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> 349164bb4aSIngo Molnar 3507b20889SRam Pai #include "pnode.h" 36948730b0SAdrian Bunk #include "internal.h" 371da177e4SLinus Torvalds 38d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 39d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 40d2921684SEric W. Biederman 410818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 420818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 430818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 440818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 450818bf27SAl Viro 460818bf27SAl Viro static __initdata unsigned long mhash_entries; 470818bf27SAl Viro static int __init set_mhash_entries(char *str) 480818bf27SAl Viro { 490818bf27SAl Viro if (!str) 500818bf27SAl Viro return 0; 510818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 520818bf27SAl Viro return 1; 530818bf27SAl Viro } 540818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 550818bf27SAl Viro 560818bf27SAl Viro static __initdata unsigned long mphash_entries; 570818bf27SAl Viro static int __init set_mphash_entries(char *str) 580818bf27SAl Viro { 590818bf27SAl Viro if (!str) 600818bf27SAl Viro return 0; 610818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 620818bf27SAl Viro return 1; 630818bf27SAl Viro } 640818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6513f14b4dSEric Dumazet 66c7999c36SAl Viro static u64 event; 6773cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 68719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 695addc5ddSAl Viro 7038129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 710818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 72e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 7359aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 744edbe133SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 754edbe133SAl Viro static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ 761da177e4SLinus Torvalds 772a186721SChristian Brauner struct mount_kattr { 782a186721SChristian Brauner unsigned int attr_set; 792a186721SChristian Brauner unsigned int attr_clr; 802a186721SChristian Brauner unsigned int propagation; 812a186721SChristian Brauner unsigned int lookup_flags; 822a186721SChristian Brauner bool recurse; 839caccd41SChristian Brauner struct user_namespace *mnt_userns; 842a186721SChristian Brauner }; 852a186721SChristian Brauner 86f87fd4c2SMiklos Szeredi /* /sys/fs */ 8700d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 8800d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 89f87fd4c2SMiklos Szeredi 9099b7db7bSNick Piggin /* 9199b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 9299b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 9399b7db7bSNick Piggin * up the tree. 9499b7db7bSNick Piggin * 9599b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 9699b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 9799b7db7bSNick Piggin */ 9848a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 9999b7db7bSNick Piggin 100d033cb67SChristian Brauner static inline void lock_mount_hash(void) 101d033cb67SChristian Brauner { 102d033cb67SChristian Brauner write_seqlock(&mount_lock); 103d033cb67SChristian Brauner } 104d033cb67SChristian Brauner 105d033cb67SChristian Brauner static inline void unlock_mount_hash(void) 106d033cb67SChristian Brauner { 107d033cb67SChristian Brauner write_sequnlock(&mount_lock); 108d033cb67SChristian Brauner } 109d033cb67SChristian Brauner 11038129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 1111da177e4SLinus Torvalds { 1121da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 1131da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 1140818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 1150818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 1160818bf27SAl Viro } 1170818bf27SAl Viro 1180818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 1190818bf27SAl Viro { 1200818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 1210818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 1220818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 1231da177e4SLinus Torvalds } 1241da177e4SLinus Torvalds 125b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 12673cd49ecSMiklos Szeredi { 127169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 12873cd49ecSMiklos Szeredi 129169b480eSMatthew Wilcox if (res < 0) 13073cd49ecSMiklos Szeredi return res; 131169b480eSMatthew Wilcox mnt->mnt_id = res; 132169b480eSMatthew Wilcox return 0; 13373cd49ecSMiklos Szeredi } 13473cd49ecSMiklos Szeredi 135b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 13673cd49ecSMiklos Szeredi { 137169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 13873cd49ecSMiklos Szeredi } 13973cd49ecSMiklos Szeredi 140719f5d7fSMiklos Szeredi /* 141719f5d7fSMiklos Szeredi * Allocate a new peer group ID 142719f5d7fSMiklos Szeredi */ 1434b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 144719f5d7fSMiklos Szeredi { 145169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 146f21f6220SAl Viro 147169b480eSMatthew Wilcox if (res < 0) 148f21f6220SAl Viro return res; 149169b480eSMatthew Wilcox mnt->mnt_group_id = res; 150169b480eSMatthew Wilcox return 0; 151719f5d7fSMiklos Szeredi } 152719f5d7fSMiklos Szeredi 153719f5d7fSMiklos Szeredi /* 154719f5d7fSMiklos Szeredi * Release a peer group ID 155719f5d7fSMiklos Szeredi */ 1564b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 157719f5d7fSMiklos Szeredi { 158169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 15915169fe7SAl Viro mnt->mnt_group_id = 0; 160719f5d7fSMiklos Szeredi } 161719f5d7fSMiklos Szeredi 162b3e19d92SNick Piggin /* 163b3e19d92SNick Piggin * vfsmount lock must be held for read 164b3e19d92SNick Piggin */ 16583adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 166b3e19d92SNick Piggin { 167b3e19d92SNick Piggin #ifdef CONFIG_SMP 16868e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 169b3e19d92SNick Piggin #else 170b3e19d92SNick Piggin preempt_disable(); 17168e8a9feSAl Viro mnt->mnt_count += n; 172b3e19d92SNick Piggin preempt_enable(); 173b3e19d92SNick Piggin #endif 174b3e19d92SNick Piggin } 175b3e19d92SNick Piggin 176b3e19d92SNick Piggin /* 177b3e19d92SNick Piggin * vfsmount lock must be held for write 178b3e19d92SNick Piggin */ 179edf7ddbfSEric Biggers int mnt_get_count(struct mount *mnt) 180b3e19d92SNick Piggin { 181b3e19d92SNick Piggin #ifdef CONFIG_SMP 182edf7ddbfSEric Biggers int count = 0; 183b3e19d92SNick Piggin int cpu; 184b3e19d92SNick Piggin 185b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 18668e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 187b3e19d92SNick Piggin } 188b3e19d92SNick Piggin 189b3e19d92SNick Piggin return count; 190b3e19d92SNick Piggin #else 19168e8a9feSAl Viro return mnt->mnt_count; 192b3e19d92SNick Piggin #endif 193b3e19d92SNick Piggin } 194b3e19d92SNick Piggin 195b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1961da177e4SLinus Torvalds { 197c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 198c63181e6SAl Viro if (mnt) { 19973cd49ecSMiklos Szeredi int err; 20073cd49ecSMiklos Szeredi 201c63181e6SAl Viro err = mnt_alloc_id(mnt); 20288b38782SLi Zefan if (err) 20388b38782SLi Zefan goto out_free_cache; 20488b38782SLi Zefan 20588b38782SLi Zefan if (name) { 206fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 207c63181e6SAl Viro if (!mnt->mnt_devname) 20888b38782SLi Zefan goto out_free_id; 20973cd49ecSMiklos Szeredi } 21073cd49ecSMiklos Szeredi 211b3e19d92SNick Piggin #ifdef CONFIG_SMP 212c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 213c63181e6SAl Viro if (!mnt->mnt_pcp) 214b3e19d92SNick Piggin goto out_free_devname; 215b3e19d92SNick Piggin 216c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 217b3e19d92SNick Piggin #else 218c63181e6SAl Viro mnt->mnt_count = 1; 219c63181e6SAl Viro mnt->mnt_writers = 0; 220b3e19d92SNick Piggin #endif 221b3e19d92SNick Piggin 22238129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 223c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 224c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 225c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 226c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 227c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 228c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 229c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2300a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 23199b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 23256cbb429SAl Viro INIT_HLIST_HEAD(&mnt->mnt_stuck_children); 233a6435940SChristian Brauner mnt->mnt.mnt_userns = &init_user_ns; 2341da177e4SLinus Torvalds } 235c63181e6SAl Viro return mnt; 23688b38782SLi Zefan 237d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 238d3ef3d73Snpiggin@suse.de out_free_devname: 239fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 240d3ef3d73Snpiggin@suse.de #endif 24188b38782SLi Zefan out_free_id: 242c63181e6SAl Viro mnt_free_id(mnt); 24388b38782SLi Zefan out_free_cache: 244c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 24588b38782SLi Zefan return NULL; 2461da177e4SLinus Torvalds } 2471da177e4SLinus Torvalds 2488366025eSDave Hansen /* 2498366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2508366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2518366025eSDave Hansen * We must keep track of when those operations start 2528366025eSDave Hansen * (for permission checks) and when they end, so that 2538366025eSDave Hansen * we can determine when writes are able to occur to 2548366025eSDave Hansen * a filesystem. 2558366025eSDave Hansen */ 2563d733633SDave Hansen /* 2573d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2583d733633SDave Hansen * @mnt: the mount to check for its write status 2593d733633SDave Hansen * 2603d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2613d733633SDave Hansen * It does not guarantee that the filesystem will stay 2623d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2633d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2643d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2653d733633SDave Hansen * r/w. 2663d733633SDave Hansen */ 26743f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 2683d733633SDave Hansen { 26943f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 2703d733633SDave Hansen } 2713d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2723d733633SDave Hansen 27383adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2743d733633SDave Hansen { 275d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 27668e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 277d3ef3d73Snpiggin@suse.de #else 27868e8a9feSAl Viro mnt->mnt_writers++; 279d3ef3d73Snpiggin@suse.de #endif 2803d733633SDave Hansen } 2813d733633SDave Hansen 28283adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2833d733633SDave Hansen { 284d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 28568e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 286d3ef3d73Snpiggin@suse.de #else 28768e8a9feSAl Viro mnt->mnt_writers--; 288d3ef3d73Snpiggin@suse.de #endif 289d3ef3d73Snpiggin@suse.de } 290d3ef3d73Snpiggin@suse.de 29183adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 292d3ef3d73Snpiggin@suse.de { 293d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 294d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2953d733633SDave Hansen int cpu; 2963d733633SDave Hansen 2973d733633SDave Hansen for_each_possible_cpu(cpu) { 29868e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2993d733633SDave Hansen } 3003d733633SDave Hansen 301d3ef3d73Snpiggin@suse.de return count; 302d3ef3d73Snpiggin@suse.de #else 303d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 304d3ef3d73Snpiggin@suse.de #endif 3053d733633SDave Hansen } 3063d733633SDave Hansen 3074ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3084ed5e82fSMiklos Szeredi { 3094ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3104ed5e82fSMiklos Szeredi return 1; 3114ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3124ed5e82fSMiklos Szeredi smp_rmb(); 3134ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3144ed5e82fSMiklos Szeredi } 3154ed5e82fSMiklos Szeredi 3163d733633SDave Hansen /* 317eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 318eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 319eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 320eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3213d733633SDave Hansen */ 3228366025eSDave Hansen /** 323eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 32483adc753SAl Viro * @m: the mount on which to take a write 3258366025eSDave Hansen * 326eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 327eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 328eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 329eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 330eb04c282SJan Kara * called. This is effectively a refcount. 3318366025eSDave Hansen */ 332eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3338366025eSDave Hansen { 33483adc753SAl Viro struct mount *mnt = real_mount(m); 3353d733633SDave Hansen int ret = 0; 3363d733633SDave Hansen 337d3ef3d73Snpiggin@suse.de preempt_disable(); 338c6653a83SNick Piggin mnt_inc_writers(mnt); 339d3ef3d73Snpiggin@suse.de /* 340c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 341d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 342d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 343d3ef3d73Snpiggin@suse.de */ 344d3ef3d73Snpiggin@suse.de smp_mb(); 3456aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 346d3ef3d73Snpiggin@suse.de cpu_relax(); 347d3ef3d73Snpiggin@suse.de /* 348d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 349d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 350d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 351d3ef3d73Snpiggin@suse.de */ 352d3ef3d73Snpiggin@suse.de smp_rmb(); 3534ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 354c6653a83SNick Piggin mnt_dec_writers(mnt); 3553d733633SDave Hansen ret = -EROFS; 3563d733633SDave Hansen } 357d3ef3d73Snpiggin@suse.de preempt_enable(); 358eb04c282SJan Kara 359eb04c282SJan Kara return ret; 360eb04c282SJan Kara } 361eb04c282SJan Kara 362eb04c282SJan Kara /** 363eb04c282SJan Kara * mnt_want_write - get write access to a mount 364eb04c282SJan Kara * @m: the mount on which to take a write 365eb04c282SJan Kara * 366eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 367eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 368eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 369eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 370eb04c282SJan Kara */ 371eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 372eb04c282SJan Kara { 373eb04c282SJan Kara int ret; 374eb04c282SJan Kara 375eb04c282SJan Kara sb_start_write(m->mnt_sb); 376eb04c282SJan Kara ret = __mnt_want_write(m); 377eb04c282SJan Kara if (ret) 378eb04c282SJan Kara sb_end_write(m->mnt_sb); 3793d733633SDave Hansen return ret; 3808366025eSDave Hansen } 3818366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3828366025eSDave Hansen 3838366025eSDave Hansen /** 384eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 385eb04c282SJan Kara * @file: the file who's mount on which to take a write 386eb04c282SJan Kara * 38714e43bf4SEric Biggers * This is like __mnt_want_write, but if the file is already open for writing it 38814e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 38914e43bf4SEric Biggers * and instead only does the check for emergency r/o remounts. This must be 39014e43bf4SEric Biggers * paired with __mnt_drop_write_file. 391eb04c282SJan Kara */ 392eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 393eb04c282SJan Kara { 39414e43bf4SEric Biggers if (file->f_mode & FMODE_WRITER) { 39514e43bf4SEric Biggers /* 39614e43bf4SEric Biggers * Superblock may have become readonly while there are still 39714e43bf4SEric Biggers * writable fd's, e.g. due to a fs error with errors=remount-ro 39814e43bf4SEric Biggers */ 39914e43bf4SEric Biggers if (__mnt_is_readonly(file->f_path.mnt)) 40014e43bf4SEric Biggers return -EROFS; 40114e43bf4SEric Biggers return 0; 40214e43bf4SEric Biggers } 403eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 404eb04c282SJan Kara } 405eb04c282SJan Kara 406eb04c282SJan Kara /** 4077c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4087c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4097c6893e3SMiklos Szeredi * 41014e43bf4SEric Biggers * This is like mnt_want_write, but if the file is already open for writing it 41114e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 41214e43bf4SEric Biggers * and instead only does the freeze protection and the check for emergency r/o 41314e43bf4SEric Biggers * remounts. This must be paired with mnt_drop_write_file. 4147c6893e3SMiklos Szeredi */ 4157c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4167c6893e3SMiklos Szeredi { 4177c6893e3SMiklos Szeredi int ret; 4187c6893e3SMiklos Szeredi 4197c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4207c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4217c6893e3SMiklos Szeredi if (ret) 4227c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4237c6893e3SMiklos Szeredi return ret; 4247c6893e3SMiklos Szeredi } 42596029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 42696029c4eSnpiggin@suse.de 42796029c4eSnpiggin@suse.de /** 428eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4298366025eSDave Hansen * @mnt: the mount on which to give up write access 4308366025eSDave Hansen * 4318366025eSDave Hansen * Tells the low-level filesystem that we are done 4328366025eSDave Hansen * performing writes to it. Must be matched with 433eb04c282SJan Kara * __mnt_want_write() call above. 4348366025eSDave Hansen */ 435eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4368366025eSDave Hansen { 437d3ef3d73Snpiggin@suse.de preempt_disable(); 43883adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 439d3ef3d73Snpiggin@suse.de preempt_enable(); 4408366025eSDave Hansen } 441eb04c282SJan Kara 442eb04c282SJan Kara /** 443eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 444eb04c282SJan Kara * @mnt: the mount on which to give up write access 445eb04c282SJan Kara * 446eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 447eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 448eb04c282SJan Kara * mnt_want_write() call above. 449eb04c282SJan Kara */ 450eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 451eb04c282SJan Kara { 452eb04c282SJan Kara __mnt_drop_write(mnt); 453eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 454eb04c282SJan Kara } 4558366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4568366025eSDave Hansen 457eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 458eb04c282SJan Kara { 45914e43bf4SEric Biggers if (!(file->f_mode & FMODE_WRITER)) 460eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 461eb04c282SJan Kara } 462eb04c282SJan Kara 4637c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4647c6893e3SMiklos Szeredi { 465a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4667c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4677c6893e3SMiklos Szeredi } 4682a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4692a79f17eSAl Viro 470fbdc2f6cSChristian Brauner static inline int mnt_hold_writers(struct mount *mnt) 4718366025eSDave Hansen { 47283adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 473d3ef3d73Snpiggin@suse.de /* 474d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 475d3ef3d73Snpiggin@suse.de * should be visible before we do. 476d3ef3d73Snpiggin@suse.de */ 477d3ef3d73Snpiggin@suse.de smp_mb(); 478d3ef3d73Snpiggin@suse.de 479d3ef3d73Snpiggin@suse.de /* 480d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 481d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 482d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 483d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 484d3ef3d73Snpiggin@suse.de * 485d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 486d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 487d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 488d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 489d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 490d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 491d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 492d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 493d3ef3d73Snpiggin@suse.de * we're counting up here. 494d3ef3d73Snpiggin@suse.de */ 495c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 496fbdc2f6cSChristian Brauner return -EBUSY; 497fbdc2f6cSChristian Brauner 498fbdc2f6cSChristian Brauner return 0; 499fbdc2f6cSChristian Brauner } 500fbdc2f6cSChristian Brauner 501fbdc2f6cSChristian Brauner static inline void mnt_unhold_writers(struct mount *mnt) 502fbdc2f6cSChristian Brauner { 503d3ef3d73Snpiggin@suse.de /* 504d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 505d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 506d3ef3d73Snpiggin@suse.de */ 507d3ef3d73Snpiggin@suse.de smp_wmb(); 50883adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 509fbdc2f6cSChristian Brauner } 510fbdc2f6cSChristian Brauner 511fbdc2f6cSChristian Brauner static int mnt_make_readonly(struct mount *mnt) 512fbdc2f6cSChristian Brauner { 513fbdc2f6cSChristian Brauner int ret; 514fbdc2f6cSChristian Brauner 515fbdc2f6cSChristian Brauner ret = mnt_hold_writers(mnt); 516fbdc2f6cSChristian Brauner if (!ret) 517fbdc2f6cSChristian Brauner mnt->mnt.mnt_flags |= MNT_READONLY; 518fbdc2f6cSChristian Brauner mnt_unhold_writers(mnt); 5193d733633SDave Hansen return ret; 5203d733633SDave Hansen } 5218366025eSDave Hansen 5224ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5234ed5e82fSMiklos Szeredi { 5244ed5e82fSMiklos Szeredi struct mount *mnt; 5254ed5e82fSMiklos Szeredi int err = 0; 5264ed5e82fSMiklos Szeredi 5278e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5288e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5298e8b8796SMiklos Szeredi return -EBUSY; 5308e8b8796SMiklos Szeredi 531719ea2fbSAl Viro lock_mount_hash(); 5324ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5334ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5344ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5354ed5e82fSMiklos Szeredi smp_mb(); 5364ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5374ed5e82fSMiklos Szeredi err = -EBUSY; 5384ed5e82fSMiklos Szeredi break; 5394ed5e82fSMiklos Szeredi } 5404ed5e82fSMiklos Szeredi } 5414ed5e82fSMiklos Szeredi } 5428e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5438e8b8796SMiklos Szeredi err = -EBUSY; 5448e8b8796SMiklos Szeredi 5454ed5e82fSMiklos Szeredi if (!err) { 5464ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5474ed5e82fSMiklos Szeredi smp_wmb(); 5484ed5e82fSMiklos Szeredi } 5494ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5504ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5514ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5524ed5e82fSMiklos Szeredi } 553719ea2fbSAl Viro unlock_mount_hash(); 5544ed5e82fSMiklos Szeredi 5554ed5e82fSMiklos Szeredi return err; 5564ed5e82fSMiklos Szeredi } 5574ed5e82fSMiklos Szeredi 558b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5591da177e4SLinus Torvalds { 560a6435940SChristian Brauner struct user_namespace *mnt_userns; 561a6435940SChristian Brauner 562a6435940SChristian Brauner mnt_userns = mnt_user_ns(&mnt->mnt); 563a6435940SChristian Brauner if (mnt_userns != &init_user_ns) 564a6435940SChristian Brauner put_user_ns(mnt_userns); 565fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 566d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 56768e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 568d3ef3d73Snpiggin@suse.de #endif 569b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5701da177e4SLinus Torvalds } 5711da177e4SLinus Torvalds 5728ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5738ffcb32eSDavid Howells { 5748ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5758ffcb32eSDavid Howells } 5768ffcb32eSDavid Howells 57748a066e7SAl Viro /* call under rcu_read_lock */ 578294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 57948a066e7SAl Viro { 58048a066e7SAl Viro struct mount *mnt; 58148a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 582294d71ffSAl Viro return 1; 58348a066e7SAl Viro if (bastard == NULL) 584294d71ffSAl Viro return 0; 58548a066e7SAl Viro mnt = real_mount(bastard); 58648a066e7SAl Viro mnt_add_count(mnt, 1); 587119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 58848a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 589294d71ffSAl Viro return 0; 59048a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 59148a066e7SAl Viro mnt_add_count(mnt, -1); 592294d71ffSAl Viro return 1; 59348a066e7SAl Viro } 594119e1ef8SAl Viro lock_mount_hash(); 595119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 596119e1ef8SAl Viro mnt_add_count(mnt, -1); 597119e1ef8SAl Viro unlock_mount_hash(); 598119e1ef8SAl Viro return 1; 599119e1ef8SAl Viro } 600119e1ef8SAl Viro unlock_mount_hash(); 601119e1ef8SAl Viro /* caller will mntput() */ 602294d71ffSAl Viro return -1; 603294d71ffSAl Viro } 604294d71ffSAl Viro 605294d71ffSAl Viro /* call under rcu_read_lock */ 606294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 607294d71ffSAl Viro { 608294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 609294d71ffSAl Viro if (likely(!res)) 610294d71ffSAl Viro return true; 611294d71ffSAl Viro if (unlikely(res < 0)) { 61248a066e7SAl Viro rcu_read_unlock(); 61348a066e7SAl Viro mntput(bastard); 61448a066e7SAl Viro rcu_read_lock(); 615294d71ffSAl Viro } 61648a066e7SAl Viro return false; 61748a066e7SAl Viro } 61848a066e7SAl Viro 6191da177e4SLinus Torvalds /* 620474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 62148a066e7SAl Viro * call under rcu_read_lock() 6221da177e4SLinus Torvalds */ 623474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6241da177e4SLinus Torvalds { 62538129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 626474279dcSAl Viro struct mount *p; 6271da177e4SLinus Torvalds 62838129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 629474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 630474279dcSAl Viro return p; 631474279dcSAl Viro return NULL; 6321da177e4SLinus Torvalds } 633474279dcSAl Viro 634474279dcSAl Viro /* 635f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 636f015f126SDavid Howells * 637f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 638f015f126SDavid Howells * following mounts: 639f015f126SDavid Howells * 640f015f126SDavid Howells * mount /dev/sda1 /mnt 641f015f126SDavid Howells * mount /dev/sda2 /mnt 642f015f126SDavid Howells * mount /dev/sda3 /mnt 643f015f126SDavid Howells * 644f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 645f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 646f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 647f015f126SDavid Howells * 648f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 649a05964f3SRam Pai */ 650ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 651a05964f3SRam Pai { 652c7105365SAl Viro struct mount *child_mnt; 65348a066e7SAl Viro struct vfsmount *m; 65448a066e7SAl Viro unsigned seq; 65599b7db7bSNick Piggin 65648a066e7SAl Viro rcu_read_lock(); 65748a066e7SAl Viro do { 65848a066e7SAl Viro seq = read_seqbegin(&mount_lock); 659474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 66048a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 66148a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 66248a066e7SAl Viro rcu_read_unlock(); 66348a066e7SAl Viro return m; 664a05964f3SRam Pai } 665a05964f3SRam Pai 6669f6c61f9SMiklos Szeredi static inline void lock_ns_list(struct mnt_namespace *ns) 6679f6c61f9SMiklos Szeredi { 6689f6c61f9SMiklos Szeredi spin_lock(&ns->ns_lock); 6699f6c61f9SMiklos Szeredi } 6709f6c61f9SMiklos Szeredi 6719f6c61f9SMiklos Szeredi static inline void unlock_ns_list(struct mnt_namespace *ns) 6729f6c61f9SMiklos Szeredi { 6739f6c61f9SMiklos Szeredi spin_unlock(&ns->ns_lock); 6749f6c61f9SMiklos Szeredi } 6759f6c61f9SMiklos Szeredi 6769f6c61f9SMiklos Szeredi static inline bool mnt_is_cursor(struct mount *mnt) 6779f6c61f9SMiklos Szeredi { 6789f6c61f9SMiklos Szeredi return mnt->mnt.mnt_flags & MNT_CURSOR; 6799f6c61f9SMiklos Szeredi } 6809f6c61f9SMiklos Szeredi 6817af1364fSEric W. Biederman /* 6827af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6837af1364fSEric W. Biederman * current mount namespace. 6847af1364fSEric W. Biederman * 6857af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6867af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6877af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6887af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6897af1364fSEric W. Biederman * is a mountpoint. 6907af1364fSEric W. Biederman * 6917af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6927af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6937af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6947af1364fSEric W. Biederman * parent mount. 6957af1364fSEric W. Biederman */ 6967af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6977af1364fSEric W. Biederman { 6987af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 6997af1364fSEric W. Biederman struct mount *mnt; 7007af1364fSEric W. Biederman bool is_covered = false; 7017af1364fSEric W. Biederman 7027af1364fSEric W. Biederman down_read(&namespace_sem); 7039f6c61f9SMiklos Szeredi lock_ns_list(ns); 7047af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7059f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 7069f6c61f9SMiklos Szeredi continue; 7077af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7087af1364fSEric W. Biederman if (is_covered) 7097af1364fSEric W. Biederman break; 7107af1364fSEric W. Biederman } 7119f6c61f9SMiklos Szeredi unlock_ns_list(ns); 7127af1364fSEric W. Biederman up_read(&namespace_sem); 7135ad05cc8SNikolay Borisov 7147af1364fSEric W. Biederman return is_covered; 7157af1364fSEric W. Biederman } 7167af1364fSEric W. Biederman 717e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 71884d17192SAl Viro { 7190818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 72084d17192SAl Viro struct mountpoint *mp; 72184d17192SAl Viro 7220818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 72384d17192SAl Viro if (mp->m_dentry == dentry) { 72484d17192SAl Viro mp->m_count++; 72584d17192SAl Viro return mp; 72684d17192SAl Viro } 72784d17192SAl Viro } 728e2dfa935SEric W. Biederman return NULL; 729e2dfa935SEric W. Biederman } 730e2dfa935SEric W. Biederman 7313895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 732e2dfa935SEric W. Biederman { 7333895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 734e2dfa935SEric W. Biederman int ret; 73584d17192SAl Viro 7363895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7371e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7381e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7391e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7403895dbf8SEric W. Biederman mountpoint: 7413895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7423895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7433895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7443895dbf8SEric W. Biederman if (mp) 7453895dbf8SEric W. Biederman goto done; 74684d17192SAl Viro } 747eed81007SMiklos Szeredi 7483895dbf8SEric W. Biederman if (!new) 7493895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7503895dbf8SEric W. Biederman if (!new) 7513895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7523895dbf8SEric W. Biederman 7533895dbf8SEric W. Biederman 7543895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7553895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7563895dbf8SEric W. Biederman 7573895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7583895dbf8SEric W. Biederman if (ret == -EBUSY) 7593895dbf8SEric W. Biederman goto mountpoint; 7603895dbf8SEric W. Biederman 7613895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7623895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7633895dbf8SEric W. Biederman if (ret) 7643895dbf8SEric W. Biederman goto done; 7653895dbf8SEric W. Biederman 7663895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7673895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7684edbe133SAl Viro new->m_dentry = dget(dentry); 7693895dbf8SEric W. Biederman new->m_count = 1; 7703895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7713895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7723895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7733895dbf8SEric W. Biederman 7743895dbf8SEric W. Biederman mp = new; 7753895dbf8SEric W. Biederman new = NULL; 7763895dbf8SEric W. Biederman done: 7773895dbf8SEric W. Biederman kfree(new); 77884d17192SAl Viro return mp; 77984d17192SAl Viro } 78084d17192SAl Viro 7814edbe133SAl Viro /* 7824edbe133SAl Viro * vfsmount lock must be held. Additionally, the caller is responsible 7834edbe133SAl Viro * for serializing calls for given disposal list. 7844edbe133SAl Viro */ 7854edbe133SAl Viro static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) 78684d17192SAl Viro { 78784d17192SAl Viro if (!--mp->m_count) { 78884d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7890a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 79084d17192SAl Viro spin_lock(&dentry->d_lock); 79184d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 79284d17192SAl Viro spin_unlock(&dentry->d_lock); 7934edbe133SAl Viro dput_to_list(dentry, list); 7940818bf27SAl Viro hlist_del(&mp->m_hash); 79584d17192SAl Viro kfree(mp); 79684d17192SAl Viro } 79784d17192SAl Viro } 79884d17192SAl Viro 7994edbe133SAl Viro /* called with namespace_lock and vfsmount lock */ 8004edbe133SAl Viro static void put_mountpoint(struct mountpoint *mp) 8014edbe133SAl Viro { 8024edbe133SAl Viro __put_mountpoint(mp, &ex_mountpoints); 8034edbe133SAl Viro } 8044edbe133SAl Viro 805143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 8061da177e4SLinus Torvalds { 8076b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 8081da177e4SLinus Torvalds } 8091da177e4SLinus Torvalds 81099b7db7bSNick Piggin /* 81199b7db7bSNick Piggin * vfsmount lock must be held for write 81299b7db7bSNick Piggin */ 8136b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 8145addc5ddSAl Viro { 8155addc5ddSAl Viro if (ns) { 8165addc5ddSAl Viro ns->event = ++event; 8175addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8185addc5ddSAl Viro } 8195addc5ddSAl Viro } 8205addc5ddSAl Viro 82199b7db7bSNick Piggin /* 82299b7db7bSNick Piggin * vfsmount lock must be held for write 82399b7db7bSNick Piggin */ 8246b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8255addc5ddSAl Viro { 8265addc5ddSAl Viro if (ns && ns->event != event) { 8275addc5ddSAl Viro ns->event = event; 8285addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8295addc5ddSAl Viro } 8305addc5ddSAl Viro } 8315addc5ddSAl Viro 83299b7db7bSNick Piggin /* 83399b7db7bSNick Piggin * vfsmount lock must be held for write 83499b7db7bSNick Piggin */ 835e4e59906SAl Viro static struct mountpoint *unhash_mnt(struct mount *mnt) 8361da177e4SLinus Torvalds { 837e4e59906SAl Viro struct mountpoint *mp; 8380714a533SAl Viro mnt->mnt_parent = mnt; 839a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8406b41d536SAl Viro list_del_init(&mnt->mnt_child); 84138129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8420a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 843e4e59906SAl Viro mp = mnt->mnt_mp; 84484d17192SAl Viro mnt->mnt_mp = NULL; 845e4e59906SAl Viro return mp; 8467bdb11deSEric W. Biederman } 8477bdb11deSEric W. Biederman 8487bdb11deSEric W. Biederman /* 8497bdb11deSEric W. Biederman * vfsmount lock must be held for write 8507bdb11deSEric W. Biederman */ 8516a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8526a46c573SEric W. Biederman { 853e4e59906SAl Viro put_mountpoint(unhash_mnt(mnt)); 8546a46c573SEric W. Biederman } 8556a46c573SEric W. Biederman 8566a46c573SEric W. Biederman /* 8576a46c573SEric W. Biederman * vfsmount lock must be held for write 8586a46c573SEric W. Biederman */ 85984d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 86084d17192SAl Viro struct mountpoint *mp, 86144d964d6SAl Viro struct mount *child_mnt) 862b90fa9aeSRam Pai { 86384d17192SAl Viro mp->m_count++; 8643a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 8654edbe133SAl Viro child_mnt->mnt_mountpoint = mp->m_dentry; 8663a2393d7SAl Viro child_mnt->mnt_parent = mnt; 86784d17192SAl Viro child_mnt->mnt_mp = mp; 8680a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 869b90fa9aeSRam Pai } 870b90fa9aeSRam Pai 8711064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8721064f874SEric W. Biederman { 8731064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8741064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8751064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8761064f874SEric W. Biederman } 8771064f874SEric W. Biederman 87899b7db7bSNick Piggin /* 87999b7db7bSNick Piggin * vfsmount lock must be held for write 88099b7db7bSNick Piggin */ 88184d17192SAl Viro static void attach_mnt(struct mount *mnt, 88284d17192SAl Viro struct mount *parent, 88384d17192SAl Viro struct mountpoint *mp) 8841da177e4SLinus Torvalds { 88584d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8861064f874SEric W. Biederman __attach_mnt(mnt, parent); 887b90fa9aeSRam Pai } 888b90fa9aeSRam Pai 8891064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 89012a5b529SAl Viro { 8911064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8921064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8931064f874SEric W. Biederman 8941064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8951064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8961064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8971064f874SEric W. Biederman 8981064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 8991064f874SEric W. Biederman 9001064f874SEric W. Biederman put_mountpoint(old_mp); 9011064f874SEric W. Biederman mnt_add_count(old_parent, -1); 90212a5b529SAl Viro } 90312a5b529SAl Viro 904b90fa9aeSRam Pai /* 90599b7db7bSNick Piggin * vfsmount lock must be held for write 906b90fa9aeSRam Pai */ 9071064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 908b90fa9aeSRam Pai { 9090714a533SAl Viro struct mount *parent = mnt->mnt_parent; 91083adc753SAl Viro struct mount *m; 911b90fa9aeSRam Pai LIST_HEAD(head); 912143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 913b90fa9aeSRam Pai 9140714a533SAl Viro BUG_ON(parent == mnt); 915b90fa9aeSRam Pai 9161a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 917f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 918143c8c91SAl Viro m->mnt_ns = n; 919f03c6599SAl Viro 920b90fa9aeSRam Pai list_splice(&head, n->list.prev); 921b90fa9aeSRam Pai 922d2921684SEric W. Biederman n->mounts += n->pending_mounts; 923d2921684SEric W. Biederman n->pending_mounts = 0; 924d2921684SEric W. Biederman 9251064f874SEric W. Biederman __attach_mnt(mnt, parent); 9266b3286edSKirill Korotaev touch_mnt_namespace(n); 9271da177e4SLinus Torvalds } 9281da177e4SLinus Torvalds 929909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9301da177e4SLinus Torvalds { 9316b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9326b41d536SAl Viro if (next == &p->mnt_mounts) { 9331da177e4SLinus Torvalds while (1) { 934909b0a88SAl Viro if (p == root) 9351da177e4SLinus Torvalds return NULL; 9366b41d536SAl Viro next = p->mnt_child.next; 9376b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9381da177e4SLinus Torvalds break; 9390714a533SAl Viro p = p->mnt_parent; 9401da177e4SLinus Torvalds } 9411da177e4SLinus Torvalds } 9426b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9431da177e4SLinus Torvalds } 9441da177e4SLinus Torvalds 945315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9469676f0c6SRam Pai { 9476b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9486b41d536SAl Viro while (prev != &p->mnt_mounts) { 9496b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9506b41d536SAl Viro prev = p->mnt_mounts.prev; 9519676f0c6SRam Pai } 9529676f0c6SRam Pai return p; 9539676f0c6SRam Pai } 9549676f0c6SRam Pai 9558f291889SAl Viro /** 9568f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 9578f291889SAl Viro * @fc: The configuration context with the superblock attached 9588f291889SAl Viro * 9598f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 9608f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 9618f291889SAl Viro * 9628f291889SAl Viro * Note that this does not attach the mount to anything. 9638f291889SAl Viro */ 9648f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 9659d412a43SAl Viro { 966b105e270SAl Viro struct mount *mnt; 9679d412a43SAl Viro 9688f291889SAl Viro if (!fc->root) 9698f291889SAl Viro return ERR_PTR(-EINVAL); 9709d412a43SAl Viro 9718f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 9729d412a43SAl Viro if (!mnt) 9739d412a43SAl Viro return ERR_PTR(-ENOMEM); 9749d412a43SAl Viro 9758f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 976b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9779d412a43SAl Viro 9788f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 9798f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 9808f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 981a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9820714a533SAl Viro mnt->mnt_parent = mnt; 9838f291889SAl Viro 984719ea2fbSAl Viro lock_mount_hash(); 9858f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 986719ea2fbSAl Viro unlock_mount_hash(); 987b105e270SAl Viro return &mnt->mnt; 9889d412a43SAl Viro } 9898f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 9908f291889SAl Viro 9918f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 9928f291889SAl Viro { 9938f291889SAl Viro int err = vfs_get_tree(fc); 9948f291889SAl Viro if (!err) { 9958f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 9968f291889SAl Viro return vfs_create_mount(fc); 9978f291889SAl Viro } 9988f291889SAl Viro return ERR_PTR(err); 9998f291889SAl Viro } 10008f291889SAl Viro EXPORT_SYMBOL(fc_mount); 10018f291889SAl Viro 10029bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 10039bc61ab1SDavid Howells int flags, const char *name, 10049bc61ab1SDavid Howells void *data) 10051da177e4SLinus Torvalds { 10069bc61ab1SDavid Howells struct fs_context *fc; 10078f291889SAl Viro struct vfsmount *mnt; 10089bc61ab1SDavid Howells int ret = 0; 10099d412a43SAl Viro 10109d412a43SAl Viro if (!type) 10113e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 10129d412a43SAl Viro 10139bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 10149bc61ab1SDavid Howells if (IS_ERR(fc)) 10159bc61ab1SDavid Howells return ERR_CAST(fc); 10169bc61ab1SDavid Howells 10173e1aeb00SDavid Howells if (name) 10183e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 10193e1aeb00SDavid Howells name, strlen(name)); 10209bc61ab1SDavid Howells if (!ret) 10219bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 10229bc61ab1SDavid Howells if (!ret) 10238f291889SAl Viro mnt = fc_mount(fc); 10248f291889SAl Viro else 10258f291889SAl Viro mnt = ERR_PTR(ret); 10269d412a43SAl Viro 10279bc61ab1SDavid Howells put_fs_context(fc); 10288f291889SAl Viro return mnt; 10299d412a43SAl Viro } 10309d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10319d412a43SAl Viro 103293faccbbSEric W. Biederman struct vfsmount * 103393faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 103493faccbbSEric W. Biederman const char *name, void *data) 103593faccbbSEric W. Biederman { 103693faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 103793faccbbSEric W. Biederman * through from the parent mount to the submount don't support 103893faccbbSEric W. Biederman * unprivileged mounts with submounts. 103993faccbbSEric W. Biederman */ 104093faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 104193faccbbSEric W. Biederman return ERR_PTR(-EPERM); 104293faccbbSEric W. Biederman 1043e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 104493faccbbSEric W. Biederman } 104593faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 104693faccbbSEric W. Biederman 104787129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 104836341f64SRam Pai int flag) 10491da177e4SLinus Torvalds { 105087129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1051be34d1a3SDavid Howells struct mount *mnt; 1052be34d1a3SDavid Howells int err; 10531da177e4SLinus Torvalds 1054be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1055be34d1a3SDavid Howells if (!mnt) 1056be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1057be34d1a3SDavid Howells 10587a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 105915169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1060719f5d7fSMiklos Szeredi else 106115169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1062719f5d7fSMiklos Szeredi 106315169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1064be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1065719f5d7fSMiklos Szeredi if (err) 1066719f5d7fSMiklos Szeredi goto out_free; 1067719f5d7fSMiklos Szeredi } 1068719f5d7fSMiklos Szeredi 106916a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 107016a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 10715ff9d8a6SEric W. Biederman 10721da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1073a6435940SChristian Brauner mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt); 1074a6435940SChristian Brauner if (mnt->mnt.mnt_userns != &init_user_ns) 1075a6435940SChristian Brauner mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns); 1076b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1077b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1078a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10790714a533SAl Viro mnt->mnt_parent = mnt; 1080719ea2fbSAl Viro lock_mount_hash(); 108139f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1082719ea2fbSAl Viro unlock_mount_hash(); 1083b90fa9aeSRam Pai 10847a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10857a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10866776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 108732301920SAl Viro mnt->mnt_master = old; 1088fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10898aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1090fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10916776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1092d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10936776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1094d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10955235d448SAl Viro } else { 10965235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10975afe0022SRam Pai } 1098b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10990f0afb1dSAl Viro set_mnt_shared(mnt); 11001da177e4SLinus Torvalds 11011da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 11021da177e4SLinus Torvalds * as the original if that was on one */ 110336341f64SRam Pai if (flag & CL_EXPIRE) { 11046776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 11056776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 11061da177e4SLinus Torvalds } 1107be34d1a3SDavid Howells 1108cb338d06SAl Viro return mnt; 1109719f5d7fSMiklos Szeredi 1110719f5d7fSMiklos Szeredi out_free: 11118ffcb32eSDavid Howells mnt_free_id(mnt); 1112719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1113be34d1a3SDavid Howells return ERR_PTR(err); 11141da177e4SLinus Torvalds } 11151da177e4SLinus Torvalds 11169ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11179ea459e1SAl Viro { 111856cbb429SAl Viro struct hlist_node *p; 111956cbb429SAl Viro struct mount *m; 11209ea459e1SAl Viro /* 112156cbb429SAl Viro * The warning here probably indicates that somebody messed 112256cbb429SAl Viro * up a mnt_want/drop_write() pair. If this happens, the 112356cbb429SAl Viro * filesystem was probably unable to make r/w->r/o transitions. 11249ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11259ea459e1SAl Viro * so mnt_get_writers() below is safe. 11269ea459e1SAl Viro */ 11279ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11289ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11299ea459e1SAl Viro mnt_pin_kill(mnt); 113056cbb429SAl Viro hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { 113156cbb429SAl Viro hlist_del(&m->mnt_umount); 113256cbb429SAl Viro mntput(&m->mnt); 113356cbb429SAl Viro } 11349ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11359ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11369ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11379ea459e1SAl Viro mnt_free_id(mnt); 11389ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11399ea459e1SAl Viro } 11409ea459e1SAl Viro 11419ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11429ea459e1SAl Viro { 11439ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11449ea459e1SAl Viro } 11459ea459e1SAl Viro 11469ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11479ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11489ea459e1SAl Viro { 11499ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 115029785735SByungchul Park struct mount *m, *t; 11519ea459e1SAl Viro 115229785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 115329785735SByungchul Park cleanup_mnt(m); 11549ea459e1SAl Viro } 11559ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11569ea459e1SAl Viro 1157900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11587b7b1aceSAl Viro { 11594edbe133SAl Viro LIST_HEAD(list); 1160edf7ddbfSEric Biggers int count; 11614edbe133SAl Viro 116248a066e7SAl Viro rcu_read_lock(); 11639ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11649ea0a46cSAl Viro /* 11659ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11669ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11679ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11689ea0a46cSAl Viro * be dropped until after an RCU delay done after 11699ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11709ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11719ea0a46cSAl Viro * we are dropping is not the final one. 11729ea0a46cSAl Viro */ 1173aa9c0e07SAl Viro mnt_add_count(mnt, -1); 117448a066e7SAl Viro rcu_read_unlock(); 117599b7db7bSNick Piggin return; 1176b3e19d92SNick Piggin } 1177719ea2fbSAl Viro lock_mount_hash(); 1178119e1ef8SAl Viro /* 1179119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1180119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1181119e1ef8SAl Viro */ 1182119e1ef8SAl Viro smp_mb(); 11839ea0a46cSAl Viro mnt_add_count(mnt, -1); 1184edf7ddbfSEric Biggers count = mnt_get_count(mnt); 1185edf7ddbfSEric Biggers if (count != 0) { 1186edf7ddbfSEric Biggers WARN_ON(count < 0); 118748a066e7SAl Viro rcu_read_unlock(); 1188719ea2fbSAl Viro unlock_mount_hash(); 118999b7db7bSNick Piggin return; 119099b7db7bSNick Piggin } 119148a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 119248a066e7SAl Viro rcu_read_unlock(); 119348a066e7SAl Viro unlock_mount_hash(); 119448a066e7SAl Viro return; 119548a066e7SAl Viro } 119648a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 119748a066e7SAl Viro rcu_read_unlock(); 1198962830dfSAndi Kleen 119939f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1200ce07d891SEric W. Biederman 1201ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1202ce07d891SEric W. Biederman struct mount *p, *tmp; 1203ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 12044edbe133SAl Viro __put_mountpoint(unhash_mnt(p), &list); 120556cbb429SAl Viro hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); 1206ce07d891SEric W. Biederman } 1207ce07d891SEric W. Biederman } 1208719ea2fbSAl Viro unlock_mount_hash(); 12094edbe133SAl Viro shrink_dentry_list(&list); 1210649a795aSAl Viro 12119ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 12129ea459e1SAl Viro struct task_struct *task = current; 12139ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 12149ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 121591989c70SJens Axboe if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) 12169ea459e1SAl Viro return; 12179ea459e1SAl Viro } 12189ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 12199ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 12209ea459e1SAl Viro return; 12219ea459e1SAl Viro } 12229ea459e1SAl Viro cleanup_mnt(mnt); 1223b3e19d92SNick Piggin } 1224b3e19d92SNick Piggin 1225b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1226b3e19d92SNick Piggin { 1227b3e19d92SNick Piggin if (mnt) { 1228863d684fSAl Viro struct mount *m = real_mount(mnt); 1229b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1230863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1231863d684fSAl Viro m->mnt_expiry_mark = 0; 1232863d684fSAl Viro mntput_no_expire(m); 1233b3e19d92SNick Piggin } 1234b3e19d92SNick Piggin } 1235b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1236b3e19d92SNick Piggin 1237b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1238b3e19d92SNick Piggin { 1239b3e19d92SNick Piggin if (mnt) 124083adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1241b3e19d92SNick Piggin return mnt; 1242b3e19d92SNick Piggin } 1243b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1244b3e19d92SNick Piggin 12451f287bc4SRandy Dunlap /** 12461f287bc4SRandy Dunlap * path_is_mountpoint() - Check if path is a mount in the current namespace. 12471f287bc4SRandy Dunlap * @path: path to check 1248c6609c0aSIan Kent * 1249c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1250c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1251c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1252c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1253c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1254c6609c0aSIan Kent * alone. 1255c6609c0aSIan Kent */ 1256c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1257c6609c0aSIan Kent { 1258c6609c0aSIan Kent unsigned seq; 1259c6609c0aSIan Kent bool res; 1260c6609c0aSIan Kent 1261c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1262c6609c0aSIan Kent return false; 1263c6609c0aSIan Kent 1264c6609c0aSIan Kent rcu_read_lock(); 1265c6609c0aSIan Kent do { 1266c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1267c6609c0aSIan Kent res = __path_is_mountpoint(path); 1268c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1269c6609c0aSIan Kent rcu_read_unlock(); 1270c6609c0aSIan Kent 1271c6609c0aSIan Kent return res; 1272c6609c0aSIan Kent } 1273c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1274c6609c0aSIan Kent 1275ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12767b7b1aceSAl Viro { 12773064c356SAl Viro struct mount *p; 12783064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12793064c356SAl Viro if (IS_ERR(p)) 12803064c356SAl Viro return ERR_CAST(p); 12813064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12823064c356SAl Viro return &p->mnt; 12837b7b1aceSAl Viro } 12841da177e4SLinus Torvalds 1285a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12869f6c61f9SMiklos Szeredi static struct mount *mnt_list_next(struct mnt_namespace *ns, 12879f6c61f9SMiklos Szeredi struct list_head *p) 12889f6c61f9SMiklos Szeredi { 12899f6c61f9SMiklos Szeredi struct mount *mnt, *ret = NULL; 12909f6c61f9SMiklos Szeredi 12919f6c61f9SMiklos Szeredi lock_ns_list(ns); 12929f6c61f9SMiklos Szeredi list_for_each_continue(p, &ns->list) { 12939f6c61f9SMiklos Szeredi mnt = list_entry(p, typeof(*mnt), mnt_list); 12949f6c61f9SMiklos Szeredi if (!mnt_is_cursor(mnt)) { 12959f6c61f9SMiklos Szeredi ret = mnt; 12969f6c61f9SMiklos Szeredi break; 12979f6c61f9SMiklos Szeredi } 12989f6c61f9SMiklos Szeredi } 12999f6c61f9SMiklos Szeredi unlock_ns_list(ns); 13009f6c61f9SMiklos Szeredi 13019f6c61f9SMiklos Szeredi return ret; 13029f6c61f9SMiklos Szeredi } 13039f6c61f9SMiklos Szeredi 13040226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 13051da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 13061da177e4SLinus Torvalds { 1307ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13089f6c61f9SMiklos Szeredi struct list_head *prev; 13091da177e4SLinus Torvalds 1310390c6843SRam Pai down_read(&namespace_sem); 13119f6c61f9SMiklos Szeredi if (!*pos) { 13129f6c61f9SMiklos Szeredi prev = &p->ns->list; 13139f6c61f9SMiklos Szeredi } else { 13149f6c61f9SMiklos Szeredi prev = &p->cursor.mnt_list; 13159f6c61f9SMiklos Szeredi 13169f6c61f9SMiklos Szeredi /* Read after we'd reached the end? */ 13179f6c61f9SMiklos Szeredi if (list_empty(prev)) 13189f6c61f9SMiklos Szeredi return NULL; 1319c7999c36SAl Viro } 1320c7999c36SAl Viro 13219f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, prev); 13221da177e4SLinus Torvalds } 13231da177e4SLinus Torvalds 13241da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13251da177e4SLinus Torvalds { 1326ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13279f6c61f9SMiklos Szeredi struct mount *mnt = v; 1328b0765fb8SPavel Emelianov 13299f6c61f9SMiklos Szeredi ++*pos; 13309f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, &mnt->mnt_list); 13311da177e4SLinus Torvalds } 13321da177e4SLinus Torvalds 13331da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13341da177e4SLinus Torvalds { 13359f6c61f9SMiklos Szeredi struct proc_mounts *p = m->private; 13369f6c61f9SMiklos Szeredi struct mount *mnt = v; 13379f6c61f9SMiklos Szeredi 13389f6c61f9SMiklos Szeredi lock_ns_list(p->ns); 13399f6c61f9SMiklos Szeredi if (mnt) 13409f6c61f9SMiklos Szeredi list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); 13419f6c61f9SMiklos Szeredi else 13429f6c61f9SMiklos Szeredi list_del_init(&p->cursor.mnt_list); 13439f6c61f9SMiklos Szeredi unlock_ns_list(p->ns); 1344390c6843SRam Pai up_read(&namespace_sem); 13451da177e4SLinus Torvalds } 13461da177e4SLinus Torvalds 13470226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13489f5596afSAl Viro { 1349ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13509f6c61f9SMiklos Szeredi struct mount *r = v; 13510226f492SAl Viro return p->show(m, &r->mnt); 13521da177e4SLinus Torvalds } 13531da177e4SLinus Torvalds 1354a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13551da177e4SLinus Torvalds .start = m_start, 13561da177e4SLinus Torvalds .next = m_next, 13571da177e4SLinus Torvalds .stop = m_stop, 13580226f492SAl Viro .show = m_show, 1359b4629fe2SChuck Lever }; 13609f6c61f9SMiklos Szeredi 13619f6c61f9SMiklos Szeredi void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) 13629f6c61f9SMiklos Szeredi { 13639f6c61f9SMiklos Szeredi down_read(&namespace_sem); 13649f6c61f9SMiklos Szeredi lock_ns_list(ns); 13659f6c61f9SMiklos Szeredi list_del(&cursor->mnt_list); 13669f6c61f9SMiklos Szeredi unlock_ns_list(ns); 13679f6c61f9SMiklos Szeredi up_read(&namespace_sem); 13689f6c61f9SMiklos Szeredi } 1369a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1370b4629fe2SChuck Lever 13711da177e4SLinus Torvalds /** 13721da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 13731f287bc4SRandy Dunlap * @m: root of mount tree 13741da177e4SLinus Torvalds * 13751da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 13761da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 13771da177e4SLinus Torvalds * busy. 13781da177e4SLinus Torvalds */ 1379909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13801da177e4SLinus Torvalds { 1381909b0a88SAl Viro struct mount *mnt = real_mount(m); 138236341f64SRam Pai int actual_refs = 0; 138336341f64SRam Pai int minimum_refs = 0; 1384315fc83eSAl Viro struct mount *p; 1385909b0a88SAl Viro BUG_ON(!m); 13861da177e4SLinus Torvalds 1387b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1388719ea2fbSAl Viro lock_mount_hash(); 1389909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 139083adc753SAl Viro actual_refs += mnt_get_count(p); 13911da177e4SLinus Torvalds minimum_refs += 2; 13921da177e4SLinus Torvalds } 1393719ea2fbSAl Viro unlock_mount_hash(); 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13961da177e4SLinus Torvalds return 0; 1397e3474a8eSIan Kent 1398e3474a8eSIan Kent return 1; 13991da177e4SLinus Torvalds } 14001da177e4SLinus Torvalds 14011da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 14021da177e4SLinus Torvalds 14031da177e4SLinus Torvalds /** 14041da177e4SLinus Torvalds * may_umount - check if a mount point is busy 14051da177e4SLinus Torvalds * @mnt: root of mount 14061da177e4SLinus Torvalds * 14071da177e4SLinus Torvalds * This is called to check if a mount point has any 14081da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 14091da177e4SLinus Torvalds * mount has sub mounts this will return busy 14101da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 14111da177e4SLinus Torvalds * 14121da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 14131da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 14141da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 14151da177e4SLinus Torvalds */ 14161da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 14171da177e4SLinus Torvalds { 1418e3474a8eSIan Kent int ret = 1; 14198ad08d8aSAl Viro down_read(&namespace_sem); 1420719ea2fbSAl Viro lock_mount_hash(); 14211ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1422e3474a8eSIan Kent ret = 0; 1423719ea2fbSAl Viro unlock_mount_hash(); 14248ad08d8aSAl Viro up_read(&namespace_sem); 1425a05964f3SRam Pai return ret; 14261da177e4SLinus Torvalds } 14271da177e4SLinus Torvalds 14281da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14291da177e4SLinus Torvalds 143097216be0SAl Viro static void namespace_unlock(void) 14311da177e4SLinus Torvalds { 1432a3b3c562SEric W. Biederman struct hlist_head head; 143356cbb429SAl Viro struct hlist_node *p; 143456cbb429SAl Viro struct mount *m; 14354edbe133SAl Viro LIST_HEAD(list); 143697216be0SAl Viro 1437a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 14384edbe133SAl Viro list_splice_init(&ex_mountpoints, &list); 1439a3b3c562SEric W. Biederman 144097216be0SAl Viro up_write(&namespace_sem); 1441a3b3c562SEric W. Biederman 14424edbe133SAl Viro shrink_dentry_list(&list); 14434edbe133SAl Viro 1444a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 144597216be0SAl Viro return; 144697216be0SAl Viro 144722cb7405SNeilBrown synchronize_rcu_expedited(); 144848a066e7SAl Viro 144956cbb429SAl Viro hlist_for_each_entry_safe(m, p, &head, mnt_umount) { 145056cbb429SAl Viro hlist_del(&m->mnt_umount); 145156cbb429SAl Viro mntput(&m->mnt); 145256cbb429SAl Viro } 145370fbcdf4SRam Pai } 145470fbcdf4SRam Pai 145597216be0SAl Viro static inline void namespace_lock(void) 1456e3197d83SAl Viro { 145797216be0SAl Viro down_write(&namespace_sem); 1458e3197d83SAl Viro } 1459e3197d83SAl Viro 1460e819f152SEric W. Biederman enum umount_tree_flags { 1461e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1462e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1463e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1464e819f152SEric W. Biederman }; 1465f2d0a123SEric W. Biederman 1466f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1467f2d0a123SEric W. Biederman { 1468f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1469f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1470f2d0a123SEric W. Biederman return true; 1471f2d0a123SEric W. Biederman 1472f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1473f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1474f2d0a123SEric W. Biederman return true; 1475f2d0a123SEric W. Biederman 1476f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1477f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1478f2d0a123SEric W. Biederman * connected to mounted mounts. 1479f2d0a123SEric W. Biederman */ 1480f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1481f2d0a123SEric W. Biederman return true; 1482f2d0a123SEric W. Biederman 1483f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1484f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1485f2d0a123SEric W. Biederman return false; 1486f2d0a123SEric W. Biederman 1487f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1488f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1489f2d0a123SEric W. Biederman return false; 1490f2d0a123SEric W. Biederman 1491f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1492f2d0a123SEric W. Biederman return true; 1493f2d0a123SEric W. Biederman } 1494f2d0a123SEric W. Biederman 149599b7db7bSNick Piggin /* 149648a066e7SAl Viro * mount_lock must be held 149799b7db7bSNick Piggin * namespace_sem must be held for write 149899b7db7bSNick Piggin */ 1499e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 150070fbcdf4SRam Pai { 1501c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1502315fc83eSAl Viro struct mount *p; 150370fbcdf4SRam Pai 15045d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 15055d88457eSEric W. Biederman propagate_mount_unlock(mnt); 15065d88457eSEric W. Biederman 1507c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1508590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1509590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1510c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1511590ce4bcSEric W. Biederman } 1512c003b26fSEric W. Biederman 1513411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1514c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1515c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 151638129a13SAl Viro } 151770fbcdf4SRam Pai 1518c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1519e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 15207b8a53fdSAl Viro propagate_umount(&tmp_list); 1521a05964f3SRam Pai 1522c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1523d2921684SEric W. Biederman struct mnt_namespace *ns; 1524ce07d891SEric W. Biederman bool disconnect; 1525c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 15266776db3dSAl Viro list_del_init(&p->mnt_expire); 15271a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1528d2921684SEric W. Biederman ns = p->mnt_ns; 1529d2921684SEric W. Biederman if (ns) { 1530d2921684SEric W. Biederman ns->mounts--; 1531d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1532d2921684SEric W. Biederman } 153363d37a84SAl Viro p->mnt_ns = NULL; 1534e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 153548a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 153687b95ce0SAl Viro 1537f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1538676da58dSAl Viro if (mnt_has_parent(p)) { 153981b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1540ce07d891SEric W. Biederman if (!disconnect) { 1541ce07d891SEric W. Biederman /* Don't forget about p */ 1542ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1543ce07d891SEric W. Biederman } else { 15446a46c573SEric W. Biederman umount_mnt(p); 15457c4b93d8SAl Viro } 1546ce07d891SEric W. Biederman } 15470f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 154819a1c409SAl Viro if (disconnect) 154919a1c409SAl Viro hlist_add_head(&p->mnt_umount, &unmounted); 155038129a13SAl Viro } 15511da177e4SLinus Torvalds } 15521da177e4SLinus Torvalds 1553b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1554c35038beSAl Viro 15558d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 15568d0347f6SDavid Howells { 15578d0347f6SDavid Howells int ret = 0; 15588d0347f6SDavid Howells 15598d0347f6SDavid Howells down_write(&sb->s_umount); 15608d0347f6SDavid Howells if (!sb_rdonly(sb)) { 15618d0347f6SDavid Howells struct fs_context *fc; 15628d0347f6SDavid Howells 15638d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 15648d0347f6SDavid Howells SB_RDONLY); 15658d0347f6SDavid Howells if (IS_ERR(fc)) { 15668d0347f6SDavid Howells ret = PTR_ERR(fc); 15678d0347f6SDavid Howells } else { 15688d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 15698d0347f6SDavid Howells if (!ret) 15708d0347f6SDavid Howells ret = reconfigure_super(fc); 15718d0347f6SDavid Howells put_fs_context(fc); 15728d0347f6SDavid Howells } 15738d0347f6SDavid Howells } 15748d0347f6SDavid Howells up_write(&sb->s_umount); 15758d0347f6SDavid Howells return ret; 15768d0347f6SDavid Howells } 15778d0347f6SDavid Howells 15781ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 15791da177e4SLinus Torvalds { 15801ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 15811da177e4SLinus Torvalds int retval; 15821da177e4SLinus Torvalds 15831ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 15841da177e4SLinus Torvalds if (retval) 15851da177e4SLinus Torvalds return retval; 15861da177e4SLinus Torvalds 15871da177e4SLinus Torvalds /* 15881da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 15891da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 15901da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 15911da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 15921da177e4SLinus Torvalds */ 15931da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 15941ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 15951da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 15961da177e4SLinus Torvalds return -EINVAL; 15971da177e4SLinus Torvalds 1598b3e19d92SNick Piggin /* 1599b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1600b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1601b3e19d92SNick Piggin */ 1602719ea2fbSAl Viro lock_mount_hash(); 160383adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1604719ea2fbSAl Viro unlock_mount_hash(); 16051da177e4SLinus Torvalds return -EBUSY; 1606b3e19d92SNick Piggin } 1607719ea2fbSAl Viro unlock_mount_hash(); 16081da177e4SLinus Torvalds 1609863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 16101da177e4SLinus Torvalds return -EAGAIN; 16111da177e4SLinus Torvalds } 16121da177e4SLinus Torvalds 16131da177e4SLinus Torvalds /* 16141da177e4SLinus Torvalds * If we may have to abort operations to get out of this 16151da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 16161da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 16171da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 16181da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 16191da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 16201da177e4SLinus Torvalds * about for the moment. 16211da177e4SLinus Torvalds */ 16221da177e4SLinus Torvalds 162342faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 162442faad99SAl Viro sb->s_op->umount_begin(sb); 162542faad99SAl Viro } 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds /* 16281da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 16291da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 16301da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 16311da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 16321da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 16331da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 16341da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 16351da177e4SLinus Torvalds */ 16361ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 16371da177e4SLinus Torvalds /* 16381da177e4SLinus Torvalds * Special case for "unmounting" root ... 16391da177e4SLinus Torvalds * we just try to remount it readonly. 16401da177e4SLinus Torvalds */ 1641bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1642a1480dccSAndy Lutomirski return -EPERM; 16438d0347f6SDavid Howells return do_umount_root(sb); 16441da177e4SLinus Torvalds } 16451da177e4SLinus Torvalds 164697216be0SAl Viro namespace_lock(); 1647719ea2fbSAl Viro lock_mount_hash(); 16481da177e4SLinus Torvalds 164925d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 165025d202edSEric W. Biederman retval = -EINVAL; 165125d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 165225d202edSEric W. Biederman goto out; 165325d202edSEric W. Biederman 165425d202edSEric W. Biederman event++; 165548a066e7SAl Viro if (flags & MNT_DETACH) { 165648a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1657e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 165848a066e7SAl Viro retval = 0; 165948a066e7SAl Viro } else { 1660b54b9be7SAl Viro shrink_submounts(mnt); 16611da177e4SLinus Torvalds retval = -EBUSY; 166248a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 16631a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1664e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 16651da177e4SLinus Torvalds retval = 0; 16661da177e4SLinus Torvalds } 166748a066e7SAl Viro } 166825d202edSEric W. Biederman out: 1669719ea2fbSAl Viro unlock_mount_hash(); 1670e3197d83SAl Viro namespace_unlock(); 16711da177e4SLinus Torvalds return retval; 16721da177e4SLinus Torvalds } 16731da177e4SLinus Torvalds 16741da177e4SLinus Torvalds /* 167580b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 167680b5dce8SEric W. Biederman * 167780b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 167880b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 167980b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 168080b5dce8SEric W. Biederman * leaking them. 168180b5dce8SEric W. Biederman * 168280b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 168380b5dce8SEric W. Biederman */ 168480b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 168580b5dce8SEric W. Biederman { 168680b5dce8SEric W. Biederman struct mountpoint *mp; 168780b5dce8SEric W. Biederman struct mount *mnt; 168880b5dce8SEric W. Biederman 168980b5dce8SEric W. Biederman namespace_lock(); 16903895dbf8SEric W. Biederman lock_mount_hash(); 169180b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1692adc9b5c0SAl Viro if (!mp) 169380b5dce8SEric W. Biederman goto out_unlock; 169480b5dce8SEric W. Biederman 1695e06b933eSAndrey Ulanov event++; 169680b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 169780b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1698ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1699fe78fcc8SEric W. Biederman umount_mnt(mnt); 170056cbb429SAl Viro hlist_add_head(&mnt->mnt_umount, &unmounted); 1701ce07d891SEric W. Biederman } 1702e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 170380b5dce8SEric W. Biederman } 170480b5dce8SEric W. Biederman put_mountpoint(mp); 170580b5dce8SEric W. Biederman out_unlock: 17063895dbf8SEric W. Biederman unlock_mount_hash(); 170780b5dce8SEric W. Biederman namespace_unlock(); 170880b5dce8SEric W. Biederman } 170980b5dce8SEric W. Biederman 171080b5dce8SEric W. Biederman /* 17119b40bc90SAl Viro * Is the caller allowed to modify his namespace? 17129b40bc90SAl Viro */ 17139b40bc90SAl Viro static inline bool may_mount(void) 17149b40bc90SAl Viro { 17159b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 17169b40bc90SAl Viro } 17179b40bc90SAl Viro 1718*f7e33bdbSJeff Layton static void warn_mandlock(void) 17199e8925b6SJeff Layton { 1720*f7e33bdbSJeff Layton pr_warn_once("=======================================================\n" 1721*f7e33bdbSJeff Layton "WARNING: The mand mount option has been deprecated and\n" 1722*f7e33bdbSJeff Layton " and is ignored by this kernel. Remove the mand\n" 1723*f7e33bdbSJeff Layton " option from the mount to silence this warning.\n" 1724*f7e33bdbSJeff Layton "=======================================================\n"); 17259e8925b6SJeff Layton } 17269e8925b6SJeff Layton 172725ccd24fSChristoph Hellwig static int can_umount(const struct path *path, int flags) 17281da177e4SLinus Torvalds { 172925ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 17301da177e4SLinus Torvalds 17319b40bc90SAl Viro if (!may_mount()) 17329b40bc90SAl Viro return -EPERM; 173341525f56SChristoph Hellwig if (path->dentry != path->mnt->mnt_root) 173425ccd24fSChristoph Hellwig return -EINVAL; 1735143c8c91SAl Viro if (!check_mnt(mnt)) 173625ccd24fSChristoph Hellwig return -EINVAL; 173725d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 173825ccd24fSChristoph Hellwig return -EINVAL; 1739b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 174025ccd24fSChristoph Hellwig return -EPERM; 174125ccd24fSChristoph Hellwig return 0; 174225ccd24fSChristoph Hellwig } 17431da177e4SLinus Torvalds 1744a0a6df9aSAl Viro // caller is responsible for flags being sane 174525ccd24fSChristoph Hellwig int path_umount(struct path *path, int flags) 174625ccd24fSChristoph Hellwig { 174725ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 174825ccd24fSChristoph Hellwig int ret; 174925ccd24fSChristoph Hellwig 175025ccd24fSChristoph Hellwig ret = can_umount(path, flags); 175125ccd24fSChristoph Hellwig if (!ret) 175225ccd24fSChristoph Hellwig ret = do_umount(mnt, flags); 175325ccd24fSChristoph Hellwig 1754429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 175541525f56SChristoph Hellwig dput(path->dentry); 1756900148dcSAl Viro mntput_no_expire(mnt); 175725ccd24fSChristoph Hellwig return ret; 17581da177e4SLinus Torvalds } 17591da177e4SLinus Torvalds 176009267defSChristoph Hellwig static int ksys_umount(char __user *name, int flags) 176141525f56SChristoph Hellwig { 176241525f56SChristoph Hellwig int lookup_flags = LOOKUP_MOUNTPOINT; 176341525f56SChristoph Hellwig struct path path; 176441525f56SChristoph Hellwig int ret; 176541525f56SChristoph Hellwig 1766a0a6df9aSAl Viro // basic validity checks done first 1767a0a6df9aSAl Viro if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1768a0a6df9aSAl Viro return -EINVAL; 1769a0a6df9aSAl Viro 177041525f56SChristoph Hellwig if (!(flags & UMOUNT_NOFOLLOW)) 177141525f56SChristoph Hellwig lookup_flags |= LOOKUP_FOLLOW; 177241525f56SChristoph Hellwig ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); 177341525f56SChristoph Hellwig if (ret) 177441525f56SChristoph Hellwig return ret; 177541525f56SChristoph Hellwig return path_umount(&path, flags); 177641525f56SChristoph Hellwig } 177741525f56SChristoph Hellwig 17783a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 17793a18ef5cSDominik Brodowski { 17803a18ef5cSDominik Brodowski return ksys_umount(name, flags); 17813a18ef5cSDominik Brodowski } 17823a18ef5cSDominik Brodowski 17831da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 17841da177e4SLinus Torvalds 17851da177e4SLinus Torvalds /* 17861da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17871da177e4SLinus Torvalds */ 1788bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17891da177e4SLinus Torvalds { 17903a18ef5cSDominik Brodowski return ksys_umount(name, 0); 17911da177e4SLinus Torvalds } 17921da177e4SLinus Torvalds 17931da177e4SLinus Torvalds #endif 17941da177e4SLinus Torvalds 17954ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 17968823c079SEric W. Biederman { 17974ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1798e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1799e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 18004ce5d2b1SEric W. Biederman } 18014ce5d2b1SEric W. Biederman 1802213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 180358be2825SAl Viro { 180458be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 180558be2825SAl Viro } 180658be2825SAl Viro 1807303cc571SChristian Brauner struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) 1808303cc571SChristian Brauner { 1809303cc571SChristian Brauner return &mnt->ns; 1810303cc571SChristian Brauner } 1811303cc571SChristian Brauner 18124ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 18134ce5d2b1SEric W. Biederman { 18144ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 18154ce5d2b1SEric W. Biederman * mount namespace loop? 18164ce5d2b1SEric W. Biederman */ 18174ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 18184ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 18194ce5d2b1SEric W. Biederman return false; 18204ce5d2b1SEric W. Biederman 1821f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 18228823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 18238823c079SEric W. Biederman } 18248823c079SEric W. Biederman 182587129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 182636341f64SRam Pai int flag) 18271da177e4SLinus Torvalds { 182884d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 18291da177e4SLinus Torvalds 18304ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 18314ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 18324ce5d2b1SEric W. Biederman 18334ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1834be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 18359676f0c6SRam Pai 183636341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1837be34d1a3SDavid Howells if (IS_ERR(q)) 1838be34d1a3SDavid Howells return q; 1839be34d1a3SDavid Howells 1840a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 18411da177e4SLinus Torvalds 18421da177e4SLinus Torvalds p = mnt; 18436b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1844315fc83eSAl Viro struct mount *s; 18457ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 18461da177e4SLinus Torvalds continue; 18471da177e4SLinus Torvalds 1848909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 18494ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 18504ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1851df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1852df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1853df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1854df7342b2SEric W. Biederman goto out; 1855df7342b2SEric W. Biederman } else { 18564ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 18574ce5d2b1SEric W. Biederman continue; 18584ce5d2b1SEric W. Biederman } 1859df7342b2SEric W. Biederman } 18604ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 18614ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 18629676f0c6SRam Pai s = skip_mnt_tree(s); 18639676f0c6SRam Pai continue; 18649676f0c6SRam Pai } 18650714a533SAl Viro while (p != s->mnt_parent) { 18660714a533SAl Viro p = p->mnt_parent; 18670714a533SAl Viro q = q->mnt_parent; 18681da177e4SLinus Torvalds } 186987129cc0SAl Viro p = s; 187084d17192SAl Viro parent = q; 187187129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1872be34d1a3SDavid Howells if (IS_ERR(q)) 1873be34d1a3SDavid Howells goto out; 1874719ea2fbSAl Viro lock_mount_hash(); 18751a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 18761064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1877719ea2fbSAl Viro unlock_mount_hash(); 18781da177e4SLinus Torvalds } 18791da177e4SLinus Torvalds } 18801da177e4SLinus Torvalds return res; 1881be34d1a3SDavid Howells out: 18821da177e4SLinus Torvalds if (res) { 1883719ea2fbSAl Viro lock_mount_hash(); 1884e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1885719ea2fbSAl Viro unlock_mount_hash(); 18861da177e4SLinus Torvalds } 1887be34d1a3SDavid Howells return q; 18881da177e4SLinus Torvalds } 18891da177e4SLinus Torvalds 1890be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1891be34d1a3SDavid Howells 1892ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 18938aec0809SAl Viro { 1894cb338d06SAl Viro struct mount *tree; 189597216be0SAl Viro namespace_lock(); 1896cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1897cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1898cd4a4017SEric W. Biederman else 189987129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 190087129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1901328e6d90SAl Viro namespace_unlock(); 1902be34d1a3SDavid Howells if (IS_ERR(tree)) 190352e220d3SDan Carpenter return ERR_CAST(tree); 1904be34d1a3SDavid Howells return &tree->mnt; 19058aec0809SAl Viro } 19068aec0809SAl Viro 1907a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1908a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1909a07b2000SAl Viro 1910a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1911a07b2000SAl Viro { 1912a07b2000SAl Viro struct mnt_namespace *ns; 1913a07b2000SAl Viro namespace_lock(); 1914a07b2000SAl Viro lock_mount_hash(); 1915a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 191644dfd84aSDavid Howells if (ns) { 191744dfd84aSDavid Howells if (is_anon_ns(ns)) 1918a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 191944dfd84aSDavid Howells else 192044dfd84aSDavid Howells ns = NULL; 192144dfd84aSDavid Howells } 1922a07b2000SAl Viro unlock_mount_hash(); 1923a07b2000SAl Viro namespace_unlock(); 192444dfd84aSDavid Howells if (ns) 1925a07b2000SAl Viro free_mnt_ns(ns); 1926a07b2000SAl Viro } 1927a07b2000SAl Viro 19288aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 19298aec0809SAl Viro { 193097216be0SAl Viro namespace_lock(); 1931719ea2fbSAl Viro lock_mount_hash(); 19329c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1933719ea2fbSAl Viro unlock_mount_hash(); 19343ab6abeeSAl Viro namespace_unlock(); 19358aec0809SAl Viro } 19368aec0809SAl Viro 1937427215d8SMiklos Szeredi static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 1938427215d8SMiklos Szeredi { 1939427215d8SMiklos Szeredi struct mount *child; 1940427215d8SMiklos Szeredi 1941427215d8SMiklos Szeredi list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 1942427215d8SMiklos Szeredi if (!is_subdir(child->mnt_mountpoint, dentry)) 1943427215d8SMiklos Szeredi continue; 1944427215d8SMiklos Szeredi 1945427215d8SMiklos Szeredi if (child->mnt.mnt_flags & MNT_LOCKED) 1946427215d8SMiklos Szeredi return true; 1947427215d8SMiklos Szeredi } 1948427215d8SMiklos Szeredi return false; 1949427215d8SMiklos Szeredi } 1950427215d8SMiklos Szeredi 1951c771d683SMiklos Szeredi /** 1952c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 19531f287bc4SRandy Dunlap * @path: path to clone 1954c771d683SMiklos Szeredi * 19551f287bc4SRandy Dunlap * This creates a new vfsmount, which will be the clone of @path. The new mount 19561f287bc4SRandy Dunlap * will not be attached anywhere in the namespace and will be private (i.e. 19571f287bc4SRandy Dunlap * changes to the originating mount won't be propagated into this). 1958c771d683SMiklos Szeredi * 1959c771d683SMiklos Szeredi * Release with mntput(). 1960c771d683SMiklos Szeredi */ 1961ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1962c771d683SMiklos Szeredi { 1963c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1964c771d683SMiklos Szeredi struct mount *new_mnt; 1965c771d683SMiklos Szeredi 1966427215d8SMiklos Szeredi down_read(&namespace_sem); 1967c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1968427215d8SMiklos Szeredi goto invalid; 1969427215d8SMiklos Szeredi 1970427215d8SMiklos Szeredi if (!check_mnt(old_mnt)) 1971427215d8SMiklos Szeredi goto invalid; 1972427215d8SMiklos Szeredi 1973427215d8SMiklos Szeredi if (has_locked_children(old_mnt, path->dentry)) 1974427215d8SMiklos Szeredi goto invalid; 1975c771d683SMiklos Szeredi 1976c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1977427215d8SMiklos Szeredi up_read(&namespace_sem); 1978427215d8SMiklos Szeredi 1979c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1980c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1981c771d683SMiklos Szeredi 1982df820f8dSMiklos Szeredi /* Longterm mount to be removed by kern_unmount*() */ 1983df820f8dSMiklos Szeredi new_mnt->mnt_ns = MNT_NS_INTERNAL; 1984df820f8dSMiklos Szeredi 1985c771d683SMiklos Szeredi return &new_mnt->mnt; 1986427215d8SMiklos Szeredi 1987427215d8SMiklos Szeredi invalid: 1988427215d8SMiklos Szeredi up_read(&namespace_sem); 1989427215d8SMiklos Szeredi return ERR_PTR(-EINVAL); 1990c771d683SMiklos Szeredi } 1991c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1992c771d683SMiklos Szeredi 19931f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 19941f707137SAl Viro struct vfsmount *root) 19951f707137SAl Viro { 19961a4eeaf2SAl Viro struct mount *mnt; 19971f707137SAl Viro int res = f(root, arg); 19981f707137SAl Viro if (res) 19991f707137SAl Viro return res; 20001a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 20011a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 20021f707137SAl Viro if (res) 20031f707137SAl Viro return res; 20041f707137SAl Viro } 20051f707137SAl Viro return 0; 20061f707137SAl Viro } 20071f707137SAl Viro 20083bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 20093bd045ccSAl Viro { 20103bd045ccSAl Viro struct mount *p; 20113bd045ccSAl Viro 20123bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 20133bd045ccSAl Viro int flags = p->mnt.mnt_flags; 20143bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 20153bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 20163bd045ccSAl Viro 20173bd045ccSAl Viro if (flags & MNT_READONLY) 20183bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 20193bd045ccSAl Viro 20203bd045ccSAl Viro if (flags & MNT_NODEV) 20213bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 20223bd045ccSAl Viro 20233bd045ccSAl Viro if (flags & MNT_NOSUID) 20243bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 20253bd045ccSAl Viro 20263bd045ccSAl Viro if (flags & MNT_NOEXEC) 20273bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 20283bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 20293bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 20303bd045ccSAl Viro flags |= MNT_LOCKED; 20313bd045ccSAl Viro p->mnt.mnt_flags = flags; 20323bd045ccSAl Viro } 20333bd045ccSAl Viro } 20343bd045ccSAl Viro 20354b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 2036719f5d7fSMiklos Szeredi { 2037315fc83eSAl Viro struct mount *p; 2038719f5d7fSMiklos Szeredi 2039909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 2040fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 20414b8b21f4SAl Viro mnt_release_group_id(p); 2042719f5d7fSMiklos Szeredi } 2043719f5d7fSMiklos Szeredi } 2044719f5d7fSMiklos Szeredi 20454b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 2046719f5d7fSMiklos Szeredi { 2047315fc83eSAl Viro struct mount *p; 2048719f5d7fSMiklos Szeredi 2049909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 2050fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 20514b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 2052719f5d7fSMiklos Szeredi if (err) { 20534b8b21f4SAl Viro cleanup_group_ids(mnt, p); 2054719f5d7fSMiklos Szeredi return err; 2055719f5d7fSMiklos Szeredi } 2056719f5d7fSMiklos Szeredi } 2057719f5d7fSMiklos Szeredi } 2058719f5d7fSMiklos Szeredi 2059719f5d7fSMiklos Szeredi return 0; 2060719f5d7fSMiklos Szeredi } 2061719f5d7fSMiklos Szeredi 2062d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 2063d2921684SEric W. Biederman { 2064d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 2065d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 2066d2921684SEric W. Biederman struct mount *p; 2067d2921684SEric W. Biederman 2068d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 2069d2921684SEric W. Biederman mounts++; 2070d2921684SEric W. Biederman 2071d2921684SEric W. Biederman old = ns->mounts; 2072d2921684SEric W. Biederman pending = ns->pending_mounts; 2073d2921684SEric W. Biederman sum = old + pending; 2074d2921684SEric W. Biederman if ((old > sum) || 2075d2921684SEric W. Biederman (pending > sum) || 2076d2921684SEric W. Biederman (max < sum) || 2077d2921684SEric W. Biederman (mounts > (max - sum))) 2078d2921684SEric W. Biederman return -ENOSPC; 2079d2921684SEric W. Biederman 2080d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 2081d2921684SEric W. Biederman return 0; 2082d2921684SEric W. Biederman } 2083d2921684SEric W. Biederman 2084b90fa9aeSRam Pai /* 2085b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 2086b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 208721444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 208821444403SRam Pai * store the parent mount and mountpoint dentry. 208921444403SRam Pai * (done when source_mnt is moved) 2090b90fa9aeSRam Pai * 2091b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 2092b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 20939676f0c6SRam Pai * --------------------------------------------------------------------------- 2094b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 20959676f0c6SRam Pai * |************************************************************************** 20969676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 20979676f0c6SRam Pai * | dest | | | | | 20989676f0c6SRam Pai * | | | | | | | 20999676f0c6SRam Pai * | v | | | | | 21009676f0c6SRam Pai * |************************************************************************** 21019676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 21025afe0022SRam Pai * | | | | | | 21039676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 21049676f0c6SRam Pai * *************************************************************************** 2105b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 2106b90fa9aeSRam Pai * destination mount. 2107b90fa9aeSRam Pai * 2108b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2109b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2110b90fa9aeSRam Pai * the peer group of the source mount. 2111b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2112b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2113b90fa9aeSRam Pai * mount. 21145afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 21155afe0022SRam Pai * of the destination mount and the cloned mount is made slave 21165afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 21175afe0022SRam Pai * is marked as 'shared and slave'. 21185afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 21195afe0022SRam Pai * source mount. 21205afe0022SRam Pai * 21219676f0c6SRam Pai * --------------------------------------------------------------------------- 212221444403SRam Pai * | MOVE MOUNT OPERATION | 21239676f0c6SRam Pai * |************************************************************************** 21249676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 21259676f0c6SRam Pai * | dest | | | | | 21269676f0c6SRam Pai * | | | | | | | 21279676f0c6SRam Pai * | v | | | | | 21289676f0c6SRam Pai * |************************************************************************** 21299676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 21305afe0022SRam Pai * | | | | | | 21319676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 21329676f0c6SRam Pai * *************************************************************************** 21335afe0022SRam Pai * 21345afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 21355afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 213621444403SRam Pai * (+*) the mount is moved to the destination. 21375afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 21385afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 21395afe0022SRam Pai * the mount is marked as 'shared and slave'. 21405afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2141b90fa9aeSRam Pai * 2142b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2143b90fa9aeSRam Pai * applied to each mount in the tree. 2144b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2145b90fa9aeSRam Pai * in allocations. 2146b90fa9aeSRam Pai */ 21470fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 214884d17192SAl Viro struct mount *dest_mnt, 214984d17192SAl Viro struct mountpoint *dest_mp, 21502763d119SAl Viro bool moving) 2151b90fa9aeSRam Pai { 21523bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 215338129a13SAl Viro HLIST_HEAD(tree_list); 2154d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 21551064f874SEric W. Biederman struct mountpoint *smp; 2156315fc83eSAl Viro struct mount *child, *p; 215738129a13SAl Viro struct hlist_node *n; 2158719f5d7fSMiklos Szeredi int err; 2159b90fa9aeSRam Pai 21601064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 21611064f874SEric W. Biederman * to be tucked under other mounts. 21621064f874SEric W. Biederman */ 21631064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 21641064f874SEric W. Biederman if (IS_ERR(smp)) 21651064f874SEric W. Biederman return PTR_ERR(smp); 21661064f874SEric W. Biederman 2167d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 21682763d119SAl Viro if (!moving) { 2169d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2170d2921684SEric W. Biederman if (err) 2171d2921684SEric W. Biederman goto out; 2172d2921684SEric W. Biederman } 2173d2921684SEric W. Biederman 2174fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 21750fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2176719f5d7fSMiklos Szeredi if (err) 2177719f5d7fSMiklos Szeredi goto out; 217884d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2179f2ebb3a9SAl Viro lock_mount_hash(); 2180719f5d7fSMiklos Szeredi if (err) 2181719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2182909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 21830f0afb1dSAl Viro set_mnt_shared(p); 21840b1b901bSAl Viro } else { 21850b1b901bSAl Viro lock_mount_hash(); 2186b90fa9aeSRam Pai } 21872763d119SAl Viro if (moving) { 21882763d119SAl Viro unhash_mnt(source_mnt); 218984d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2190143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 219121444403SRam Pai } else { 219244dfd84aSDavid Howells if (source_mnt->mnt_ns) { 219344dfd84aSDavid Howells /* move from anon - the caller will destroy */ 219444dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 219544dfd84aSDavid Howells } 219684d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 21971064f874SEric W. Biederman commit_tree(source_mnt); 219821444403SRam Pai } 2199b90fa9aeSRam Pai 220038129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 22011d6a32acSAl Viro struct mount *q; 220238129a13SAl Viro hlist_del_init(&child->mnt_hash); 22031064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 22041d6a32acSAl Viro child->mnt_mountpoint); 22051064f874SEric W. Biederman if (q) 22061064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 22073bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 22083bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 22093bd045ccSAl Viro lock_mnt_tree(child); 2210d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 22111064f874SEric W. Biederman commit_tree(child); 2212b90fa9aeSRam Pai } 22131064f874SEric W. Biederman put_mountpoint(smp); 2214719ea2fbSAl Viro unlock_mount_hash(); 221599b7db7bSNick Piggin 2216b90fa9aeSRam Pai return 0; 2217719f5d7fSMiklos Szeredi 2218719f5d7fSMiklos Szeredi out_cleanup_ids: 2219f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2220f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2221d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2222e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2223f2ebb3a9SAl Viro } 2224f2ebb3a9SAl Viro unlock_mount_hash(); 22250fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2226719f5d7fSMiklos Szeredi out: 2227d2921684SEric W. Biederman ns->pending_mounts = 0; 22281064f874SEric W. Biederman 22291064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 22301064f874SEric W. Biederman put_mountpoint(smp); 22311064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 22321064f874SEric W. Biederman 2233719f5d7fSMiklos Szeredi return err; 2234b90fa9aeSRam Pai } 2235b90fa9aeSRam Pai 223684d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2237b12cea91SAl Viro { 2238b12cea91SAl Viro struct vfsmount *mnt; 223984d17192SAl Viro struct dentry *dentry = path->dentry; 2240b12cea91SAl Viro retry: 22415955102cSAl Viro inode_lock(dentry->d_inode); 224284d17192SAl Viro if (unlikely(cant_mount(dentry))) { 22435955102cSAl Viro inode_unlock(dentry->d_inode); 224484d17192SAl Viro return ERR_PTR(-ENOENT); 2245b12cea91SAl Viro } 224697216be0SAl Viro namespace_lock(); 2247b12cea91SAl Viro mnt = lookup_mnt(path); 224884d17192SAl Viro if (likely(!mnt)) { 22493895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 225084d17192SAl Viro if (IS_ERR(mp)) { 225197216be0SAl Viro namespace_unlock(); 22525955102cSAl Viro inode_unlock(dentry->d_inode); 225384d17192SAl Viro return mp; 225484d17192SAl Viro } 225584d17192SAl Viro return mp; 225684d17192SAl Viro } 225797216be0SAl Viro namespace_unlock(); 22585955102cSAl Viro inode_unlock(path->dentry->d_inode); 2259b12cea91SAl Viro path_put(path); 2260b12cea91SAl Viro path->mnt = mnt; 226184d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2262b12cea91SAl Viro goto retry; 2263b12cea91SAl Viro } 2264b12cea91SAl Viro 226584d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2266b12cea91SAl Viro { 226784d17192SAl Viro struct dentry *dentry = where->m_dentry; 22683895dbf8SEric W. Biederman 22693895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 227084d17192SAl Viro put_mountpoint(where); 22713895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 22723895dbf8SEric W. Biederman 2273328e6d90SAl Viro namespace_unlock(); 22745955102cSAl Viro inode_unlock(dentry->d_inode); 2275b12cea91SAl Viro } 2276b12cea91SAl Viro 227784d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 22781da177e4SLinus Torvalds { 2279e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 22801da177e4SLinus Torvalds return -EINVAL; 22811da177e4SLinus Torvalds 2282e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2283e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 22841da177e4SLinus Torvalds return -ENOTDIR; 22851da177e4SLinus Torvalds 22862763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 22871da177e4SLinus Torvalds } 22881da177e4SLinus Torvalds 22891da177e4SLinus Torvalds /* 22907a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 22917a2e8a8fSValerie Aurora */ 22927a2e8a8fSValerie Aurora 2293e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 22947a2e8a8fSValerie Aurora { 2295e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 22967a2e8a8fSValerie Aurora 22977a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 22987a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 22997a2e8a8fSValerie Aurora return 0; 23007a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 23017a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 23027a2e8a8fSValerie Aurora return 0; 23037a2e8a8fSValerie Aurora return type; 23047a2e8a8fSValerie Aurora } 23057a2e8a8fSValerie Aurora 23067a2e8a8fSValerie Aurora /* 230707b20889SRam Pai * recursively change the type of the mountpoint. 230807b20889SRam Pai */ 2309e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 231007b20889SRam Pai { 2311315fc83eSAl Viro struct mount *m; 23124b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2313e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 23147a2e8a8fSValerie Aurora int type; 2315719f5d7fSMiklos Szeredi int err = 0; 231607b20889SRam Pai 23172d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 231807b20889SRam Pai return -EINVAL; 231907b20889SRam Pai 2320e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 23217a2e8a8fSValerie Aurora if (!type) 23227a2e8a8fSValerie Aurora return -EINVAL; 23237a2e8a8fSValerie Aurora 232497216be0SAl Viro namespace_lock(); 2325719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2326719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2327719f5d7fSMiklos Szeredi if (err) 2328719f5d7fSMiklos Szeredi goto out_unlock; 2329719f5d7fSMiklos Szeredi } 2330719f5d7fSMiklos Szeredi 2331719ea2fbSAl Viro lock_mount_hash(); 2332909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 23330f0afb1dSAl Viro change_mnt_propagation(m, type); 2334719ea2fbSAl Viro unlock_mount_hash(); 2335719f5d7fSMiklos Szeredi 2336719f5d7fSMiklos Szeredi out_unlock: 233797216be0SAl Viro namespace_unlock(); 2338719f5d7fSMiklos Szeredi return err; 233907b20889SRam Pai } 234007b20889SRam Pai 2341a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2342a07b2000SAl Viro { 2343a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2344a07b2000SAl Viro 2345a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2346a07b2000SAl Viro return mnt; 2347a07b2000SAl Viro 2348a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2349a07b2000SAl Viro return mnt; 2350a07b2000SAl Viro 2351a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2352a07b2000SAl Viro return mnt; 2353a07b2000SAl Viro 2354a07b2000SAl Viro if (recurse) 2355a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2356a07b2000SAl Viro else 2357a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2358a07b2000SAl Viro 2359a07b2000SAl Viro if (!IS_ERR(mnt)) 2360a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2361a07b2000SAl Viro 2362a07b2000SAl Viro return mnt; 2363a07b2000SAl Viro } 2364a07b2000SAl Viro 236507b20889SRam Pai /* 23661da177e4SLinus Torvalds * do loopback mount. 23671da177e4SLinus Torvalds */ 2368808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 23692dafe1c4SEric Sandeen int recurse) 23701da177e4SLinus Torvalds { 23712d92ab3cSAl Viro struct path old_path; 2372a07b2000SAl Viro struct mount *mnt = NULL, *parent; 237384d17192SAl Viro struct mountpoint *mp; 237457eccb83SAl Viro int err; 23751da177e4SLinus Torvalds if (!old_name || !*old_name) 23761da177e4SLinus Torvalds return -EINVAL; 2377815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 23781da177e4SLinus Torvalds if (err) 23791da177e4SLinus Torvalds return err; 23801da177e4SLinus Torvalds 23818823c079SEric W. Biederman err = -EINVAL; 23824ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 23838823c079SEric W. Biederman goto out; 23848823c079SEric W. Biederman 238584d17192SAl Viro mp = lock_mount(path); 2386a07b2000SAl Viro if (IS_ERR(mp)) { 238784d17192SAl Viro err = PTR_ERR(mp); 23889676f0c6SRam Pai goto out; 2389a07b2000SAl Viro } 23909676f0c6SRam Pai 239184d17192SAl Viro parent = real_mount(path->mnt); 2392e149ed2bSAl Viro if (!check_mnt(parent)) 2393e149ed2bSAl Viro goto out2; 2394e149ed2bSAl Viro 2395a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2396be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2397be34d1a3SDavid Howells err = PTR_ERR(mnt); 2398e9c5d8a5SAndrey Vagin goto out2; 2399be34d1a3SDavid Howells } 2400ccd48bc7SAl Viro 240184d17192SAl Viro err = graft_tree(mnt, parent, mp); 24021da177e4SLinus Torvalds if (err) { 2403719ea2fbSAl Viro lock_mount_hash(); 2404e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2405719ea2fbSAl Viro unlock_mount_hash(); 24065b83d2c5SRam Pai } 2407b12cea91SAl Viro out2: 240884d17192SAl Viro unlock_mount(mp); 2409ccd48bc7SAl Viro out: 24102d92ab3cSAl Viro path_put(&old_path); 24111da177e4SLinus Torvalds return err; 24121da177e4SLinus Torvalds } 24131da177e4SLinus Torvalds 2414a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2415a07b2000SAl Viro { 2416a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2417a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2418a07b2000SAl Viro struct mount *mnt, *p; 2419a07b2000SAl Viro struct file *file; 2420a07b2000SAl Viro 2421a07b2000SAl Viro if (IS_ERR(ns)) 2422a07b2000SAl Viro return ERR_CAST(ns); 2423a07b2000SAl Viro 2424a07b2000SAl Viro namespace_lock(); 2425a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2426a07b2000SAl Viro if (IS_ERR(mnt)) { 2427a07b2000SAl Viro namespace_unlock(); 2428a07b2000SAl Viro free_mnt_ns(ns); 2429a07b2000SAl Viro return ERR_CAST(mnt); 2430a07b2000SAl Viro } 2431a07b2000SAl Viro 2432a07b2000SAl Viro lock_mount_hash(); 2433a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2434a07b2000SAl Viro p->mnt_ns = ns; 2435a07b2000SAl Viro ns->mounts++; 2436a07b2000SAl Viro } 2437a07b2000SAl Viro ns->root = mnt; 2438a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2439a07b2000SAl Viro mntget(&mnt->mnt); 2440a07b2000SAl Viro unlock_mount_hash(); 2441a07b2000SAl Viro namespace_unlock(); 2442a07b2000SAl Viro 2443a07b2000SAl Viro mntput(path->mnt); 2444a07b2000SAl Viro path->mnt = &mnt->mnt; 2445a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2446a07b2000SAl Viro if (IS_ERR(file)) 2447a07b2000SAl Viro dissolve_on_fput(path->mnt); 2448a07b2000SAl Viro else 2449a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2450a07b2000SAl Viro return file; 2451a07b2000SAl Viro } 2452a07b2000SAl Viro 24532658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2454a07b2000SAl Viro { 2455a07b2000SAl Viro struct file *file; 2456a07b2000SAl Viro struct path path; 2457a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2458a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2459a07b2000SAl Viro int error; 2460a07b2000SAl Viro int fd; 2461a07b2000SAl Viro 2462a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2463a07b2000SAl Viro 2464a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2465a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2466a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2467a07b2000SAl Viro return -EINVAL; 2468a07b2000SAl Viro 2469a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2470a07b2000SAl Viro return -EINVAL; 2471a07b2000SAl Viro 2472a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2473a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2474a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2475a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2476a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2477a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2478a07b2000SAl Viro 2479a07b2000SAl Viro if (detached && !may_mount()) 2480a07b2000SAl Viro return -EPERM; 2481a07b2000SAl Viro 2482a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2483a07b2000SAl Viro if (fd < 0) 2484a07b2000SAl Viro return fd; 2485a07b2000SAl Viro 2486a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2487a07b2000SAl Viro if (unlikely(error)) { 2488a07b2000SAl Viro file = ERR_PTR(error); 2489a07b2000SAl Viro } else { 2490a07b2000SAl Viro if (detached) 2491a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2492a07b2000SAl Viro else 2493a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2494a07b2000SAl Viro path_put(&path); 2495a07b2000SAl Viro } 2496a07b2000SAl Viro if (IS_ERR(file)) { 2497a07b2000SAl Viro put_unused_fd(fd); 2498a07b2000SAl Viro return PTR_ERR(file); 2499a07b2000SAl Viro } 2500a07b2000SAl Viro fd_install(fd, file); 2501a07b2000SAl Viro return fd; 2502a07b2000SAl Viro } 2503a07b2000SAl Viro 250443f5e655SDavid Howells /* 250543f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 250643f5e655SDavid Howells * 250743f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 250843f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 250943f5e655SDavid Howells */ 251043f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 25112e4b7fcdSDave Hansen { 251243f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 25132e4b7fcdSDave Hansen 251443f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 251543f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 251643f5e655SDavid Howells return false; 251743f5e655SDavid Howells 251843f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 251943f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 252043f5e655SDavid Howells return false; 252143f5e655SDavid Howells 252243f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 252343f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 252443f5e655SDavid Howells return false; 252543f5e655SDavid Howells 252643f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 252743f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 252843f5e655SDavid Howells return false; 252943f5e655SDavid Howells 253043f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 253143f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 253243f5e655SDavid Howells return false; 253343f5e655SDavid Howells 253443f5e655SDavid Howells return true; 253543f5e655SDavid Howells } 253643f5e655SDavid Howells 253743f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 253843f5e655SDavid Howells { 253943f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 254043f5e655SDavid Howells 254143f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 25422e4b7fcdSDave Hansen return 0; 25432e4b7fcdSDave Hansen 25442e4b7fcdSDave Hansen if (readonly_request) 254543f5e655SDavid Howells return mnt_make_readonly(mnt); 254643f5e655SDavid Howells 254768847c94SChristian Brauner mnt->mnt.mnt_flags &= ~MNT_READONLY; 254868847c94SChristian Brauner return 0; 254943f5e655SDavid Howells } 255043f5e655SDavid Howells 255143f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 255243f5e655SDavid Howells { 255343f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 255443f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 255543f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 255643f5e655SDavid Howells } 255743f5e655SDavid Howells 2558f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2559f8b92ba6SDeepa Dinamani { 2560f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2561f8b92ba6SDeepa Dinamani 2562f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2563f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2564f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2565f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2566f8b92ba6SDeepa Dinamani struct tm tm; 2567f8b92ba6SDeepa Dinamani 2568f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2569f8b92ba6SDeepa Dinamani 25700ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 25710ecee669SEric Biggers sb->s_type->name, 25720ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 25730ecee669SEric Biggers mntpath, 2574f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2575f8b92ba6SDeepa Dinamani 2576f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2577f8b92ba6SDeepa Dinamani } 2578f8b92ba6SDeepa Dinamani } 2579f8b92ba6SDeepa Dinamani 258043f5e655SDavid Howells /* 258143f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 258243f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 258343f5e655SDavid Howells * to mount(2). 258443f5e655SDavid Howells */ 258543f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 258643f5e655SDavid Howells { 258743f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 258843f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 258943f5e655SDavid Howells int ret; 259043f5e655SDavid Howells 259143f5e655SDavid Howells if (!check_mnt(mnt)) 259243f5e655SDavid Howells return -EINVAL; 259343f5e655SDavid Howells 259443f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 259543f5e655SDavid Howells return -EINVAL; 259643f5e655SDavid Howells 259743f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 259843f5e655SDavid Howells return -EPERM; 259943f5e655SDavid Howells 2600e58ace1aSChristian Brauner /* 2601e58ace1aSChristian Brauner * We're only checking whether the superblock is read-only not 2602e58ace1aSChristian Brauner * changing it, so only take down_read(&sb->s_umount). 2603e58ace1aSChristian Brauner */ 2604e58ace1aSChristian Brauner down_read(&sb->s_umount); 260568847c94SChristian Brauner lock_mount_hash(); 260643f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 260743f5e655SDavid Howells if (ret == 0) 260843f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 260968847c94SChristian Brauner unlock_mount_hash(); 2610e58ace1aSChristian Brauner up_read(&sb->s_umount); 2611f8b92ba6SDeepa Dinamani 2612f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2613f8b92ba6SDeepa Dinamani 261443f5e655SDavid Howells return ret; 26152e4b7fcdSDave Hansen } 26162e4b7fcdSDave Hansen 26171da177e4SLinus Torvalds /* 26181da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 26191da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 26201da177e4SLinus Torvalds * on it - tough luck. 26211da177e4SLinus Torvalds */ 2622e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2623e462ec50SDavid Howells int mnt_flags, void *data) 26241da177e4SLinus Torvalds { 26251da177e4SLinus Torvalds int err; 26262d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2627143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 26288d0347f6SDavid Howells struct fs_context *fc; 26291da177e4SLinus Torvalds 2630143c8c91SAl Viro if (!check_mnt(mnt)) 26311da177e4SLinus Torvalds return -EINVAL; 26321da177e4SLinus Torvalds 26332d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 26341da177e4SLinus Torvalds return -EINVAL; 26351da177e4SLinus Torvalds 263643f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 263707b64558SEric W. Biederman return -EPERM; 26389566d674SEric W. Biederman 26398d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 26408d0347f6SDavid Howells if (IS_ERR(fc)) 26418d0347f6SDavid Howells return PTR_ERR(fc); 2642ff36fe2cSEric Paris 2643b330966fSMiklos Szeredi fc->oldapi = true; 26448d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 26458d0347f6SDavid Howells if (!err) { 26461da177e4SLinus Torvalds down_write(&sb->s_umount); 264757eccb83SAl Viro err = -EPERM; 264843f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 26498d0347f6SDavid Howells err = reconfigure_super(fc); 265068847c94SChristian Brauner if (!err) { 265168847c94SChristian Brauner lock_mount_hash(); 265243f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 265368847c94SChristian Brauner unlock_mount_hash(); 265468847c94SChristian Brauner } 26550e55a7ccSDan Williams } 26566339dab8SAl Viro up_write(&sb->s_umount); 26578d0347f6SDavid Howells } 2658f8b92ba6SDeepa Dinamani 2659f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2660f8b92ba6SDeepa Dinamani 26618d0347f6SDavid Howells put_fs_context(fc); 26621da177e4SLinus Torvalds return err; 26631da177e4SLinus Torvalds } 26641da177e4SLinus Torvalds 2665cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 26669676f0c6SRam Pai { 2667315fc83eSAl Viro struct mount *p; 2668909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2669fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 26709676f0c6SRam Pai return 1; 26719676f0c6SRam Pai } 26729676f0c6SRam Pai return 0; 26739676f0c6SRam Pai } 26749676f0c6SRam Pai 267544dfd84aSDavid Howells /* 267644dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 267744dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 267844dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 267944dfd84aSDavid Howells * cycles to be made. 268044dfd84aSDavid Howells */ 268144dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 268244dfd84aSDavid Howells { 268344dfd84aSDavid Howells struct mount *p; 268444dfd84aSDavid Howells bool ret = false; 268544dfd84aSDavid Howells 268644dfd84aSDavid Howells lock_mount_hash(); 268744dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 268844dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 268944dfd84aSDavid Howells goto out; 269044dfd84aSDavid Howells 269144dfd84aSDavid Howells ret = true; 269244dfd84aSDavid Howells out: 269344dfd84aSDavid Howells unlock_mount_hash(); 269444dfd84aSDavid Howells return ret; 269544dfd84aSDavid Howells } 269644dfd84aSDavid Howells 26972db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 26981da177e4SLinus Torvalds { 269944dfd84aSDavid Howells struct mnt_namespace *ns; 2700676da58dSAl Viro struct mount *p; 27010fb54e50SAl Viro struct mount *old; 27022763d119SAl Viro struct mount *parent; 27032763d119SAl Viro struct mountpoint *mp, *old_mp; 270457eccb83SAl Viro int err; 270544dfd84aSDavid Howells bool attached; 27061da177e4SLinus Torvalds 27072db154b3SDavid Howells mp = lock_mount(new_path); 270884d17192SAl Viro if (IS_ERR(mp)) 27092db154b3SDavid Howells return PTR_ERR(mp); 2710cc53ce53SDavid Howells 27112db154b3SDavid Howells old = real_mount(old_path->mnt); 27122db154b3SDavid Howells p = real_mount(new_path->mnt); 27132763d119SAl Viro parent = old->mnt_parent; 271444dfd84aSDavid Howells attached = mnt_has_parent(old); 27152763d119SAl Viro old_mp = old->mnt_mp; 271644dfd84aSDavid Howells ns = old->mnt_ns; 2717143c8c91SAl Viro 27181da177e4SLinus Torvalds err = -EINVAL; 271944dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 272044dfd84aSDavid Howells if (!check_mnt(p)) 27212db154b3SDavid Howells goto out; 27221da177e4SLinus Torvalds 2723570d7a98SEric Biggers /* The thing moved must be mounted... */ 2724570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 272544dfd84aSDavid Howells goto out; 272644dfd84aSDavid Howells 2727570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2728570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 27292db154b3SDavid Howells goto out; 27301da177e4SLinus Torvalds 27312db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 27322db154b3SDavid Howells goto out; 27332db154b3SDavid Howells 27342db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 27352db154b3SDavid Howells goto out; 27362db154b3SDavid Howells 27372db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 27382db154b3SDavid Howells d_is_dir(old_path->dentry)) 27392db154b3SDavid Howells goto out; 274021444403SRam Pai /* 274121444403SRam Pai * Don't move a mount residing in a shared parent. 274221444403SRam Pai */ 27432763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 27442db154b3SDavid Howells goto out; 27459676f0c6SRam Pai /* 27469676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 27479676f0c6SRam Pai * mount which is shared. 27489676f0c6SRam Pai */ 2749fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 27502db154b3SDavid Howells goto out; 27511da177e4SLinus Torvalds err = -ELOOP; 275244dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 275344dfd84aSDavid Howells goto out; 2754fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2755676da58dSAl Viro if (p == old) 27562db154b3SDavid Howells goto out; 27571da177e4SLinus Torvalds 27582db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 27592763d119SAl Viro attached); 27604ac91378SJan Blunck if (err) 27612db154b3SDavid Howells goto out; 27621da177e4SLinus Torvalds 27631da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 27641da177e4SLinus Torvalds * automatically */ 27656776db3dSAl Viro list_del_init(&old->mnt_expire); 27662763d119SAl Viro if (attached) 27672763d119SAl Viro put_mountpoint(old_mp); 27681da177e4SLinus Torvalds out: 27692db154b3SDavid Howells unlock_mount(mp); 277044dfd84aSDavid Howells if (!err) { 27712763d119SAl Viro if (attached) 27722763d119SAl Viro mntput_no_expire(parent); 27732763d119SAl Viro else 277444dfd84aSDavid Howells free_mnt_ns(ns); 277544dfd84aSDavid Howells } 27762db154b3SDavid Howells return err; 27772db154b3SDavid Howells } 27782db154b3SDavid Howells 27792db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 27802db154b3SDavid Howells { 27812db154b3SDavid Howells struct path old_path; 27822db154b3SDavid Howells int err; 27832db154b3SDavid Howells 27842db154b3SDavid Howells if (!old_name || !*old_name) 27852db154b3SDavid Howells return -EINVAL; 27862db154b3SDavid Howells 27872db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 27882db154b3SDavid Howells if (err) 27892db154b3SDavid Howells return err; 27902db154b3SDavid Howells 27912db154b3SDavid Howells err = do_move_mount(&old_path, path); 27922d92ab3cSAl Viro path_put(&old_path); 27931da177e4SLinus Torvalds return err; 27941da177e4SLinus Torvalds } 27951da177e4SLinus Torvalds 27969d412a43SAl Viro /* 27979d412a43SAl Viro * add a mount into a namespace's mount tree 27989d412a43SAl Viro */ 27998f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 28008f11538eSAl Viro struct path *path, int mnt_flags) 28019d412a43SAl Viro { 28028f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 28039d412a43SAl Viro 2804f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 28059d412a43SAl Viro 280684d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2807156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2808156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 28098f11538eSAl Viro return -EINVAL; 2810156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 281184d17192SAl Viro if (!parent->mnt_ns) 28128f11538eSAl Viro return -EINVAL; 2813156cacb1SAl Viro } 28149d412a43SAl Viro 28159d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 281695bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 28179d412a43SAl Viro path->mnt->mnt_root == path->dentry) 28188f11538eSAl Viro return -EBUSY; 28199d412a43SAl Viro 2820e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 28218f11538eSAl Viro return -EINVAL; 28229d412a43SAl Viro 282395bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 28248f11538eSAl Viro return graft_tree(newmnt, parent, mp); 28259d412a43SAl Viro } 2826b1e75df4SAl Viro 2827132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2828132e4608SDavid Howells 2829132e4608SDavid Howells /* 2830132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2831132e4608SDavid Howells * be added to the namespace tree. 2832132e4608SDavid Howells */ 2833132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2834132e4608SDavid Howells unsigned int mnt_flags) 2835132e4608SDavid Howells { 2836132e4608SDavid Howells struct vfsmount *mnt; 28378f11538eSAl Viro struct mountpoint *mp; 2838132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2839132e4608SDavid Howells int error; 2840132e4608SDavid Howells 2841c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2842c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2843c9ce29edSAl Viro error = -EPERM; 2844c9ce29edSAl Viro 2845c9ce29edSAl Viro if (unlikely(error)) { 2846c9ce29edSAl Viro fc_drop_locked(fc); 2847c9ce29edSAl Viro return error; 2848132e4608SDavid Howells } 2849132e4608SDavid Howells 2850132e4608SDavid Howells up_write(&sb->s_umount); 2851132e4608SDavid Howells 2852132e4608SDavid Howells mnt = vfs_create_mount(fc); 2853132e4608SDavid Howells if (IS_ERR(mnt)) 2854132e4608SDavid Howells return PTR_ERR(mnt); 2855132e4608SDavid Howells 2856f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 2857f8b92ba6SDeepa Dinamani 28588f11538eSAl Viro mp = lock_mount(mountpoint); 28598f11538eSAl Viro if (IS_ERR(mp)) { 28608f11538eSAl Viro mntput(mnt); 28618f11538eSAl Viro return PTR_ERR(mp); 28628f11538eSAl Viro } 28638f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 28648f11538eSAl Viro unlock_mount(mp); 28650ecee669SEric Biggers if (error < 0) 28660ecee669SEric Biggers mntput(mnt); 2867f8b92ba6SDeepa Dinamani return error; 2868f8b92ba6SDeepa Dinamani } 2869f8b92ba6SDeepa Dinamani 28701da177e4SLinus Torvalds /* 28711da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 28721da177e4SLinus Torvalds * namespace's tree 28731da177e4SLinus Torvalds */ 2874e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2875808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 28761da177e4SLinus Torvalds { 28770c55cfc4SEric W. Biederman struct file_system_type *type; 2878a0c9a8b8SAl Viro struct fs_context *fc; 2879a0c9a8b8SAl Viro const char *subtype = NULL; 2880a0c9a8b8SAl Viro int err = 0; 28811da177e4SLinus Torvalds 28820c55cfc4SEric W. Biederman if (!fstype) 28831da177e4SLinus Torvalds return -EINVAL; 28841da177e4SLinus Torvalds 28850c55cfc4SEric W. Biederman type = get_fs_type(fstype); 28860c55cfc4SEric W. Biederman if (!type) 28870c55cfc4SEric W. Biederman return -ENODEV; 28880c55cfc4SEric W. Biederman 2889a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 2890a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 2891a0c9a8b8SAl Viro if (subtype) { 2892a0c9a8b8SAl Viro subtype++; 2893a0c9a8b8SAl Viro if (!*subtype) { 28940c55cfc4SEric W. Biederman put_filesystem(type); 2895a0c9a8b8SAl Viro return -EINVAL; 2896a0c9a8b8SAl Viro } 2897a0c9a8b8SAl Viro } 28988654df4eSEric W. Biederman } 28998654df4eSEric W. Biederman 2900a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 2901a0c9a8b8SAl Viro put_filesystem(type); 2902a0c9a8b8SAl Viro if (IS_ERR(fc)) 2903a0c9a8b8SAl Viro return PTR_ERR(fc); 2904a0c9a8b8SAl Viro 29053e1aeb00SDavid Howells if (subtype) 29063e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 29073e1aeb00SDavid Howells subtype, strlen(subtype)); 29083e1aeb00SDavid Howells if (!err && name) 29093e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 2910a0c9a8b8SAl Viro if (!err) 2911a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 2912c3aabf07SAl Viro if (!err && !mount_capable(fc)) 2913c3aabf07SAl Viro err = -EPERM; 2914a0c9a8b8SAl Viro if (!err) 2915a0c9a8b8SAl Viro err = vfs_get_tree(fc); 2916132e4608SDavid Howells if (!err) 2917132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 2918a0c9a8b8SAl Viro 2919a0c9a8b8SAl Viro put_fs_context(fc); 292015f9a3f3SAl Viro return err; 29211da177e4SLinus Torvalds } 29221da177e4SLinus Torvalds 292319a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 292419a167afSAl Viro { 292526df6034SAl Viro struct dentry *dentry = path->dentry; 29268f11538eSAl Viro struct mountpoint *mp; 292725e195aaSAl Viro struct mount *mnt; 292819a167afSAl Viro int err; 292925e195aaSAl Viro 293025e195aaSAl Viro if (!m) 293125e195aaSAl Viro return 0; 293225e195aaSAl Viro if (IS_ERR(m)) 293325e195aaSAl Viro return PTR_ERR(m); 293425e195aaSAl Viro 293525e195aaSAl Viro mnt = real_mount(m); 293619a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 293719a167afSAl Viro * expired before we get a chance to add it 293819a167afSAl Viro */ 29396776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 294019a167afSAl Viro 294119a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 294226df6034SAl Viro m->mnt_root == dentry) { 2943b1e75df4SAl Viro err = -ELOOP; 294426df6034SAl Viro goto discard; 294519a167afSAl Viro } 294619a167afSAl Viro 294726df6034SAl Viro /* 294826df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 294926df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 295026df6034SAl Viro * got", not "try to mount it on top". 295126df6034SAl Viro */ 295226df6034SAl Viro inode_lock(dentry->d_inode); 295326df6034SAl Viro namespace_lock(); 295426df6034SAl Viro if (unlikely(cant_mount(dentry))) { 295526df6034SAl Viro err = -ENOENT; 295626df6034SAl Viro goto discard_locked; 295726df6034SAl Viro } 295826df6034SAl Viro rcu_read_lock(); 295926df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 296026df6034SAl Viro rcu_read_unlock(); 296126df6034SAl Viro err = 0; 296226df6034SAl Viro goto discard_locked; 296326df6034SAl Viro } 296426df6034SAl Viro rcu_read_unlock(); 296526df6034SAl Viro mp = get_mountpoint(dentry); 29668f11538eSAl Viro if (IS_ERR(mp)) { 29678f11538eSAl Viro err = PTR_ERR(mp); 296826df6034SAl Viro goto discard_locked; 29698f11538eSAl Viro } 297026df6034SAl Viro 29718f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 29728f11538eSAl Viro unlock_mount(mp); 297326df6034SAl Viro if (unlikely(err)) 297426df6034SAl Viro goto discard; 297526df6034SAl Viro mntput(m); 2976b1e75df4SAl Viro return 0; 297726df6034SAl Viro 297826df6034SAl Viro discard_locked: 297926df6034SAl Viro namespace_unlock(); 298026df6034SAl Viro inode_unlock(dentry->d_inode); 298126df6034SAl Viro discard: 2982b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 29836776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 298497216be0SAl Viro namespace_lock(); 29856776db3dSAl Viro list_del_init(&mnt->mnt_expire); 298697216be0SAl Viro namespace_unlock(); 298719a167afSAl Viro } 2988b1e75df4SAl Viro mntput(m); 2989b1e75df4SAl Viro mntput(m); 299019a167afSAl Viro return err; 299119a167afSAl Viro } 299219a167afSAl Viro 2993ea5b778aSDavid Howells /** 2994ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2995ea5b778aSDavid Howells * @mnt: The mount to list. 2996ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2997ea5b778aSDavid Howells */ 2998ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2999ea5b778aSDavid Howells { 300097216be0SAl Viro namespace_lock(); 3001ea5b778aSDavid Howells 30026776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 3003ea5b778aSDavid Howells 300497216be0SAl Viro namespace_unlock(); 3005ea5b778aSDavid Howells } 3006ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 3007ea5b778aSDavid Howells 3008ea5b778aSDavid Howells /* 30091da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 30101da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 30111da177e4SLinus Torvalds * here 30121da177e4SLinus Torvalds */ 30131da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 30141da177e4SLinus Torvalds { 3015761d5c38SAl Viro struct mount *mnt, *next; 30161da177e4SLinus Torvalds LIST_HEAD(graveyard); 30171da177e4SLinus Torvalds 30181da177e4SLinus Torvalds if (list_empty(mounts)) 30191da177e4SLinus Torvalds return; 30201da177e4SLinus Torvalds 302197216be0SAl Viro namespace_lock(); 3022719ea2fbSAl Viro lock_mount_hash(); 30231da177e4SLinus Torvalds 30241da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 30251da177e4SLinus Torvalds * following criteria: 30261da177e4SLinus Torvalds * - only referenced by its parent vfsmount 30271da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 30281da177e4SLinus Torvalds * cleared by mntput()) 30291da177e4SLinus Torvalds */ 30306776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 3031863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 30321ab59738SAl Viro propagate_mount_busy(mnt, 1)) 30331da177e4SLinus Torvalds continue; 30346776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 30351da177e4SLinus Torvalds } 3036bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 30376776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 3038143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 3039e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3040bcc5c7d2SAl Viro } 3041719ea2fbSAl Viro unlock_mount_hash(); 30423ab6abeeSAl Viro namespace_unlock(); 30431da177e4SLinus Torvalds } 30441da177e4SLinus Torvalds 30451da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 30461da177e4SLinus Torvalds 30471da177e4SLinus Torvalds /* 30485528f911STrond Myklebust * Ripoff of 'select_parent()' 30495528f911STrond Myklebust * 30505528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 30515528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 30525528f911STrond Myklebust */ 3053692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 30545528f911STrond Myklebust { 3055692afc31SAl Viro struct mount *this_parent = parent; 30565528f911STrond Myklebust struct list_head *next; 30575528f911STrond Myklebust int found = 0; 30585528f911STrond Myklebust 30595528f911STrond Myklebust repeat: 30606b41d536SAl Viro next = this_parent->mnt_mounts.next; 30615528f911STrond Myklebust resume: 30626b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 30635528f911STrond Myklebust struct list_head *tmp = next; 30646b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 30655528f911STrond Myklebust 30665528f911STrond Myklebust next = tmp->next; 3067692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 30685528f911STrond Myklebust continue; 30695528f911STrond Myklebust /* 30705528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 30715528f911STrond Myklebust */ 30726b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 30735528f911STrond Myklebust this_parent = mnt; 30745528f911STrond Myklebust goto repeat; 30755528f911STrond Myklebust } 30765528f911STrond Myklebust 30771ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 30786776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 30795528f911STrond Myklebust found++; 30805528f911STrond Myklebust } 30815528f911STrond Myklebust } 30825528f911STrond Myklebust /* 30835528f911STrond Myklebust * All done at this level ... ascend and resume the search 30845528f911STrond Myklebust */ 30855528f911STrond Myklebust if (this_parent != parent) { 30866b41d536SAl Viro next = this_parent->mnt_child.next; 30870714a533SAl Viro this_parent = this_parent->mnt_parent; 30885528f911STrond Myklebust goto resume; 30895528f911STrond Myklebust } 30905528f911STrond Myklebust return found; 30915528f911STrond Myklebust } 30925528f911STrond Myklebust 30935528f911STrond Myklebust /* 30945528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 30955528f911STrond Myklebust * submounts of a specific parent mountpoint 309699b7db7bSNick Piggin * 309748a066e7SAl Viro * mount_lock must be held for write 30985528f911STrond Myklebust */ 3099b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 31005528f911STrond Myklebust { 31015528f911STrond Myklebust LIST_HEAD(graveyard); 3102761d5c38SAl Viro struct mount *m; 31035528f911STrond Myklebust 31045528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 3105c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3106bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3107761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 31086776db3dSAl Viro mnt_expire); 3109143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3110e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3111bcc5c7d2SAl Viro } 3112bcc5c7d2SAl Viro } 31135528f911STrond Myklebust } 31145528f911STrond Myklebust 3115028abd92SChristoph Hellwig static void *copy_mount_options(const void __user * data) 31161da177e4SLinus Torvalds { 3117b40ef869SAl Viro char *copy; 3118d563d678SCatalin Marinas unsigned left, offset; 31191da177e4SLinus Torvalds 31201da177e4SLinus Torvalds if (!data) 3121b40ef869SAl Viro return NULL; 31221da177e4SLinus Torvalds 3123b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3124b40ef869SAl Viro if (!copy) 3125b40ef869SAl Viro return ERR_PTR(-ENOMEM); 31261da177e4SLinus Torvalds 3127d563d678SCatalin Marinas left = copy_from_user(copy, data, PAGE_SIZE); 31281da177e4SLinus Torvalds 3129d563d678SCatalin Marinas /* 3130d563d678SCatalin Marinas * Not all architectures have an exact copy_from_user(). Resort to 3131d563d678SCatalin Marinas * byte at a time. 3132d563d678SCatalin Marinas */ 3133d563d678SCatalin Marinas offset = PAGE_SIZE - left; 3134d563d678SCatalin Marinas while (left) { 3135d563d678SCatalin Marinas char c; 3136d563d678SCatalin Marinas if (get_user(c, (const char __user *)data + offset)) 3137d563d678SCatalin Marinas break; 3138d563d678SCatalin Marinas copy[offset] = c; 3139d563d678SCatalin Marinas left--; 3140d563d678SCatalin Marinas offset++; 3141d563d678SCatalin Marinas } 3142d563d678SCatalin Marinas 3143d563d678SCatalin Marinas if (left == PAGE_SIZE) { 3144b40ef869SAl Viro kfree(copy); 3145b40ef869SAl Viro return ERR_PTR(-EFAULT); 31461da177e4SLinus Torvalds } 3147d563d678SCatalin Marinas 3148b40ef869SAl Viro return copy; 31491da177e4SLinus Torvalds } 31501da177e4SLinus Torvalds 3151028abd92SChristoph Hellwig static char *copy_mount_string(const void __user *data) 3152eca6f534SVegard Nossum { 3153fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3154eca6f534SVegard Nossum } 3155eca6f534SVegard Nossum 31561da177e4SLinus Torvalds /* 31571da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 31581da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 31591da177e4SLinus Torvalds * 31601da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 31611da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 31621da177e4SLinus Torvalds * information (or be NULL). 31631da177e4SLinus Torvalds * 31641da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 31651da177e4SLinus Torvalds * When the flags word was introduced its top half was required 31661da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 31671da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 31681da177e4SLinus Torvalds * and must be discarded. 31691da177e4SLinus Torvalds */ 3170c60166f0SChristoph Hellwig int path_mount(const char *dev_name, struct path *path, 3171808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 31721da177e4SLinus Torvalds { 3173e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 3174a1e6aaa3SChristoph Hellwig int ret; 31751da177e4SLinus Torvalds 31761da177e4SLinus Torvalds /* Discard magic */ 31771da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 31781da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 31791da177e4SLinus Torvalds 31801da177e4SLinus Torvalds /* Basic sanity checks */ 31811da177e4SLinus Torvalds if (data_page) 31821da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 31831da177e4SLinus Torvalds 3184e462ec50SDavid Howells if (flags & MS_NOUSER) 3185e462ec50SDavid Howells return -EINVAL; 3186e462ec50SDavid Howells 3187a1e6aaa3SChristoph Hellwig ret = security_sb_mount(dev_name, path, type_page, flags, data_page); 3188a1e6aaa3SChristoph Hellwig if (ret) 3189a1e6aaa3SChristoph Hellwig return ret; 3190a1e6aaa3SChristoph Hellwig if (!may_mount()) 3191a1e6aaa3SChristoph Hellwig return -EPERM; 3192*f7e33bdbSJeff Layton if (flags & SB_MANDLOCK) 3193*f7e33bdbSJeff Layton warn_mandlock(); 3194a27ab9f2STetsuo Handa 3195613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3196613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 31970a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 31980a1c01c9SMatthew Garrett 31991da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 32001da177e4SLinus Torvalds if (flags & MS_NOSUID) 32011da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 32021da177e4SLinus Torvalds if (flags & MS_NODEV) 32031da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 32041da177e4SLinus Torvalds if (flags & MS_NOEXEC) 32051da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3206fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3207fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3208fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3209fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3210d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3211d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3212a9e5b732SDavid Howells if (flags & MS_RDONLY) 32132e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3214dab741e0SMattias Nissler if (flags & MS_NOSYMFOLLOW) 3215dab741e0SMattias Nissler mnt_flags |= MNT_NOSYMFOLLOW; 3216fc33a7bbSChristoph Hellwig 3217ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3218ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3219ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3220ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3221ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3222a1e6aaa3SChristoph Hellwig mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; 3223ffbc6f0eSEric W. Biederman } 3224ffbc6f0eSEric W. Biederman 3225e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3226e462ec50SDavid Howells SB_SYNCHRONOUS | 3227e462ec50SDavid Howells SB_MANDLOCK | 3228e462ec50SDavid Howells SB_DIRSYNC | 3229e462ec50SDavid Howells SB_SILENT | 3230917086ffSMimi Zohar SB_POSIXACL | 3231d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3232917086ffSMimi Zohar SB_I_VERSION); 32331da177e4SLinus Torvalds 323443f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 3235a1e6aaa3SChristoph Hellwig return do_reconfigure_mnt(path, mnt_flags); 3236a1e6aaa3SChristoph Hellwig if (flags & MS_REMOUNT) 3237a1e6aaa3SChristoph Hellwig return do_remount(path, flags, sb_flags, mnt_flags, data_page); 3238a1e6aaa3SChristoph Hellwig if (flags & MS_BIND) 3239a1e6aaa3SChristoph Hellwig return do_loopback(path, dev_name, flags & MS_REC); 3240a1e6aaa3SChristoph Hellwig if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 3241a1e6aaa3SChristoph Hellwig return do_change_type(path, flags); 3242a1e6aaa3SChristoph Hellwig if (flags & MS_MOVE) 3243a1e6aaa3SChristoph Hellwig return do_move_mount_old(path, dev_name); 3244a1e6aaa3SChristoph Hellwig 3245a1e6aaa3SChristoph Hellwig return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, 32461da177e4SLinus Torvalds data_page); 3247a1e6aaa3SChristoph Hellwig } 3248a1e6aaa3SChristoph Hellwig 3249a1e6aaa3SChristoph Hellwig long do_mount(const char *dev_name, const char __user *dir_name, 3250a1e6aaa3SChristoph Hellwig const char *type_page, unsigned long flags, void *data_page) 3251a1e6aaa3SChristoph Hellwig { 3252a1e6aaa3SChristoph Hellwig struct path path; 3253a1e6aaa3SChristoph Hellwig int ret; 3254a1e6aaa3SChristoph Hellwig 3255a1e6aaa3SChristoph Hellwig ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3256a1e6aaa3SChristoph Hellwig if (ret) 3257a1e6aaa3SChristoph Hellwig return ret; 3258a1e6aaa3SChristoph Hellwig ret = path_mount(dev_name, &path, type_page, flags, data_page); 32592d92ab3cSAl Viro path_put(&path); 3260a1e6aaa3SChristoph Hellwig return ret; 32611da177e4SLinus Torvalds } 32621da177e4SLinus Torvalds 3263537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3264537f7ccbSEric W. Biederman { 3265537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3266537f7ccbSEric W. Biederman } 3267537f7ccbSEric W. Biederman 3268537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3269537f7ccbSEric W. Biederman { 3270537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3271537f7ccbSEric W. Biederman } 3272537f7ccbSEric W. Biederman 3273771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3274771b1371SEric W. Biederman { 327574e83122SAl Viro if (!is_anon_ns(ns)) 32766344c433SAl Viro ns_free_inum(&ns->ns); 3277537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3278771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3279771b1371SEric W. Biederman kfree(ns); 3280771b1371SEric W. Biederman } 3281771b1371SEric W. Biederman 32828823c079SEric W. Biederman /* 32838823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 32848823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 32858823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 32868823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 32878823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 32888823c079SEric W. Biederman */ 32898823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 32908823c079SEric W. Biederman 329174e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3292cf8d2c11STrond Myklebust { 3293cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3294537f7ccbSEric W. Biederman struct ucounts *ucounts; 329598f842e6SEric W. Biederman int ret; 3296cf8d2c11STrond Myklebust 3297537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3298537f7ccbSEric W. Biederman if (!ucounts) 3299df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3300537f7ccbSEric W. Biederman 330174e83122SAl Viro new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 3302537f7ccbSEric W. Biederman if (!new_ns) { 3303537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3304cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3305537f7ccbSEric W. Biederman } 330674e83122SAl Viro if (!anon) { 33076344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 330898f842e6SEric W. Biederman if (ret) { 330998f842e6SEric W. Biederman kfree(new_ns); 3310537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 331198f842e6SEric W. Biederman return ERR_PTR(ret); 331298f842e6SEric W. Biederman } 331374e83122SAl Viro } 331433c42940SAl Viro new_ns->ns.ops = &mntns_operations; 331574e83122SAl Viro if (!anon) 33168823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 33171a7b8969SKirill Tkhai refcount_set(&new_ns->ns.count, 1); 3318cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3319cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 33209f6c61f9SMiklos Szeredi spin_lock_init(&new_ns->ns_lock); 3321771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3322537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3323cf8d2c11STrond Myklebust return new_ns; 3324cf8d2c11STrond Myklebust } 3325cf8d2c11STrond Myklebust 33260766f788SEmese Revfy __latent_entropy 33279559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 33289559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 33291da177e4SLinus Torvalds { 33306b3286edSKirill Korotaev struct mnt_namespace *new_ns; 33317f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3332315fc83eSAl Viro struct mount *p, *q; 33339559f689SAl Viro struct mount *old; 3334cb338d06SAl Viro struct mount *new; 33357a472ef4SEric W. Biederman int copy_flags; 33361da177e4SLinus Torvalds 33379559f689SAl Viro BUG_ON(!ns); 33389559f689SAl Viro 33399559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 33409559f689SAl Viro get_mnt_ns(ns); 33419559f689SAl Viro return ns; 33429559f689SAl Viro } 33439559f689SAl Viro 33449559f689SAl Viro old = ns->root; 33459559f689SAl Viro 334674e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3347cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3348cf8d2c11STrond Myklebust return new_ns; 33491da177e4SLinus Torvalds 335097216be0SAl Viro namespace_lock(); 33511da177e4SLinus Torvalds /* First pass: copy the tree topology */ 33524ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 33539559f689SAl Viro if (user_ns != ns->user_ns) 33543bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 33557a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3356be34d1a3SDavid Howells if (IS_ERR(new)) { 3357328e6d90SAl Viro namespace_unlock(); 3358771b1371SEric W. Biederman free_mnt_ns(new_ns); 3359be34d1a3SDavid Howells return ERR_CAST(new); 33601da177e4SLinus Torvalds } 33613bd045ccSAl Viro if (user_ns != ns->user_ns) { 33623bd045ccSAl Viro lock_mount_hash(); 33633bd045ccSAl Viro lock_mnt_tree(new); 33643bd045ccSAl Viro unlock_mount_hash(); 33653bd045ccSAl Viro } 3366be08d6d2SAl Viro new_ns->root = new; 33671a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 33681da177e4SLinus Torvalds 33691da177e4SLinus Torvalds /* 33701da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 33711da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 33721da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 33731da177e4SLinus Torvalds */ 3374909b0a88SAl Viro p = old; 3375cb338d06SAl Viro q = new; 33761da177e4SLinus Torvalds while (p) { 3377143c8c91SAl Viro q->mnt_ns = new_ns; 3378d2921684SEric W. Biederman new_ns->mounts++; 33799559f689SAl Viro if (new_fs) { 33809559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 33819559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3382315fc83eSAl Viro rootmnt = &p->mnt; 33831da177e4SLinus Torvalds } 33849559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 33859559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3386315fc83eSAl Viro pwdmnt = &p->mnt; 33871da177e4SLinus Torvalds } 33881da177e4SLinus Torvalds } 3389909b0a88SAl Viro p = next_mnt(p, old); 3390909b0a88SAl Viro q = next_mnt(q, new); 33914ce5d2b1SEric W. Biederman if (!q) 33924ce5d2b1SEric W. Biederman break; 33934ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 33944ce5d2b1SEric W. Biederman p = next_mnt(p, old); 33951da177e4SLinus Torvalds } 3396328e6d90SAl Viro namespace_unlock(); 33971da177e4SLinus Torvalds 33981da177e4SLinus Torvalds if (rootmnt) 3399f03c6599SAl Viro mntput(rootmnt); 34001da177e4SLinus Torvalds if (pwdmnt) 3401f03c6599SAl Viro mntput(pwdmnt); 34021da177e4SLinus Torvalds 3403741a2951SJANAK DESAI return new_ns; 3404741a2951SJANAK DESAI } 3405741a2951SJANAK DESAI 340674e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3407cf8d2c11STrond Myklebust { 34081a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3409ea441d11SAl Viro struct mnt_namespace *ns; 3410d31da0f0SAl Viro struct super_block *s; 3411ea441d11SAl Viro struct path path; 3412ea441d11SAl Viro int err; 3413ea441d11SAl Viro 341474e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 341574e83122SAl Viro if (IS_ERR(ns)) { 341674e83122SAl Viro mntput(m); 3417ea441d11SAl Viro return ERR_CAST(ns); 341874e83122SAl Viro } 341974e83122SAl Viro mnt->mnt_ns = ns; 342074e83122SAl Viro ns->root = mnt; 342174e83122SAl Viro ns->mounts++; 342274e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3423ea441d11SAl Viro 342474e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3425ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3426ea441d11SAl Viro 3427ea441d11SAl Viro put_mnt_ns(ns); 3428ea441d11SAl Viro 3429ea441d11SAl Viro if (err) 3430ea441d11SAl Viro return ERR_PTR(err); 3431ea441d11SAl Viro 3432ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3433d31da0f0SAl Viro s = path.mnt->mnt_sb; 3434d31da0f0SAl Viro atomic_inc(&s->s_active); 3435ea441d11SAl Viro mntput(path.mnt); 3436ea441d11SAl Viro /* lock the sucker */ 3437d31da0f0SAl Viro down_write(&s->s_umount); 3438ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3439ea441d11SAl Viro return path.dentry; 3440ea441d11SAl Viro } 3441ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3442ea441d11SAl Viro 3443cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3444cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 34451da177e4SLinus Torvalds { 3446eca6f534SVegard Nossum int ret; 3447eca6f534SVegard Nossum char *kernel_type; 3448eca6f534SVegard Nossum char *kernel_dev; 3449b40ef869SAl Viro void *options; 34501da177e4SLinus Torvalds 3451b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3452b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3453b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3454eca6f534SVegard Nossum goto out_type; 34551da177e4SLinus Torvalds 3456b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3457b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3458b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3459eca6f534SVegard Nossum goto out_dev; 34601da177e4SLinus Torvalds 3461b40ef869SAl Viro options = copy_mount_options(data); 3462b40ef869SAl Viro ret = PTR_ERR(options); 3463b40ef869SAl Viro if (IS_ERR(options)) 3464eca6f534SVegard Nossum goto out_data; 34651da177e4SLinus Torvalds 3466b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3467eca6f534SVegard Nossum 3468b40ef869SAl Viro kfree(options); 3469eca6f534SVegard Nossum out_data: 3470eca6f534SVegard Nossum kfree(kernel_dev); 3471eca6f534SVegard Nossum out_dev: 3472eca6f534SVegard Nossum kfree(kernel_type); 3473eca6f534SVegard Nossum out_type: 3474eca6f534SVegard Nossum return ret; 34751da177e4SLinus Torvalds } 34761da177e4SLinus Torvalds 34775b490500SChristian Brauner #define FSMOUNT_VALID_FLAGS \ 34785b490500SChristian Brauner (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ 3479dd8b477fSChristian Brauner MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ 3480dd8b477fSChristian Brauner MOUNT_ATTR_NOSYMFOLLOW) 34815b490500SChristian Brauner 34829caccd41SChristian Brauner #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) 34832a186721SChristian Brauner 34842a186721SChristian Brauner #define MOUNT_SETATTR_PROPAGATION_FLAGS \ 34852a186721SChristian Brauner (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) 34862a186721SChristian Brauner 34875b490500SChristian Brauner static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) 34885b490500SChristian Brauner { 34895b490500SChristian Brauner unsigned int mnt_flags = 0; 34905b490500SChristian Brauner 34915b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_RDONLY) 34925b490500SChristian Brauner mnt_flags |= MNT_READONLY; 34935b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOSUID) 34945b490500SChristian Brauner mnt_flags |= MNT_NOSUID; 34955b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODEV) 34965b490500SChristian Brauner mnt_flags |= MNT_NODEV; 34975b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOEXEC) 34985b490500SChristian Brauner mnt_flags |= MNT_NOEXEC; 34995b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODIRATIME) 35005b490500SChristian Brauner mnt_flags |= MNT_NODIRATIME; 3501dd8b477fSChristian Brauner if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) 3502dd8b477fSChristian Brauner mnt_flags |= MNT_NOSYMFOLLOW; 35035b490500SChristian Brauner 35045b490500SChristian Brauner return mnt_flags; 35055b490500SChristian Brauner } 35065b490500SChristian Brauner 35071da177e4SLinus Torvalds /* 350893766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 350993766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 351093766fbdSDavid Howells */ 351193766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 351293766fbdSDavid Howells unsigned int, attr_flags) 351393766fbdSDavid Howells { 351493766fbdSDavid Howells struct mnt_namespace *ns; 351593766fbdSDavid Howells struct fs_context *fc; 351693766fbdSDavid Howells struct file *file; 351793766fbdSDavid Howells struct path newmount; 351893766fbdSDavid Howells struct mount *mnt; 351993766fbdSDavid Howells struct fd f; 352093766fbdSDavid Howells unsigned int mnt_flags = 0; 352193766fbdSDavid Howells long ret; 352293766fbdSDavid Howells 352393766fbdSDavid Howells if (!may_mount()) 352493766fbdSDavid Howells return -EPERM; 352593766fbdSDavid Howells 352693766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 352793766fbdSDavid Howells return -EINVAL; 352893766fbdSDavid Howells 35295b490500SChristian Brauner if (attr_flags & ~FSMOUNT_VALID_FLAGS) 353093766fbdSDavid Howells return -EINVAL; 353193766fbdSDavid Howells 35325b490500SChristian Brauner mnt_flags = attr_flags_to_mnt_flags(attr_flags); 353393766fbdSDavid Howells 353493766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 353593766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 353693766fbdSDavid Howells break; 353793766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 353893766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 353993766fbdSDavid Howells break; 354093766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 354193766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 354293766fbdSDavid Howells break; 354393766fbdSDavid Howells default: 354493766fbdSDavid Howells return -EINVAL; 354593766fbdSDavid Howells } 354693766fbdSDavid Howells 354793766fbdSDavid Howells f = fdget(fs_fd); 354893766fbdSDavid Howells if (!f.file) 354993766fbdSDavid Howells return -EBADF; 355093766fbdSDavid Howells 355193766fbdSDavid Howells ret = -EINVAL; 355293766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 355393766fbdSDavid Howells goto err_fsfd; 355493766fbdSDavid Howells 355593766fbdSDavid Howells fc = f.file->private_data; 355693766fbdSDavid Howells 355793766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 355893766fbdSDavid Howells if (ret < 0) 355993766fbdSDavid Howells goto err_fsfd; 356093766fbdSDavid Howells 356193766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 356293766fbdSDavid Howells ret = -EINVAL; 356393766fbdSDavid Howells if (!fc->root) 356493766fbdSDavid Howells goto err_unlock; 356593766fbdSDavid Howells 356693766fbdSDavid Howells ret = -EPERM; 356793766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 356893766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 356993766fbdSDavid Howells goto err_unlock; 357093766fbdSDavid Howells } 357193766fbdSDavid Howells 357293766fbdSDavid Howells ret = -EBUSY; 357393766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 357493766fbdSDavid Howells goto err_unlock; 357593766fbdSDavid Howells 3576*f7e33bdbSJeff Layton if (fc->sb_flags & SB_MANDLOCK) 3577*f7e33bdbSJeff Layton warn_mandlock(); 357893766fbdSDavid Howells 357993766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 358093766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 358193766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 358293766fbdSDavid Howells goto err_unlock; 358393766fbdSDavid Howells } 358493766fbdSDavid Howells newmount.dentry = dget(fc->root); 358593766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 358693766fbdSDavid Howells 358793766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 358893766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 358993766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 359093766fbdSDavid Howells * don't want to have to handle any errors incurred. 359193766fbdSDavid Howells */ 359293766fbdSDavid Howells vfs_clean_context(fc); 359393766fbdSDavid Howells 359493766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 359593766fbdSDavid Howells if (IS_ERR(ns)) { 359693766fbdSDavid Howells ret = PTR_ERR(ns); 359793766fbdSDavid Howells goto err_path; 359893766fbdSDavid Howells } 359993766fbdSDavid Howells mnt = real_mount(newmount.mnt); 360093766fbdSDavid Howells mnt->mnt_ns = ns; 360193766fbdSDavid Howells ns->root = mnt; 360293766fbdSDavid Howells ns->mounts = 1; 360393766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 36041b0b9cc8SEric Biggers mntget(newmount.mnt); 360593766fbdSDavid Howells 360693766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 360793766fbdSDavid Howells * it, not just simply put it. 360893766fbdSDavid Howells */ 360993766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 361093766fbdSDavid Howells if (IS_ERR(file)) { 361193766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 361293766fbdSDavid Howells ret = PTR_ERR(file); 361393766fbdSDavid Howells goto err_path; 361493766fbdSDavid Howells } 361593766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 361693766fbdSDavid Howells 361793766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 361893766fbdSDavid Howells if (ret >= 0) 361993766fbdSDavid Howells fd_install(ret, file); 362093766fbdSDavid Howells else 362193766fbdSDavid Howells fput(file); 362293766fbdSDavid Howells 362393766fbdSDavid Howells err_path: 362493766fbdSDavid Howells path_put(&newmount); 362593766fbdSDavid Howells err_unlock: 362693766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 362793766fbdSDavid Howells err_fsfd: 362893766fbdSDavid Howells fdput(f); 362993766fbdSDavid Howells return ret; 363093766fbdSDavid Howells } 363193766fbdSDavid Howells 363293766fbdSDavid Howells /* 363393766fbdSDavid Howells * Move a mount from one place to another. In combination with 363493766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 363593766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 363693766fbdSDavid Howells * a mount subtree. 36372db154b3SDavid Howells * 36382db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 36392db154b3SDavid Howells */ 36402db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 36412658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 36422658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 36432db154b3SDavid Howells unsigned int, flags) 36442db154b3SDavid Howells { 36452db154b3SDavid Howells struct path from_path, to_path; 36462db154b3SDavid Howells unsigned int lflags; 36472db154b3SDavid Howells int ret = 0; 36482db154b3SDavid Howells 36492db154b3SDavid Howells if (!may_mount()) 36502db154b3SDavid Howells return -EPERM; 36512db154b3SDavid Howells 36522db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 36532db154b3SDavid Howells return -EINVAL; 36542db154b3SDavid Howells 36552db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 36562db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 36572db154b3SDavid Howells * to clear it afterwards. 36582db154b3SDavid Howells */ 36592db154b3SDavid Howells lflags = 0; 36602db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 36612db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 36622db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 36632db154b3SDavid Howells 36642db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 36652db154b3SDavid Howells if (ret < 0) 36662db154b3SDavid Howells return ret; 36672db154b3SDavid Howells 36682db154b3SDavid Howells lflags = 0; 36692db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 36702db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 36712db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 36722db154b3SDavid Howells 36732db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 36742db154b3SDavid Howells if (ret < 0) 36752db154b3SDavid Howells goto out_from; 36762db154b3SDavid Howells 36772db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 36782db154b3SDavid Howells if (ret < 0) 36792db154b3SDavid Howells goto out_to; 36802db154b3SDavid Howells 36812db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 36822db154b3SDavid Howells 36832db154b3SDavid Howells out_to: 36842db154b3SDavid Howells path_put(&to_path); 36852db154b3SDavid Howells out_from: 36862db154b3SDavid Howells path_put(&from_path); 36872db154b3SDavid Howells return ret; 36882db154b3SDavid Howells } 36892db154b3SDavid Howells 36902db154b3SDavid Howells /* 3691afac7cbaSAl Viro * Return true if path is reachable from root 3692afac7cbaSAl Viro * 369348a066e7SAl Viro * namespace_sem or mount_lock is held 3694afac7cbaSAl Viro */ 3695643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3696afac7cbaSAl Viro const struct path *root) 3697afac7cbaSAl Viro { 3698643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3699a73324daSAl Viro dentry = mnt->mnt_mountpoint; 37000714a533SAl Viro mnt = mnt->mnt_parent; 3701afac7cbaSAl Viro } 3702643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3703afac7cbaSAl Viro } 3704afac7cbaSAl Viro 3705640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3706afac7cbaSAl Viro { 370725ab4c9bSYaowei Bai bool res; 370848a066e7SAl Viro read_seqlock_excl(&mount_lock); 3709643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 371048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3711afac7cbaSAl Viro return res; 3712afac7cbaSAl Viro } 3713afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3714afac7cbaSAl Viro 3715afac7cbaSAl Viro /* 37161da177e4SLinus Torvalds * pivot_root Semantics: 37171da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 37181da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 37191da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 37201da177e4SLinus Torvalds * 37211da177e4SLinus Torvalds * Restrictions: 37221da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 37231da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 37241da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 37251da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 37261da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 37271da177e4SLinus Torvalds * 37284a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 37290c1bc6b8SMauro Carvalho Chehab * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives 37304a0d11faSNeil Brown * in this situation. 37314a0d11faSNeil Brown * 37321da177e4SLinus Torvalds * Notes: 37331da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 37341da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 37351da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 37361da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 37371da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 37381da177e4SLinus Torvalds * first. 37391da177e4SLinus Torvalds */ 37403480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 37413480b257SHeiko Carstens const char __user *, put_old) 37421da177e4SLinus Torvalds { 37432763d119SAl Viro struct path new, old, root; 37442763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 374584d17192SAl Viro struct mountpoint *old_mp, *root_mp; 37461da177e4SLinus Torvalds int error; 37471da177e4SLinus Torvalds 37489b40bc90SAl Viro if (!may_mount()) 37491da177e4SLinus Torvalds return -EPERM; 37501da177e4SLinus Torvalds 3751ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3752ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 37531da177e4SLinus Torvalds if (error) 37541da177e4SLinus Torvalds goto out0; 37551da177e4SLinus Torvalds 3756ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 3757ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 37581da177e4SLinus Torvalds if (error) 37591da177e4SLinus Torvalds goto out1; 37601da177e4SLinus Torvalds 37612d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3762b12cea91SAl Viro if (error) 3763b12cea91SAl Viro goto out2; 37641da177e4SLinus Torvalds 3765f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 376684d17192SAl Viro old_mp = lock_mount(&old); 376784d17192SAl Viro error = PTR_ERR(old_mp); 376884d17192SAl Viro if (IS_ERR(old_mp)) 3769b12cea91SAl Viro goto out3; 3770b12cea91SAl Viro 37711da177e4SLinus Torvalds error = -EINVAL; 3772419148daSAl Viro new_mnt = real_mount(new.mnt); 3773419148daSAl Viro root_mnt = real_mount(root.mnt); 377484d17192SAl Viro old_mnt = real_mount(old.mnt); 37752763d119SAl Viro ex_parent = new_mnt->mnt_parent; 37762763d119SAl Viro root_parent = root_mnt->mnt_parent; 377784d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 37782763d119SAl Viro IS_MNT_SHARED(ex_parent) || 37792763d119SAl Viro IS_MNT_SHARED(root_parent)) 3780b12cea91SAl Viro goto out4; 3781143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3782b12cea91SAl Viro goto out4; 37835ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 37845ff9d8a6SEric W. Biederman goto out4; 37851da177e4SLinus Torvalds error = -ENOENT; 3786f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3787b12cea91SAl Viro goto out4; 37881da177e4SLinus Torvalds error = -EBUSY; 378984d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3790b12cea91SAl Viro goto out4; /* loop, on the same file system */ 37911da177e4SLinus Torvalds error = -EINVAL; 37928c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3793b12cea91SAl Viro goto out4; /* not a mountpoint */ 3794676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3795b12cea91SAl Viro goto out4; /* not attached */ 37962d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3797b12cea91SAl Viro goto out4; /* not a mountpoint */ 3798676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3799b12cea91SAl Viro goto out4; /* not attached */ 38004ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 380184d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3802b12cea91SAl Viro goto out4; 38030d082601SEric W. Biederman /* make certain new is below the root */ 38040d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 38050d082601SEric W. Biederman goto out4; 3806719ea2fbSAl Viro lock_mount_hash(); 38072763d119SAl Viro umount_mnt(new_mnt); 38082763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 38095ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 38105ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 38115ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 38125ff9d8a6SEric W. Biederman } 38134ac91378SJan Blunck /* mount old root on put_old */ 381484d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 38154ac91378SJan Blunck /* mount new_root on / */ 38162763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 38172763d119SAl Viro mnt_add_count(root_parent, -1); 38186b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 38194fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 38204fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 38213895dbf8SEric W. Biederman put_mountpoint(root_mp); 3822719ea2fbSAl Viro unlock_mount_hash(); 38232d8f3038SAl Viro chroot_fs_refs(&root, &new); 38241da177e4SLinus Torvalds error = 0; 3825b12cea91SAl Viro out4: 382684d17192SAl Viro unlock_mount(old_mp); 38272763d119SAl Viro if (!error) 38282763d119SAl Viro mntput_no_expire(ex_parent); 3829b12cea91SAl Viro out3: 38308c3ee42eSAl Viro path_put(&root); 3831b12cea91SAl Viro out2: 38322d8f3038SAl Viro path_put(&old); 38331da177e4SLinus Torvalds out1: 38342d8f3038SAl Viro path_put(&new); 38351da177e4SLinus Torvalds out0: 38361da177e4SLinus Torvalds return error; 38371da177e4SLinus Torvalds } 38381da177e4SLinus Torvalds 38392a186721SChristian Brauner static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) 38402a186721SChristian Brauner { 38412a186721SChristian Brauner unsigned int flags = mnt->mnt.mnt_flags; 38422a186721SChristian Brauner 38432a186721SChristian Brauner /* flags to clear */ 38442a186721SChristian Brauner flags &= ~kattr->attr_clr; 38452a186721SChristian Brauner /* flags to raise */ 38462a186721SChristian Brauner flags |= kattr->attr_set; 38472a186721SChristian Brauner 38482a186721SChristian Brauner return flags; 38492a186721SChristian Brauner } 38502a186721SChristian Brauner 38519caccd41SChristian Brauner static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 38529caccd41SChristian Brauner { 38539caccd41SChristian Brauner struct vfsmount *m = &mnt->mnt; 38549caccd41SChristian Brauner 38559caccd41SChristian Brauner if (!kattr->mnt_userns) 38569caccd41SChristian Brauner return 0; 38579caccd41SChristian Brauner 38589caccd41SChristian Brauner /* 38599caccd41SChristian Brauner * Once a mount has been idmapped we don't allow it to change its 38609caccd41SChristian Brauner * mapping. It makes things simpler and callers can just create 38619caccd41SChristian Brauner * another bind-mount they can idmap if they want to. 38629caccd41SChristian Brauner */ 38639caccd41SChristian Brauner if (mnt_user_ns(m) != &init_user_ns) 38649caccd41SChristian Brauner return -EPERM; 38659caccd41SChristian Brauner 38669caccd41SChristian Brauner /* The underlying filesystem doesn't support idmapped mounts yet. */ 38679caccd41SChristian Brauner if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) 38689caccd41SChristian Brauner return -EINVAL; 38699caccd41SChristian Brauner 38702ca4dcc4SChristian Brauner /* Don't yet support filesystem mountable in user namespaces. */ 38712ca4dcc4SChristian Brauner if (m->mnt_sb->s_user_ns != &init_user_ns) 38722ca4dcc4SChristian Brauner return -EINVAL; 38732ca4dcc4SChristian Brauner 38749caccd41SChristian Brauner /* We're not controlling the superblock. */ 38752ca4dcc4SChristian Brauner if (!capable(CAP_SYS_ADMIN)) 38769caccd41SChristian Brauner return -EPERM; 38779caccd41SChristian Brauner 38789caccd41SChristian Brauner /* Mount has already been visible in the filesystem hierarchy. */ 38799caccd41SChristian Brauner if (!is_anon_ns(mnt->mnt_ns)) 38809caccd41SChristian Brauner return -EINVAL; 38819caccd41SChristian Brauner 38829caccd41SChristian Brauner return 0; 38839caccd41SChristian Brauner } 38849caccd41SChristian Brauner 38852a186721SChristian Brauner static struct mount *mount_setattr_prepare(struct mount_kattr *kattr, 38862a186721SChristian Brauner struct mount *mnt, int *err) 38872a186721SChristian Brauner { 38882a186721SChristian Brauner struct mount *m = mnt, *last = NULL; 38892a186721SChristian Brauner 38902a186721SChristian Brauner if (!is_mounted(&m->mnt)) { 38912a186721SChristian Brauner *err = -EINVAL; 38922a186721SChristian Brauner goto out; 38932a186721SChristian Brauner } 38942a186721SChristian Brauner 38952a186721SChristian Brauner if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) { 38962a186721SChristian Brauner *err = -EINVAL; 38972a186721SChristian Brauner goto out; 38982a186721SChristian Brauner } 38992a186721SChristian Brauner 39002a186721SChristian Brauner do { 39012a186721SChristian Brauner unsigned int flags; 39022a186721SChristian Brauner 39032a186721SChristian Brauner flags = recalc_flags(kattr, m); 39042a186721SChristian Brauner if (!can_change_locked_flags(m, flags)) { 39052a186721SChristian Brauner *err = -EPERM; 39062a186721SChristian Brauner goto out; 39072a186721SChristian Brauner } 39082a186721SChristian Brauner 39099caccd41SChristian Brauner *err = can_idmap_mount(kattr, m); 39109caccd41SChristian Brauner if (*err) 39119caccd41SChristian Brauner goto out; 39129caccd41SChristian Brauner 39132a186721SChristian Brauner last = m; 39142a186721SChristian Brauner 39152a186721SChristian Brauner if ((kattr->attr_set & MNT_READONLY) && 39162a186721SChristian Brauner !(m->mnt.mnt_flags & MNT_READONLY)) { 39172a186721SChristian Brauner *err = mnt_hold_writers(m); 39182a186721SChristian Brauner if (*err) 39192a186721SChristian Brauner goto out; 39202a186721SChristian Brauner } 39212a186721SChristian Brauner } while (kattr->recurse && (m = next_mnt(m, mnt))); 39222a186721SChristian Brauner 39232a186721SChristian Brauner out: 39242a186721SChristian Brauner return last; 39252a186721SChristian Brauner } 39262a186721SChristian Brauner 39279caccd41SChristian Brauner static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 39289caccd41SChristian Brauner { 39299caccd41SChristian Brauner struct user_namespace *mnt_userns; 39309caccd41SChristian Brauner 39319caccd41SChristian Brauner if (!kattr->mnt_userns) 39329caccd41SChristian Brauner return; 39339caccd41SChristian Brauner 39349caccd41SChristian Brauner mnt_userns = get_user_ns(kattr->mnt_userns); 39359caccd41SChristian Brauner /* Pairs with smp_load_acquire() in mnt_user_ns(). */ 39369caccd41SChristian Brauner smp_store_release(&mnt->mnt.mnt_userns, mnt_userns); 39379caccd41SChristian Brauner } 39389caccd41SChristian Brauner 39392a186721SChristian Brauner static void mount_setattr_commit(struct mount_kattr *kattr, 39402a186721SChristian Brauner struct mount *mnt, struct mount *last, 39412a186721SChristian Brauner int err) 39422a186721SChristian Brauner { 39432a186721SChristian Brauner struct mount *m = mnt; 39442a186721SChristian Brauner 39452a186721SChristian Brauner do { 39462a186721SChristian Brauner if (!err) { 39472a186721SChristian Brauner unsigned int flags; 39482a186721SChristian Brauner 39499caccd41SChristian Brauner do_idmap_mount(kattr, m); 39502a186721SChristian Brauner flags = recalc_flags(kattr, m); 39512a186721SChristian Brauner WRITE_ONCE(m->mnt.mnt_flags, flags); 39522a186721SChristian Brauner } 39532a186721SChristian Brauner 39542a186721SChristian Brauner /* 39552a186721SChristian Brauner * We either set MNT_READONLY above so make it visible 39562a186721SChristian Brauner * before ~MNT_WRITE_HOLD or we failed to recursively 39572a186721SChristian Brauner * apply mount options. 39582a186721SChristian Brauner */ 39592a186721SChristian Brauner if ((kattr->attr_set & MNT_READONLY) && 39602a186721SChristian Brauner (m->mnt.mnt_flags & MNT_WRITE_HOLD)) 39612a186721SChristian Brauner mnt_unhold_writers(m); 39622a186721SChristian Brauner 39632a186721SChristian Brauner if (!err && kattr->propagation) 39642a186721SChristian Brauner change_mnt_propagation(m, kattr->propagation); 39652a186721SChristian Brauner 39662a186721SChristian Brauner /* 39672a186721SChristian Brauner * On failure, only cleanup until we found the first mount 39682a186721SChristian Brauner * we failed to handle. 39692a186721SChristian Brauner */ 39702a186721SChristian Brauner if (err && m == last) 39712a186721SChristian Brauner break; 39722a186721SChristian Brauner } while (kattr->recurse && (m = next_mnt(m, mnt))); 39732a186721SChristian Brauner 39742a186721SChristian Brauner if (!err) 39752a186721SChristian Brauner touch_mnt_namespace(mnt->mnt_ns); 39762a186721SChristian Brauner } 39772a186721SChristian Brauner 39782a186721SChristian Brauner static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) 39792a186721SChristian Brauner { 39802a186721SChristian Brauner struct mount *mnt = real_mount(path->mnt), *last = NULL; 39812a186721SChristian Brauner int err = 0; 39822a186721SChristian Brauner 39832a186721SChristian Brauner if (path->dentry != mnt->mnt.mnt_root) 39842a186721SChristian Brauner return -EINVAL; 39852a186721SChristian Brauner 39862a186721SChristian Brauner if (kattr->propagation) { 39872a186721SChristian Brauner /* 39882a186721SChristian Brauner * Only take namespace_lock() if we're actually changing 39892a186721SChristian Brauner * propagation. 39902a186721SChristian Brauner */ 39912a186721SChristian Brauner namespace_lock(); 39922a186721SChristian Brauner if (kattr->propagation == MS_SHARED) { 39932a186721SChristian Brauner err = invent_group_ids(mnt, kattr->recurse); 39942a186721SChristian Brauner if (err) { 39952a186721SChristian Brauner namespace_unlock(); 39962a186721SChristian Brauner return err; 39972a186721SChristian Brauner } 39982a186721SChristian Brauner } 39992a186721SChristian Brauner } 40002a186721SChristian Brauner 40012a186721SChristian Brauner lock_mount_hash(); 40022a186721SChristian Brauner 40032a186721SChristian Brauner /* 40042a186721SChristian Brauner * Get the mount tree in a shape where we can change mount 40052a186721SChristian Brauner * properties without failure. 40062a186721SChristian Brauner */ 40072a186721SChristian Brauner last = mount_setattr_prepare(kattr, mnt, &err); 40082a186721SChristian Brauner if (last) /* Commit all changes or revert to the old state. */ 40092a186721SChristian Brauner mount_setattr_commit(kattr, mnt, last, err); 40102a186721SChristian Brauner 40112a186721SChristian Brauner unlock_mount_hash(); 40122a186721SChristian Brauner 40132a186721SChristian Brauner if (kattr->propagation) { 40142a186721SChristian Brauner namespace_unlock(); 40152a186721SChristian Brauner if (err) 40162a186721SChristian Brauner cleanup_group_ids(mnt, NULL); 40172a186721SChristian Brauner } 40182a186721SChristian Brauner 40192a186721SChristian Brauner return err; 40202a186721SChristian Brauner } 40212a186721SChristian Brauner 40229caccd41SChristian Brauner static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, 40239caccd41SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 40249caccd41SChristian Brauner { 40259caccd41SChristian Brauner int err = 0; 40269caccd41SChristian Brauner struct ns_common *ns; 40279caccd41SChristian Brauner struct user_namespace *mnt_userns; 40289caccd41SChristian Brauner struct file *file; 40299caccd41SChristian Brauner 40309caccd41SChristian Brauner if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) 40319caccd41SChristian Brauner return 0; 40329caccd41SChristian Brauner 40339caccd41SChristian Brauner /* 40349caccd41SChristian Brauner * We currently do not support clearing an idmapped mount. If this ever 40359caccd41SChristian Brauner * is a use-case we can revisit this but for now let's keep it simple 40369caccd41SChristian Brauner * and not allow it. 40379caccd41SChristian Brauner */ 40389caccd41SChristian Brauner if (attr->attr_clr & MOUNT_ATTR_IDMAP) 40399caccd41SChristian Brauner return -EINVAL; 40409caccd41SChristian Brauner 40419caccd41SChristian Brauner if (attr->userns_fd > INT_MAX) 40429caccd41SChristian Brauner return -EINVAL; 40439caccd41SChristian Brauner 40449caccd41SChristian Brauner file = fget(attr->userns_fd); 40459caccd41SChristian Brauner if (!file) 40469caccd41SChristian Brauner return -EBADF; 40479caccd41SChristian Brauner 40489caccd41SChristian Brauner if (!proc_ns_file(file)) { 40499caccd41SChristian Brauner err = -EINVAL; 40509caccd41SChristian Brauner goto out_fput; 40519caccd41SChristian Brauner } 40529caccd41SChristian Brauner 40539caccd41SChristian Brauner ns = get_proc_ns(file_inode(file)); 40549caccd41SChristian Brauner if (ns->ops->type != CLONE_NEWUSER) { 40559caccd41SChristian Brauner err = -EINVAL; 40569caccd41SChristian Brauner goto out_fput; 40579caccd41SChristian Brauner } 40589caccd41SChristian Brauner 40599caccd41SChristian Brauner /* 40609caccd41SChristian Brauner * The init_user_ns is used to indicate that a vfsmount is not idmapped. 40619caccd41SChristian Brauner * This is simpler than just having to treat NULL as unmapped. Users 40629caccd41SChristian Brauner * wanting to idmap a mount to init_user_ns can just use a namespace 40639caccd41SChristian Brauner * with an identity mapping. 40649caccd41SChristian Brauner */ 40659caccd41SChristian Brauner mnt_userns = container_of(ns, struct user_namespace, ns); 40669caccd41SChristian Brauner if (mnt_userns == &init_user_ns) { 40679caccd41SChristian Brauner err = -EPERM; 40689caccd41SChristian Brauner goto out_fput; 40699caccd41SChristian Brauner } 40709caccd41SChristian Brauner kattr->mnt_userns = get_user_ns(mnt_userns); 40719caccd41SChristian Brauner 40729caccd41SChristian Brauner out_fput: 40739caccd41SChristian Brauner fput(file); 40749caccd41SChristian Brauner return err; 40759caccd41SChristian Brauner } 40769caccd41SChristian Brauner 40779caccd41SChristian Brauner static int build_mount_kattr(const struct mount_attr *attr, size_t usize, 40782a186721SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 40792a186721SChristian Brauner { 40802a186721SChristian Brauner unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 40812a186721SChristian Brauner 40822a186721SChristian Brauner if (flags & AT_NO_AUTOMOUNT) 40832a186721SChristian Brauner lookup_flags &= ~LOOKUP_AUTOMOUNT; 40842a186721SChristian Brauner if (flags & AT_SYMLINK_NOFOLLOW) 40852a186721SChristian Brauner lookup_flags &= ~LOOKUP_FOLLOW; 40862a186721SChristian Brauner if (flags & AT_EMPTY_PATH) 40872a186721SChristian Brauner lookup_flags |= LOOKUP_EMPTY; 40882a186721SChristian Brauner 40892a186721SChristian Brauner *kattr = (struct mount_kattr) { 40902a186721SChristian Brauner .lookup_flags = lookup_flags, 40912a186721SChristian Brauner .recurse = !!(flags & AT_RECURSIVE), 40922a186721SChristian Brauner }; 40932a186721SChristian Brauner 40942a186721SChristian Brauner if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) 40952a186721SChristian Brauner return -EINVAL; 40962a186721SChristian Brauner if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) 40972a186721SChristian Brauner return -EINVAL; 40982a186721SChristian Brauner kattr->propagation = attr->propagation; 40992a186721SChristian Brauner 41002a186721SChristian Brauner if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) 41012a186721SChristian Brauner return -EINVAL; 41022a186721SChristian Brauner 41032a186721SChristian Brauner kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); 41042a186721SChristian Brauner kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); 41052a186721SChristian Brauner 41062a186721SChristian Brauner /* 41072a186721SChristian Brauner * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, 41082a186721SChristian Brauner * users wanting to transition to a different atime setting cannot 41092a186721SChristian Brauner * simply specify the atime setting in @attr_set, but must also 41102a186721SChristian Brauner * specify MOUNT_ATTR__ATIME in the @attr_clr field. 41112a186721SChristian Brauner * So ensure that MOUNT_ATTR__ATIME can't be partially set in 41122a186721SChristian Brauner * @attr_clr and that @attr_set can't have any atime bits set if 41132a186721SChristian Brauner * MOUNT_ATTR__ATIME isn't set in @attr_clr. 41142a186721SChristian Brauner */ 41152a186721SChristian Brauner if (attr->attr_clr & MOUNT_ATTR__ATIME) { 41162a186721SChristian Brauner if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) 41172a186721SChristian Brauner return -EINVAL; 41182a186721SChristian Brauner 41192a186721SChristian Brauner /* 41202a186721SChristian Brauner * Clear all previous time settings as they are mutually 41212a186721SChristian Brauner * exclusive. 41222a186721SChristian Brauner */ 41232a186721SChristian Brauner kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; 41242a186721SChristian Brauner switch (attr->attr_set & MOUNT_ATTR__ATIME) { 41252a186721SChristian Brauner case MOUNT_ATTR_RELATIME: 41262a186721SChristian Brauner kattr->attr_set |= MNT_RELATIME; 41272a186721SChristian Brauner break; 41282a186721SChristian Brauner case MOUNT_ATTR_NOATIME: 41292a186721SChristian Brauner kattr->attr_set |= MNT_NOATIME; 41302a186721SChristian Brauner break; 41312a186721SChristian Brauner case MOUNT_ATTR_STRICTATIME: 41322a186721SChristian Brauner break; 41332a186721SChristian Brauner default: 41342a186721SChristian Brauner return -EINVAL; 41352a186721SChristian Brauner } 41362a186721SChristian Brauner } else { 41372a186721SChristian Brauner if (attr->attr_set & MOUNT_ATTR__ATIME) 41382a186721SChristian Brauner return -EINVAL; 41392a186721SChristian Brauner } 41402a186721SChristian Brauner 41419caccd41SChristian Brauner return build_mount_idmapped(attr, usize, kattr, flags); 41429caccd41SChristian Brauner } 41439caccd41SChristian Brauner 41449caccd41SChristian Brauner static void finish_mount_kattr(struct mount_kattr *kattr) 41459caccd41SChristian Brauner { 41469caccd41SChristian Brauner put_user_ns(kattr->mnt_userns); 41479caccd41SChristian Brauner kattr->mnt_userns = NULL; 41482a186721SChristian Brauner } 41492a186721SChristian Brauner 41502a186721SChristian Brauner SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, 41512a186721SChristian Brauner unsigned int, flags, struct mount_attr __user *, uattr, 41522a186721SChristian Brauner size_t, usize) 41532a186721SChristian Brauner { 41542a186721SChristian Brauner int err; 41552a186721SChristian Brauner struct path target; 41562a186721SChristian Brauner struct mount_attr attr; 41572a186721SChristian Brauner struct mount_kattr kattr; 41582a186721SChristian Brauner 41592a186721SChristian Brauner BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); 41602a186721SChristian Brauner 41612a186721SChristian Brauner if (flags & ~(AT_EMPTY_PATH | 41622a186721SChristian Brauner AT_RECURSIVE | 41632a186721SChristian Brauner AT_SYMLINK_NOFOLLOW | 41642a186721SChristian Brauner AT_NO_AUTOMOUNT)) 41652a186721SChristian Brauner return -EINVAL; 41662a186721SChristian Brauner 41672a186721SChristian Brauner if (unlikely(usize > PAGE_SIZE)) 41682a186721SChristian Brauner return -E2BIG; 41692a186721SChristian Brauner if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) 41702a186721SChristian Brauner return -EINVAL; 41712a186721SChristian Brauner 41722a186721SChristian Brauner if (!may_mount()) 41732a186721SChristian Brauner return -EPERM; 41742a186721SChristian Brauner 41752a186721SChristian Brauner err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); 41762a186721SChristian Brauner if (err) 41772a186721SChristian Brauner return err; 41782a186721SChristian Brauner 41792a186721SChristian Brauner /* Don't bother walking through the mounts if this is a nop. */ 41802a186721SChristian Brauner if (attr.attr_set == 0 && 41812a186721SChristian Brauner attr.attr_clr == 0 && 41822a186721SChristian Brauner attr.propagation == 0) 41832a186721SChristian Brauner return 0; 41842a186721SChristian Brauner 41859caccd41SChristian Brauner err = build_mount_kattr(&attr, usize, &kattr, flags); 41862a186721SChristian Brauner if (err) 41872a186721SChristian Brauner return err; 41882a186721SChristian Brauner 41892a186721SChristian Brauner err = user_path_at(dfd, path, kattr.lookup_flags, &target); 41902a186721SChristian Brauner if (err) 41912a186721SChristian Brauner return err; 41922a186721SChristian Brauner 41932a186721SChristian Brauner err = do_mount_setattr(&target, &kattr); 41949caccd41SChristian Brauner finish_mount_kattr(&kattr); 41952a186721SChristian Brauner path_put(&target); 41962a186721SChristian Brauner return err; 41972a186721SChristian Brauner } 41982a186721SChristian Brauner 41991da177e4SLinus Torvalds static void __init init_mount_tree(void) 42001da177e4SLinus Torvalds { 42011da177e4SLinus Torvalds struct vfsmount *mnt; 420274e83122SAl Viro struct mount *m; 42036b3286edSKirill Korotaev struct mnt_namespace *ns; 4204ac748a09SJan Blunck struct path root; 42051da177e4SLinus Torvalds 4206fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 42071da177e4SLinus Torvalds if (IS_ERR(mnt)) 42081da177e4SLinus Torvalds panic("Can't create rootfs"); 4209b3e19d92SNick Piggin 421074e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 42113b22edc5STrond Myklebust if (IS_ERR(ns)) 42121da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 421374e83122SAl Viro m = real_mount(mnt); 421474e83122SAl Viro m->mnt_ns = ns; 421574e83122SAl Viro ns->root = m; 421674e83122SAl Viro ns->mounts = 1; 421774e83122SAl Viro list_add(&m->mnt_list, &ns->list); 42186b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 42196b3286edSKirill Korotaev get_mnt_ns(ns); 42201da177e4SLinus Torvalds 4221be08d6d2SAl Viro root.mnt = mnt; 4222be08d6d2SAl Viro root.dentry = mnt->mnt_root; 4223da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 4224ac748a09SJan Blunck 4225ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 4226ac748a09SJan Blunck set_fs_root(current->fs, &root); 42271da177e4SLinus Torvalds } 42281da177e4SLinus Torvalds 422974bf17cfSDenis Cheng void __init mnt_init(void) 42301da177e4SLinus Torvalds { 423115a67dd8SRandy Dunlap int err; 42321da177e4SLinus Torvalds 42337d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 423420c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 42351da177e4SLinus Torvalds 42360818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 423738129a13SAl Viro sizeof(struct hlist_head), 42380818bf27SAl Viro mhash_entries, 19, 42393d375d78SPavel Tatashin HASH_ZERO, 42400818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 42410818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 42420818bf27SAl Viro sizeof(struct hlist_head), 42430818bf27SAl Viro mphash_entries, 19, 42443d375d78SPavel Tatashin HASH_ZERO, 42450818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 42461da177e4SLinus Torvalds 424784d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 42481da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 42491da177e4SLinus Torvalds 42504b93dc9bSTejun Heo kernfs_init(); 42514b93dc9bSTejun Heo 425215a67dd8SRandy Dunlap err = sysfs_init(); 425315a67dd8SRandy Dunlap if (err) 425415a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 42558e24eea7SHarvey Harrison __func__, err); 425600d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 425700d26666SGreg Kroah-Hartman if (!fs_kobj) 42588e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 4259037f11b4SAl Viro shmem_init(); 42601da177e4SLinus Torvalds init_rootfs(); 42611da177e4SLinus Torvalds init_mount_tree(); 42621da177e4SLinus Torvalds } 42631da177e4SLinus Torvalds 4264616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 42651da177e4SLinus Torvalds { 42661a7b8969SKirill Tkhai if (!refcount_dec_and_test(&ns->ns.count)) 4267616511d0STrond Myklebust return; 42687b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 4269771b1371SEric W. Biederman free_mnt_ns(ns); 42701da177e4SLinus Torvalds } 42719d412a43SAl Viro 4272d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 42739d412a43SAl Viro { 4274423e0ab0STim Chen struct vfsmount *mnt; 4275d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 4276423e0ab0STim Chen if (!IS_ERR(mnt)) { 4277423e0ab0STim Chen /* 4278423e0ab0STim Chen * it is a longterm mount, don't release mnt until 4279423e0ab0STim Chen * we unmount before file sys is unregistered 4280423e0ab0STim Chen */ 4281f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 4282423e0ab0STim Chen } 4283423e0ab0STim Chen return mnt; 42849d412a43SAl Viro } 4285d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 4286423e0ab0STim Chen 4287423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 4288423e0ab0STim Chen { 4289423e0ab0STim Chen /* release long term mount so mount point can be released */ 4290423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 4291f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 429248a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 4293423e0ab0STim Chen mntput(mnt); 4294423e0ab0STim Chen } 4295423e0ab0STim Chen } 4296423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 429702125a82SAl Viro 4298df820f8dSMiklos Szeredi void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) 4299df820f8dSMiklos Szeredi { 4300df820f8dSMiklos Szeredi unsigned int i; 4301df820f8dSMiklos Szeredi 4302df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4303df820f8dSMiklos Szeredi if (mnt[i]) 4304df820f8dSMiklos Szeredi real_mount(mnt[i])->mnt_ns = NULL; 4305df820f8dSMiklos Szeredi synchronize_rcu_expedited(); 4306df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4307df820f8dSMiklos Szeredi mntput(mnt[i]); 4308df820f8dSMiklos Szeredi } 4309df820f8dSMiklos Szeredi EXPORT_SYMBOL(kern_unmount_array); 4310df820f8dSMiklos Szeredi 431102125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 431202125a82SAl Viro { 4313143c8c91SAl Viro return check_mnt(real_mount(mnt)); 431402125a82SAl Viro } 43158823c079SEric W. Biederman 43163151527eSEric W. Biederman bool current_chrooted(void) 43173151527eSEric W. Biederman { 43183151527eSEric W. Biederman /* Does the current process have a non-standard root */ 43193151527eSEric W. Biederman struct path ns_root; 43203151527eSEric W. Biederman struct path fs_root; 43213151527eSEric W. Biederman bool chrooted; 43223151527eSEric W. Biederman 43233151527eSEric W. Biederman /* Find the namespace root */ 43243151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 43253151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 43263151527eSEric W. Biederman path_get(&ns_root); 43273151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 43283151527eSEric W. Biederman ; 43293151527eSEric W. Biederman 43303151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 43313151527eSEric W. Biederman 43323151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 43333151527eSEric W. Biederman 43343151527eSEric W. Biederman path_put(&fs_root); 43353151527eSEric W. Biederman path_put(&ns_root); 43363151527eSEric W. Biederman 43373151527eSEric W. Biederman return chrooted; 43383151527eSEric W. Biederman } 43393151527eSEric W. Biederman 4340132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 4341132e4608SDavid Howells const struct super_block *sb, 43428654df4eSEric W. Biederman int *new_mnt_flags) 434387a8ebd6SEric W. Biederman { 43448c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 434587a8ebd6SEric W. Biederman struct mount *mnt; 4346e51db735SEric W. Biederman bool visible = false; 434787a8ebd6SEric W. Biederman 434844bb4385SAl Viro down_read(&namespace_sem); 43499f6c61f9SMiklos Szeredi lock_ns_list(ns); 435087a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 4351e51db735SEric W. Biederman struct mount *child; 435277b1a97dSEric W. Biederman int mnt_flags; 435377b1a97dSEric W. Biederman 43549f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 43559f6c61f9SMiklos Szeredi continue; 43569f6c61f9SMiklos Szeredi 4357132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 4358e51db735SEric W. Biederman continue; 4359e51db735SEric W. Biederman 43607e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 43617e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 43627e96c1b0SEric W. Biederman */ 43637e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 43647e96c1b0SEric W. Biederman continue; 43657e96c1b0SEric W. Biederman 4366a1935c17SEric W. Biederman /* A local view of the mount flags */ 436777b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 436877b1a97dSEric W. Biederman 4369695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 4370bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 4371695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 4372695e9df0SEric W. Biederman 43738c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 43748c6cf9ccSEric W. Biederman * than the proposed new mount. 43758c6cf9ccSEric W. Biederman */ 437677b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 43778c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 43788c6cf9ccSEric W. Biederman continue; 437977b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 438077b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 43818c6cf9ccSEric W. Biederman continue; 43828c6cf9ccSEric W. Biederman 4383ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 4384ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 4385ceeb0e5dSEric W. Biederman * empty directories. 4386e51db735SEric W. Biederman */ 4387e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 4388e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 4389ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 4390d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 4391ceeb0e5dSEric W. Biederman continue; 43927236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 43937236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 4394e51db735SEric W. Biederman goto next; 439587a8ebd6SEric W. Biederman } 43968c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 439777b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 43988c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 4399e51db735SEric W. Biederman visible = true; 4400e51db735SEric W. Biederman goto found; 4401e51db735SEric W. Biederman next: ; 440287a8ebd6SEric W. Biederman } 4403e51db735SEric W. Biederman found: 44049f6c61f9SMiklos Szeredi unlock_ns_list(ns); 440544bb4385SAl Viro up_read(&namespace_sem); 4406e51db735SEric W. Biederman return visible; 440787a8ebd6SEric W. Biederman } 440887a8ebd6SEric W. Biederman 4409132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 44108654df4eSEric W. Biederman { 4411a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 44128654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 44138654df4eSEric W. Biederman unsigned long s_iflags; 44148654df4eSEric W. Biederman 44158654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 44168654df4eSEric W. Biederman return false; 44178654df4eSEric W. Biederman 44188654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 4419132e4608SDavid Howells s_iflags = sb->s_iflags; 44208654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 44218654df4eSEric W. Biederman return false; 44228654df4eSEric W. Biederman 4423a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 4424a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 4425a1935c17SEric W. Biederman required_iflags); 4426a1935c17SEric W. Biederman return true; 4427a1935c17SEric W. Biederman } 4428a1935c17SEric W. Biederman 4429132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 44308654df4eSEric W. Biederman } 44318654df4eSEric W. Biederman 4432380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 4433380cf5baSAndy Lutomirski { 4434380cf5baSAndy Lutomirski /* 4435380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 4436380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 4437380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 4438380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 4439380cf5baSAndy Lutomirski * in other namespaces. 4440380cf5baSAndy Lutomirski */ 4441380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 4442380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 4443380cf5baSAndy Lutomirski } 4444380cf5baSAndy Lutomirski 444564964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 44468823c079SEric W. Biederman { 444758be2825SAl Viro struct ns_common *ns = NULL; 44488823c079SEric W. Biederman struct nsproxy *nsproxy; 44498823c079SEric W. Biederman 4450728dba3aSEric W. Biederman task_lock(task); 4451728dba3aSEric W. Biederman nsproxy = task->nsproxy; 44528823c079SEric W. Biederman if (nsproxy) { 445358be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 445458be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 44558823c079SEric W. Biederman } 4456728dba3aSEric W. Biederman task_unlock(task); 44578823c079SEric W. Biederman 44588823c079SEric W. Biederman return ns; 44598823c079SEric W. Biederman } 44608823c079SEric W. Biederman 446164964528SAl Viro static void mntns_put(struct ns_common *ns) 44628823c079SEric W. Biederman { 446358be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 44648823c079SEric W. Biederman } 44658823c079SEric W. Biederman 4466f2a8d52eSChristian Brauner static int mntns_install(struct nsset *nsset, struct ns_common *ns) 44678823c079SEric W. Biederman { 4468f2a8d52eSChristian Brauner struct nsproxy *nsproxy = nsset->nsproxy; 4469f2a8d52eSChristian Brauner struct fs_struct *fs = nsset->fs; 44704f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 4471f2a8d52eSChristian Brauner struct user_namespace *user_ns = nsset->cred->user_ns; 44728823c079SEric W. Biederman struct path root; 44734f757f3cSAl Viro int err; 44748823c079SEric W. Biederman 44750c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 4476f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_CHROOT) || 4477f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_ADMIN)) 4478ae11e0f1SZhao Hongjiang return -EPERM; 44798823c079SEric W. Biederman 448074e83122SAl Viro if (is_anon_ns(mnt_ns)) 448174e83122SAl Viro return -EINVAL; 448274e83122SAl Viro 44838823c079SEric W. Biederman if (fs->users != 1) 44848823c079SEric W. Biederman return -EINVAL; 44858823c079SEric W. Biederman 44868823c079SEric W. Biederman get_mnt_ns(mnt_ns); 44874f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 44888823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 44898823c079SEric W. Biederman 44908823c079SEric W. Biederman /* Find the root */ 44914f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 44924f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 44934f757f3cSAl Viro if (err) { 44944f757f3cSAl Viro /* revert to old namespace */ 44954f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 44964f757f3cSAl Viro put_mnt_ns(mnt_ns); 44974f757f3cSAl Viro return err; 44984f757f3cSAl Viro } 44998823c079SEric W. Biederman 45004068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 45014068367cSAndrei Vagin 45028823c079SEric W. Biederman /* Update the pwd and root */ 45038823c079SEric W. Biederman set_fs_pwd(fs, &root); 45048823c079SEric W. Biederman set_fs_root(fs, &root); 45058823c079SEric W. Biederman 45068823c079SEric W. Biederman path_put(&root); 45078823c079SEric W. Biederman return 0; 45088823c079SEric W. Biederman } 45098823c079SEric W. Biederman 4510bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 4511bcac25a5SAndrey Vagin { 4512bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 4513bcac25a5SAndrey Vagin } 4514bcac25a5SAndrey Vagin 45158823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 45168823c079SEric W. Biederman .name = "mnt", 45178823c079SEric W. Biederman .type = CLONE_NEWNS, 45188823c079SEric W. Biederman .get = mntns_get, 45198823c079SEric W. Biederman .put = mntns_put, 45208823c079SEric W. Biederman .install = mntns_install, 4521bcac25a5SAndrey Vagin .owner = mntns_owner, 45228823c079SEric W. Biederman }; 4523