAl Viro return error; 2959132e4608SDavid Howells } 2960132e4608SDavid Howells 2961132e4608SDavid Howells up_write(&sb->s_umount); 2962132e4608SDavid Howells 2963132e4608SDavid Howells mnt = vfs_create_mount(fc); 2964132e4608SDavid Howells if (IS_ERR(mnt)) 2965132e4608SDavid Howells return PTR_ERR(mnt); 2966132e4608SDavid Howells 2967f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 2968f8b92ba6SDeepa Dinamani 29698f11538eSAl Viro mp = lock_mount(mountpoint); 29708f11538eSAl Viro if (IS_ERR(mp)) { 29718f11538eSAl Viro mntput(mnt); 29728f11538eSAl Viro return PTR_ERR(mp); 29738f11538eSAl Viro } 29748f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 29758f11538eSAl Viro unlock_mount(mp); 29760ecee669SEric Biggers if (error < 0) 29770ecee669SEric Biggers mntput(mnt); 2978f8b92ba6SDeepa Dinamani return error; 2979f8b92ba6SDeepa Dinamani } 2980f8b92ba6SDeepa Dinamani 29811da177e4SLinus Torvalds /* 29821da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 29831da177e4SLinus Torvalds * namespace's tree 29841da177e4SLinus Torvalds */ 2985e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2986808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 29871da177e4SLinus Torvalds { 29880c55cfc4SEric W. Biederman struct file_system_type *type; 2989a0c9a8b8SAl Viro struct fs_context *fc; 2990a0c9a8b8SAl Viro const char *subtype = NULL; 2991a0c9a8b8SAl Viro int err = 0; 29921da177e4SLinus Torvalds 29930c55cfc4SEric W. Biederman if (!fstype) 29941da177e4SLinus Torvalds return -EINVAL; 29951da177e4SLinus Torvalds 29960c55cfc4SEric W. Biederman type = get_fs_type(fstype); 29970c55cfc4SEric W. Biederman if (!type) 29980c55cfc4SEric W. Biederman return -ENODEV; 29990c55cfc4SEric W. Biederman 3000a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 3001a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 3002a0c9a8b8SAl Viro if (subtype) { 3003a0c9a8b8SAl Viro subtype++; 3004a0c9a8b8SAl Viro if (!*subtype) { 30050c55cfc4SEric W. Biederman put_filesystem(type); 3006a0c9a8b8SAl Viro return -EINVAL; 3007a0c9a8b8SAl Viro } 3008a0c9a8b8SAl Viro } 30098654df4eSEric W. Biederman } 30108654df4eSEric W. Biederman 3011a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 3012a0c9a8b8SAl Viro put_filesystem(type); 3013a0c9a8b8SAl Viro if (IS_ERR(fc)) 3014a0c9a8b8SAl Viro return PTR_ERR(fc); 3015a0c9a8b8SAl Viro 30163e1aeb00SDavid Howells if (subtype) 30173e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 30183e1aeb00SDavid Howells subtype, strlen(subtype)); 30193e1aeb00SDavid Howells if (!err && name) 30203e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 3021a0c9a8b8SAl Viro if (!err) 3022a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 3023c3aabf07SAl Viro if (!err && !mount_capable(fc)) 3024c3aabf07SAl Viro err = -EPERM; 3025a0c9a8b8SAl Viro if (!err) 3026a0c9a8b8SAl Viro err = vfs_get_tree(fc); 3027132e4608SDavid Howells if (!err) 3028132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 3029a0c9a8b8SAl Viro 3030a0c9a8b8SAl Viro put_fs_context(fc); 303115f9a3f3SAl Viro return err; 30321da177e4SLinus Torvalds } 30331da177e4SLinus Torvalds 303419a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 303519a167afSAl Viro { 303626df6034SAl Viro struct dentry *dentry = path->dentry; 30378f11538eSAl Viro struct mountpoint *mp; 303825e195aaSAl Viro struct mount *mnt; 303919a167afSAl Viro int err; 304025e195aaSAl Viro 304125e195aaSAl Viro if (!m) 304225e195aaSAl Viro return 0; 304325e195aaSAl Viro if (IS_ERR(m)) 304425e195aaSAl Viro return PTR_ERR(m); 304525e195aaSAl Viro 304625e195aaSAl Viro mnt = real_mount(m); 304719a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 304819a167afSAl Viro * expired before we get a chance to add it 304919a167afSAl Viro */ 30506776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 305119a167afSAl Viro 305219a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 305326df6034SAl Viro m->mnt_root == dentry) { 3054b1e75df4SAl Viro err = -ELOOP; 305526df6034SAl Viro goto discard; 305619a167afSAl Viro } 305719a167afSAl Viro 305826df6034SAl Viro /* 305926df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 306026df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 306126df6034SAl Viro * got", not "try to mount it on top". 306226df6034SAl Viro */ 306326df6034SAl Viro inode_lock(dentry->d_inode); 306426df6034SAl Viro namespace_lock(); 306526df6034SAl Viro if (unlikely(cant_mount(dentry))) { 306626df6034SAl Viro err = -ENOENT; 306726df6034SAl Viro goto discard_locked; 306826df6034SAl Viro } 306926df6034SAl Viro rcu_read_lock(); 307026df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 307126df6034SAl Viro rcu_read_unlock(); 307226df6034SAl Viro err = 0; 307326df6034SAl Viro goto discard_locked; 307426df6034SAl Viro } 307526df6034SAl Viro rcu_read_unlock(); 307626df6034SAl Viro mp = get_mountpoint(dentry); 30778f11538eSAl Viro if (IS_ERR(mp)) { 30788f11538eSAl Viro err = PTR_ERR(mp); 307926df6034SAl Viro goto discard_locked; 30808f11538eSAl Viro } 308126df6034SAl Viro 30828f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 30838f11538eSAl Viro unlock_mount(mp); 308426df6034SAl Viro if (unlikely(err)) 308526df6034SAl Viro goto discard; 308626df6034SAl Viro mntput(m); 3087b1e75df4SAl Viro return 0; 308826df6034SAl Viro 308926df6034SAl Viro discard_locked: 309026df6034SAl Viro namespace_unlock(); 309126df6034SAl Viro inode_unlock(dentry->d_inode); 309226df6034SAl Viro discard: 3093b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 30946776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 309597216be0SAl Viro namespace_lock(); 30966776db3dSAl Viro list_del_init(&mnt->mnt_expire); 309797216be0SAl Viro namespace_unlock(); 309819a167afSAl Viro } 3099b1e75df4SAl Viro mntput(m); 3100b1e75df4SAl Viro mntput(m); 310119a167afSAl Viro return err; 310219a167afSAl Viro } 310319a167afSAl Viro 3104ea5b778aSDavid Howells /** 3105ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 3106ea5b778aSDavid Howells * @mnt: The mount to list. 3107ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 3108ea5b778aSDavid Howells */ 3109ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 3110ea5b778aSDavid Howells { 311197216be0SAl Viro namespace_lock(); 3112ea5b778aSDavid Howells 31136776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 3114ea5b778aSDavid Howells 311597216be0SAl Viro namespace_unlock(); 3116ea5b778aSDavid Howells } 3117ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 3118ea5b778aSDavid Howells 3119ea5b778aSDavid Howells /* 31201da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 31211da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 31221da177e4SLinus Torvalds * here 31231da177e4SLinus Torvalds */ 31241da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 31251da177e4SLinus Torvalds { 3126761d5c38SAl Viro struct mount *mnt, *next; 31271da177e4SLinus Torvalds LIST_HEAD(graveyard); 31281da177e4SLinus Torvalds 31291da177e4SLinus Torvalds if (list_empty(mounts)) 31301da177e4SLinus Torvalds return; 31311da177e4SLinus Torvalds 313297216be0SAl Viro namespace_lock(); 3133719ea2fbSAl Viro lock_mount_hash(); 31341da177e4SLinus Torvalds 31351da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 31361da177e4SLinus Torvalds * following criteria: 31371da177e4SLinus Torvalds * - only referenced by its parent vfsmount 31381da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 31391da177e4SLinus Torvalds * cleared by mntput()) 31401da177e4SLinus Torvalds */ 31416776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 3142863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 31431ab59738SAl Viro propagate_mount_busy(mnt, 1)) 31441da177e4SLinus Torvalds continue; 31456776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 31461da177e4SLinus Torvalds } 3147bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 31486776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 3149143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 3150e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3151bcc5c7d2SAl Viro } 3152719ea2fbSAl Viro unlock_mount_hash(); 31533ab6abeeSAl Viro namespace_unlock(); 31541da177e4SLinus Torvalds } 31551da177e4SLinus Torvalds 31561da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 31571da177e4SLinus Torvalds 31581da177e4SLinus Torvalds /* 31595528f911STrond Myklebust * Ripoff of 'select_parent()' 31605528f911STrond Myklebust * 31615528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 31625528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 31635528f911STrond Myklebust */ 3164692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 31655528f911STrond Myklebust { 3166692afc31SAl Viro struct mount *this_parent = parent; 31675528f911STrond Myklebust struct list_head *next; 31685528f911STrond Myklebust int found = 0; 31695528f911STrond Myklebust 31705528f911STrond Myklebust repeat: 31716b41d536SAl Viro next = this_parent->mnt_mounts.next; 31725528f911STrond Myklebust resume: 31736b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 31745528f911STrond Myklebust struct list_head *tmp = next; 31756b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 31765528f911STrond Myklebust 31775528f911STrond Myklebust next = tmp->next; 3178692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 31795528f911STrond Myklebust continue; 31805528f911STrond Myklebust /* 31815528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 31825528f911STrond Myklebust */ 31836b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 31845528f911STrond Myklebust this_parent = mnt; 31855528f911STrond Myklebust goto repeat; 31865528f911STrond Myklebust } 31875528f911STrond Myklebust 31881ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 31896776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 31905528f911STrond Myklebust found++; 31915528f911STrond Myklebust } 31925528f911STrond Myklebust } 31935528f911STrond Myklebust /* 31945528f911STrond Myklebust * All done at this level ... ascend and resume the search 31955528f911STrond Myklebust */ 31965528f911STrond Myklebust if (this_parent != parent) { 31976b41d536SAl Viro next = this_parent->mnt_child.next; 31980714a533SAl Viro this_parent = this_parent->mnt_parent; 31995528f911STrond Myklebust goto resume; 32005528f911STrond Myklebust } 32015528f911STrond Myklebust return found; 32025528f911STrond Myklebust } 32035528f911STrond Myklebust 32045528f911STrond Myklebust /* 32055528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 32065528f911STrond Myklebust * submounts of a specific parent mountpoint 320799b7db7bSNick Piggin * 320848a066e7SAl Viro * mount_lock must be held for write 32095528f911STrond Myklebust */ 3210b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 32115528f911STrond Myklebust { 32125528f911STrond Myklebust LIST_HEAD(graveyard); 3213761d5c38SAl Viro struct mount *m; 32145528f911STrond Myklebust 32155528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 3216c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3217bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3218761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 32196776db3dSAl Viro mnt_expire); 3220143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3221e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3222bcc5c7d2SAl Viro } 3223bcc5c7d2SAl Viro } 32245528f911STrond Myklebust } 32255528f911STrond Myklebust 3226028abd92SChristoph Hellwig static void *copy_mount_options(const void __user * data) 32271da177e4SLinus Torvalds { 3228b40ef869SAl Viro char *copy; 3229d563d678SCatalin Marinas unsigned left, offset; 32301da177e4SLinus Torvalds 32311da177e4SLinus Torvalds if (!data) 3232b40ef869SAl Viro return NULL; 32331da177e4SLinus Torvalds 3234b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3235b40ef869SAl Viro if (!copy) 3236b40ef869SAl Viro return ERR_PTR(-ENOMEM); 32371da177e4SLinus Torvalds 3238d563d678SCatalin Marinas left = copy_from_user(copy, data, PAGE_SIZE); 32391da177e4SLinus Torvalds 3240d563d678SCatalin Marinas /* 3241d563d678SCatalin Marinas * Not all architectures have an exact copy_from_user(). Resort to 3242d563d678SCatalin Marinas * byte at a time. 3243d563d678SCatalin Marinas */ 3244d563d678SCatalin Marinas offset = PAGE_SIZE - left; 3245d563d678SCatalin Marinas while (left) { 3246d563d678SCatalin Marinas char c; 3247d563d678SCatalin Marinas if (get_user(c, (const char __user *)data + offset)) 3248d563d678SCatalin Marinas break; 3249d563d678SCatalin Marinas copy[offset] = c; 3250d563d678SCatalin Marinas left--; 3251d563d678SCatalin Marinas offset++; 3252d563d678SCatalin Marinas } 3253d563d678SCatalin Marinas 3254d563d678SCatalin Marinas if (left == PAGE_SIZE) { 3255b40ef869SAl Viro kfree(copy); 3256b40ef869SAl Viro return ERR_PTR(-EFAULT); 32571da177e4SLinus Torvalds } 3258d563d678SCatalin Marinas 3259b40ef869SAl Viro return copy; 32601da177e4SLinus Torvalds } 32611da177e4SLinus Torvalds 3262028abd92SChristoph Hellwig static char *copy_mount_string(const void __user *data) 3263eca6f534SVegard Nossum { 3264fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3265eca6f534SVegard Nossum } 3266eca6f534SVegard Nossum 32671da177e4SLinus Torvalds /* 32681da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 32691da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 32701da177e4SLinus Torvalds * 32711da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 32721da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 32731da177e4SLinus Torvalds * information (or be NULL). 32741da177e4SLinus Torvalds * 32751da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 32761da177e4SLinus Torvalds * When the flags word was introduced its top half was required 32771da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 32781da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 32791da177e4SLinus Torvalds * and must be discarded. 32801da177e4SLinus Torvalds */ 3281c60166f0SChristoph Hellwig int path_mount(const char *dev_name, struct path *path, 3282808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 32831da177e4SLinus Torvalds { 3284e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 3285a1e6aaa3SChristoph Hellwig int ret; 32861da177e4SLinus Torvalds 32871da177e4SLinus Torvalds /* Discard magic */ 32881da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 32891da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 32901da177e4SLinus Torvalds 32911da177e4SLinus Torvalds /* Basic sanity checks */ 32921da177e4SLinus Torvalds if (data_page) 32931da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 32941da177e4SLinus Torvalds 3295e462ec50SDavid Howells if (flags & MS_NOUSER) 3296e462ec50SDavid Howells return -EINVAL; 3297e462ec50SDavid Howells 3298a1e6aaa3SChristoph Hellwig ret = security_sb_mount(dev_name, path, type_page, flags, data_page); 3299a1e6aaa3SChristoph Hellwig if (ret) 3300a1e6aaa3SChristoph Hellwig return ret; 3301a1e6aaa3SChristoph Hellwig if (!may_mount()) 3302a1e6aaa3SChristoph Hellwig return -EPERM; 3303f7e33bdbSJeff Layton if (flags & SB_MANDLOCK) 3304f7e33bdbSJeff Layton warn_mandlock(); 3305a27ab9f2STetsuo Handa 3306613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3307613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 33080a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 33090a1c01c9SMatthew Garrett 33101da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 33111da177e4SLinus Torvalds if (flags & MS_NOSUID) 33121da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 33131da177e4SLinus Torvalds if (flags & MS_NODEV) 33141da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 33151da177e4SLinus Torvalds if (flags & MS_NOEXEC) 33161da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3317fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3318fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3319fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3320fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3321d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3322d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3323a9e5b732SDavid Howells if (flags & MS_RDONLY) 33242e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3325dab741e0SMattias Nissler if (flags & MS_NOSYMFOLLOW) 3326dab741e0SMattias Nissler mnt_flags |= MNT_NOSYMFOLLOW; 3327fc33a7bbSChristoph Hellwig 3328ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3329ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3330ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3331ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3332ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3333a1e6aaa3SChristoph Hellwig mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; 3334ffbc6f0eSEric W. Biederman } 3335ffbc6f0eSEric W. Biederman 3336e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3337e462ec50SDavid Howells SB_SYNCHRONOUS | 3338e462ec50SDavid Howells SB_MANDLOCK | 3339e462ec50SDavid Howells SB_DIRSYNC | 3340e462ec50SDavid Howells SB_SILENT | 3341917086ffSMimi Zohar SB_POSIXACL | 3342d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3343917086ffSMimi Zohar SB_I_VERSION); 33441da177e4SLinus Torvalds 334543f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 3346a1e6aaa3SChristoph Hellwig return do_reconfigure_mnt(path, mnt_flags); 3347a1e6aaa3SChristoph Hellwig if (flags & MS_REMOUNT) 3348a1e6aaa3SChristoph Hellwig return do_remount(path, flags, sb_flags, mnt_flags, data_page); 3349a1e6aaa3SChristoph Hellwig if (flags & MS_BIND) 3350a1e6aaa3SChristoph Hellwig return do_loopback(path, dev_name, flags & MS_REC); 3351a1e6aaa3SChristoph Hellwig if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 3352a1e6aaa3SChristoph Hellwig return do_change_type(path, flags); 3353a1e6aaa3SChristoph Hellwig if (flags & MS_MOVE) 3354a1e6aaa3SChristoph Hellwig return do_move_mount_old(path, dev_name); 3355a1e6aaa3SChristoph Hellwig 3356a1e6aaa3SChristoph Hellwig return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, 33571da177e4SLinus Torvalds data_page); 3358a1e6aaa3SChristoph Hellwig } 3359a1e6aaa3SChristoph Hellwig 3360a1e6aaa3SChristoph Hellwig long do_mount(const char *dev_name, const char __user *dir_name, 3361a1e6aaa3SChristoph Hellwig const char *type_page, unsigned long flags, void *data_page) 3362a1e6aaa3SChristoph Hellwig { 3363a1e6aaa3SChristoph Hellwig struct path path; 3364a1e6aaa3SChristoph Hellwig int ret; 3365a1e6aaa3SChristoph Hellwig 3366a1e6aaa3SChristoph Hellwig ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3367a1e6aaa3SChristoph Hellwig if (ret) 3368a1e6aaa3SChristoph Hellwig return ret; 3369a1e6aaa3SChristoph Hellwig ret = path_mount(dev_name, &path, type_page, flags, data_page); 33702d92ab3cSAl Viro path_put(&path); 3371a1e6aaa3SChristoph Hellwig return ret; 33721da177e4SLinus Torvalds } 33731da177e4SLinus Torvalds 3374537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3375537f7ccbSEric W. Biederman { 3376537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3377537f7ccbSEric W. Biederman } 3378537f7ccbSEric W. Biederman 3379537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3380537f7ccbSEric W. Biederman { 3381537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3382537f7ccbSEric W. Biederman } 3383537f7ccbSEric W. Biederman 3384771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3385771b1371SEric W. Biederman { 338674e83122SAl Viro if (!is_anon_ns(ns)) 33876344c433SAl Viro ns_free_inum(&ns->ns); 3388537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3389771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3390771b1371SEric W. Biederman kfree(ns); 3391771b1371SEric W. Biederman } 3392771b1371SEric W. Biederman 33938823c079SEric W. Biederman /* 33948823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 33958823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 33968823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 33978823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 33988823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 33998823c079SEric W. Biederman */ 34008823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 34018823c079SEric W. Biederman 340274e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3403cf8d2c11STrond Myklebust { 3404cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3405537f7ccbSEric W. Biederman struct ucounts *ucounts; 340698f842e6SEric W. Biederman int ret; 3407cf8d2c11STrond Myklebust 3408537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3409537f7ccbSEric W. Biederman if (!ucounts) 3410df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3411537f7ccbSEric W. Biederman 341230acd0bdSVasily Averin new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); 3413537f7ccbSEric W. Biederman if (!new_ns) { 3414537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3415cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3416537f7ccbSEric W. Biederman } 341774e83122SAl Viro if (!anon) { 34186344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 341998f842e6SEric W. Biederman if (ret) { 342098f842e6SEric W. Biederman kfree(new_ns); 3421537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 342298f842e6SEric W. Biederman return ERR_PTR(ret); 342398f842e6SEric W. Biederman } 342474e83122SAl Viro } 342533c42940SAl Viro new_ns->ns.ops = &mntns_operations; 342674e83122SAl Viro if (!anon) 34278823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 34281a7b8969SKirill Tkhai refcount_set(&new_ns->ns.count, 1); 3429cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3430cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 34319f6c61f9SMiklos Szeredi spin_lock_init(&new_ns->ns_lock); 3432771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3433537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3434cf8d2c11STrond Myklebust return new_ns; 3435cf8d2c11STrond Myklebust } 3436cf8d2c11STrond Myklebust 34370766f788SEmese Revfy __latent_entropy 34389559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 34399559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 34401da177e4SLinus Torvalds { 34416b3286edSKirill Korotaev struct mnt_namespace *new_ns; 34427f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3443315fc83eSAl Viro struct mount *p, *q; 34449559f689SAl Viro struct mount *old; 3445cb338d06SAl Viro struct mount *new; 34467a472ef4SEric W. Biederman int copy_flags; 34471da177e4SLinus Torvalds 34489559f689SAl Viro BUG_ON(!ns); 34499559f689SAl Viro 34509559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 34519559f689SAl Viro get_mnt_ns(ns); 34529559f689SAl Viro return ns; 34539559f689SAl Viro } 34549559f689SAl Viro 34559559f689SAl Viro old = ns->root; 34569559f689SAl Viro 345774e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3458cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3459cf8d2c11STrond Myklebust return new_ns; 34601da177e4SLinus Torvalds 346197216be0SAl Viro namespace_lock(); 34621da177e4SLinus Torvalds /* First pass: copy the tree topology */ 34634ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 34649559f689SAl Viro if (user_ns != ns->user_ns) 34653bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 34667a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3467be34d1a3SDavid Howells if (IS_ERR(new)) { 3468328e6d90SAl Viro namespace_unlock(); 3469771b1371SEric W. Biederman free_mnt_ns(new_ns); 3470be34d1a3SDavid Howells return ERR_CAST(new); 34711da177e4SLinus Torvalds } 34723bd045ccSAl Viro if (user_ns != ns->user_ns) { 34733bd045ccSAl Viro lock_mount_hash(); 34743bd045ccSAl Viro lock_mnt_tree(new); 34753bd045ccSAl Viro unlock_mount_hash(); 34763bd045ccSAl Viro } 3477be08d6d2SAl Viro new_ns->root = new; 34781a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 34791da177e4SLinus Torvalds 34801da177e4SLinus Torvalds /* 34811da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 34821da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 34831da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 34841da177e4SLinus Torvalds */ 3485909b0a88SAl Viro p = old; 3486cb338d06SAl Viro q = new; 34871da177e4SLinus Torvalds while (p) { 3488143c8c91SAl Viro q->mnt_ns = new_ns; 3489d2921684SEric W. Biederman new_ns->mounts++; 34909559f689SAl Viro if (new_fs) { 34919559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 34929559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3493315fc83eSAl Viro rootmnt = &p->mnt; 34941da177e4SLinus Torvalds } 34959559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 34969559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3497315fc83eSAl Viro pwdmnt = &p->mnt; 34981da177e4SLinus Torvalds } 34991da177e4SLinus Torvalds } 3500909b0a88SAl Viro p = next_mnt(p, old); 3501909b0a88SAl Viro q = next_mnt(q, new); 35024ce5d2b1SEric W. Biederman if (!q) 35034ce5d2b1SEric W. Biederman break; 35044ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 35054ce5d2b1SEric W. Biederman p = next_mnt(p, old); 35061da177e4SLinus Torvalds } 3507328e6d90SAl Viro namespace_unlock(); 35081da177e4SLinus Torvalds 35091da177e4SLinus Torvalds if (rootmnt) 3510f03c6599SAl Viro mntput(rootmnt); 35111da177e4SLinus Torvalds if (pwdmnt) 3512f03c6599SAl Viro mntput(pwdmnt); 35131da177e4SLinus Torvalds 3514741a2951SJANAK DESAI return new_ns; 3515741a2951SJANAK DESAI } 3516741a2951SJANAK DESAI 351774e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3518cf8d2c11STrond Myklebust { 35191a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3520ea441d11SAl Viro struct mnt_namespace *ns; 3521d31da0f0SAl Viro struct super_block *s; 3522ea441d11SAl Viro struct path path; 3523ea441d11SAl Viro int err; 3524ea441d11SAl Viro 352574e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 352674e83122SAl Viro if (IS_ERR(ns)) { 352774e83122SAl Viro mntput(m); 3528ea441d11SAl Viro return ERR_CAST(ns); 352974e83122SAl Viro } 353074e83122SAl Viro mnt->mnt_ns = ns; 353174e83122SAl Viro ns->root = mnt; 353274e83122SAl Viro ns->mounts++; 353374e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3534ea441d11SAl Viro 353574e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3536ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3537ea441d11SAl Viro 3538ea441d11SAl Viro put_mnt_ns(ns); 3539ea441d11SAl Viro 3540ea441d11SAl Viro if (err) 3541ea441d11SAl Viro return ERR_PTR(err); 3542ea441d11SAl Viro 3543ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3544d31da0f0SAl Viro s = path.mnt->mnt_sb; 3545d31da0f0SAl Viro atomic_inc(&s->s_active); 3546ea441d11SAl Viro mntput(path.mnt); 3547ea441d11SAl Viro /* lock the sucker */ 3548d31da0f0SAl Viro down_write(&s->s_umount); 3549ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3550ea441d11SAl Viro return path.dentry; 3551ea441d11SAl Viro } 3552ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3553ea441d11SAl Viro 3554cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3555cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 35561da177e4SLinus Torvalds { 3557eca6f534SVegard Nossum int ret; 3558eca6f534SVegard Nossum char *kernel_type; 3559eca6f534SVegard Nossum char *kernel_dev; 3560b40ef869SAl Viro void *options; 35611da177e4SLinus Torvalds 3562b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3563b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3564b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3565eca6f534SVegard Nossum goto out_type; 35661da177e4SLinus Torvalds 3567b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3568b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3569b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3570eca6f534SVegard Nossum goto out_dev; 35711da177e4SLinus Torvalds 3572b40ef869SAl Viro options = copy_mount_options(data); 3573b40ef869SAl Viro ret = PTR_ERR(options); 3574b40ef869SAl Viro if (IS_ERR(options)) 3575eca6f534SVegard Nossum goto out_data; 35761da177e4SLinus Torvalds 3577b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3578eca6f534SVegard Nossum 3579b40ef869SAl Viro kfree(options); 3580eca6f534SVegard Nossum out_data: 3581eca6f534SVegard Nossum kfree(kernel_dev); 3582eca6f534SVegard Nossum out_dev: 3583eca6f534SVegard Nossum kfree(kernel_type); 3584eca6f534SVegard Nossum out_type: 3585eca6f534SVegard Nossum return ret; 35861da177e4SLinus Torvalds } 35871da177e4SLinus Torvalds 35885b490500SChristian Brauner #define FSMOUNT_VALID_FLAGS \ 35895b490500SChristian Brauner (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ 3590dd8b477fSChristian Brauner MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ 3591dd8b477fSChristian Brauner MOUNT_ATTR_NOSYMFOLLOW) 35925b490500SChristian Brauner 35939caccd41SChristian Brauner #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) 35942a186721SChristian Brauner 35952a186721SChristian Brauner #define MOUNT_SETATTR_PROPAGATION_FLAGS \ 35962a186721SChristian Brauner (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) 35972a186721SChristian Brauner 35985b490500SChristian Brauner static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) 35995b490500SChristian Brauner { 36005b490500SChristian Brauner unsigned int mnt_flags = 0; 36015b490500SChristian Brauner 36025b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_RDONLY) 36035b490500SChristian Brauner mnt_flags |= MNT_READONLY; 36045b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOSUID) 36055b490500SChristian Brauner mnt_flags |= MNT_NOSUID; 36065b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODEV) 36075b490500SChristian Brauner mnt_flags |= MNT_NODEV; 36085b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOEXEC) 36095b490500SChristian Brauner mnt_flags |= MNT_NOEXEC; 36105b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODIRATIME) 36115b490500SChristian Brauner mnt_flags |= MNT_NODIRATIME; 3612dd8b477fSChristian Brauner if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) 3613dd8b477fSChristian Brauner mnt_flags |= MNT_NOSYMFOLLOW; 36145b490500SChristian Brauner 36155b490500SChristian Brauner return mnt_flags; 36165b490500SChristian Brauner } 36175b490500SChristian Brauner 36181da177e4SLinus Torvalds /* 361993766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 362093766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 362193766fbdSDavid Howells */ 362293766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 362393766fbdSDavid Howells unsigned int, attr_flags) 362493766fbdSDavid Howells { 362593766fbdSDavid Howells struct mnt_namespace *ns; 362693766fbdSDavid Howells struct fs_context *fc; 362793766fbdSDavid Howells struct file *file; 362893766fbdSDavid Howells struct path newmount; 362993766fbdSDavid Howells struct mount *mnt; 363093766fbdSDavid Howells struct fd f; 363193766fbdSDavid Howells unsigned int mnt_flags = 0; 363293766fbdSDavid Howells long ret; 363393766fbdSDavid Howells 363493766fbdSDavid Howells if (!may_mount()) 363593766fbdSDavid Howells return -EPERM; 363693766fbdSDavid Howells 363793766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 363893766fbdSDavid Howells return -EINVAL; 363993766fbdSDavid Howells 36405b490500SChristian Brauner if (attr_flags & ~FSMOUNT_VALID_FLAGS) 364193766fbdSDavid Howells return -EINVAL; 364293766fbdSDavid Howells 36435b490500SChristian Brauner mnt_flags = attr_flags_to_mnt_flags(attr_flags); 364493766fbdSDavid Howells 364593766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 364693766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 364793766fbdSDavid Howells break; 364893766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 364993766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 365093766fbdSDavid Howells break; 365193766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 365293766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 365393766fbdSDavid Howells break; 365493766fbdSDavid Howells default: 365593766fbdSDavid Howells return -EINVAL; 365693766fbdSDavid Howells } 365793766fbdSDavid Howells 365893766fbdSDavid Howells f = fdget(fs_fd); 365993766fbdSDavid Howells if (!f.file) 366093766fbdSDavid Howells return -EBADF; 366193766fbdSDavid Howells 366293766fbdSDavid Howells ret = -EINVAL; 366393766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 366493766fbdSDavid Howells goto err_fsfd; 366593766fbdSDavid Howells 366693766fbdSDavid Howells fc = f.file->private_data; 366793766fbdSDavid Howells 366893766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 366993766fbdSDavid Howells if (ret < 0) 367093766fbdSDavid Howells goto err_fsfd; 367193766fbdSDavid Howells 367293766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 367393766fbdSDavid Howells ret = -EINVAL; 367493766fbdSDavid Howells if (!fc->root) 367593766fbdSDavid Howells goto err_unlock; 367693766fbdSDavid Howells 367793766fbdSDavid Howells ret = -EPERM; 367893766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 367993766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 368093766fbdSDavid Howells goto err_unlock; 368193766fbdSDavid Howells } 368293766fbdSDavid Howells 368393766fbdSDavid Howells ret = -EBUSY; 368493766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 368593766fbdSDavid Howells goto err_unlock; 368693766fbdSDavid Howells 3687f7e33bdbSJeff Layton if (fc->sb_flags & SB_MANDLOCK) 3688f7e33bdbSJeff Layton warn_mandlock(); 368993766fbdSDavid Howells 369093766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 369193766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 369293766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 369393766fbdSDavid Howells goto err_unlock; 369493766fbdSDavid Howells } 369593766fbdSDavid Howells newmount.dentry = dget(fc->root); 369693766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 369793766fbdSDavid Howells 369893766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 369993766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 370093766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 370193766fbdSDavid Howells * don't want to have to handle any errors incurred. 370293766fbdSDavid Howells */ 370393766fbdSDavid Howells vfs_clean_context(fc); 370493766fbdSDavid Howells 370593766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 370693766fbdSDavid Howells if (IS_ERR(ns)) { 370793766fbdSDavid Howells ret = PTR_ERR(ns); 370893766fbdSDavid Howells goto err_path; 370993766fbdSDavid Howells } 371093766fbdSDavid Howells mnt = real_mount(newmount.mnt); 371193766fbdSDavid Howells mnt->mnt_ns = ns; 371293766fbdSDavid Howells ns->root = mnt; 371393766fbdSDavid Howells ns->mounts = 1; 371493766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 37151b0b9cc8SEric Biggers mntget(newmount.mnt); 371693766fbdSDavid Howells 371793766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 371893766fbdSDavid Howells * it, not just simply put it. 371993766fbdSDavid Howells */ 372093766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 372193766fbdSDavid Howells if (IS_ERR(file)) { 372293766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 372393766fbdSDavid Howells ret = PTR_ERR(file); 372493766fbdSDavid Howells goto err_path; 372593766fbdSDavid Howells } 372693766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 372793766fbdSDavid Howells 372893766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 372993766fbdSDavid Howells if (ret >= 0) 373093766fbdSDavid Howells fd_install(ret, file); 373193766fbdSDavid Howells else 373293766fbdSDavid Howells fput(file); 373393766fbdSDavid Howells 373493766fbdSDavid Howells err_path: 373593766fbdSDavid Howells path_put(&newmount); 373693766fbdSDavid Howells err_unlock: 373793766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 373893766fbdSDavid Howells err_fsfd: 373993766fbdSDavid Howells fdput(f); 374093766fbdSDavid Howells return ret; 374193766fbdSDavid Howells } 374293766fbdSDavid Howells 374393766fbdSDavid Howells /* 374493766fbdSDavid Howells * Move a mount from one place to another. In combination with 374593766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 374693766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 374793766fbdSDavid Howells * a mount subtree. 37482db154b3SDavid Howells * 37492db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 37502db154b3SDavid Howells */ 37512db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 37522658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 37532658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 37542db154b3SDavid Howells unsigned int, flags) 37552db154b3SDavid Howells { 37562db154b3SDavid Howells struct path from_path, to_path; 37572db154b3SDavid Howells unsigned int lflags; 37582db154b3SDavid Howells int ret = 0; 37592db154b3SDavid Howells 37602db154b3SDavid Howells if (!may_mount()) 37612db154b3SDavid Howells return -EPERM; 37622db154b3SDavid Howells 37632db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 37642db154b3SDavid Howells return -EINVAL; 37652db154b3SDavid Howells 37662db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 37672db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 37682db154b3SDavid Howells * to clear it afterwards. 37692db154b3SDavid Howells */ 37702db154b3SDavid Howells lflags = 0; 37712db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37722db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37732db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 37742db154b3SDavid Howells 37752db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 37762db154b3SDavid Howells if (ret < 0) 37772db154b3SDavid Howells return ret; 37782db154b3SDavid Howells 37792db154b3SDavid Howells lflags = 0; 37802db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 37812db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 37822db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 37832db154b3SDavid Howells 37842db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 37852db154b3SDavid Howells if (ret < 0) 37862db154b3SDavid Howells goto out_from; 37872db154b3SDavid Howells 37882db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 37892db154b3SDavid Howells if (ret < 0) 37902db154b3SDavid Howells goto out_to; 37912db154b3SDavid Howells 37929ffb14efSPavel Tikhomirov if (flags & MOVE_MOUNT_SET_GROUP) 37939ffb14efSPavel Tikhomirov ret = do_set_group(&from_path, &to_path); 37949ffb14efSPavel Tikhomirov else 37952db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 37962db154b3SDavid Howells 37972db154b3SDavid Howells out_to: 37982db154b3SDavid Howells path_put(&to_path); 37992db154b3SDavid Howells out_from: 38002db154b3SDavid Howells path_put(&from_path); 38012db154b3SDavid Howells return ret; 38022db154b3SDavid Howells } 38032db154b3SDavid Howells 38042db154b3SDavid Howells /* 3805afac7cbaSAl Viro * Return true if path is reachable from root 3806afac7cbaSAl Viro * 380748a066e7SAl Viro * namespace_sem or mount_lock is held 3808afac7cbaSAl Viro */ 3809643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3810afac7cbaSAl Viro const struct path *root) 3811afac7cbaSAl Viro { 3812643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3813a73324daSAl Viro dentry = mnt->mnt_mountpoint; 38140714a533SAl Viro mnt = mnt->mnt_parent; 3815afac7cbaSAl Viro } 3816643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3817afac7cbaSAl Viro } 3818afac7cbaSAl Viro 3819640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3820afac7cbaSAl Viro { 382125ab4c9bSYaowei Bai bool res; 382248a066e7SAl Viro read_seqlock_excl(&mount_lock); 3823643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 382448a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3825afac7cbaSAl Viro return res; 3826afac7cbaSAl Viro } 3827afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3828afac7cbaSAl Viro 3829afac7cbaSAl Viro /* 38301da177e4SLinus Torvalds * pivot_root Semantics: 38311da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 38321da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 38331da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 38341da177e4SLinus Torvalds * 38351da177e4SLinus Torvalds * Restrictions: 38361da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 38371da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 38381da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 38391da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 38401da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 38411da177e4SLinus Torvalds * 38424a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 38430c1bc6b8SMauro Carvalho Chehab * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives 38444a0d11faSNeil Brown * in this situation. 38454a0d11faSNeil Brown * 38461da177e4SLinus Torvalds * Notes: 38471da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 38481da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 38491da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 38501da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 38511da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 38521da177e4SLinus Torvalds * first. 38531da177e4SLinus Torvalds */ 38543480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 38553480b257SHeiko Carstens const char __user *, put_old) 38561da177e4SLinus Torvalds { 38572763d119SAl Viro struct path new, old, root; 38582763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 385984d17192SAl Viro struct mountpoint *old_mp, *root_mp; 38601da177e4SLinus Torvalds int error; 38611da177e4SLinus Torvalds 38629b40bc90SAl Viro if (!may_mount()) 38631da177e4SLinus Torvalds return -EPERM; 38641da177e4SLinus Torvalds 3865ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3866ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 38671da177e4SLinus Torvalds if (error) 38681da177e4SLinus Torvalds goto out0; 38691da177e4SLinus Torvalds 3870ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 3871ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 38721da177e4SLinus Torvalds if (error) 38731da177e4SLinus Torvalds goto out1; 38741da177e4SLinus Torvalds 38752d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3876b12cea91SAl Viro if (error) 3877b12cea91SAl Viro goto out2; 38781da177e4SLinus Torvalds 3879f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 388084d17192SAl Viro old_mp = lock_mount(&old); 388184d17192SAl Viro error = PTR_ERR(old_mp); 388284d17192SAl Viro if (IS_ERR(old_mp)) 3883b12cea91SAl Viro goto out3; 3884b12cea91SAl Viro 38851da177e4SLinus Torvalds error = -EINVAL; 3886419148daSAl Viro new_mnt = real_mount(new.mnt); 3887419148daSAl Viro root_mnt = real_mount(root.mnt); 388884d17192SAl Viro old_mnt = real_mount(old.mnt); 38892763d119SAl Viro ex_parent = new_mnt->mnt_parent; 38902763d119SAl Viro root_parent = root_mnt->mnt_parent; 389184d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 38922763d119SAl Viro IS_MNT_SHARED(ex_parent) || 38932763d119SAl Viro IS_MNT_SHARED(root_parent)) 3894b12cea91SAl Viro goto out4; 3895143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3896b12cea91SAl Viro goto out4; 38975ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 38985ff9d8a6SEric W. Biederman goto out4; 38991da177e4SLinus Torvalds error = -ENOENT; 3900f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3901b12cea91SAl Viro goto out4; 39021da177e4SLinus Torvalds error = -EBUSY; 390384d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3904b12cea91SAl Viro goto out4; /* loop, on the same file system */ 39051da177e4SLinus Torvalds error = -EINVAL; 39068c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3907b12cea91SAl Viro goto out4; /* not a mountpoint */ 3908676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3909b12cea91SAl Viro goto out4; /* not attached */ 39102d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3911b12cea91SAl Viro goto out4; /* not a mountpoint */ 3912676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3913b12cea91SAl Viro goto out4; /* not attached */ 39144ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 391584d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3916b12cea91SAl Viro goto out4; 39170d082601SEric W. Biederman /* make certain new is below the root */ 39180d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 39190d082601SEric W. Biederman goto out4; 3920719ea2fbSAl Viro lock_mount_hash(); 39212763d119SAl Viro umount_mnt(new_mnt); 39222763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 39235ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 39245ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 39255ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 39265ff9d8a6SEric W. Biederman } 39274ac91378SJan Blunck /* mount old root on put_old */ 392884d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 39294ac91378SJan Blunck /* mount new_root on / */ 39302763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 39312763d119SAl Viro mnt_add_count(root_parent, -1); 39326b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 39334fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 39344fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 39353895dbf8SEric W. Biederman put_mountpoint(root_mp); 3936719ea2fbSAl Viro unlock_mount_hash(); 39372d8f3038SAl Viro chroot_fs_refs(&root, &new); 39381da177e4SLinus Torvalds error = 0; 3939b12cea91SAl Viro out4: 394084d17192SAl Viro unlock_mount(old_mp); 39412763d119SAl Viro if (!error) 39422763d119SAl Viro mntput_no_expire(ex_parent); 3943b12cea91SAl Viro out3: 39448c3ee42eSAl Viro path_put(&root); 3945b12cea91SAl Viro out2: 39462d8f3038SAl Viro path_put(&old); 39471da177e4SLinus Torvalds out1: 39482d8f3038SAl Viro path_put(&new); 39491da177e4SLinus Torvalds out0: 39501da177e4SLinus Torvalds return error; 39511da177e4SLinus Torvalds } 39521da177e4SLinus Torvalds 39532a186721SChristian Brauner static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) 39542a186721SChristian Brauner { 39552a186721SChristian Brauner unsigned int flags = mnt->mnt.mnt_flags; 39562a186721SChristian Brauner 39572a186721SChristian Brauner /* flags to clear */ 39582a186721SChristian Brauner flags &= ~kattr->attr_clr; 39592a186721SChristian Brauner /* flags to raise */ 39602a186721SChristian Brauner flags |= kattr->attr_set; 39612a186721SChristian Brauner 39622a186721SChristian Brauner return flags; 39632a186721SChristian Brauner } 39642a186721SChristian Brauner 39659caccd41SChristian Brauner static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 39669caccd41SChristian Brauner { 39679caccd41SChristian Brauner struct vfsmount *m = &mnt->mnt; 3968bd303368SChristian Brauner struct user_namespace *fs_userns = m->mnt_sb->s_user_ns; 39699caccd41SChristian Brauner 39709caccd41SChristian Brauner if (!kattr->mnt_userns) 39719caccd41SChristian Brauner return 0; 39729caccd41SChristian Brauner 39739caccd41SChristian Brauner /* 3974bd303368SChristian Brauner * Creating an idmapped mount with the filesystem wide idmapping 3975bd303368SChristian Brauner * doesn't make sense so block that. We don't allow mushy semantics. 3976bd303368SChristian Brauner */ 3977bd303368SChristian Brauner if (kattr->mnt_userns == fs_userns) 3978bd303368SChristian Brauner return -EINVAL; 3979bd303368SChristian Brauner 3980bd303368SChristian Brauner /* 39819caccd41SChristian Brauner * Once a mount has been idmapped we don't allow it to change its 39829caccd41SChristian Brauner * mapping. It makes things simpler and callers can just create 39839caccd41SChristian Brauner * another bind-mount they can idmap if they want to. 39849caccd41SChristian Brauner */ 3985bb49e9e7SChristian Brauner if (is_idmapped_mnt(m)) 39869caccd41SChristian Brauner return -EPERM; 39879caccd41SChristian Brauner 39889caccd41SChristian Brauner /* The underlying filesystem doesn't support idmapped mounts yet. */ 39899caccd41SChristian Brauner if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) 39909caccd41SChristian Brauner return -EINVAL; 39919caccd41SChristian Brauner 39929caccd41SChristian Brauner /* We're not controlling the superblock. */ 3993bd303368SChristian Brauner if (!ns_capable(fs_userns, CAP_SYS_ADMIN)) 39949caccd41SChristian Brauner return -EPERM; 39959caccd41SChristian Brauner 39969caccd41SChristian Brauner /* Mount has already been visible in the filesystem hierarchy. */ 39979caccd41SChristian Brauner if (!is_anon_ns(mnt->mnt_ns)) 39989caccd41SChristian Brauner return -EINVAL; 39999caccd41SChristian Brauner 40009caccd41SChristian Brauner return 0; 40019caccd41SChristian Brauner } 40029caccd41SChristian Brauner 40032a186721SChristian Brauner static struct mount *mount_setattr_prepare(struct mount_kattr *kattr, 40042a186721SChristian Brauner struct mount *mnt, int *err) 40052a186721SChristian Brauner { 40062a186721SChristian Brauner struct mount *m = mnt, *last = NULL; 40072a186721SChristian Brauner 40082a186721SChristian Brauner if (!is_mounted(&m->mnt)) { 40092a186721SChristian Brauner *err = -EINVAL; 40102a186721SChristian Brauner goto out; 40112a186721SChristian Brauner } 40122a186721SChristian Brauner 40132a186721SChristian Brauner if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) { 40142a186721SChristian Brauner *err = -EINVAL; 40152a186721SChristian Brauner goto out; 40162a186721SChristian Brauner } 40172a186721SChristian Brauner 40182a186721SChristian Brauner do { 40192a186721SChristian Brauner unsigned int flags; 40202a186721SChristian Brauner 40212a186721SChristian Brauner flags = recalc_flags(kattr, m); 40222a186721SChristian Brauner if (!can_change_locked_flags(m, flags)) { 40232a186721SChristian Brauner *err = -EPERM; 40242a186721SChristian Brauner goto out; 40252a186721SChristian Brauner } 40262a186721SChristian Brauner 40279caccd41SChristian Brauner *err = can_idmap_mount(kattr, m); 40289caccd41SChristian Brauner if (*err) 40299caccd41SChristian Brauner goto out; 40309caccd41SChristian Brauner 40312a186721SChristian Brauner last = m; 40322a186721SChristian Brauner 40332a186721SChristian Brauner if ((kattr->attr_set & MNT_READONLY) && 40342a186721SChristian Brauner !(m->mnt.mnt_flags & MNT_READONLY)) { 40352a186721SChristian Brauner *err = mnt_hold_writers(m); 40362a186721SChristian Brauner if (*err) 40372a186721SChristian Brauner goto out; 40382a186721SChristian Brauner } 40392a186721SChristian Brauner } while (kattr->recurse && (m = next_mnt(m, mnt))); 40402a186721SChristian Brauner 40412a186721SChristian Brauner out: 40422a186721SChristian Brauner return last; 40432a186721SChristian Brauner } 40442a186721SChristian Brauner 40459caccd41SChristian Brauner static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 40469caccd41SChristian Brauner { 4047bd303368SChristian Brauner struct user_namespace *mnt_userns, *old_mnt_userns; 40489caccd41SChristian Brauner 40499caccd41SChristian Brauner if (!kattr->mnt_userns) 40509caccd41SChristian Brauner return; 40519caccd41SChristian Brauner 4052bd303368SChristian Brauner /* 4053bd303368SChristian Brauner * We're the only ones able to change the mount's idmapping. So 4054bd303368SChristian Brauner * mnt->mnt.mnt_userns is stable and we can retrieve it directly. 4055bd303368SChristian Brauner */ 4056bd303368SChristian Brauner old_mnt_userns = mnt->mnt.mnt_userns; 4057bd303368SChristian Brauner 40589caccd41SChristian Brauner mnt_userns = get_user_ns(kattr->mnt_userns); 40599caccd41SChristian Brauner /* Pairs with smp_load_acquire() in mnt_user_ns(). */ 40609caccd41SChristian Brauner smp_store_release(&mnt->mnt.mnt_userns, mnt_userns); 4061bd303368SChristian Brauner 4062bd303368SChristian Brauner /* 4063bd303368SChristian Brauner * If this is an idmapped filesystem drop the reference we've taken 4064bd303368SChristian Brauner * in vfs_create_mount() before. 4065bd303368SChristian Brauner */ 4066bd303368SChristian Brauner if (!initial_idmapping(old_mnt_userns)) 4067bd303368SChristian Brauner put_user_ns(old_mnt_userns); 40689caccd41SChristian Brauner } 40699caccd41SChristian Brauner 40702a186721SChristian Brauner static void mount_setattr_commit(struct mount_kattr *kattr, 40712a186721SChristian Brauner struct mount *mnt, struct mount *last, 40722a186721SChristian Brauner int err) 40732a186721SChristian Brauner { 40742a186721SChristian Brauner struct mount *m = mnt; 40752a186721SChristian Brauner 40762a186721SChristian Brauner do { 40772a186721SChristian Brauner if (!err) { 40782a186721SChristian Brauner unsigned int flags; 40792a186721SChristian Brauner 40809caccd41SChristian Brauner do_idmap_mount(kattr, m); 40812a186721SChristian Brauner flags = recalc_flags(kattr, m); 40822a186721SChristian Brauner WRITE_ONCE(m->mnt.mnt_flags, flags); 40832a186721SChristian Brauner } 40842a186721SChristian Brauner 40852a186721SChristian Brauner /* 40862a186721SChristian Brauner * We either set MNT_READONLY above so make it visible 40872a186721SChristian Brauner * before ~MNT_WRITE_HOLD or we failed to recursively 40882a186721SChristian Brauner * apply mount options. 40892a186721SChristian Brauner */ 40902a186721SChristian Brauner if ((kattr->attr_set & MNT_READONLY) && 40912a186721SChristian Brauner (m->mnt.mnt_flags & MNT_WRITE_HOLD)) 40922a186721SChristian Brauner mnt_unhold_writers(m); 40932a186721SChristian Brauner 40942a186721SChristian Brauner if (!err && kattr->propagation) 40952a186721SChristian Brauner change_mnt_propagation(m, kattr->propagation); 40962a186721SChristian Brauner 40972a186721SChristian Brauner /* 40982a186721SChristian Brauner * On failure, only cleanup until we found the first mount 40992a186721SChristian Brauner * we failed to handle. 41002a186721SChristian Brauner */ 41012a186721SChristian Brauner if (err && m == last) 41022a186721SChristian Brauner break; 41032a186721SChristian Brauner } while (kattr->recurse && (m = next_mnt(m, mnt))); 41042a186721SChristian Brauner 41052a186721SChristian Brauner if (!err) 41062a186721SChristian Brauner touch_mnt_namespace(mnt->mnt_ns); 41072a186721SChristian Brauner } 41082a186721SChristian Brauner 41092a186721SChristian Brauner static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) 41102a186721SChristian Brauner { 41112a186721SChristian Brauner struct mount *mnt = real_mount(path->mnt), *last = NULL; 41122a186721SChristian Brauner int err = 0; 41132a186721SChristian Brauner 41142a186721SChristian Brauner if (path->dentry != mnt->mnt.mnt_root) 41152a186721SChristian Brauner return -EINVAL; 41162a186721SChristian Brauner 41172a186721SChristian Brauner if (kattr->propagation) { 41182a186721SChristian Brauner /* 41192a186721SChristian Brauner * Only take namespace_lock() if we're actually changing 41202a186721SChristian Brauner * propagation. 41212a186721SChristian Brauner */ 41222a186721SChristian Brauner namespace_lock(); 41232a186721SChristian Brauner if (kattr->propagation == MS_SHARED) { 41242a186721SChristian Brauner err = invent_group_ids(mnt, kattr->recurse); 41252a186721SChristian Brauner if (err) { 41262a186721SChristian Brauner namespace_unlock(); 41272a186721SChristian Brauner return err; 41282a186721SChristian Brauner } 41292a186721SChristian Brauner } 41302a186721SChristian Brauner } 41312a186721SChristian Brauner 41322a186721SChristian Brauner lock_mount_hash(); 41332a186721SChristian Brauner 41342a186721SChristian Brauner /* 41352a186721SChristian Brauner * Get the mount tree in a shape where we can change mount 41362a186721SChristian Brauner * properties without failure. 41372a186721SChristian Brauner */ 41382a186721SChristian Brauner last = mount_setattr_prepare(kattr, mnt, &err); 41392a186721SChristian Brauner if (last) /* Commit all changes or revert to the old state. */ 41402a186721SChristian Brauner mount_setattr_commit(kattr, mnt, last, err); 41412a186721SChristian Brauner 41422a186721SChristian Brauner unlock_mount_hash(); 41432a186721SChristian Brauner 41442a186721SChristian Brauner if (kattr->propagation) { 41452a186721SChristian Brauner namespace_unlock(); 41462a186721SChristian Brauner if (err) 41472a186721SChristian Brauner cleanup_group_ids(mnt, NULL); 41482a186721SChristian Brauner } 41492a186721SChristian Brauner 41502a186721SChristian Brauner return err; 41512a186721SChristian Brauner } 41522a186721SChristian Brauner 41539caccd41SChristian Brauner static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, 41549caccd41SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 41559caccd41SChristian Brauner { 41569caccd41SChristian Brauner int err = 0; 41579caccd41SChristian Brauner struct ns_common *ns; 41589caccd41SChristian Brauner struct user_namespace *mnt_userns; 41599caccd41SChristian Brauner struct file *file; 41609caccd41SChristian Brauner 41619caccd41SChristian Brauner if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) 41629caccd41SChristian Brauner return 0; 41639caccd41SChristian Brauner 41649caccd41SChristian Brauner /* 41659caccd41SChristian Brauner * We currently do not support clearing an idmapped mount. If this ever 41669caccd41SChristian Brauner * is a use-case we can revisit this but for now let's keep it simple 41679caccd41SChristian Brauner * and not allow it. 41689caccd41SChristian Brauner */ 41699caccd41SChristian Brauner if (attr->attr_clr & MOUNT_ATTR_IDMAP) 41709caccd41SChristian Brauner return -EINVAL; 41719caccd41SChristian Brauner 41729caccd41SChristian Brauner if (attr->userns_fd > INT_MAX) 41739caccd41SChristian Brauner return -EINVAL; 41749caccd41SChristian Brauner 41759caccd41SChristian Brauner file = fget(attr->userns_fd); 41769caccd41SChristian Brauner if (!file) 41779caccd41SChristian Brauner return -EBADF; 41789caccd41SChristian Brauner 41799caccd41SChristian Brauner if (!proc_ns_file(file)) { 41809caccd41SChristian Brauner err = -EINVAL; 41819caccd41SChristian Brauner goto out_fput; 41829caccd41SChristian Brauner } 41839caccd41SChristian Brauner 41849caccd41SChristian Brauner ns = get_proc_ns(file_inode(file)); 41859caccd41SChristian Brauner if (ns->ops->type != CLONE_NEWUSER) { 41869caccd41SChristian Brauner err = -EINVAL; 41879caccd41SChristian Brauner goto out_fput; 41889caccd41SChristian Brauner } 41899caccd41SChristian Brauner 41909caccd41SChristian Brauner /* 4191bd303368SChristian Brauner * The initial idmapping cannot be used to create an idmapped 4192bd303368SChristian Brauner * mount. We use the initial idmapping as an indicator of a mount 4193bd303368SChristian Brauner * that is not idmapped. It can simply be passed into helpers that 4194bd303368SChristian Brauner * are aware of idmapped mounts as a convenient shortcut. A user 4195bd303368SChristian Brauner * can just create a dedicated identity mapping to achieve the same 4196bd303368SChristian Brauner * result. 41979caccd41SChristian Brauner */ 41989caccd41SChristian Brauner mnt_userns = container_of(ns, struct user_namespace, ns); 4199bd303368SChristian Brauner if (initial_idmapping(mnt_userns)) { 42009caccd41SChristian Brauner err = -EPERM; 42019caccd41SChristian Brauner goto out_fput; 42029caccd41SChristian Brauner } 42039caccd41SChristian Brauner kattr->mnt_userns = get_user_ns(mnt_userns); 42049caccd41SChristian Brauner 42059caccd41SChristian Brauner out_fput: 42069caccd41SChristian Brauner fput(file); 42079caccd41SChristian Brauner return err; 42089caccd41SChristian Brauner } 42099caccd41SChristian Brauner 42109caccd41SChristian Brauner static int build_mount_kattr(const struct mount_attr *attr, size_t usize, 42112a186721SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 42122a186721SChristian Brauner { 42132a186721SChristian Brauner unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 42142a186721SChristian Brauner 42152a186721SChristian Brauner if (flags & AT_NO_AUTOMOUNT) 42162a186721SChristian Brauner lookup_flags &= ~LOOKUP_AUTOMOUNT; 42172a186721SChristian Brauner if (flags & AT_SYMLINK_NOFOLLOW) 42182a186721SChristian Brauner lookup_flags &= ~LOOKUP_FOLLOW; 42192a186721SChristian Brauner if (flags & AT_EMPTY_PATH) 42202a186721SChristian Brauner lookup_flags |= LOOKUP_EMPTY; 42212a186721SChristian Brauner 42222a186721SChristian Brauner *kattr = (struct mount_kattr) { 42232a186721SChristian Brauner .lookup_flags = lookup_flags, 42242a186721SChristian Brauner .recurse = !!(flags & AT_RECURSIVE), 42252a186721SChristian Brauner }; 42262a186721SChristian Brauner 42272a186721SChristian Brauner if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) 42282a186721SChristian Brauner return -EINVAL; 42292a186721SChristian Brauner if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) 42302a186721SChristian Brauner return -EINVAL; 42312a186721SChristian Brauner kattr->propagation = attr->propagation; 42322a186721SChristian Brauner 42332a186721SChristian Brauner if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) 42342a186721SChristian Brauner return -EINVAL; 42352a186721SChristian Brauner 42362a186721SChristian Brauner kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); 42372a186721SChristian Brauner kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); 42382a186721SChristian Brauner 42392a186721SChristian Brauner /* 42402a186721SChristian Brauner * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, 42412a186721SChristian Brauner * users wanting to transition to a different atime setting cannot 42422a186721SChristian Brauner * simply specify the atime setting in @attr_set, but must also 42432a186721SChristian Brauner * specify MOUNT_ATTR__ATIME in the @attr_clr field. 42442a186721SChristian Brauner * So ensure that MOUNT_ATTR__ATIME can't be partially set in 42452a186721SChristian Brauner * @attr_clr and that @attr_set can't have any atime bits set if 42462a186721SChristian Brauner * MOUNT_ATTR__ATIME isn't set in @attr_clr. 42472a186721SChristian Brauner */ 42482a186721SChristian Brauner if (attr->attr_clr & MOUNT_ATTR__ATIME) { 42492a186721SChristian Brauner if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) 42502a186721SChristian Brauner return -EINVAL; 42512a186721SChristian Brauner 42522a186721SChristian Brauner /* 42532a186721SChristian Brauner * Clear all previous time settings as they are mutually 42542a186721SChristian Brauner * exclusive. 42552a186721SChristian Brauner */ 42562a186721SChristian Brauner kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; 42572a186721SChristian Brauner switch (attr->attr_set & MOUNT_ATTR__ATIME) { 42582a186721SChristian Brauner case MOUNT_ATTR_RELATIME: 42592a186721SChristian Brauner kattr->attr_set |= MNT_RELATIME; 42602a186721SChristian Brauner break; 42612a186721SChristian Brauner case MOUNT_ATTR_NOATIME: 42622a186721SChristian Brauner kattr->attr_set |= MNT_NOATIME; 42632a186721SChristian Brauner break; 42642a186721SChristian Brauner case MOUNT_ATTR_STRICTATIME: 42652a186721SChristian Brauner break; 42662a186721SChristian Brauner default: 42672a186721SChristian Brauner return -EINVAL; 42682a186721SChristian Brauner } 42692a186721SChristian Brauner } else { 42702a186721SChristian Brauner if (attr->attr_set & MOUNT_ATTR__ATIME) 42712a186721SChristian Brauner return -EINVAL; 42722a186721SChristian Brauner } 42732a186721SChristian Brauner 42749caccd41SChristian Brauner return build_mount_idmapped(attr, usize, kattr, flags); 42759caccd41SChristian Brauner } 42769caccd41SChristian Brauner 42779caccd41SChristian Brauner static void finish_mount_kattr(struct mount_kattr *kattr) 42789caccd41SChristian Brauner { 42799caccd41SChristian Brauner put_user_ns(kattr->mnt_userns); 42809caccd41SChristian Brauner kattr->mnt_userns = NULL; 42812a186721SChristian Brauner } 42822a186721SChristian Brauner 42832a186721SChristian Brauner SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, 42842a186721SChristian Brauner unsigned int, flags, struct mount_attr __user *, uattr, 42852a186721SChristian Brauner size_t, usize) 42862a186721SChristian Brauner { 42872a186721SChristian Brauner int err; 42882a186721SChristian Brauner struct path target; 42892a186721SChristian Brauner struct mount_attr attr; 42902a186721SChristian Brauner struct mount_kattr kattr; 42912a186721SChristian Brauner 42922a186721SChristian Brauner BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); 42932a186721SChristian Brauner 42942a186721SChristian Brauner if (flags & ~(AT_EMPTY_PATH | 42952a186721SChristian Brauner AT_RECURSIVE | 42962a186721SChristian Brauner AT_SYMLINK_NOFOLLOW | 42972a186721SChristian Brauner AT_NO_AUTOMOUNT)) 42982a186721SChristian Brauner return -EINVAL; 42992a186721SChristian Brauner 43002a186721SChristian Brauner if (unlikely(usize > PAGE_SIZE)) 43012a186721SChristian Brauner return -E2BIG; 43022a186721SChristian Brauner if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) 43032a186721SChristian Brauner return -EINVAL; 43042a186721SChristian Brauner 43052a186721SChristian Brauner if (!may_mount()) 43062a186721SChristian Brauner return -EPERM; 43072a186721SChristian Brauner 43082a186721SChristian Brauner err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); 43092a186721SChristian Brauner if (err) 43102a186721SChristian Brauner return err; 43112a186721SChristian Brauner 43122a186721SChristian Brauner /* Don't bother walking through the mounts if this is a nop. */ 43132a186721SChristian Brauner if (attr.attr_set == 0 && 43142a186721SChristian Brauner attr.attr_clr == 0 && 43152a186721SChristian Brauner attr.propagation == 0) 43162a186721SChristian Brauner return 0; 43172a186721SChristian Brauner 43189caccd41SChristian Brauner err = build_mount_kattr(&attr, usize, &kattr, flags); 43192a186721SChristian Brauner if (err) 43202a186721SChristian Brauner return err; 43212a186721SChristian Brauner 43222a186721SChristian Brauner err = user_path_at(dfd, path, kattr.lookup_flags, &target); 4323012e3322SChristian Brauner if (!err) { 43242a186721SChristian Brauner err = do_mount_setattr(&target, &kattr); 43252a186721SChristian Brauner path_put(&target); 4326012e3322SChristian Brauner } 4327012e3322SChristian Brauner finish_mount_kattr(&kattr); 43282a186721SChristian Brauner return err; 43292a186721SChristian Brauner } 43302a186721SChristian Brauner 43311da177e4SLinus Torvalds static void __init init_mount_tree(void) 43321da177e4SLinus Torvalds { 43331da177e4SLinus Torvalds struct vfsmount *mnt; 433474e83122SAl Viro struct mount *m; 43356b3286edSKirill Korotaev struct mnt_namespace *ns; 4336ac748a09SJan Blunck struct path root; 43371da177e4SLinus Torvalds 4338fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 43391da177e4SLinus Torvalds if (IS_ERR(mnt)) 43401da177e4SLinus Torvalds panic("Can't create rootfs"); 4341b3e19d92SNick Piggin 434274e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 43433b22edc5STrond Myklebust if (IS_ERR(ns)) 43441da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 434574e83122SAl Viro m = real_mount(mnt); 434674e83122SAl Viro m->mnt_ns = ns; 434774e83122SAl Viro ns->root = m; 434874e83122SAl Viro ns->mounts = 1; 434974e83122SAl Viro list_add(&m->mnt_list, &ns->list); 43506b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 43516b3286edSKirill Korotaev get_mnt_ns(ns); 43521da177e4SLinus Torvalds 4353be08d6d2SAl Viro root.mnt = mnt; 4354be08d6d2SAl Viro root.dentry = mnt->mnt_root; 4355da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 4356ac748a09SJan Blunck 4357ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 4358ac748a09SJan Blunck set_fs_root(current->fs, &root); 43591da177e4SLinus Torvalds } 43601da177e4SLinus Torvalds 436174bf17cfSDenis Cheng void __init mnt_init(void) 43621da177e4SLinus Torvalds { 436315a67dd8SRandy Dunlap int err; 43641da177e4SLinus Torvalds 43657d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 436679f6540bSVasily Averin 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); 43671da177e4SLinus Torvalds 43680818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 436938129a13SAl Viro sizeof(struct hlist_head), 43700818bf27SAl Viro mhash_entries, 19, 43713d375d78SPavel Tatashin HASH_ZERO, 43720818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 43730818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 43740818bf27SAl Viro sizeof(struct hlist_head), 43750818bf27SAl Viro mphash_entries, 19, 43763d375d78SPavel Tatashin HASH_ZERO, 43770818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 43781da177e4SLinus Torvalds 437984d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 43801da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 43811da177e4SLinus Torvalds 43824b93dc9bSTejun Heo kernfs_init(); 43834b93dc9bSTejun Heo 438415a67dd8SRandy Dunlap err = sysfs_init(); 438515a67dd8SRandy Dunlap if (err) 438615a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 43878e24eea7SHarvey Harrison __func__, err); 438800d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 438900d26666SGreg Kroah-Hartman if (!fs_kobj) 43908e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 4391037f11b4SAl Viro shmem_init(); 43921da177e4SLinus Torvalds init_rootfs(); 43931da177e4SLinus Torvalds init_mount_tree(); 43941da177e4SLinus Torvalds } 43951da177e4SLinus Torvalds 4396616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 43971da177e4SLinus Torvalds { 43981a7b8969SKirill Tkhai if (!refcount_dec_and_test(&ns->ns.count)) 4399616511d0STrond Myklebust return; 44007b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 4401771b1371SEric W. Biederman free_mnt_ns(ns); 44021da177e4SLinus Torvalds } 44039d412a43SAl Viro 4404d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 44059d412a43SAl Viro { 4406423e0ab0STim Chen struct vfsmount *mnt; 4407d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 4408423e0ab0STim Chen if (!IS_ERR(mnt)) { 4409423e0ab0STim Chen /* 4410423e0ab0STim Chen * it is a longterm mount, don't release mnt until 4411423e0ab0STim Chen * we unmount before file sys is unregistered 4412423e0ab0STim Chen */ 4413f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 4414423e0ab0STim Chen } 4415423e0ab0STim Chen return mnt; 44169d412a43SAl Viro } 4417d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 4418423e0ab0STim Chen 4419423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 4420423e0ab0STim Chen { 4421423e0ab0STim Chen /* release long term mount so mount point can be released */ 4422423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 4423f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 442448a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 4425423e0ab0STim Chen mntput(mnt); 4426423e0ab0STim Chen } 4427423e0ab0STim Chen } 4428423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 442902125a82SAl Viro 4430df820f8dSMiklos Szeredi void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) 4431df820f8dSMiklos Szeredi { 4432df820f8dSMiklos Szeredi unsigned int i; 4433df820f8dSMiklos Szeredi 4434df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4435df820f8dSMiklos Szeredi if (mnt[i]) 4436df820f8dSMiklos Szeredi real_mount(mnt[i])->mnt_ns = NULL; 4437df820f8dSMiklos Szeredi synchronize_rcu_expedited(); 4438df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4439df820f8dSMiklos Szeredi mntput(mnt[i]); 4440df820f8dSMiklos Szeredi } 4441df820f8dSMiklos Szeredi EXPORT_SYMBOL(kern_unmount_array); 4442df820f8dSMiklos Szeredi 444302125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 444402125a82SAl Viro { 4445143c8c91SAl Viro return check_mnt(real_mount(mnt)); 444602125a82SAl Viro } 44478823c079SEric W. Biederman 44483151527eSEric W. Biederman bool current_chrooted(void) 44493151527eSEric W. Biederman { 44503151527eSEric W. Biederman /* Does the current process have a non-standard root */ 44513151527eSEric W. Biederman struct path ns_root; 44523151527eSEric W. Biederman struct path fs_root; 44533151527eSEric W. Biederman bool chrooted; 44543151527eSEric W. Biederman 44553151527eSEric W. Biederman /* Find the namespace root */ 44563151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 44573151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 44583151527eSEric W. Biederman path_get(&ns_root); 44593151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 44603151527eSEric W. Biederman ; 44613151527eSEric W. Biederman 44623151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 44633151527eSEric W. Biederman 44643151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 44653151527eSEric W. Biederman 44663151527eSEric W. Biederman path_put(&fs_root); 44673151527eSEric W. Biederman path_put(&ns_root); 44683151527eSEric W. Biederman 44693151527eSEric W. Biederman return chrooted; 44703151527eSEric W. Biederman } 44713151527eSEric W. Biederman 4472132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 4473132e4608SDavid Howells const struct super_block *sb, 44748654df4eSEric W. Biederman int *new_mnt_flags) 447587a8ebd6SEric W. Biederman { 44768c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 447787a8ebd6SEric W. Biederman struct mount *mnt; 4478e51db735SEric W. Biederman bool visible = false; 447987a8ebd6SEric W. Biederman 448044bb4385SAl Viro down_read(&namespace_sem); 44819f6c61f9SMiklos Szeredi lock_ns_list(ns); 448287a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 4483e51db735SEric W. Biederman struct mount *child; 448477b1a97dSEric W. Biederman int mnt_flags; 448577b1a97dSEric W. Biederman 44869f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 44879f6c61f9SMiklos Szeredi continue; 44889f6c61f9SMiklos Szeredi 4489132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 4490e51db735SEric W. Biederman continue; 4491e51db735SEric W. Biederman 44927e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 44937e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 44947e96c1b0SEric W. Biederman */ 44957e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 44967e96c1b0SEric W. Biederman continue; 44977e96c1b0SEric W. Biederman 4498a1935c17SEric W. Biederman /* A local view of the mount flags */ 449977b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 450077b1a97dSEric W. Biederman 4501695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 4502bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 4503695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 4504695e9df0SEric W. Biederman 45058c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 45068c6cf9ccSEric W. Biederman * than the proposed new mount. 45078c6cf9ccSEric W. Biederman */ 450877b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 45098c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 45108c6cf9ccSEric W. Biederman continue; 451177b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 451277b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 45138c6cf9ccSEric W. Biederman continue; 45148c6cf9ccSEric W. Biederman 4515ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 4516ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 4517ceeb0e5dSEric W. Biederman * empty directories. 4518e51db735SEric W. Biederman */ 4519e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 4520e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 4521ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 4522d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 4523ceeb0e5dSEric W. Biederman continue; 45247236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 45257236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 4526e51db735SEric W. Biederman goto next; 452787a8ebd6SEric W. Biederman } 45288c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 452977b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 45308c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 4531e51db735SEric W. Biederman visible = true; 4532e51db735SEric W. Biederman goto found; 4533e51db735SEric W. Biederman next: ; 453487a8ebd6SEric W. Biederman } 4535e51db735SEric W. Biederman found: 45369f6c61f9SMiklos Szeredi unlock_ns_list(ns); 453744bb4385SAl Viro up_read(&namespace_sem); 4538e51db735SEric W. Biederman return visible; 453987a8ebd6SEric W. Biederman } 454087a8ebd6SEric W. Biederman 4541132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 45428654df4eSEric W. Biederman { 4543a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 45448654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 45458654df4eSEric W. Biederman unsigned long s_iflags; 45468654df4eSEric W. Biederman 45478654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 45488654df4eSEric W. Biederman return false; 45498654df4eSEric W. Biederman 45508654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 4551132e4608SDavid Howells s_iflags = sb->s_iflags; 45528654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 45538654df4eSEric W. Biederman return false; 45548654df4eSEric W. Biederman 4555a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 4556a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 4557a1935c17SEric W. Biederman required_iflags); 4558a1935c17SEric W. Biederman return true; 4559a1935c17SEric W. Biederman } 4560a1935c17SEric W. Biederman 4561132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 45628654df4eSEric W. Biederman } 45638654df4eSEric W. Biederman 4564380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 4565380cf5baSAndy Lutomirski { 4566380cf5baSAndy Lutomirski /* 4567380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 4568380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 4569380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 4570380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 4571380cf5baSAndy Lutomirski * in other namespaces. 4572380cf5baSAndy Lutomirski */ 4573380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 4574380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 4575380cf5baSAndy Lutomirski } 4576380cf5baSAndy Lutomirski 457764964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 45788823c079SEric W. Biederman { 457958be2825SAl Viro struct ns_common *ns = NULL; 45808823c079SEric W. Biederman struct nsproxy *nsproxy; 45818823c079SEric W. Biederman 4582728dba3aSEric W. Biederman task_lock(task); 4583728dba3aSEric W. Biederman nsproxy = task->nsproxy; 45848823c079SEric W. Biederman if (nsproxy) { 458558be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 458658be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 45878823c079SEric W. Biederman } 4588728dba3aSEric W. Biederman task_unlock(task); 45898823c079SEric W. Biederman 45908823c079SEric W. Biederman return ns; 45918823c079SEric W. Biederman } 45928823c079SEric W. Biederman 459364964528SAl Viro static void mntns_put(struct ns_common *ns) 45948823c079SEric W. Biederman { 459558be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 45968823c079SEric W. Biederman } 45978823c079SEric W. Biederman 4598f2a8d52eSChristian Brauner static int mntns_install(struct nsset *nsset, struct ns_common *ns) 45998823c079SEric W. Biederman { 4600f2a8d52eSChristian Brauner struct nsproxy *nsproxy = nsset->nsproxy; 4601f2a8d52eSChristian Brauner struct fs_struct *fs = nsset->fs; 46024f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 4603f2a8d52eSChristian Brauner struct user_namespace *user_ns = nsset->cred->user_ns; 46048823c079SEric W. Biederman struct path root; 46054f757f3cSAl Viro int err; 46068823c079SEric W. Biederman 46070c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 4608f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_CHROOT) || 4609f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_ADMIN)) 4610ae11e0f1SZhao Hongjiang return -EPERM; 46118823c079SEric W. Biederman 461274e83122SAl Viro if (is_anon_ns(mnt_ns)) 461374e83122SAl Viro return -EINVAL; 461474e83122SAl Viro 46158823c079SEric W. Biederman if (fs->users != 1) 46168823c079SEric W. Biederman return -EINVAL; 46178823c079SEric W. Biederman 46188823c079SEric W. Biederman get_mnt_ns(mnt_ns); 46194f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 46208823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 46218823c079SEric W. Biederman 46228823c079SEric W. Biederman /* Find the root */ 46234f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 46244f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 46254f757f3cSAl Viro if (err) { 46264f757f3cSAl Viro /* revert to old namespace */ 46274f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 46284f757f3cSAl Viro put_mnt_ns(mnt_ns); 46294f757f3cSAl Viro return err; 46304f757f3cSAl Viro } 46318823c079SEric W. Biederman 46324068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 46334068367cSAndrei Vagin 46348823c079SEric W. Biederman /* Update the pwd and root */ 46358823c079SEric W. Biederman set_fs_pwd(fs, &root); 46368823c079SEric W. Biederman set_fs_root(fs, &root); 46378823c079SEric W. Biederman 46388823c079SEric W. Biederman path_put(&root); 46398823c079SEric W. Biederman return 0; 46408823c079SEric W. Biederman } 46418823c079SEric W. Biederman 4642bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 4643bcac25a5SAndrey Vagin { 4644bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 4645bcac25a5SAndrey Vagin } 4646bcac25a5SAndrey Vagin 46478823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 46488823c079SEric W. Biederman .name = "mnt", 46498823c079SEric W. Biederman .type = CLONE_NEWNS, 46508823c079SEric W. Biederman .get = mntns_get, 46518823c079SEric W. Biederman .put = mntns_put, 46528823c079SEric W. Biederman .install = mntns_install, 4653bcac25a5SAndrey Vagin .owner = mntns_owner, 46548823c079SEric W. Biederman }; 4655ab171b95SLuis Chamberlain 4656ab171b95SLuis Chamberlain #ifdef CONFIG_SYSCTL 4657ab171b95SLuis Chamberlain static struct ctl_table fs_namespace_sysctls[] = { 4658ab171b95SLuis Chamberlain { 4659ab171b95SLuis Chamberlain .procname = "mount-max", 4660ab171b95SLuis Chamberlain .data = &sysctl_mount_max, 4661ab171b95SLuis Chamberlain .maxlen = sizeof(unsigned int), 4662ab171b95SLuis Chamberlain .mode = 0644, 4663ab171b95SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 4664ab171b95SLuis Chamberlain .extra1 = SYSCTL_ONE, 4665ab171b95SLuis Chamberlain }, 4666ab171b95SLuis Chamberlain { } 4667ab171b95SLuis Chamberlain }; 4668ab171b95SLuis Chamberlain 4669ab171b95SLuis Chamberlain static int __init init_fs_namespace_sysctls(void) 4670ab171b95SLuis Chamberlain { 4671ab171b95SLuis Chamberlain register_sysctl_init("fs", fs_namespace_sysctls); 4672ab171b95SLuis Chamberlain return 0; 4673ab171b95SLuis Chamberlain } 4674ab171b95SLuis Chamberlain fs_initcall(init_fs_namespace_sysctls); 4675ab171b95SLuis Chamberlain 4676ab171b95SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 4677