11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 185b825c3aSIngo Molnar #include <linux/cred.h> 1973cd49ecSMiklos Szeredi #include <linux/idr.h> 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23d10577a8SAl Viro #include <linux/uaccess.h> 240bb80f24SDavid Howells #include <linux/proc_ns.h> 2520b4fb48SLinus Torvalds #include <linux/magic.h> 2657c8a661SMike Rapoport #include <linux/memblock.h> 279ea459e1SAl Viro #include <linux/task_work.h> 289164bb4aSIngo Molnar #include <linux/sched/task.h> 29e262e32dSDavid Howells #include <uapi/linux/mount.h> 309164bb4aSIngo Molnar 3107b20889SRam Pai #include "pnode.h" 32948730b0SAdrian Bunk #include "internal.h" 331da177e4SLinus Torvalds 34d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 35d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 36d2921684SEric W. Biederman 370818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 380818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 390818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 400818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 410818bf27SAl Viro 420818bf27SAl Viro static __initdata unsigned long mhash_entries; 430818bf27SAl Viro static int __init set_mhash_entries(char *str) 440818bf27SAl Viro { 450818bf27SAl Viro if (!str) 460818bf27SAl Viro return 0; 470818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 480818bf27SAl Viro return 1; 490818bf27SAl Viro } 500818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 510818bf27SAl Viro 520818bf27SAl Viro static __initdata unsigned long mphash_entries; 530818bf27SAl Viro static int __init set_mphash_entries(char *str) 540818bf27SAl Viro { 550818bf27SAl Viro if (!str) 560818bf27SAl Viro return 0; 570818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 580818bf27SAl Viro return 1; 590818bf27SAl Viro } 600818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6113f14b4dSEric Dumazet 62c7999c36SAl Viro static u64 event; 6373cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 64719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 655addc5ddSAl Viro 6638129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 670818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 68e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 6959aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 701da177e4SLinus Torvalds 71f87fd4c2SMiklos Szeredi /* /sys/fs */ 7200d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7300d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 74f87fd4c2SMiklos Szeredi 7599b7db7bSNick Piggin /* 7699b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7799b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 7899b7db7bSNick Piggin * up the tree. 7999b7db7bSNick Piggin * 8099b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8199b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8299b7db7bSNick Piggin */ 8348a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8499b7db7bSNick Piggin 8538129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 861da177e4SLinus Torvalds { 871da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 881da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 890818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 900818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 910818bf27SAl Viro } 920818bf27SAl Viro 930818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 940818bf27SAl Viro { 950818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 960818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 970818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 981da177e4SLinus Torvalds } 991da177e4SLinus Torvalds 100b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10173cd49ecSMiklos Szeredi { 102169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 10373cd49ecSMiklos Szeredi 104169b480eSMatthew Wilcox if (res < 0) 10573cd49ecSMiklos Szeredi return res; 106169b480eSMatthew Wilcox mnt->mnt_id = res; 107169b480eSMatthew Wilcox return 0; 10873cd49ecSMiklos Szeredi } 10973cd49ecSMiklos Szeredi 110b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11173cd49ecSMiklos Szeredi { 112169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 11373cd49ecSMiklos Szeredi } 11473cd49ecSMiklos Szeredi 115719f5d7fSMiklos Szeredi /* 116719f5d7fSMiklos Szeredi * Allocate a new peer group ID 117719f5d7fSMiklos Szeredi */ 1184b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 119719f5d7fSMiklos Szeredi { 120169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 121f21f6220SAl Viro 122169b480eSMatthew Wilcox if (res < 0) 123f21f6220SAl Viro return res; 124169b480eSMatthew Wilcox mnt->mnt_group_id = res; 125169b480eSMatthew Wilcox return 0; 126719f5d7fSMiklos Szeredi } 127719f5d7fSMiklos Szeredi 128719f5d7fSMiklos Szeredi /* 129719f5d7fSMiklos Szeredi * Release a peer group ID 130719f5d7fSMiklos Szeredi */ 1314b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 132719f5d7fSMiklos Szeredi { 133169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 13415169fe7SAl Viro mnt->mnt_group_id = 0; 135719f5d7fSMiklos Szeredi } 136719f5d7fSMiklos Szeredi 137b3e19d92SNick Piggin /* 138b3e19d92SNick Piggin * vfsmount lock must be held for read 139b3e19d92SNick Piggin */ 14083adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 141b3e19d92SNick Piggin { 142b3e19d92SNick Piggin #ifdef CONFIG_SMP 14368e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 144b3e19d92SNick Piggin #else 145b3e19d92SNick Piggin preempt_disable(); 14668e8a9feSAl Viro mnt->mnt_count += n; 147b3e19d92SNick Piggin preempt_enable(); 148b3e19d92SNick Piggin #endif 149b3e19d92SNick Piggin } 150b3e19d92SNick Piggin 151b3e19d92SNick Piggin /* 152b3e19d92SNick Piggin * vfsmount lock must be held for write 153b3e19d92SNick Piggin */ 15483adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 155b3e19d92SNick Piggin { 156b3e19d92SNick Piggin #ifdef CONFIG_SMP 157f03c6599SAl Viro unsigned int count = 0; 158b3e19d92SNick Piggin int cpu; 159b3e19d92SNick Piggin 160b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 16168e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 162b3e19d92SNick Piggin } 163b3e19d92SNick Piggin 164b3e19d92SNick Piggin return count; 165b3e19d92SNick Piggin #else 16668e8a9feSAl Viro return mnt->mnt_count; 167b3e19d92SNick Piggin #endif 168b3e19d92SNick Piggin } 169b3e19d92SNick Piggin 17087b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 17187b95ce0SAl Viro { 17287b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 17387b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 17487b95ce0SAl Viro pin_remove(p); 17587b95ce0SAl Viro mntput(&m->mnt); 17687b95ce0SAl Viro } 17787b95ce0SAl Viro 178b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1791da177e4SLinus Torvalds { 180c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 181c63181e6SAl Viro if (mnt) { 18273cd49ecSMiklos Szeredi int err; 18373cd49ecSMiklos Szeredi 184c63181e6SAl Viro err = mnt_alloc_id(mnt); 18588b38782SLi Zefan if (err) 18688b38782SLi Zefan goto out_free_cache; 18788b38782SLi Zefan 18888b38782SLi Zefan if (name) { 189fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 190c63181e6SAl Viro if (!mnt->mnt_devname) 19188b38782SLi Zefan goto out_free_id; 19273cd49ecSMiklos Szeredi } 19373cd49ecSMiklos Szeredi 194b3e19d92SNick Piggin #ifdef CONFIG_SMP 195c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 196c63181e6SAl Viro if (!mnt->mnt_pcp) 197b3e19d92SNick Piggin goto out_free_devname; 198b3e19d92SNick Piggin 199c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 200b3e19d92SNick Piggin #else 201c63181e6SAl Viro mnt->mnt_count = 1; 202c63181e6SAl Viro mnt->mnt_writers = 0; 203b3e19d92SNick Piggin #endif 204b3e19d92SNick Piggin 20538129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 206c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 207c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 208c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 209c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 210c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 211c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 212c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2130a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 21499b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 21587b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2161da177e4SLinus Torvalds } 217c63181e6SAl Viro return mnt; 21888b38782SLi Zefan 219d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 220d3ef3d73Snpiggin@suse.de out_free_devname: 221fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 222d3ef3d73Snpiggin@suse.de #endif 22388b38782SLi Zefan out_free_id: 224c63181e6SAl Viro mnt_free_id(mnt); 22588b38782SLi Zefan out_free_cache: 226c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 22788b38782SLi Zefan return NULL; 2281da177e4SLinus Torvalds } 2291da177e4SLinus Torvalds 2308366025eSDave Hansen /* 2318366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2328366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2338366025eSDave Hansen * We must keep track of when those operations start 2348366025eSDave Hansen * (for permission checks) and when they end, so that 2358366025eSDave Hansen * we can determine when writes are able to occur to 2368366025eSDave Hansen * a filesystem. 2378366025eSDave Hansen */ 2383d733633SDave Hansen /* 2393d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2403d733633SDave Hansen * @mnt: the mount to check for its write status 2413d733633SDave Hansen * 2423d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2433d733633SDave Hansen * It does not guarantee that the filesystem will stay 2443d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2453d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2463d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2473d733633SDave Hansen * r/w. 2483d733633SDave Hansen */ 24943f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 2503d733633SDave Hansen { 25143f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 2523d733633SDave Hansen } 2533d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2543d733633SDave Hansen 25583adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2563d733633SDave Hansen { 257d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 25868e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 259d3ef3d73Snpiggin@suse.de #else 26068e8a9feSAl Viro mnt->mnt_writers++; 261d3ef3d73Snpiggin@suse.de #endif 2623d733633SDave Hansen } 2633d733633SDave Hansen 26483adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2653d733633SDave Hansen { 266d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 26768e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 268d3ef3d73Snpiggin@suse.de #else 26968e8a9feSAl Viro mnt->mnt_writers--; 270d3ef3d73Snpiggin@suse.de #endif 271d3ef3d73Snpiggin@suse.de } 272d3ef3d73Snpiggin@suse.de 27383adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 274d3ef3d73Snpiggin@suse.de { 275d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 276d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2773d733633SDave Hansen int cpu; 2783d733633SDave Hansen 2793d733633SDave Hansen for_each_possible_cpu(cpu) { 28068e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2813d733633SDave Hansen } 2823d733633SDave Hansen 283d3ef3d73Snpiggin@suse.de return count; 284d3ef3d73Snpiggin@suse.de #else 285d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 286d3ef3d73Snpiggin@suse.de #endif 2873d733633SDave Hansen } 2883d733633SDave Hansen 2894ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2904ed5e82fSMiklos Szeredi { 2914ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2924ed5e82fSMiklos Szeredi return 1; 2934ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2944ed5e82fSMiklos Szeredi smp_rmb(); 2954ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2964ed5e82fSMiklos Szeredi } 2974ed5e82fSMiklos Szeredi 2983d733633SDave Hansen /* 299eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 300eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 301eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 302eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3033d733633SDave Hansen */ 3048366025eSDave Hansen /** 305eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 30683adc753SAl Viro * @m: the mount on which to take a write 3078366025eSDave Hansen * 308eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 309eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 310eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 311eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 312eb04c282SJan Kara * called. This is effectively a refcount. 3138366025eSDave Hansen */ 314eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3158366025eSDave Hansen { 31683adc753SAl Viro struct mount *mnt = real_mount(m); 3173d733633SDave Hansen int ret = 0; 3183d733633SDave Hansen 319d3ef3d73Snpiggin@suse.de preempt_disable(); 320c6653a83SNick Piggin mnt_inc_writers(mnt); 321d3ef3d73Snpiggin@suse.de /* 322c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 323d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 324d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 325d3ef3d73Snpiggin@suse.de */ 326d3ef3d73Snpiggin@suse.de smp_mb(); 3276aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 328d3ef3d73Snpiggin@suse.de cpu_relax(); 329d3ef3d73Snpiggin@suse.de /* 330d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 331d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 332d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 333d3ef3d73Snpiggin@suse.de */ 334d3ef3d73Snpiggin@suse.de smp_rmb(); 3354ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 336c6653a83SNick Piggin mnt_dec_writers(mnt); 3373d733633SDave Hansen ret = -EROFS; 3383d733633SDave Hansen } 339d3ef3d73Snpiggin@suse.de preempt_enable(); 340eb04c282SJan Kara 341eb04c282SJan Kara return ret; 342eb04c282SJan Kara } 343eb04c282SJan Kara 344eb04c282SJan Kara /** 345eb04c282SJan Kara * mnt_want_write - get write access to a mount 346eb04c282SJan Kara * @m: the mount on which to take a write 347eb04c282SJan Kara * 348eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 349eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 350eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 351eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 352eb04c282SJan Kara */ 353eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 354eb04c282SJan Kara { 355eb04c282SJan Kara int ret; 356eb04c282SJan Kara 357eb04c282SJan Kara sb_start_write(m->mnt_sb); 358eb04c282SJan Kara ret = __mnt_want_write(m); 359eb04c282SJan Kara if (ret) 360eb04c282SJan Kara sb_end_write(m->mnt_sb); 3613d733633SDave Hansen return ret; 3628366025eSDave Hansen } 3638366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3648366025eSDave Hansen 3658366025eSDave Hansen /** 36696029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 36796029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 36896029c4eSnpiggin@suse.de * 36996029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 37096029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 37196029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 37296029c4eSnpiggin@suse.de * on it. This allows some optimisation. 37396029c4eSnpiggin@suse.de * 37496029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 37596029c4eSnpiggin@suse.de * drop the reference. 37696029c4eSnpiggin@suse.de */ 37796029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 37896029c4eSnpiggin@suse.de { 37996029c4eSnpiggin@suse.de /* superblock may be r/o */ 38096029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 38196029c4eSnpiggin@suse.de return -EROFS; 38296029c4eSnpiggin@suse.de preempt_disable(); 38383adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 38496029c4eSnpiggin@suse.de preempt_enable(); 38596029c4eSnpiggin@suse.de return 0; 38696029c4eSnpiggin@suse.de } 38796029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 38896029c4eSnpiggin@suse.de 38996029c4eSnpiggin@suse.de /** 390eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 391eb04c282SJan Kara * @file: the file who's mount on which to take a write 392eb04c282SJan Kara * 393eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 394eb04c282SJan Kara * do some optimisations if the file is open for write already 395eb04c282SJan Kara */ 396eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 397eb04c282SJan Kara { 39883f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 399eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 400eb04c282SJan Kara else 401eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 402eb04c282SJan Kara } 403eb04c282SJan Kara 404eb04c282SJan Kara /** 4057c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4067c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4077c6893e3SMiklos Szeredi * 4087c6893e3SMiklos Szeredi * This is like mnt_want_write, but it takes a file and can 4097c6893e3SMiklos Szeredi * do some optimisations if the file is open for write already 4107c6893e3SMiklos Szeredi */ 4117c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4127c6893e3SMiklos Szeredi { 4137c6893e3SMiklos Szeredi int ret; 4147c6893e3SMiklos Szeredi 4157c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4167c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4177c6893e3SMiklos Szeredi if (ret) 4187c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4197c6893e3SMiklos Szeredi return ret; 4207c6893e3SMiklos Szeredi } 42196029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 42296029c4eSnpiggin@suse.de 42396029c4eSnpiggin@suse.de /** 424eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4258366025eSDave Hansen * @mnt: the mount on which to give up write access 4268366025eSDave Hansen * 4278366025eSDave Hansen * Tells the low-level filesystem that we are done 4288366025eSDave Hansen * performing writes to it. Must be matched with 429eb04c282SJan Kara * __mnt_want_write() call above. 4308366025eSDave Hansen */ 431eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4328366025eSDave Hansen { 433d3ef3d73Snpiggin@suse.de preempt_disable(); 43483adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 435d3ef3d73Snpiggin@suse.de preempt_enable(); 4368366025eSDave Hansen } 437eb04c282SJan Kara 438eb04c282SJan Kara /** 439eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 440eb04c282SJan Kara * @mnt: the mount on which to give up write access 441eb04c282SJan Kara * 442eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 443eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 444eb04c282SJan Kara * mnt_want_write() call above. 445eb04c282SJan Kara */ 446eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 447eb04c282SJan Kara { 448eb04c282SJan Kara __mnt_drop_write(mnt); 449eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 450eb04c282SJan Kara } 4518366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4528366025eSDave Hansen 453eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 454eb04c282SJan Kara { 455eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 456eb04c282SJan Kara } 457eb04c282SJan Kara 4587c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4597c6893e3SMiklos Szeredi { 460a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4617c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4627c6893e3SMiklos Szeredi } 4632a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4642a79f17eSAl Viro 46583adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4668366025eSDave Hansen { 4673d733633SDave Hansen int ret = 0; 4683d733633SDave Hansen 469719ea2fbSAl Viro lock_mount_hash(); 47083adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 471d3ef3d73Snpiggin@suse.de /* 472d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 473d3ef3d73Snpiggin@suse.de * should be visible before we do. 474d3ef3d73Snpiggin@suse.de */ 475d3ef3d73Snpiggin@suse.de smp_mb(); 476d3ef3d73Snpiggin@suse.de 477d3ef3d73Snpiggin@suse.de /* 478d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 479d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 480d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 481d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 482d3ef3d73Snpiggin@suse.de * 483d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 484d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 485d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 486d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 487d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 488d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 489d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 490d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 491d3ef3d73Snpiggin@suse.de * we're counting up here. 492d3ef3d73Snpiggin@suse.de */ 493c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 494d3ef3d73Snpiggin@suse.de ret = -EBUSY; 495d3ef3d73Snpiggin@suse.de else 49683adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 497d3ef3d73Snpiggin@suse.de /* 498d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 499d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 500d3ef3d73Snpiggin@suse.de */ 501d3ef3d73Snpiggin@suse.de smp_wmb(); 50283adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 503719ea2fbSAl Viro unlock_mount_hash(); 5043d733633SDave Hansen return ret; 5053d733633SDave Hansen } 5068366025eSDave Hansen 50743f5e655SDavid Howells static int __mnt_unmake_readonly(struct mount *mnt) 5082e4b7fcdSDave Hansen { 509719ea2fbSAl Viro lock_mount_hash(); 51083adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 511719ea2fbSAl Viro unlock_mount_hash(); 51243f5e655SDavid Howells return 0; 5132e4b7fcdSDave Hansen } 5142e4b7fcdSDave Hansen 5154ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5164ed5e82fSMiklos Szeredi { 5174ed5e82fSMiklos Szeredi struct mount *mnt; 5184ed5e82fSMiklos Szeredi int err = 0; 5194ed5e82fSMiklos Szeredi 5208e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5218e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5228e8b8796SMiklos Szeredi return -EBUSY; 5238e8b8796SMiklos Szeredi 524719ea2fbSAl Viro lock_mount_hash(); 5254ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5264ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5274ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5284ed5e82fSMiklos Szeredi smp_mb(); 5294ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5304ed5e82fSMiklos Szeredi err = -EBUSY; 5314ed5e82fSMiklos Szeredi break; 5324ed5e82fSMiklos Szeredi } 5334ed5e82fSMiklos Szeredi } 5344ed5e82fSMiklos Szeredi } 5358e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5368e8b8796SMiklos Szeredi err = -EBUSY; 5378e8b8796SMiklos Szeredi 5384ed5e82fSMiklos Szeredi if (!err) { 5394ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5404ed5e82fSMiklos Szeredi smp_wmb(); 5414ed5e82fSMiklos Szeredi } 5424ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5434ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5444ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5454ed5e82fSMiklos Szeredi } 546719ea2fbSAl Viro unlock_mount_hash(); 5474ed5e82fSMiklos Szeredi 5484ed5e82fSMiklos Szeredi return err; 5494ed5e82fSMiklos Szeredi } 5504ed5e82fSMiklos Szeredi 551b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5521da177e4SLinus Torvalds { 553fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 554d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 55568e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 556d3ef3d73Snpiggin@suse.de #endif 557b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5581da177e4SLinus Torvalds } 5591da177e4SLinus Torvalds 5608ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5618ffcb32eSDavid Howells { 5628ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5638ffcb32eSDavid Howells } 5648ffcb32eSDavid Howells 56548a066e7SAl Viro /* call under rcu_read_lock */ 566294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 56748a066e7SAl Viro { 56848a066e7SAl Viro struct mount *mnt; 56948a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 570294d71ffSAl Viro return 1; 57148a066e7SAl Viro if (bastard == NULL) 572294d71ffSAl Viro return 0; 57348a066e7SAl Viro mnt = real_mount(bastard); 57448a066e7SAl Viro mnt_add_count(mnt, 1); 575119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 57648a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 577294d71ffSAl Viro return 0; 57848a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 57948a066e7SAl Viro mnt_add_count(mnt, -1); 580294d71ffSAl Viro return 1; 58148a066e7SAl Viro } 582119e1ef8SAl Viro lock_mount_hash(); 583119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 584119e1ef8SAl Viro mnt_add_count(mnt, -1); 585119e1ef8SAl Viro unlock_mount_hash(); 586119e1ef8SAl Viro return 1; 587119e1ef8SAl Viro } 588119e1ef8SAl Viro unlock_mount_hash(); 589119e1ef8SAl Viro /* caller will mntput() */ 590294d71ffSAl Viro return -1; 591294d71ffSAl Viro } 592294d71ffSAl Viro 593294d71ffSAl Viro /* call under rcu_read_lock */ 594294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 595294d71ffSAl Viro { 596294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 597294d71ffSAl Viro if (likely(!res)) 598294d71ffSAl Viro return true; 599294d71ffSAl Viro if (unlikely(res < 0)) { 60048a066e7SAl Viro rcu_read_unlock(); 60148a066e7SAl Viro mntput(bastard); 60248a066e7SAl Viro rcu_read_lock(); 603294d71ffSAl Viro } 60448a066e7SAl Viro return false; 60548a066e7SAl Viro } 60648a066e7SAl Viro 6071da177e4SLinus Torvalds /* 608474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 60948a066e7SAl Viro * call under rcu_read_lock() 6101da177e4SLinus Torvalds */ 611474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6121da177e4SLinus Torvalds { 61338129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 614474279dcSAl Viro struct mount *p; 6151da177e4SLinus Torvalds 61638129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 617474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 618474279dcSAl Viro return p; 619474279dcSAl Viro return NULL; 6201da177e4SLinus Torvalds } 621474279dcSAl Viro 622474279dcSAl Viro /* 623f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 624f015f126SDavid Howells * 625f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 626f015f126SDavid Howells * following mounts: 627f015f126SDavid Howells * 628f015f126SDavid Howells * mount /dev/sda1 /mnt 629f015f126SDavid Howells * mount /dev/sda2 /mnt 630f015f126SDavid Howells * mount /dev/sda3 /mnt 631f015f126SDavid Howells * 632f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 633f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 634f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 635f015f126SDavid Howells * 636f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 637a05964f3SRam Pai */ 638ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 639a05964f3SRam Pai { 640c7105365SAl Viro struct mount *child_mnt; 64148a066e7SAl Viro struct vfsmount *m; 64248a066e7SAl Viro unsigned seq; 64399b7db7bSNick Piggin 64448a066e7SAl Viro rcu_read_lock(); 64548a066e7SAl Viro do { 64648a066e7SAl Viro seq = read_seqbegin(&mount_lock); 647474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 64848a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 64948a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 65048a066e7SAl Viro rcu_read_unlock(); 65148a066e7SAl Viro return m; 652a05964f3SRam Pai } 653a05964f3SRam Pai 6547af1364fSEric W. Biederman /* 6557af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6567af1364fSEric W. Biederman * current mount namespace. 6577af1364fSEric W. Biederman * 6587af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6597af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6607af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6617af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6627af1364fSEric W. Biederman * is a mountpoint. 6637af1364fSEric W. Biederman * 6647af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6657af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6667af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6677af1364fSEric W. Biederman * parent mount. 6687af1364fSEric W. Biederman */ 6697af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6707af1364fSEric W. Biederman { 6717af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 6727af1364fSEric W. Biederman struct mount *mnt; 6737af1364fSEric W. Biederman bool is_covered = false; 6747af1364fSEric W. Biederman 6757af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 6767af1364fSEric W. Biederman goto out; 6777af1364fSEric W. Biederman 6787af1364fSEric W. Biederman down_read(&namespace_sem); 6797af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 6807af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 6817af1364fSEric W. Biederman if (is_covered) 6827af1364fSEric W. Biederman break; 6837af1364fSEric W. Biederman } 6847af1364fSEric W. Biederman up_read(&namespace_sem); 6857af1364fSEric W. Biederman out: 6867af1364fSEric W. Biederman return is_covered; 6877af1364fSEric W. Biederman } 6887af1364fSEric W. Biederman 689e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 69084d17192SAl Viro { 6910818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 69284d17192SAl Viro struct mountpoint *mp; 69384d17192SAl Viro 6940818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 69584d17192SAl Viro if (mp->m_dentry == dentry) { 69684d17192SAl Viro mp->m_count++; 69784d17192SAl Viro return mp; 69884d17192SAl Viro } 69984d17192SAl Viro } 700e2dfa935SEric W. Biederman return NULL; 701e2dfa935SEric W. Biederman } 702e2dfa935SEric W. Biederman 7033895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 704e2dfa935SEric W. Biederman { 7053895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 706e2dfa935SEric W. Biederman int ret; 70784d17192SAl Viro 7083895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7091e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7101e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7111e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7123895dbf8SEric W. Biederman mountpoint: 7133895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7143895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7153895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7163895dbf8SEric W. Biederman if (mp) 7173895dbf8SEric W. Biederman goto done; 71884d17192SAl Viro } 719eed81007SMiklos Szeredi 7203895dbf8SEric W. Biederman if (!new) 7213895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7223895dbf8SEric W. Biederman if (!new) 7233895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7243895dbf8SEric W. Biederman 7253895dbf8SEric W. Biederman 7263895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7273895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7283895dbf8SEric W. Biederman 7293895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7303895dbf8SEric W. Biederman if (ret == -EBUSY) 7313895dbf8SEric W. Biederman goto mountpoint; 7323895dbf8SEric W. Biederman 7333895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7343895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7353895dbf8SEric W. Biederman if (ret) 7363895dbf8SEric W. Biederman goto done; 7373895dbf8SEric W. Biederman 7383895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7393895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7403895dbf8SEric W. Biederman new->m_dentry = dentry; 7413895dbf8SEric W. Biederman new->m_count = 1; 7423895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7433895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7443895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7453895dbf8SEric W. Biederman 7463895dbf8SEric W. Biederman mp = new; 7473895dbf8SEric W. Biederman new = NULL; 7483895dbf8SEric W. Biederman done: 7493895dbf8SEric W. Biederman kfree(new); 75084d17192SAl Viro return mp; 75184d17192SAl Viro } 75284d17192SAl Viro 75384d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 75484d17192SAl Viro { 75584d17192SAl Viro if (!--mp->m_count) { 75684d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7570a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 75884d17192SAl Viro spin_lock(&dentry->d_lock); 75984d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 76084d17192SAl Viro spin_unlock(&dentry->d_lock); 7610818bf27SAl Viro hlist_del(&mp->m_hash); 76284d17192SAl Viro kfree(mp); 76384d17192SAl Viro } 76484d17192SAl Viro } 76584d17192SAl Viro 766143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7671da177e4SLinus Torvalds { 7686b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7691da177e4SLinus Torvalds } 7701da177e4SLinus Torvalds 77199b7db7bSNick Piggin /* 77299b7db7bSNick Piggin * vfsmount lock must be held for write 77399b7db7bSNick Piggin */ 7746b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7755addc5ddSAl Viro { 7765addc5ddSAl Viro if (ns) { 7775addc5ddSAl Viro ns->event = ++event; 7785addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7795addc5ddSAl Viro } 7805addc5ddSAl Viro } 7815addc5ddSAl Viro 78299b7db7bSNick Piggin /* 78399b7db7bSNick Piggin * vfsmount lock must be held for write 78499b7db7bSNick Piggin */ 7856b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 7865addc5ddSAl Viro { 7875addc5ddSAl Viro if (ns && ns->event != event) { 7885addc5ddSAl Viro ns->event = event; 7895addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7905addc5ddSAl Viro } 7915addc5ddSAl Viro } 7925addc5ddSAl Viro 79399b7db7bSNick Piggin /* 79499b7db7bSNick Piggin * vfsmount lock must be held for write 79599b7db7bSNick Piggin */ 7967bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 7971da177e4SLinus Torvalds { 7980714a533SAl Viro mnt->mnt_parent = mnt; 799a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8006b41d536SAl Viro list_del_init(&mnt->mnt_child); 80138129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8020a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 80384d17192SAl Viro put_mountpoint(mnt->mnt_mp); 80484d17192SAl Viro mnt->mnt_mp = NULL; 8051da177e4SLinus Torvalds } 8061da177e4SLinus Torvalds 80799b7db7bSNick Piggin /* 80899b7db7bSNick Piggin * vfsmount lock must be held for write 80999b7db7bSNick Piggin */ 8107bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8117bdb11deSEric W. Biederman { 8127bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8137bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8147bdb11deSEric W. Biederman unhash_mnt(mnt); 8157bdb11deSEric W. Biederman } 8167bdb11deSEric W. Biederman 8177bdb11deSEric W. Biederman /* 8187bdb11deSEric W. Biederman * vfsmount lock must be held for write 8197bdb11deSEric W. Biederman */ 8206a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8216a46c573SEric W. Biederman { 8226a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8236a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8246a46c573SEric W. Biederman unhash_mnt(mnt); 8256a46c573SEric W. Biederman } 8266a46c573SEric W. Biederman 8276a46c573SEric W. Biederman /* 8286a46c573SEric W. Biederman * vfsmount lock must be held for write 8296a46c573SEric W. Biederman */ 83084d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 83184d17192SAl Viro struct mountpoint *mp, 83244d964d6SAl Viro struct mount *child_mnt) 833b90fa9aeSRam Pai { 83484d17192SAl Viro mp->m_count++; 8353a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 83684d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8373a2393d7SAl Viro child_mnt->mnt_parent = mnt; 83884d17192SAl Viro child_mnt->mnt_mp = mp; 8390a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 840b90fa9aeSRam Pai } 841b90fa9aeSRam Pai 8421064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8431064f874SEric W. Biederman { 8441064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8451064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8461064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8471064f874SEric W. Biederman } 8481064f874SEric W. Biederman 84999b7db7bSNick Piggin /* 85099b7db7bSNick Piggin * vfsmount lock must be held for write 85199b7db7bSNick Piggin */ 85284d17192SAl Viro static void attach_mnt(struct mount *mnt, 85384d17192SAl Viro struct mount *parent, 85484d17192SAl Viro struct mountpoint *mp) 8551da177e4SLinus Torvalds { 85684d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8571064f874SEric W. Biederman __attach_mnt(mnt, parent); 858b90fa9aeSRam Pai } 859b90fa9aeSRam Pai 8601064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 86112a5b529SAl Viro { 8621064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8631064f874SEric W. Biederman struct dentry *old_mountpoint = mnt->mnt_mountpoint; 8641064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8651064f874SEric W. Biederman 8661064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8671064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8681064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8691064f874SEric W. Biederman 8701064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 8711064f874SEric W. Biederman 8721064f874SEric W. Biederman put_mountpoint(old_mp); 8731064f874SEric W. Biederman 8741064f874SEric W. Biederman /* 8751064f874SEric W. Biederman * Safely avoid even the suggestion this code might sleep or 8761064f874SEric W. Biederman * lock the mount hash by taking advantage of the knowledge that 8771064f874SEric W. Biederman * mnt_change_mountpoint will not release the final reference 8781064f874SEric W. Biederman * to a mountpoint. 8791064f874SEric W. Biederman * 8801064f874SEric W. Biederman * During mounting, the mount passed in as the parent mount will 8811064f874SEric W. Biederman * continue to use the old mountpoint and during unmounting, the 8821064f874SEric W. Biederman * old mountpoint will continue to exist until namespace_unlock, 8831064f874SEric W. Biederman * which happens well after mnt_change_mountpoint. 8841064f874SEric W. Biederman */ 8851064f874SEric W. Biederman spin_lock(&old_mountpoint->d_lock); 8861064f874SEric W. Biederman old_mountpoint->d_lockref.count--; 8871064f874SEric W. Biederman spin_unlock(&old_mountpoint->d_lock); 8881064f874SEric W. Biederman 8891064f874SEric W. Biederman mnt_add_count(old_parent, -1); 89012a5b529SAl Viro } 89112a5b529SAl Viro 892b90fa9aeSRam Pai /* 89399b7db7bSNick Piggin * vfsmount lock must be held for write 894b90fa9aeSRam Pai */ 8951064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 896b90fa9aeSRam Pai { 8970714a533SAl Viro struct mount *parent = mnt->mnt_parent; 89883adc753SAl Viro struct mount *m; 899b90fa9aeSRam Pai LIST_HEAD(head); 900143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 901b90fa9aeSRam Pai 9020714a533SAl Viro BUG_ON(parent == mnt); 903b90fa9aeSRam Pai 9041a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 905f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 906143c8c91SAl Viro m->mnt_ns = n; 907f03c6599SAl Viro 908b90fa9aeSRam Pai list_splice(&head, n->list.prev); 909b90fa9aeSRam Pai 910d2921684SEric W. Biederman n->mounts += n->pending_mounts; 911d2921684SEric W. Biederman n->pending_mounts = 0; 912d2921684SEric W. Biederman 9131064f874SEric W. Biederman __attach_mnt(mnt, parent); 9146b3286edSKirill Korotaev touch_mnt_namespace(n); 9151da177e4SLinus Torvalds } 9161da177e4SLinus Torvalds 917909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9181da177e4SLinus Torvalds { 9196b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9206b41d536SAl Viro if (next == &p->mnt_mounts) { 9211da177e4SLinus Torvalds while (1) { 922909b0a88SAl Viro if (p == root) 9231da177e4SLinus Torvalds return NULL; 9246b41d536SAl Viro next = p->mnt_child.next; 9256b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9261da177e4SLinus Torvalds break; 9270714a533SAl Viro p = p->mnt_parent; 9281da177e4SLinus Torvalds } 9291da177e4SLinus Torvalds } 9306b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9311da177e4SLinus Torvalds } 9321da177e4SLinus Torvalds 933315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9349676f0c6SRam Pai { 9356b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9366b41d536SAl Viro while (prev != &p->mnt_mounts) { 9376b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9386b41d536SAl Viro prev = p->mnt_mounts.prev; 9399676f0c6SRam Pai } 9409676f0c6SRam Pai return p; 9419676f0c6SRam Pai } 9429676f0c6SRam Pai 9439d412a43SAl Viro struct vfsmount * 9449d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 9459d412a43SAl Viro { 946b105e270SAl Viro struct mount *mnt; 9479d412a43SAl Viro struct dentry *root; 9489d412a43SAl Viro 9499d412a43SAl Viro if (!type) 9509d412a43SAl Viro return ERR_PTR(-ENODEV); 9519d412a43SAl Viro 9529d412a43SAl Viro mnt = alloc_vfsmnt(name); 9539d412a43SAl Viro if (!mnt) 9549d412a43SAl Viro return ERR_PTR(-ENOMEM); 9559d412a43SAl Viro 956e462ec50SDavid Howells if (flags & SB_KERNMOUNT) 957b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9589d412a43SAl Viro 9599d412a43SAl Viro root = mount_fs(type, flags, name, data); 9609d412a43SAl Viro if (IS_ERR(root)) { 9618ffcb32eSDavid Howells mnt_free_id(mnt); 9629d412a43SAl Viro free_vfsmnt(mnt); 9639d412a43SAl Viro return ERR_CAST(root); 9649d412a43SAl Viro } 9659d412a43SAl Viro 966b105e270SAl Viro mnt->mnt.mnt_root = root; 967b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 968a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9690714a533SAl Viro mnt->mnt_parent = mnt; 970719ea2fbSAl Viro lock_mount_hash(); 97139f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 972719ea2fbSAl Viro unlock_mount_hash(); 973b105e270SAl Viro return &mnt->mnt; 9749d412a43SAl Viro } 9759d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 9769d412a43SAl Viro 97793faccbbSEric W. Biederman struct vfsmount * 97893faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 97993faccbbSEric W. Biederman const char *name, void *data) 98093faccbbSEric W. Biederman { 98193faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 98293faccbbSEric W. Biederman * through from the parent mount to the submount don't support 98393faccbbSEric W. Biederman * unprivileged mounts with submounts. 98493faccbbSEric W. Biederman */ 98593faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 98693faccbbSEric W. Biederman return ERR_PTR(-EPERM); 98793faccbbSEric W. Biederman 988e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 98993faccbbSEric W. Biederman } 99093faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 99193faccbbSEric W. Biederman 99287129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 99336341f64SRam Pai int flag) 9941da177e4SLinus Torvalds { 99587129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 996be34d1a3SDavid Howells struct mount *mnt; 997be34d1a3SDavid Howells int err; 9981da177e4SLinus Torvalds 999be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1000be34d1a3SDavid Howells if (!mnt) 1001be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1002be34d1a3SDavid Howells 10037a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 100415169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1005719f5d7fSMiklos Szeredi else 100615169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1007719f5d7fSMiklos Szeredi 100815169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1009be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1010719f5d7fSMiklos Szeredi if (err) 1011719f5d7fSMiklos Szeredi goto out_free; 1012719f5d7fSMiklos Szeredi } 1013719f5d7fSMiklos Szeredi 101416a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 101516a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 10165ff9d8a6SEric W. Biederman 10171da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1018b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1019b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1020a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10210714a533SAl Viro mnt->mnt_parent = mnt; 1022719ea2fbSAl Viro lock_mount_hash(); 102339f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1024719ea2fbSAl Viro unlock_mount_hash(); 1025b90fa9aeSRam Pai 10267a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10277a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10286776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 102932301920SAl Viro mnt->mnt_master = old; 1030fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10318aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1032fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10336776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1034d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10356776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1036d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10375235d448SAl Viro } else { 10385235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10395afe0022SRam Pai } 1040b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10410f0afb1dSAl Viro set_mnt_shared(mnt); 10421da177e4SLinus Torvalds 10431da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10441da177e4SLinus Torvalds * as the original if that was on one */ 104536341f64SRam Pai if (flag & CL_EXPIRE) { 10466776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10476776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10481da177e4SLinus Torvalds } 1049be34d1a3SDavid Howells 1050cb338d06SAl Viro return mnt; 1051719f5d7fSMiklos Szeredi 1052719f5d7fSMiklos Szeredi out_free: 10538ffcb32eSDavid Howells mnt_free_id(mnt); 1054719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1055be34d1a3SDavid Howells return ERR_PTR(err); 10561da177e4SLinus Torvalds } 10571da177e4SLinus Torvalds 10589ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10599ea459e1SAl Viro { 10609ea459e1SAl Viro /* 10619ea459e1SAl Viro * This probably indicates that somebody messed 10629ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 10639ea459e1SAl Viro * happens, the filesystem was probably unable 10649ea459e1SAl Viro * to make r/w->r/o transitions. 10659ea459e1SAl Viro */ 10669ea459e1SAl Viro /* 10679ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 10689ea459e1SAl Viro * so mnt_get_writers() below is safe. 10699ea459e1SAl Viro */ 10709ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 10719ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 10729ea459e1SAl Viro mnt_pin_kill(mnt); 10739ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 10749ea459e1SAl Viro dput(mnt->mnt.mnt_root); 10759ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 10769ea459e1SAl Viro mnt_free_id(mnt); 10779ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 10789ea459e1SAl Viro } 10799ea459e1SAl Viro 10809ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 10819ea459e1SAl Viro { 10829ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 10839ea459e1SAl Viro } 10849ea459e1SAl Viro 10859ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 10869ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 10879ea459e1SAl Viro { 10889ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 108929785735SByungchul Park struct mount *m, *t; 10909ea459e1SAl Viro 109129785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 109229785735SByungchul Park cleanup_mnt(m); 10939ea459e1SAl Viro } 10949ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 10959ea459e1SAl Viro 1096900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 10977b7b1aceSAl Viro { 109848a066e7SAl Viro rcu_read_lock(); 10999ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11009ea0a46cSAl Viro /* 11019ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11029ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11039ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11049ea0a46cSAl Viro * be dropped until after an RCU delay done after 11059ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11069ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11079ea0a46cSAl Viro * we are dropping is not the final one. 11089ea0a46cSAl Viro */ 1109aa9c0e07SAl Viro mnt_add_count(mnt, -1); 111048a066e7SAl Viro rcu_read_unlock(); 111199b7db7bSNick Piggin return; 1112b3e19d92SNick Piggin } 1113719ea2fbSAl Viro lock_mount_hash(); 1114119e1ef8SAl Viro /* 1115119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1116119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1117119e1ef8SAl Viro */ 1118119e1ef8SAl Viro smp_mb(); 11199ea0a46cSAl Viro mnt_add_count(mnt, -1); 1120b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 112148a066e7SAl Viro rcu_read_unlock(); 1122719ea2fbSAl Viro unlock_mount_hash(); 112399b7db7bSNick Piggin return; 112499b7db7bSNick Piggin } 112548a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 112648a066e7SAl Viro rcu_read_unlock(); 112748a066e7SAl Viro unlock_mount_hash(); 112848a066e7SAl Viro return; 112948a066e7SAl Viro } 113048a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 113148a066e7SAl Viro rcu_read_unlock(); 1132962830dfSAndi Kleen 113339f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1134ce07d891SEric W. Biederman 1135ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1136ce07d891SEric W. Biederman struct mount *p, *tmp; 1137ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1138ce07d891SEric W. Biederman umount_mnt(p); 1139ce07d891SEric W. Biederman } 1140ce07d891SEric W. Biederman } 1141719ea2fbSAl Viro unlock_mount_hash(); 1142649a795aSAl Viro 11439ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11449ea459e1SAl Viro struct task_struct *task = current; 11459ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11469ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11479ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11489ea459e1SAl Viro return; 11499ea459e1SAl Viro } 11509ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11519ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11529ea459e1SAl Viro return; 11539ea459e1SAl Viro } 11549ea459e1SAl Viro cleanup_mnt(mnt); 1155b3e19d92SNick Piggin } 1156b3e19d92SNick Piggin 1157b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1158b3e19d92SNick Piggin { 1159b3e19d92SNick Piggin if (mnt) { 1160863d684fSAl Viro struct mount *m = real_mount(mnt); 1161b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1162863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1163863d684fSAl Viro m->mnt_expiry_mark = 0; 1164863d684fSAl Viro mntput_no_expire(m); 1165b3e19d92SNick Piggin } 1166b3e19d92SNick Piggin } 1167b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1168b3e19d92SNick Piggin 1169b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1170b3e19d92SNick Piggin { 1171b3e19d92SNick Piggin if (mnt) 117283adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1173b3e19d92SNick Piggin return mnt; 1174b3e19d92SNick Piggin } 1175b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1176b3e19d92SNick Piggin 1177c6609c0aSIan Kent /* path_is_mountpoint() - Check if path is a mount in the current 1178c6609c0aSIan Kent * namespace. 1179c6609c0aSIan Kent * 1180c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1181c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1182c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1183c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1184c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1185c6609c0aSIan Kent * alone. 1186c6609c0aSIan Kent */ 1187c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1188c6609c0aSIan Kent { 1189c6609c0aSIan Kent unsigned seq; 1190c6609c0aSIan Kent bool res; 1191c6609c0aSIan Kent 1192c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1193c6609c0aSIan Kent return false; 1194c6609c0aSIan Kent 1195c6609c0aSIan Kent rcu_read_lock(); 1196c6609c0aSIan Kent do { 1197c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1198c6609c0aSIan Kent res = __path_is_mountpoint(path); 1199c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1200c6609c0aSIan Kent rcu_read_unlock(); 1201c6609c0aSIan Kent 1202c6609c0aSIan Kent return res; 1203c6609c0aSIan Kent } 1204c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1205c6609c0aSIan Kent 1206ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12077b7b1aceSAl Viro { 12083064c356SAl Viro struct mount *p; 12093064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12103064c356SAl Viro if (IS_ERR(p)) 12113064c356SAl Viro return ERR_CAST(p); 12123064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12133064c356SAl Viro return &p->mnt; 12147b7b1aceSAl Viro } 12151da177e4SLinus Torvalds 1216a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12170226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12181da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12191da177e4SLinus Torvalds { 1220ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12211da177e4SLinus Torvalds 1222390c6843SRam Pai down_read(&namespace_sem); 1223c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1224c7999c36SAl Viro void *v = p->cached_mount; 1225c7999c36SAl Viro if (*pos == p->cached_index) 1226c7999c36SAl Viro return v; 1227c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1228c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1229c7999c36SAl Viro return p->cached_mount = v; 1230c7999c36SAl Viro } 1231c7999c36SAl Viro } 1232c7999c36SAl Viro 1233c7999c36SAl Viro p->cached_event = p->ns->event; 1234c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1235c7999c36SAl Viro p->cached_index = *pos; 1236c7999c36SAl Viro return p->cached_mount; 12371da177e4SLinus Torvalds } 12381da177e4SLinus Torvalds 12391da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12401da177e4SLinus Torvalds { 1241ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1242b0765fb8SPavel Emelianov 1243c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1244c7999c36SAl Viro p->cached_index = *pos; 1245c7999c36SAl Viro return p->cached_mount; 12461da177e4SLinus Torvalds } 12471da177e4SLinus Torvalds 12481da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12491da177e4SLinus Torvalds { 1250390c6843SRam Pai up_read(&namespace_sem); 12511da177e4SLinus Torvalds } 12521da177e4SLinus Torvalds 12530226f492SAl Viro static int m_show(struct seq_file *m, void *v) 12549f5596afSAl Viro { 1255ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12561a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 12570226f492SAl Viro return p->show(m, &r->mnt); 12581da177e4SLinus Torvalds } 12591da177e4SLinus Torvalds 1260a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 12611da177e4SLinus Torvalds .start = m_start, 12621da177e4SLinus Torvalds .next = m_next, 12631da177e4SLinus Torvalds .stop = m_stop, 12640226f492SAl Viro .show = m_show, 1265b4629fe2SChuck Lever }; 1266a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1267b4629fe2SChuck Lever 12681da177e4SLinus Torvalds /** 12691da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 12701da177e4SLinus Torvalds * @mnt: root of mount tree 12711da177e4SLinus Torvalds * 12721da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 12731da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 12741da177e4SLinus Torvalds * busy. 12751da177e4SLinus Torvalds */ 1276909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 12771da177e4SLinus Torvalds { 1278909b0a88SAl Viro struct mount *mnt = real_mount(m); 127936341f64SRam Pai int actual_refs = 0; 128036341f64SRam Pai int minimum_refs = 0; 1281315fc83eSAl Viro struct mount *p; 1282909b0a88SAl Viro BUG_ON(!m); 12831da177e4SLinus Torvalds 1284b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1285719ea2fbSAl Viro lock_mount_hash(); 1286909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 128783adc753SAl Viro actual_refs += mnt_get_count(p); 12881da177e4SLinus Torvalds minimum_refs += 2; 12891da177e4SLinus Torvalds } 1290719ea2fbSAl Viro unlock_mount_hash(); 12911da177e4SLinus Torvalds 12921da177e4SLinus Torvalds if (actual_refs > minimum_refs) 12931da177e4SLinus Torvalds return 0; 1294e3474a8eSIan Kent 1295e3474a8eSIan Kent return 1; 12961da177e4SLinus Torvalds } 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds /** 13011da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13021da177e4SLinus Torvalds * @mnt: root of mount 13031da177e4SLinus Torvalds * 13041da177e4SLinus Torvalds * This is called to check if a mount point has any 13051da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13061da177e4SLinus Torvalds * mount has sub mounts this will return busy 13071da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13081da177e4SLinus Torvalds * 13091da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13101da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13111da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13121da177e4SLinus Torvalds */ 13131da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13141da177e4SLinus Torvalds { 1315e3474a8eSIan Kent int ret = 1; 13168ad08d8aSAl Viro down_read(&namespace_sem); 1317719ea2fbSAl Viro lock_mount_hash(); 13181ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1319e3474a8eSIan Kent ret = 0; 1320719ea2fbSAl Viro unlock_mount_hash(); 13218ad08d8aSAl Viro up_read(&namespace_sem); 1322a05964f3SRam Pai return ret; 13231da177e4SLinus Torvalds } 13241da177e4SLinus Torvalds 13251da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13261da177e4SLinus Torvalds 132738129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1328e3197d83SAl Viro 132997216be0SAl Viro static void namespace_unlock(void) 13301da177e4SLinus Torvalds { 1331a3b3c562SEric W. Biederman struct hlist_head head; 133297216be0SAl Viro 1333a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1334a3b3c562SEric W. Biederman 133597216be0SAl Viro up_write(&namespace_sem); 1336a3b3c562SEric W. Biederman 1337a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 133897216be0SAl Viro return; 133997216be0SAl Viro 134022cb7405SNeilBrown synchronize_rcu_expedited(); 134148a066e7SAl Viro 134287b95ce0SAl Viro group_pin_kill(&head); 134370fbcdf4SRam Pai } 134470fbcdf4SRam Pai 134597216be0SAl Viro static inline void namespace_lock(void) 1346e3197d83SAl Viro { 134797216be0SAl Viro down_write(&namespace_sem); 1348e3197d83SAl Viro } 1349e3197d83SAl Viro 1350e819f152SEric W. Biederman enum umount_tree_flags { 1351e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1352e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1353e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1354e819f152SEric W. Biederman }; 1355f2d0a123SEric W. Biederman 1356f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1357f2d0a123SEric W. Biederman { 1358f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1359f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1360f2d0a123SEric W. Biederman return true; 1361f2d0a123SEric W. Biederman 1362f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1363f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1364f2d0a123SEric W. Biederman return true; 1365f2d0a123SEric W. Biederman 1366f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1367f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1368f2d0a123SEric W. Biederman * connected to mounted mounts. 1369f2d0a123SEric W. Biederman */ 1370f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1371f2d0a123SEric W. Biederman return true; 1372f2d0a123SEric W. Biederman 1373f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1374f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1375f2d0a123SEric W. Biederman return false; 1376f2d0a123SEric W. Biederman 1377f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1378f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1379f2d0a123SEric W. Biederman return false; 1380f2d0a123SEric W. Biederman 1381f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1382f2d0a123SEric W. Biederman return true; 1383f2d0a123SEric W. Biederman } 1384f2d0a123SEric W. Biederman 138599b7db7bSNick Piggin /* 138648a066e7SAl Viro * mount_lock must be held 138799b7db7bSNick Piggin * namespace_sem must be held for write 138899b7db7bSNick Piggin */ 1389e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 139070fbcdf4SRam Pai { 1391c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1392315fc83eSAl Viro struct mount *p; 139370fbcdf4SRam Pai 13945d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 13955d88457eSEric W. Biederman propagate_mount_unlock(mnt); 13965d88457eSEric W. Biederman 1397c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1398590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1399590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1400c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1401590ce4bcSEric W. Biederman } 1402c003b26fSEric W. Biederman 1403411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1404c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1405c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 140638129a13SAl Viro } 140770fbcdf4SRam Pai 1408c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1409e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14107b8a53fdSAl Viro propagate_umount(&tmp_list); 1411a05964f3SRam Pai 1412c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1413d2921684SEric W. Biederman struct mnt_namespace *ns; 1414ce07d891SEric W. Biederman bool disconnect; 1415c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 14166776db3dSAl Viro list_del_init(&p->mnt_expire); 14171a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1418d2921684SEric W. Biederman ns = p->mnt_ns; 1419d2921684SEric W. Biederman if (ns) { 1420d2921684SEric W. Biederman ns->mounts--; 1421d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1422d2921684SEric W. Biederman } 142363d37a84SAl Viro p->mnt_ns = NULL; 1424e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 142548a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 142687b95ce0SAl Viro 1427f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1428ce07d891SEric W. Biederman 1429ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1430ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1431676da58dSAl Viro if (mnt_has_parent(p)) { 143281b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1433ce07d891SEric W. Biederman if (!disconnect) { 1434ce07d891SEric W. Biederman /* Don't forget about p */ 1435ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1436ce07d891SEric W. Biederman } else { 14376a46c573SEric W. Biederman umount_mnt(p); 14387c4b93d8SAl Viro } 1439ce07d891SEric W. Biederman } 14400f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 144138129a13SAl Viro } 14421da177e4SLinus Torvalds } 14431da177e4SLinus Torvalds 1444b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1445c35038beSAl Viro 14461ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 14471da177e4SLinus Torvalds { 14481ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 14491da177e4SLinus Torvalds int retval; 14501da177e4SLinus Torvalds 14511ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 14521da177e4SLinus Torvalds if (retval) 14531da177e4SLinus Torvalds return retval; 14541da177e4SLinus Torvalds 14551da177e4SLinus Torvalds /* 14561da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 14571da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 14581da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 14591da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 14601da177e4SLinus Torvalds */ 14611da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 14621ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 14631da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 14641da177e4SLinus Torvalds return -EINVAL; 14651da177e4SLinus Torvalds 1466b3e19d92SNick Piggin /* 1467b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1468b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1469b3e19d92SNick Piggin */ 1470719ea2fbSAl Viro lock_mount_hash(); 147183adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1472719ea2fbSAl Viro unlock_mount_hash(); 14731da177e4SLinus Torvalds return -EBUSY; 1474b3e19d92SNick Piggin } 1475719ea2fbSAl Viro unlock_mount_hash(); 14761da177e4SLinus Torvalds 1477863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 14781da177e4SLinus Torvalds return -EAGAIN; 14791da177e4SLinus Torvalds } 14801da177e4SLinus Torvalds 14811da177e4SLinus Torvalds /* 14821da177e4SLinus Torvalds * If we may have to abort operations to get out of this 14831da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 14841da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 14851da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 14861da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 14871da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 14881da177e4SLinus Torvalds * about for the moment. 14891da177e4SLinus Torvalds */ 14901da177e4SLinus Torvalds 149142faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 149242faad99SAl Viro sb->s_op->umount_begin(sb); 149342faad99SAl Viro } 14941da177e4SLinus Torvalds 14951da177e4SLinus Torvalds /* 14961da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 14971da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 14981da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 14991da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15001da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15011da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15021da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15031da177e4SLinus Torvalds */ 15041ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15051da177e4SLinus Torvalds /* 15061da177e4SLinus Torvalds * Special case for "unmounting" root ... 15071da177e4SLinus Torvalds * we just try to remount it readonly. 15081da177e4SLinus Torvalds */ 1509bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1510a1480dccSAndy Lutomirski return -EPERM; 15111da177e4SLinus Torvalds down_write(&sb->s_umount); 1512bc98a42cSDavid Howells if (!sb_rdonly(sb)) 1513e462ec50SDavid Howells retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); 15141da177e4SLinus Torvalds up_write(&sb->s_umount); 15151da177e4SLinus Torvalds return retval; 15161da177e4SLinus Torvalds } 15171da177e4SLinus Torvalds 151897216be0SAl Viro namespace_lock(); 1519719ea2fbSAl Viro lock_mount_hash(); 15201da177e4SLinus Torvalds 152125d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 152225d202edSEric W. Biederman retval = -EINVAL; 152325d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 152425d202edSEric W. Biederman goto out; 152525d202edSEric W. Biederman 152625d202edSEric W. Biederman event++; 152748a066e7SAl Viro if (flags & MNT_DETACH) { 152848a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1529e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 153048a066e7SAl Viro retval = 0; 153148a066e7SAl Viro } else { 1532b54b9be7SAl Viro shrink_submounts(mnt); 15331da177e4SLinus Torvalds retval = -EBUSY; 153448a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 15351a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1536e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 15371da177e4SLinus Torvalds retval = 0; 15381da177e4SLinus Torvalds } 153948a066e7SAl Viro } 154025d202edSEric W. Biederman out: 1541719ea2fbSAl Viro unlock_mount_hash(); 1542e3197d83SAl Viro namespace_unlock(); 15431da177e4SLinus Torvalds return retval; 15441da177e4SLinus Torvalds } 15451da177e4SLinus Torvalds 15461da177e4SLinus Torvalds /* 154780b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 154880b5dce8SEric W. Biederman * 154980b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 155080b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 155180b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 155280b5dce8SEric W. Biederman * leaking them. 155380b5dce8SEric W. Biederman * 155480b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 155580b5dce8SEric W. Biederman */ 155680b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 155780b5dce8SEric W. Biederman { 155880b5dce8SEric W. Biederman struct mountpoint *mp; 155980b5dce8SEric W. Biederman struct mount *mnt; 156080b5dce8SEric W. Biederman 156180b5dce8SEric W. Biederman namespace_lock(); 15623895dbf8SEric W. Biederman lock_mount_hash(); 156380b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1564f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 156580b5dce8SEric W. Biederman goto out_unlock; 156680b5dce8SEric W. Biederman 1567e06b933eSAndrey Ulanov event++; 156880b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 156980b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1570ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1571fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1572fe78fcc8SEric W. Biederman umount_mnt(mnt); 1573ce07d891SEric W. Biederman } 1574e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 157580b5dce8SEric W. Biederman } 157680b5dce8SEric W. Biederman put_mountpoint(mp); 157780b5dce8SEric W. Biederman out_unlock: 15783895dbf8SEric W. Biederman unlock_mount_hash(); 157980b5dce8SEric W. Biederman namespace_unlock(); 158080b5dce8SEric W. Biederman } 158180b5dce8SEric W. Biederman 158280b5dce8SEric W. Biederman /* 15839b40bc90SAl Viro * Is the caller allowed to modify his namespace? 15849b40bc90SAl Viro */ 15859b40bc90SAl Viro static inline bool may_mount(void) 15869b40bc90SAl Viro { 15879b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 15889b40bc90SAl Viro } 15899b40bc90SAl Viro 15909e8925b6SJeff Layton static inline bool may_mandlock(void) 15919e8925b6SJeff Layton { 15929e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 15939e8925b6SJeff Layton return false; 15949e8925b6SJeff Layton #endif 159595ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 15969e8925b6SJeff Layton } 15979e8925b6SJeff Layton 15989b40bc90SAl Viro /* 15991da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16001da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16011da177e4SLinus Torvalds * 16021da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16031da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16041da177e4SLinus Torvalds */ 16051da177e4SLinus Torvalds 16063a18ef5cSDominik Brodowski int ksys_umount(char __user *name, int flags) 16071da177e4SLinus Torvalds { 16082d8f3038SAl Viro struct path path; 1609900148dcSAl Viro struct mount *mnt; 16101da177e4SLinus Torvalds int retval; 1611db1f05bbSMiklos Szeredi int lookup_flags = 0; 16121da177e4SLinus Torvalds 1613db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1614db1f05bbSMiklos Szeredi return -EINVAL; 1615db1f05bbSMiklos Szeredi 16169b40bc90SAl Viro if (!may_mount()) 16179b40bc90SAl Viro return -EPERM; 16189b40bc90SAl Viro 1619db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1620db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1621db1f05bbSMiklos Szeredi 1622197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 16231da177e4SLinus Torvalds if (retval) 16241da177e4SLinus Torvalds goto out; 1625900148dcSAl Viro mnt = real_mount(path.mnt); 16261da177e4SLinus Torvalds retval = -EINVAL; 16272d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 16281da177e4SLinus Torvalds goto dput_and_out; 1629143c8c91SAl Viro if (!check_mnt(mnt)) 16301da177e4SLinus Torvalds goto dput_and_out; 163125d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 16325ff9d8a6SEric W. Biederman goto dput_and_out; 1633b2f5d4dcSEric W. Biederman retval = -EPERM; 1634b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1635b2f5d4dcSEric W. Biederman goto dput_and_out; 16361da177e4SLinus Torvalds 1637900148dcSAl Viro retval = do_umount(mnt, flags); 16381da177e4SLinus Torvalds dput_and_out: 1639429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 16402d8f3038SAl Viro dput(path.dentry); 1641900148dcSAl Viro mntput_no_expire(mnt); 16421da177e4SLinus Torvalds out: 16431da177e4SLinus Torvalds return retval; 16441da177e4SLinus Torvalds } 16451da177e4SLinus Torvalds 16463a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 16473a18ef5cSDominik Brodowski { 16483a18ef5cSDominik Brodowski return ksys_umount(name, flags); 16493a18ef5cSDominik Brodowski } 16503a18ef5cSDominik Brodowski 16511da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 16521da177e4SLinus Torvalds 16531da177e4SLinus Torvalds /* 16541da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 16551da177e4SLinus Torvalds */ 1656bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 16571da177e4SLinus Torvalds { 16583a18ef5cSDominik Brodowski return ksys_umount(name, 0); 16591da177e4SLinus Torvalds } 16601da177e4SLinus Torvalds 16611da177e4SLinus Torvalds #endif 16621da177e4SLinus Torvalds 16634ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 16648823c079SEric W. Biederman { 16654ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1666e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1667e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 16684ce5d2b1SEric W. Biederman } 16694ce5d2b1SEric W. Biederman 167058be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 167158be2825SAl Viro { 167258be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 167358be2825SAl Viro } 167458be2825SAl Viro 16754ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 16764ce5d2b1SEric W. Biederman { 16774ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 16784ce5d2b1SEric W. Biederman * mount namespace loop? 16794ce5d2b1SEric W. Biederman */ 16804ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 16814ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 16824ce5d2b1SEric W. Biederman return false; 16834ce5d2b1SEric W. Biederman 1684f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 16858823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 16868823c079SEric W. Biederman } 16878823c079SEric W. Biederman 168887129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 168936341f64SRam Pai int flag) 16901da177e4SLinus Torvalds { 169184d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 16921da177e4SLinus Torvalds 16934ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 16944ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 16954ce5d2b1SEric W. Biederman 16964ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1697be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 16989676f0c6SRam Pai 169936341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1700be34d1a3SDavid Howells if (IS_ERR(q)) 1701be34d1a3SDavid Howells return q; 1702be34d1a3SDavid Howells 1703a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17041da177e4SLinus Torvalds 17051da177e4SLinus Torvalds p = mnt; 17066b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1707315fc83eSAl Viro struct mount *s; 17087ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17091da177e4SLinus Torvalds continue; 17101da177e4SLinus Torvalds 1711909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 17124ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17134ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1714df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1715df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1716df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1717df7342b2SEric W. Biederman goto out; 1718df7342b2SEric W. Biederman } else { 17194ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17204ce5d2b1SEric W. Biederman continue; 17214ce5d2b1SEric W. Biederman } 1722df7342b2SEric W. Biederman } 17234ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17244ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17259676f0c6SRam Pai s = skip_mnt_tree(s); 17269676f0c6SRam Pai continue; 17279676f0c6SRam Pai } 17280714a533SAl Viro while (p != s->mnt_parent) { 17290714a533SAl Viro p = p->mnt_parent; 17300714a533SAl Viro q = q->mnt_parent; 17311da177e4SLinus Torvalds } 173287129cc0SAl Viro p = s; 173384d17192SAl Viro parent = q; 173487129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1735be34d1a3SDavid Howells if (IS_ERR(q)) 1736be34d1a3SDavid Howells goto out; 1737719ea2fbSAl Viro lock_mount_hash(); 17381a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 17391064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1740719ea2fbSAl Viro unlock_mount_hash(); 17411da177e4SLinus Torvalds } 17421da177e4SLinus Torvalds } 17431da177e4SLinus Torvalds return res; 1744be34d1a3SDavid Howells out: 17451da177e4SLinus Torvalds if (res) { 1746719ea2fbSAl Viro lock_mount_hash(); 1747e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1748719ea2fbSAl Viro unlock_mount_hash(); 17491da177e4SLinus Torvalds } 1750be34d1a3SDavid Howells return q; 17511da177e4SLinus Torvalds } 17521da177e4SLinus Torvalds 1753be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1754be34d1a3SDavid Howells 1755ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 17568aec0809SAl Viro { 1757cb338d06SAl Viro struct mount *tree; 175897216be0SAl Viro namespace_lock(); 1759cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1760cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1761cd4a4017SEric W. Biederman else 176287129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 176387129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1764328e6d90SAl Viro namespace_unlock(); 1765be34d1a3SDavid Howells if (IS_ERR(tree)) 176652e220d3SDan Carpenter return ERR_CAST(tree); 1767be34d1a3SDavid Howells return &tree->mnt; 17688aec0809SAl Viro } 17698aec0809SAl Viro 17708aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 17718aec0809SAl Viro { 177297216be0SAl Viro namespace_lock(); 1773719ea2fbSAl Viro lock_mount_hash(); 17749c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1775719ea2fbSAl Viro unlock_mount_hash(); 17763ab6abeeSAl Viro namespace_unlock(); 17778aec0809SAl Viro } 17788aec0809SAl Viro 1779c771d683SMiklos Szeredi /** 1780c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1781c771d683SMiklos Szeredi * 1782c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1783c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1784c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1785c771d683SMiklos Szeredi * 1786c771d683SMiklos Szeredi * Release with mntput(). 1787c771d683SMiklos Szeredi */ 1788ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1789c771d683SMiklos Szeredi { 1790c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1791c771d683SMiklos Szeredi struct mount *new_mnt; 1792c771d683SMiklos Szeredi 1793c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1794c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1795c771d683SMiklos Szeredi 1796c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1797c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1798c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1799c771d683SMiklos Szeredi 1800c771d683SMiklos Szeredi return &new_mnt->mnt; 1801c771d683SMiklos Szeredi } 1802c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1803c771d683SMiklos Szeredi 18041f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18051f707137SAl Viro struct vfsmount *root) 18061f707137SAl Viro { 18071a4eeaf2SAl Viro struct mount *mnt; 18081f707137SAl Viro int res = f(root, arg); 18091f707137SAl Viro if (res) 18101f707137SAl Viro return res; 18111a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18121a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18131f707137SAl Viro if (res) 18141f707137SAl Viro return res; 18151f707137SAl Viro } 18161f707137SAl Viro return 0; 18171f707137SAl Viro } 18181f707137SAl Viro 18193bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 18203bd045ccSAl Viro { 18213bd045ccSAl Viro struct mount *p; 18223bd045ccSAl Viro 18233bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 18243bd045ccSAl Viro int flags = p->mnt.mnt_flags; 18253bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 18263bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 18273bd045ccSAl Viro 18283bd045ccSAl Viro if (flags & MNT_READONLY) 18293bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 18303bd045ccSAl Viro 18313bd045ccSAl Viro if (flags & MNT_NODEV) 18323bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 18333bd045ccSAl Viro 18343bd045ccSAl Viro if (flags & MNT_NOSUID) 18353bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 18363bd045ccSAl Viro 18373bd045ccSAl Viro if (flags & MNT_NOEXEC) 18383bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 18393bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 18403bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 18413bd045ccSAl Viro flags |= MNT_LOCKED; 18423bd045ccSAl Viro p->mnt.mnt_flags = flags; 18433bd045ccSAl Viro } 18443bd045ccSAl Viro } 18453bd045ccSAl Viro 18464b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1847719f5d7fSMiklos Szeredi { 1848315fc83eSAl Viro struct mount *p; 1849719f5d7fSMiklos Szeredi 1850909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1851fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 18524b8b21f4SAl Viro mnt_release_group_id(p); 1853719f5d7fSMiklos Szeredi } 1854719f5d7fSMiklos Szeredi } 1855719f5d7fSMiklos Szeredi 18564b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1857719f5d7fSMiklos Szeredi { 1858315fc83eSAl Viro struct mount *p; 1859719f5d7fSMiklos Szeredi 1860909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1861fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 18624b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1863719f5d7fSMiklos Szeredi if (err) { 18644b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1865719f5d7fSMiklos Szeredi return err; 1866719f5d7fSMiklos Szeredi } 1867719f5d7fSMiklos Szeredi } 1868719f5d7fSMiklos Szeredi } 1869719f5d7fSMiklos Szeredi 1870719f5d7fSMiklos Szeredi return 0; 1871719f5d7fSMiklos Szeredi } 1872719f5d7fSMiklos Szeredi 1873d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1874d2921684SEric W. Biederman { 1875d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1876d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1877d2921684SEric W. Biederman struct mount *p; 1878d2921684SEric W. Biederman 1879d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1880d2921684SEric W. Biederman mounts++; 1881d2921684SEric W. Biederman 1882d2921684SEric W. Biederman old = ns->mounts; 1883d2921684SEric W. Biederman pending = ns->pending_mounts; 1884d2921684SEric W. Biederman sum = old + pending; 1885d2921684SEric W. Biederman if ((old > sum) || 1886d2921684SEric W. Biederman (pending > sum) || 1887d2921684SEric W. Biederman (max < sum) || 1888d2921684SEric W. Biederman (mounts > (max - sum))) 1889d2921684SEric W. Biederman return -ENOSPC; 1890d2921684SEric W. Biederman 1891d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1892d2921684SEric W. Biederman return 0; 1893d2921684SEric W. Biederman } 1894d2921684SEric W. Biederman 1895b90fa9aeSRam Pai /* 1896b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1897b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 189821444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 189921444403SRam Pai * store the parent mount and mountpoint dentry. 190021444403SRam Pai * (done when source_mnt is moved) 1901b90fa9aeSRam Pai * 1902b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1903b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 19049676f0c6SRam Pai * --------------------------------------------------------------------------- 1905b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 19069676f0c6SRam Pai * |************************************************************************** 19079676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19089676f0c6SRam Pai * | dest | | | | | 19099676f0c6SRam Pai * | | | | | | | 19109676f0c6SRam Pai * | v | | | | | 19119676f0c6SRam Pai * |************************************************************************** 19129676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 19135afe0022SRam Pai * | | | | | | 19149676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 19159676f0c6SRam Pai * *************************************************************************** 1916b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1917b90fa9aeSRam Pai * destination mount. 1918b90fa9aeSRam Pai * 1919b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1920b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1921b90fa9aeSRam Pai * the peer group of the source mount. 1922b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1923b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1924b90fa9aeSRam Pai * mount. 19255afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 19265afe0022SRam Pai * of the destination mount and the cloned mount is made slave 19275afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 19285afe0022SRam Pai * is marked as 'shared and slave'. 19295afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 19305afe0022SRam Pai * source mount. 19315afe0022SRam Pai * 19329676f0c6SRam Pai * --------------------------------------------------------------------------- 193321444403SRam Pai * | MOVE MOUNT OPERATION | 19349676f0c6SRam Pai * |************************************************************************** 19359676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19369676f0c6SRam Pai * | dest | | | | | 19379676f0c6SRam Pai * | | | | | | | 19389676f0c6SRam Pai * | v | | | | | 19399676f0c6SRam Pai * |************************************************************************** 19409676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 19415afe0022SRam Pai * | | | | | | 19429676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 19439676f0c6SRam Pai * *************************************************************************** 19445afe0022SRam Pai * 19455afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 19465afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 194721444403SRam Pai * (+*) the mount is moved to the destination. 19485afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 19495afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 19505afe0022SRam Pai * the mount is marked as 'shared and slave'. 19515afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1952b90fa9aeSRam Pai * 1953b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1954b90fa9aeSRam Pai * applied to each mount in the tree. 1955b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1956b90fa9aeSRam Pai * in allocations. 1957b90fa9aeSRam Pai */ 19580fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 195984d17192SAl Viro struct mount *dest_mnt, 196084d17192SAl Viro struct mountpoint *dest_mp, 196184d17192SAl Viro struct path *parent_path) 1962b90fa9aeSRam Pai { 19633bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 196438129a13SAl Viro HLIST_HEAD(tree_list); 1965d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 19661064f874SEric W. Biederman struct mountpoint *smp; 1967315fc83eSAl Viro struct mount *child, *p; 196838129a13SAl Viro struct hlist_node *n; 1969719f5d7fSMiklos Szeredi int err; 1970b90fa9aeSRam Pai 19711064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 19721064f874SEric W. Biederman * to be tucked under other mounts. 19731064f874SEric W. Biederman */ 19741064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 19751064f874SEric W. Biederman if (IS_ERR(smp)) 19761064f874SEric W. Biederman return PTR_ERR(smp); 19771064f874SEric W. Biederman 1978d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 1979d2921684SEric W. Biederman if (!parent_path) { 1980d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 1981d2921684SEric W. Biederman if (err) 1982d2921684SEric W. Biederman goto out; 1983d2921684SEric W. Biederman } 1984d2921684SEric W. Biederman 1985fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 19860fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1987719f5d7fSMiklos Szeredi if (err) 1988719f5d7fSMiklos Szeredi goto out; 198984d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1990f2ebb3a9SAl Viro lock_mount_hash(); 1991719f5d7fSMiklos Szeredi if (err) 1992719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1993909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 19940f0afb1dSAl Viro set_mnt_shared(p); 19950b1b901bSAl Viro } else { 19960b1b901bSAl Viro lock_mount_hash(); 1997b90fa9aeSRam Pai } 19981a390689SAl Viro if (parent_path) { 19990fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 200084d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2001143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 200221444403SRam Pai } else { 200384d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 20041064f874SEric W. Biederman commit_tree(source_mnt); 200521444403SRam Pai } 2006b90fa9aeSRam Pai 200738129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 20081d6a32acSAl Viro struct mount *q; 200938129a13SAl Viro hlist_del_init(&child->mnt_hash); 20101064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 20111d6a32acSAl Viro child->mnt_mountpoint); 20121064f874SEric W. Biederman if (q) 20131064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 20143bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 20153bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 20163bd045ccSAl Viro lock_mnt_tree(child); 20171064f874SEric W. Biederman commit_tree(child); 2018b90fa9aeSRam Pai } 20191064f874SEric W. Biederman put_mountpoint(smp); 2020719ea2fbSAl Viro unlock_mount_hash(); 202199b7db7bSNick Piggin 2022b90fa9aeSRam Pai return 0; 2023719f5d7fSMiklos Szeredi 2024719f5d7fSMiklos Szeredi out_cleanup_ids: 2025f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2026f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2027d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2028e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2029f2ebb3a9SAl Viro } 2030f2ebb3a9SAl Viro unlock_mount_hash(); 20310fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2032719f5d7fSMiklos Szeredi out: 2033d2921684SEric W. Biederman ns->pending_mounts = 0; 20341064f874SEric W. Biederman 20351064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 20361064f874SEric W. Biederman put_mountpoint(smp); 20371064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 20381064f874SEric W. Biederman 2039719f5d7fSMiklos Szeredi return err; 2040b90fa9aeSRam Pai } 2041b90fa9aeSRam Pai 204284d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2043b12cea91SAl Viro { 2044b12cea91SAl Viro struct vfsmount *mnt; 204584d17192SAl Viro struct dentry *dentry = path->dentry; 2046b12cea91SAl Viro retry: 20475955102cSAl Viro inode_lock(dentry->d_inode); 204884d17192SAl Viro if (unlikely(cant_mount(dentry))) { 20495955102cSAl Viro inode_unlock(dentry->d_inode); 205084d17192SAl Viro return ERR_PTR(-ENOENT); 2051b12cea91SAl Viro } 205297216be0SAl Viro namespace_lock(); 2053b12cea91SAl Viro mnt = lookup_mnt(path); 205484d17192SAl Viro if (likely(!mnt)) { 20553895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 205684d17192SAl Viro if (IS_ERR(mp)) { 205797216be0SAl Viro namespace_unlock(); 20585955102cSAl Viro inode_unlock(dentry->d_inode); 205984d17192SAl Viro return mp; 206084d17192SAl Viro } 206184d17192SAl Viro return mp; 206284d17192SAl Viro } 206397216be0SAl Viro namespace_unlock(); 20645955102cSAl Viro inode_unlock(path->dentry->d_inode); 2065b12cea91SAl Viro path_put(path); 2066b12cea91SAl Viro path->mnt = mnt; 206784d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2068b12cea91SAl Viro goto retry; 2069b12cea91SAl Viro } 2070b12cea91SAl Viro 207184d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2072b12cea91SAl Viro { 207384d17192SAl Viro struct dentry *dentry = where->m_dentry; 20743895dbf8SEric W. Biederman 20753895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 207684d17192SAl Viro put_mountpoint(where); 20773895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 20783895dbf8SEric W. Biederman 2079328e6d90SAl Viro namespace_unlock(); 20805955102cSAl Viro inode_unlock(dentry->d_inode); 2081b12cea91SAl Viro } 2082b12cea91SAl Viro 208384d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 20841da177e4SLinus Torvalds { 2085e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 20861da177e4SLinus Torvalds return -EINVAL; 20871da177e4SLinus Torvalds 2088e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2089e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 20901da177e4SLinus Torvalds return -ENOTDIR; 20911da177e4SLinus Torvalds 209284d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 20931da177e4SLinus Torvalds } 20941da177e4SLinus Torvalds 20951da177e4SLinus Torvalds /* 20967a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 20977a2e8a8fSValerie Aurora */ 20987a2e8a8fSValerie Aurora 2099e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 21007a2e8a8fSValerie Aurora { 2101e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 21027a2e8a8fSValerie Aurora 21037a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 21047a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 21057a2e8a8fSValerie Aurora return 0; 21067a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 21077a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 21087a2e8a8fSValerie Aurora return 0; 21097a2e8a8fSValerie Aurora return type; 21107a2e8a8fSValerie Aurora } 21117a2e8a8fSValerie Aurora 21127a2e8a8fSValerie Aurora /* 211307b20889SRam Pai * recursively change the type of the mountpoint. 211407b20889SRam Pai */ 2115e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 211607b20889SRam Pai { 2117315fc83eSAl Viro struct mount *m; 21184b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2119e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 21207a2e8a8fSValerie Aurora int type; 2121719f5d7fSMiklos Szeredi int err = 0; 212207b20889SRam Pai 21232d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 212407b20889SRam Pai return -EINVAL; 212507b20889SRam Pai 2126e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 21277a2e8a8fSValerie Aurora if (!type) 21287a2e8a8fSValerie Aurora return -EINVAL; 21297a2e8a8fSValerie Aurora 213097216be0SAl Viro namespace_lock(); 2131719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2132719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2133719f5d7fSMiklos Szeredi if (err) 2134719f5d7fSMiklos Szeredi goto out_unlock; 2135719f5d7fSMiklos Szeredi } 2136719f5d7fSMiklos Szeredi 2137719ea2fbSAl Viro lock_mount_hash(); 2138909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 21390f0afb1dSAl Viro change_mnt_propagation(m, type); 2140719ea2fbSAl Viro unlock_mount_hash(); 2141719f5d7fSMiklos Szeredi 2142719f5d7fSMiklos Szeredi out_unlock: 214397216be0SAl Viro namespace_unlock(); 2144719f5d7fSMiklos Szeredi return err; 214507b20889SRam Pai } 214607b20889SRam Pai 21475ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 21485ff9d8a6SEric W. Biederman { 21495ff9d8a6SEric W. Biederman struct mount *child; 21505ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 21515ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 21525ff9d8a6SEric W. Biederman continue; 21535ff9d8a6SEric W. Biederman 21545ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 21555ff9d8a6SEric W. Biederman return true; 21565ff9d8a6SEric W. Biederman } 21575ff9d8a6SEric W. Biederman return false; 21585ff9d8a6SEric W. Biederman } 21595ff9d8a6SEric W. Biederman 216007b20889SRam Pai /* 21611da177e4SLinus Torvalds * do loopback mount. 21621da177e4SLinus Torvalds */ 2163808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 21642dafe1c4SEric Sandeen int recurse) 21651da177e4SLinus Torvalds { 21662d92ab3cSAl Viro struct path old_path; 216784d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 216884d17192SAl Viro struct mountpoint *mp; 216957eccb83SAl Viro int err; 21701da177e4SLinus Torvalds if (!old_name || !*old_name) 21711da177e4SLinus Torvalds return -EINVAL; 2172815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 21731da177e4SLinus Torvalds if (err) 21741da177e4SLinus Torvalds return err; 21751da177e4SLinus Torvalds 21768823c079SEric W. Biederman err = -EINVAL; 21774ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 21788823c079SEric W. Biederman goto out; 21798823c079SEric W. Biederman 218084d17192SAl Viro mp = lock_mount(path); 218184d17192SAl Viro err = PTR_ERR(mp); 218284d17192SAl Viro if (IS_ERR(mp)) 21839676f0c6SRam Pai goto out; 21849676f0c6SRam Pai 218587129cc0SAl Viro old = real_mount(old_path.mnt); 218684d17192SAl Viro parent = real_mount(path->mnt); 218787129cc0SAl Viro 2188b12cea91SAl Viro err = -EINVAL; 2189fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2190b12cea91SAl Viro goto out2; 2191b12cea91SAl Viro 2192e149ed2bSAl Viro if (!check_mnt(parent)) 2193e149ed2bSAl Viro goto out2; 2194e149ed2bSAl Viro 2195e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2196b12cea91SAl Viro goto out2; 2197ccd48bc7SAl Viro 21985ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 21995ff9d8a6SEric W. Biederman goto out2; 22005ff9d8a6SEric W. Biederman 22011da177e4SLinus Torvalds if (recurse) 22024ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 22031da177e4SLinus Torvalds else 220487129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 22051da177e4SLinus Torvalds 2206be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2207be34d1a3SDavid Howells err = PTR_ERR(mnt); 2208e9c5d8a5SAndrey Vagin goto out2; 2209be34d1a3SDavid Howells } 2210ccd48bc7SAl Viro 22115ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 22125ff9d8a6SEric W. Biederman 221384d17192SAl Viro err = graft_tree(mnt, parent, mp); 22141da177e4SLinus Torvalds if (err) { 2215719ea2fbSAl Viro lock_mount_hash(); 2216e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2217719ea2fbSAl Viro unlock_mount_hash(); 22185b83d2c5SRam Pai } 2219b12cea91SAl Viro out2: 222084d17192SAl Viro unlock_mount(mp); 2221ccd48bc7SAl Viro out: 22222d92ab3cSAl Viro path_put(&old_path); 22231da177e4SLinus Torvalds return err; 22241da177e4SLinus Torvalds } 22251da177e4SLinus Torvalds 222643f5e655SDavid Howells /* 222743f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 222843f5e655SDavid Howells * 222943f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 223043f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 223143f5e655SDavid Howells */ 223243f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 22332e4b7fcdSDave Hansen { 223443f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 22352e4b7fcdSDave Hansen 223643f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 223743f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 223843f5e655SDavid Howells return false; 223943f5e655SDavid Howells 224043f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 224143f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 224243f5e655SDavid Howells return false; 224343f5e655SDavid Howells 224443f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 224543f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 224643f5e655SDavid Howells return false; 224743f5e655SDavid Howells 224843f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 224943f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 225043f5e655SDavid Howells return false; 225143f5e655SDavid Howells 225243f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 225343f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 225443f5e655SDavid Howells return false; 225543f5e655SDavid Howells 225643f5e655SDavid Howells return true; 225743f5e655SDavid Howells } 225843f5e655SDavid Howells 225943f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 226043f5e655SDavid Howells { 226143f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 226243f5e655SDavid Howells 226343f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 22642e4b7fcdSDave Hansen return 0; 22652e4b7fcdSDave Hansen 22662e4b7fcdSDave Hansen if (readonly_request) 226743f5e655SDavid Howells return mnt_make_readonly(mnt); 226843f5e655SDavid Howells 226943f5e655SDavid Howells return __mnt_unmake_readonly(mnt); 227043f5e655SDavid Howells } 227143f5e655SDavid Howells 227243f5e655SDavid Howells /* 227343f5e655SDavid Howells * Update the user-settable attributes on a mount. The caller must hold 227443f5e655SDavid Howells * sb->s_umount for writing. 227543f5e655SDavid Howells */ 227643f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 227743f5e655SDavid Howells { 227843f5e655SDavid Howells lock_mount_hash(); 227943f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 228043f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 228143f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 228243f5e655SDavid Howells unlock_mount_hash(); 228343f5e655SDavid Howells } 228443f5e655SDavid Howells 228543f5e655SDavid Howells /* 228643f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 228743f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 228843f5e655SDavid Howells * to mount(2). 228943f5e655SDavid Howells */ 229043f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 229143f5e655SDavid Howells { 229243f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 229343f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 229443f5e655SDavid Howells int ret; 229543f5e655SDavid Howells 229643f5e655SDavid Howells if (!check_mnt(mnt)) 229743f5e655SDavid Howells return -EINVAL; 229843f5e655SDavid Howells 229943f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 230043f5e655SDavid Howells return -EINVAL; 230143f5e655SDavid Howells 230243f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 230343f5e655SDavid Howells return -EPERM; 230443f5e655SDavid Howells 230543f5e655SDavid Howells down_write(&sb->s_umount); 230643f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 230743f5e655SDavid Howells if (ret == 0) 230843f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 230943f5e655SDavid Howells up_write(&sb->s_umount); 231043f5e655SDavid Howells return ret; 23112e4b7fcdSDave Hansen } 23122e4b7fcdSDave Hansen 23131da177e4SLinus Torvalds /* 23141da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 23151da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 23161da177e4SLinus Torvalds * on it - tough luck. 23171da177e4SLinus Torvalds */ 2318e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2319e462ec50SDavid Howells int mnt_flags, void *data) 23201da177e4SLinus Torvalds { 23211da177e4SLinus Torvalds int err; 23222d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2323143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 2324204cc0ccSAl Viro void *sec_opts = NULL; 23251da177e4SLinus Torvalds 2326143c8c91SAl Viro if (!check_mnt(mnt)) 23271da177e4SLinus Torvalds return -EINVAL; 23281da177e4SLinus Torvalds 23292d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 23301da177e4SLinus Torvalds return -EINVAL; 23311da177e4SLinus Torvalds 233243f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 233307b64558SEric W. Biederman return -EPERM; 23349566d674SEric W. Biederman 2335c039bc3cSAl Viro if (data && !(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)) { 2336204cc0ccSAl Viro err = security_sb_eat_lsm_opts(data, &sec_opts); 2337c039bc3cSAl Viro if (err) 2338c039bc3cSAl Viro return err; 2339c039bc3cSAl Viro } 2340204cc0ccSAl Viro err = security_sb_remount(sb, sec_opts); 2341204cc0ccSAl Viro security_free_mnt_opts(&sec_opts); 2342ff36fe2cSEric Paris if (err) 2343ff36fe2cSEric Paris return err; 2344ff36fe2cSEric Paris 23451da177e4SLinus Torvalds down_write(&sb->s_umount); 234657eccb83SAl Viro err = -EPERM; 234743f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 2348e462ec50SDavid Howells err = do_remount_sb(sb, sb_flags, data, 0); 234943f5e655SDavid Howells if (!err) 235043f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 23510e55a7ccSDan Williams } 23526339dab8SAl Viro up_write(&sb->s_umount); 23531da177e4SLinus Torvalds return err; 23541da177e4SLinus Torvalds } 23551da177e4SLinus Torvalds 2356cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 23579676f0c6SRam Pai { 2358315fc83eSAl Viro struct mount *p; 2359909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2360fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 23619676f0c6SRam Pai return 1; 23629676f0c6SRam Pai } 23639676f0c6SRam Pai return 0; 23649676f0c6SRam Pai } 23659676f0c6SRam Pai 2366808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 23671da177e4SLinus Torvalds { 23682d92ab3cSAl Viro struct path old_path, parent_path; 2369676da58dSAl Viro struct mount *p; 23700fb54e50SAl Viro struct mount *old; 237184d17192SAl Viro struct mountpoint *mp; 237257eccb83SAl Viro int err; 23731da177e4SLinus Torvalds if (!old_name || !*old_name) 23741da177e4SLinus Torvalds return -EINVAL; 23752d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 23761da177e4SLinus Torvalds if (err) 23771da177e4SLinus Torvalds return err; 23781da177e4SLinus Torvalds 237984d17192SAl Viro mp = lock_mount(path); 238084d17192SAl Viro err = PTR_ERR(mp); 238184d17192SAl Viro if (IS_ERR(mp)) 2382cc53ce53SDavid Howells goto out; 2383cc53ce53SDavid Howells 2384143c8c91SAl Viro old = real_mount(old_path.mnt); 2385fc7be130SAl Viro p = real_mount(path->mnt); 2386143c8c91SAl Viro 23871da177e4SLinus Torvalds err = -EINVAL; 2388fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 23891da177e4SLinus Torvalds goto out1; 23901da177e4SLinus Torvalds 23915ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 23925ff9d8a6SEric W. Biederman goto out1; 23935ff9d8a6SEric W. Biederman 23941da177e4SLinus Torvalds err = -EINVAL; 23952d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 239621444403SRam Pai goto out1; 23971da177e4SLinus Torvalds 2398676da58dSAl Viro if (!mnt_has_parent(old)) 239921444403SRam Pai goto out1; 24001da177e4SLinus Torvalds 2401e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2402e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 240321444403SRam Pai goto out1; 240421444403SRam Pai /* 240521444403SRam Pai * Don't move a mount residing in a shared parent. 240621444403SRam Pai */ 2407fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 240821444403SRam Pai goto out1; 24099676f0c6SRam Pai /* 24109676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 24119676f0c6SRam Pai * mount which is shared. 24129676f0c6SRam Pai */ 2413fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 24149676f0c6SRam Pai goto out1; 24151da177e4SLinus Torvalds err = -ELOOP; 2416fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2417676da58dSAl Viro if (p == old) 241821444403SRam Pai goto out1; 24191da177e4SLinus Torvalds 242084d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 24214ac91378SJan Blunck if (err) 242221444403SRam Pai goto out1; 24231da177e4SLinus Torvalds 24241da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 24251da177e4SLinus Torvalds * automatically */ 24266776db3dSAl Viro list_del_init(&old->mnt_expire); 24271da177e4SLinus Torvalds out1: 242884d17192SAl Viro unlock_mount(mp); 24291da177e4SLinus Torvalds out: 24301da177e4SLinus Torvalds if (!err) 24311a390689SAl Viro path_put(&parent_path); 24322d92ab3cSAl Viro path_put(&old_path); 24331da177e4SLinus Torvalds return err; 24341da177e4SLinus Torvalds } 24351da177e4SLinus Torvalds 24369d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 24379d412a43SAl Viro { 24389d412a43SAl Viro int err; 24399d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 24409d412a43SAl Viro if (subtype) { 24419d412a43SAl Viro subtype++; 24429d412a43SAl Viro err = -EINVAL; 24439d412a43SAl Viro if (!subtype[0]) 24449d412a43SAl Viro goto err; 24459d412a43SAl Viro } else 24469d412a43SAl Viro subtype = ""; 24479d412a43SAl Viro 24489d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 24499d412a43SAl Viro err = -ENOMEM; 24509d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 24519d412a43SAl Viro goto err; 24529d412a43SAl Viro return mnt; 24539d412a43SAl Viro 24549d412a43SAl Viro err: 24559d412a43SAl Viro mntput(mnt); 24569d412a43SAl Viro return ERR_PTR(err); 24579d412a43SAl Viro } 24589d412a43SAl Viro 24599d412a43SAl Viro /* 24609d412a43SAl Viro * add a mount into a namespace's mount tree 24619d412a43SAl Viro */ 246295bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 24639d412a43SAl Viro { 246484d17192SAl Viro struct mountpoint *mp; 246584d17192SAl Viro struct mount *parent; 24669d412a43SAl Viro int err; 24679d412a43SAl Viro 2468f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 24699d412a43SAl Viro 247084d17192SAl Viro mp = lock_mount(path); 247184d17192SAl Viro if (IS_ERR(mp)) 247284d17192SAl Viro return PTR_ERR(mp); 24739d412a43SAl Viro 247484d17192SAl Viro parent = real_mount(path->mnt); 24759d412a43SAl Viro err = -EINVAL; 247684d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2477156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2478156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 24799d412a43SAl Viro goto unlock; 2480156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 248184d17192SAl Viro if (!parent->mnt_ns) 2482156cacb1SAl Viro goto unlock; 2483156cacb1SAl Viro } 24849d412a43SAl Viro 24859d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 24869d412a43SAl Viro err = -EBUSY; 248795bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 24889d412a43SAl Viro path->mnt->mnt_root == path->dentry) 24899d412a43SAl Viro goto unlock; 24909d412a43SAl Viro 24919d412a43SAl Viro err = -EINVAL; 2492e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 24939d412a43SAl Viro goto unlock; 24949d412a43SAl Viro 249595bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 249684d17192SAl Viro err = graft_tree(newmnt, parent, mp); 24979d412a43SAl Viro 24989d412a43SAl Viro unlock: 249984d17192SAl Viro unlock_mount(mp); 25009d412a43SAl Viro return err; 25019d412a43SAl Viro } 2502b1e75df4SAl Viro 25038654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); 25041b852bceSEric W. Biederman 25051da177e4SLinus Torvalds /* 25061da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 25071da177e4SLinus Torvalds * namespace's tree 25081da177e4SLinus Torvalds */ 2509e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2510808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 25111da177e4SLinus Torvalds { 25120c55cfc4SEric W. Biederman struct file_system_type *type; 25131da177e4SLinus Torvalds struct vfsmount *mnt; 251415f9a3f3SAl Viro int err; 25151da177e4SLinus Torvalds 25160c55cfc4SEric W. Biederman if (!fstype) 25171da177e4SLinus Torvalds return -EINVAL; 25181da177e4SLinus Torvalds 25190c55cfc4SEric W. Biederman type = get_fs_type(fstype); 25200c55cfc4SEric W. Biederman if (!type) 25210c55cfc4SEric W. Biederman return -ENODEV; 25220c55cfc4SEric W. Biederman 2523e462ec50SDavid Howells mnt = vfs_kern_mount(type, sb_flags, name, data); 25240c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 25250c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 25260c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 25270c55cfc4SEric W. Biederman 25280c55cfc4SEric W. Biederman put_filesystem(type); 25291da177e4SLinus Torvalds if (IS_ERR(mnt)) 25301da177e4SLinus Torvalds return PTR_ERR(mnt); 25311da177e4SLinus Torvalds 25328654df4eSEric W. Biederman if (mount_too_revealing(mnt, &mnt_flags)) { 25338654df4eSEric W. Biederman mntput(mnt); 25348654df4eSEric W. Biederman return -EPERM; 25358654df4eSEric W. Biederman } 25368654df4eSEric W. Biederman 253795bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 253815f9a3f3SAl Viro if (err) 253915f9a3f3SAl Viro mntput(mnt); 254015f9a3f3SAl Viro return err; 25411da177e4SLinus Torvalds } 25421da177e4SLinus Torvalds 254319a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 254419a167afSAl Viro { 25456776db3dSAl Viro struct mount *mnt = real_mount(m); 254619a167afSAl Viro int err; 254719a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 254819a167afSAl Viro * expired before we get a chance to add it 254919a167afSAl Viro */ 25506776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 255119a167afSAl Viro 255219a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 255319a167afSAl Viro m->mnt_root == path->dentry) { 2554b1e75df4SAl Viro err = -ELOOP; 2555b1e75df4SAl Viro goto fail; 255619a167afSAl Viro } 255719a167afSAl Viro 255895bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2559b1e75df4SAl Viro if (!err) 2560b1e75df4SAl Viro return 0; 2561b1e75df4SAl Viro fail: 2562b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 25636776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 256497216be0SAl Viro namespace_lock(); 25656776db3dSAl Viro list_del_init(&mnt->mnt_expire); 256697216be0SAl Viro namespace_unlock(); 256719a167afSAl Viro } 2568b1e75df4SAl Viro mntput(m); 2569b1e75df4SAl Viro mntput(m); 257019a167afSAl Viro return err; 257119a167afSAl Viro } 257219a167afSAl Viro 2573ea5b778aSDavid Howells /** 2574ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2575ea5b778aSDavid Howells * @mnt: The mount to list. 2576ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2577ea5b778aSDavid Howells */ 2578ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2579ea5b778aSDavid Howells { 258097216be0SAl Viro namespace_lock(); 2581ea5b778aSDavid Howells 25826776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2583ea5b778aSDavid Howells 258497216be0SAl Viro namespace_unlock(); 2585ea5b778aSDavid Howells } 2586ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2587ea5b778aSDavid Howells 2588ea5b778aSDavid Howells /* 25891da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 25901da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 25911da177e4SLinus Torvalds * here 25921da177e4SLinus Torvalds */ 25931da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 25941da177e4SLinus Torvalds { 2595761d5c38SAl Viro struct mount *mnt, *next; 25961da177e4SLinus Torvalds LIST_HEAD(graveyard); 25971da177e4SLinus Torvalds 25981da177e4SLinus Torvalds if (list_empty(mounts)) 25991da177e4SLinus Torvalds return; 26001da177e4SLinus Torvalds 260197216be0SAl Viro namespace_lock(); 2602719ea2fbSAl Viro lock_mount_hash(); 26031da177e4SLinus Torvalds 26041da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 26051da177e4SLinus Torvalds * following criteria: 26061da177e4SLinus Torvalds * - only referenced by its parent vfsmount 26071da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 26081da177e4SLinus Torvalds * cleared by mntput()) 26091da177e4SLinus Torvalds */ 26106776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2611863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 26121ab59738SAl Viro propagate_mount_busy(mnt, 1)) 26131da177e4SLinus Torvalds continue; 26146776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 26151da177e4SLinus Torvalds } 2616bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 26176776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2618143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2619e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2620bcc5c7d2SAl Viro } 2621719ea2fbSAl Viro unlock_mount_hash(); 26223ab6abeeSAl Viro namespace_unlock(); 26231da177e4SLinus Torvalds } 26241da177e4SLinus Torvalds 26251da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 26261da177e4SLinus Torvalds 26271da177e4SLinus Torvalds /* 26285528f911STrond Myklebust * Ripoff of 'select_parent()' 26295528f911STrond Myklebust * 26305528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 26315528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 26325528f911STrond Myklebust */ 2633692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 26345528f911STrond Myklebust { 2635692afc31SAl Viro struct mount *this_parent = parent; 26365528f911STrond Myklebust struct list_head *next; 26375528f911STrond Myklebust int found = 0; 26385528f911STrond Myklebust 26395528f911STrond Myklebust repeat: 26406b41d536SAl Viro next = this_parent->mnt_mounts.next; 26415528f911STrond Myklebust resume: 26426b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 26435528f911STrond Myklebust struct list_head *tmp = next; 26446b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 26455528f911STrond Myklebust 26465528f911STrond Myklebust next = tmp->next; 2647692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 26485528f911STrond Myklebust continue; 26495528f911STrond Myklebust /* 26505528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 26515528f911STrond Myklebust */ 26526b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 26535528f911STrond Myklebust this_parent = mnt; 26545528f911STrond Myklebust goto repeat; 26555528f911STrond Myklebust } 26565528f911STrond Myklebust 26571ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 26586776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 26595528f911STrond Myklebust found++; 26605528f911STrond Myklebust } 26615528f911STrond Myklebust } 26625528f911STrond Myklebust /* 26635528f911STrond Myklebust * All done at this level ... ascend and resume the search 26645528f911STrond Myklebust */ 26655528f911STrond Myklebust if (this_parent != parent) { 26666b41d536SAl Viro next = this_parent->mnt_child.next; 26670714a533SAl Viro this_parent = this_parent->mnt_parent; 26685528f911STrond Myklebust goto resume; 26695528f911STrond Myklebust } 26705528f911STrond Myklebust return found; 26715528f911STrond Myklebust } 26725528f911STrond Myklebust 26735528f911STrond Myklebust /* 26745528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 26755528f911STrond Myklebust * submounts of a specific parent mountpoint 267699b7db7bSNick Piggin * 267748a066e7SAl Viro * mount_lock must be held for write 26785528f911STrond Myklebust */ 2679b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 26805528f911STrond Myklebust { 26815528f911STrond Myklebust LIST_HEAD(graveyard); 2682761d5c38SAl Viro struct mount *m; 26835528f911STrond Myklebust 26845528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2685c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2686bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2687761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 26886776db3dSAl Viro mnt_expire); 2689143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2690e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2691bcc5c7d2SAl Viro } 2692bcc5c7d2SAl Viro } 26935528f911STrond Myklebust } 26945528f911STrond Myklebust 26955528f911STrond Myklebust /* 26961da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 26971da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 26981da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 26991da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 27001da177e4SLinus Torvalds */ 2701b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2702b58fed8bSRam Pai unsigned long n) 27031da177e4SLinus Torvalds { 27041da177e4SLinus Torvalds char *t = to; 27051da177e4SLinus Torvalds const char __user *f = from; 27061da177e4SLinus Torvalds char c; 27071da177e4SLinus Torvalds 270896d4f267SLinus Torvalds if (!access_ok(from, n)) 27091da177e4SLinus Torvalds return n; 27101da177e4SLinus Torvalds 27119da3f2b7SJann Horn current->kernel_uaccess_faults_ok++; 27121da177e4SLinus Torvalds while (n) { 27131da177e4SLinus Torvalds if (__get_user(c, f)) { 27141da177e4SLinus Torvalds memset(t, 0, n); 27151da177e4SLinus Torvalds break; 27161da177e4SLinus Torvalds } 27171da177e4SLinus Torvalds *t++ = c; 27181da177e4SLinus Torvalds f++; 27191da177e4SLinus Torvalds n--; 27201da177e4SLinus Torvalds } 27219da3f2b7SJann Horn current->kernel_uaccess_faults_ok--; 27221da177e4SLinus Torvalds return n; 27231da177e4SLinus Torvalds } 27241da177e4SLinus Torvalds 2725b40ef869SAl Viro void *copy_mount_options(const void __user * data) 27261da177e4SLinus Torvalds { 27271da177e4SLinus Torvalds int i; 27281da177e4SLinus Torvalds unsigned long size; 2729b40ef869SAl Viro char *copy; 27301da177e4SLinus Torvalds 27311da177e4SLinus Torvalds if (!data) 2732b40ef869SAl Viro return NULL; 27331da177e4SLinus Torvalds 2734b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2735b40ef869SAl Viro if (!copy) 2736b40ef869SAl Viro return ERR_PTR(-ENOMEM); 27371da177e4SLinus Torvalds 27381da177e4SLinus Torvalds /* We only care that *some* data at the address the user 27391da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 27401da177e4SLinus Torvalds * the remainder of the page. 27411da177e4SLinus Torvalds */ 27421da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 27431da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 27441da177e4SLinus Torvalds if (size > PAGE_SIZE) 27451da177e4SLinus Torvalds size = PAGE_SIZE; 27461da177e4SLinus Torvalds 2747b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 27481da177e4SLinus Torvalds if (!i) { 2749b40ef869SAl Viro kfree(copy); 2750b40ef869SAl Viro return ERR_PTR(-EFAULT); 27511da177e4SLinus Torvalds } 27521da177e4SLinus Torvalds if (i != PAGE_SIZE) 2753b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2754b40ef869SAl Viro return copy; 27551da177e4SLinus Torvalds } 27561da177e4SLinus Torvalds 2757b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2758eca6f534SVegard Nossum { 2759b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2760eca6f534SVegard Nossum } 2761eca6f534SVegard Nossum 27621da177e4SLinus Torvalds /* 27631da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 27641da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 27651da177e4SLinus Torvalds * 27661da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 27671da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 27681da177e4SLinus Torvalds * information (or be NULL). 27691da177e4SLinus Torvalds * 27701da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 27711da177e4SLinus Torvalds * When the flags word was introduced its top half was required 27721da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 27731da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 27741da177e4SLinus Torvalds * and must be discarded. 27751da177e4SLinus Torvalds */ 27765e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2777808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 27781da177e4SLinus Torvalds { 27792d92ab3cSAl Viro struct path path; 2780e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 27811da177e4SLinus Torvalds int retval = 0; 27821da177e4SLinus Torvalds 27831da177e4SLinus Torvalds /* Discard magic */ 27841da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 27851da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 27861da177e4SLinus Torvalds 27871da177e4SLinus Torvalds /* Basic sanity checks */ 27881da177e4SLinus Torvalds if (data_page) 27891da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 27901da177e4SLinus Torvalds 2791e462ec50SDavid Howells if (flags & MS_NOUSER) 2792e462ec50SDavid Howells return -EINVAL; 2793e462ec50SDavid Howells 2794a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 27955e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2796a27ab9f2STetsuo Handa if (retval) 2797a27ab9f2STetsuo Handa return retval; 2798a27ab9f2STetsuo Handa 2799a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2800a27ab9f2STetsuo Handa type_page, flags, data_page); 28010d5cadb8SAl Viro if (!retval && !may_mount()) 28020d5cadb8SAl Viro retval = -EPERM; 2803e462ec50SDavid Howells if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) 28049e8925b6SJeff Layton retval = -EPERM; 2805a27ab9f2STetsuo Handa if (retval) 2806a27ab9f2STetsuo Handa goto dput_out; 2807a27ab9f2STetsuo Handa 2808613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2809613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 28100a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 28110a1c01c9SMatthew Garrett 28121da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 28131da177e4SLinus Torvalds if (flags & MS_NOSUID) 28141da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 28151da177e4SLinus Torvalds if (flags & MS_NODEV) 28161da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 28171da177e4SLinus Torvalds if (flags & MS_NOEXEC) 28181da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2819fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2820fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2821fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2822fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2823d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2824d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 2825a9e5b732SDavid Howells if (flags & MS_RDONLY) 28262e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2827fc33a7bbSChristoph Hellwig 2828ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2829ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2830ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2831ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2832ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2833ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2834ffbc6f0eSEric W. Biederman } 2835ffbc6f0eSEric W. Biederman 2836e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 2837e462ec50SDavid Howells SB_SYNCHRONOUS | 2838e462ec50SDavid Howells SB_MANDLOCK | 2839e462ec50SDavid Howells SB_DIRSYNC | 2840e462ec50SDavid Howells SB_SILENT | 2841917086ffSMimi Zohar SB_POSIXACL | 2842d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 2843917086ffSMimi Zohar SB_I_VERSION); 28441da177e4SLinus Torvalds 284543f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 284643f5e655SDavid Howells retval = do_reconfigure_mnt(&path, mnt_flags); 284743f5e655SDavid Howells else if (flags & MS_REMOUNT) 2848e462ec50SDavid Howells retval = do_remount(&path, flags, sb_flags, mnt_flags, 28491da177e4SLinus Torvalds data_page); 28501da177e4SLinus Torvalds else if (flags & MS_BIND) 28512d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 28529676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 28532d92ab3cSAl Viro retval = do_change_type(&path, flags); 28541da177e4SLinus Torvalds else if (flags & MS_MOVE) 28552d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 28561da177e4SLinus Torvalds else 2857e462ec50SDavid Howells retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, 28581da177e4SLinus Torvalds dev_name, data_page); 28591da177e4SLinus Torvalds dput_out: 28602d92ab3cSAl Viro path_put(&path); 28611da177e4SLinus Torvalds return retval; 28621da177e4SLinus Torvalds } 28631da177e4SLinus Torvalds 2864537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 2865537f7ccbSEric W. Biederman { 2866537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 2867537f7ccbSEric W. Biederman } 2868537f7ccbSEric W. Biederman 2869537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 2870537f7ccbSEric W. Biederman { 2871537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 2872537f7ccbSEric W. Biederman } 2873537f7ccbSEric W. Biederman 2874771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2875771b1371SEric W. Biederman { 287674e83122SAl Viro if (!is_anon_ns(ns)) 28776344c433SAl Viro ns_free_inum(&ns->ns); 2878537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 2879771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2880771b1371SEric W. Biederman kfree(ns); 2881771b1371SEric W. Biederman } 2882771b1371SEric W. Biederman 28838823c079SEric W. Biederman /* 28848823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 28858823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 28868823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 28878823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 28888823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 28898823c079SEric W. Biederman */ 28908823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 28918823c079SEric W. Biederman 289274e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 2893cf8d2c11STrond Myklebust { 2894cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2895537f7ccbSEric W. Biederman struct ucounts *ucounts; 289698f842e6SEric W. Biederman int ret; 2897cf8d2c11STrond Myklebust 2898537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 2899537f7ccbSEric W. Biederman if (!ucounts) 2900df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 2901537f7ccbSEric W. Biederman 290274e83122SAl Viro new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2903537f7ccbSEric W. Biederman if (!new_ns) { 2904537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 2905cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2906537f7ccbSEric W. Biederman } 290774e83122SAl Viro if (!anon) { 29086344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 290998f842e6SEric W. Biederman if (ret) { 291098f842e6SEric W. Biederman kfree(new_ns); 2911537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 291298f842e6SEric W. Biederman return ERR_PTR(ret); 291398f842e6SEric W. Biederman } 291474e83122SAl Viro } 291533c42940SAl Viro new_ns->ns.ops = &mntns_operations; 291674e83122SAl Viro if (!anon) 29178823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2918cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2919cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2920cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2921771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2922537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 2923cf8d2c11STrond Myklebust return new_ns; 2924cf8d2c11STrond Myklebust } 2925cf8d2c11STrond Myklebust 29260766f788SEmese Revfy __latent_entropy 29279559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 29289559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 29291da177e4SLinus Torvalds { 29306b3286edSKirill Korotaev struct mnt_namespace *new_ns; 29317f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2932315fc83eSAl Viro struct mount *p, *q; 29339559f689SAl Viro struct mount *old; 2934cb338d06SAl Viro struct mount *new; 29357a472ef4SEric W. Biederman int copy_flags; 29361da177e4SLinus Torvalds 29379559f689SAl Viro BUG_ON(!ns); 29389559f689SAl Viro 29399559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 29409559f689SAl Viro get_mnt_ns(ns); 29419559f689SAl Viro return ns; 29429559f689SAl Viro } 29439559f689SAl Viro 29449559f689SAl Viro old = ns->root; 29459559f689SAl Viro 294674e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 2947cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2948cf8d2c11STrond Myklebust return new_ns; 29491da177e4SLinus Torvalds 295097216be0SAl Viro namespace_lock(); 29511da177e4SLinus Torvalds /* First pass: copy the tree topology */ 29524ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 29539559f689SAl Viro if (user_ns != ns->user_ns) 29543bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 29557a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2956be34d1a3SDavid Howells if (IS_ERR(new)) { 2957328e6d90SAl Viro namespace_unlock(); 2958771b1371SEric W. Biederman free_mnt_ns(new_ns); 2959be34d1a3SDavid Howells return ERR_CAST(new); 29601da177e4SLinus Torvalds } 29613bd045ccSAl Viro if (user_ns != ns->user_ns) { 29623bd045ccSAl Viro lock_mount_hash(); 29633bd045ccSAl Viro lock_mnt_tree(new); 29643bd045ccSAl Viro unlock_mount_hash(); 29653bd045ccSAl Viro } 2966be08d6d2SAl Viro new_ns->root = new; 29671a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 29681da177e4SLinus Torvalds 29691da177e4SLinus Torvalds /* 29701da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 29711da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 29721da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 29731da177e4SLinus Torvalds */ 2974909b0a88SAl Viro p = old; 2975cb338d06SAl Viro q = new; 29761da177e4SLinus Torvalds while (p) { 2977143c8c91SAl Viro q->mnt_ns = new_ns; 2978d2921684SEric W. Biederman new_ns->mounts++; 29799559f689SAl Viro if (new_fs) { 29809559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 29819559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2982315fc83eSAl Viro rootmnt = &p->mnt; 29831da177e4SLinus Torvalds } 29849559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 29859559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2986315fc83eSAl Viro pwdmnt = &p->mnt; 29871da177e4SLinus Torvalds } 29881da177e4SLinus Torvalds } 2989909b0a88SAl Viro p = next_mnt(p, old); 2990909b0a88SAl Viro q = next_mnt(q, new); 29914ce5d2b1SEric W. Biederman if (!q) 29924ce5d2b1SEric W. Biederman break; 29934ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 29944ce5d2b1SEric W. Biederman p = next_mnt(p, old); 29951da177e4SLinus Torvalds } 2996328e6d90SAl Viro namespace_unlock(); 29971da177e4SLinus Torvalds 29981da177e4SLinus Torvalds if (rootmnt) 2999f03c6599SAl Viro mntput(rootmnt); 30001da177e4SLinus Torvalds if (pwdmnt) 3001f03c6599SAl Viro mntput(pwdmnt); 30021da177e4SLinus Torvalds 3003741a2951SJANAK DESAI return new_ns; 3004741a2951SJANAK DESAI } 3005741a2951SJANAK DESAI 300674e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3007cf8d2c11STrond Myklebust { 30081a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3009ea441d11SAl Viro struct mnt_namespace *ns; 3010d31da0f0SAl Viro struct super_block *s; 3011ea441d11SAl Viro struct path path; 3012ea441d11SAl Viro int err; 3013ea441d11SAl Viro 301474e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 301574e83122SAl Viro if (IS_ERR(ns)) { 301674e83122SAl Viro mntput(m); 3017ea441d11SAl Viro return ERR_CAST(ns); 301874e83122SAl Viro } 301974e83122SAl Viro mnt->mnt_ns = ns; 302074e83122SAl Viro ns->root = mnt; 302174e83122SAl Viro ns->mounts++; 302274e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3023ea441d11SAl Viro 302474e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3025ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3026ea441d11SAl Viro 3027ea441d11SAl Viro put_mnt_ns(ns); 3028ea441d11SAl Viro 3029ea441d11SAl Viro if (err) 3030ea441d11SAl Viro return ERR_PTR(err); 3031ea441d11SAl Viro 3032ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3033d31da0f0SAl Viro s = path.mnt->mnt_sb; 3034d31da0f0SAl Viro atomic_inc(&s->s_active); 3035ea441d11SAl Viro mntput(path.mnt); 3036ea441d11SAl Viro /* lock the sucker */ 3037d31da0f0SAl Viro down_write(&s->s_umount); 3038ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3039ea441d11SAl Viro return path.dentry; 3040ea441d11SAl Viro } 3041ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3042ea441d11SAl Viro 3043312db1aaSDominik Brodowski int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, 3044312db1aaSDominik Brodowski unsigned long flags, void __user *data) 30451da177e4SLinus Torvalds { 3046eca6f534SVegard Nossum int ret; 3047eca6f534SVegard Nossum char *kernel_type; 3048eca6f534SVegard Nossum char *kernel_dev; 3049b40ef869SAl Viro void *options; 30501da177e4SLinus Torvalds 3051b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3052b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3053b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3054eca6f534SVegard Nossum goto out_type; 30551da177e4SLinus Torvalds 3056b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3057b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3058b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3059eca6f534SVegard Nossum goto out_dev; 30601da177e4SLinus Torvalds 3061b40ef869SAl Viro options = copy_mount_options(data); 3062b40ef869SAl Viro ret = PTR_ERR(options); 3063b40ef869SAl Viro if (IS_ERR(options)) 3064eca6f534SVegard Nossum goto out_data; 30651da177e4SLinus Torvalds 3066b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3067eca6f534SVegard Nossum 3068b40ef869SAl Viro kfree(options); 3069eca6f534SVegard Nossum out_data: 3070eca6f534SVegard Nossum kfree(kernel_dev); 3071eca6f534SVegard Nossum out_dev: 3072eca6f534SVegard Nossum kfree(kernel_type); 3073eca6f534SVegard Nossum out_type: 3074eca6f534SVegard Nossum return ret; 30751da177e4SLinus Torvalds } 30761da177e4SLinus Torvalds 3077312db1aaSDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3078312db1aaSDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 3079312db1aaSDominik Brodowski { 3080312db1aaSDominik Brodowski return ksys_mount(dev_name, dir_name, type, flags, data); 3081312db1aaSDominik Brodowski } 3082312db1aaSDominik Brodowski 30831da177e4SLinus Torvalds /* 3084afac7cbaSAl Viro * Return true if path is reachable from root 3085afac7cbaSAl Viro * 308648a066e7SAl Viro * namespace_sem or mount_lock is held 3087afac7cbaSAl Viro */ 3088643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3089afac7cbaSAl Viro const struct path *root) 3090afac7cbaSAl Viro { 3091643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3092a73324daSAl Viro dentry = mnt->mnt_mountpoint; 30930714a533SAl Viro mnt = mnt->mnt_parent; 3094afac7cbaSAl Viro } 3095643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3096afac7cbaSAl Viro } 3097afac7cbaSAl Viro 3098640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3099afac7cbaSAl Viro { 310025ab4c9bSYaowei Bai bool res; 310148a066e7SAl Viro read_seqlock_excl(&mount_lock); 3102643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 310348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3104afac7cbaSAl Viro return res; 3105afac7cbaSAl Viro } 3106afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3107afac7cbaSAl Viro 3108afac7cbaSAl Viro /* 31091da177e4SLinus Torvalds * pivot_root Semantics: 31101da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 31111da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 31121da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 31131da177e4SLinus Torvalds * 31141da177e4SLinus Torvalds * Restrictions: 31151da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 31161da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 31171da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 31181da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 31191da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 31201da177e4SLinus Torvalds * 31214a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 31224a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 31234a0d11faSNeil Brown * in this situation. 31244a0d11faSNeil Brown * 31251da177e4SLinus Torvalds * Notes: 31261da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 31271da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 31281da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 31291da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 31301da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 31311da177e4SLinus Torvalds * first. 31321da177e4SLinus Torvalds */ 31333480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 31343480b257SHeiko Carstens const char __user *, put_old) 31351da177e4SLinus Torvalds { 31362d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 313784d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 313884d17192SAl Viro struct mountpoint *old_mp, *root_mp; 31391da177e4SLinus Torvalds int error; 31401da177e4SLinus Torvalds 31419b40bc90SAl Viro if (!may_mount()) 31421da177e4SLinus Torvalds return -EPERM; 31431da177e4SLinus Torvalds 31442d8f3038SAl Viro error = user_path_dir(new_root, &new); 31451da177e4SLinus Torvalds if (error) 31461da177e4SLinus Torvalds goto out0; 31471da177e4SLinus Torvalds 31482d8f3038SAl Viro error = user_path_dir(put_old, &old); 31491da177e4SLinus Torvalds if (error) 31501da177e4SLinus Torvalds goto out1; 31511da177e4SLinus Torvalds 31522d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3153b12cea91SAl Viro if (error) 3154b12cea91SAl Viro goto out2; 31551da177e4SLinus Torvalds 3156f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 315784d17192SAl Viro old_mp = lock_mount(&old); 315884d17192SAl Viro error = PTR_ERR(old_mp); 315984d17192SAl Viro if (IS_ERR(old_mp)) 3160b12cea91SAl Viro goto out3; 3161b12cea91SAl Viro 31621da177e4SLinus Torvalds error = -EINVAL; 3163419148daSAl Viro new_mnt = real_mount(new.mnt); 3164419148daSAl Viro root_mnt = real_mount(root.mnt); 316584d17192SAl Viro old_mnt = real_mount(old.mnt); 316684d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3167fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3168fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3169b12cea91SAl Viro goto out4; 3170143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3171b12cea91SAl Viro goto out4; 31725ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 31735ff9d8a6SEric W. Biederman goto out4; 31741da177e4SLinus Torvalds error = -ENOENT; 3175f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3176b12cea91SAl Viro goto out4; 31771da177e4SLinus Torvalds error = -EBUSY; 317884d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3179b12cea91SAl Viro goto out4; /* loop, on the same file system */ 31801da177e4SLinus Torvalds error = -EINVAL; 31818c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3182b12cea91SAl Viro goto out4; /* not a mountpoint */ 3183676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3184b12cea91SAl Viro goto out4; /* not attached */ 318584d17192SAl Viro root_mp = root_mnt->mnt_mp; 31862d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3187b12cea91SAl Viro goto out4; /* not a mountpoint */ 3188676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3189b12cea91SAl Viro goto out4; /* not attached */ 31904ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 319184d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3192b12cea91SAl Viro goto out4; 31930d082601SEric W. Biederman /* make certain new is below the root */ 31940d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 31950d082601SEric W. Biederman goto out4; 319684d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3197719ea2fbSAl Viro lock_mount_hash(); 3198419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3199419148daSAl Viro detach_mnt(root_mnt, &root_parent); 32005ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 32015ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 32025ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 32035ff9d8a6SEric W. Biederman } 32044ac91378SJan Blunck /* mount old root on put_old */ 320584d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 32064ac91378SJan Blunck /* mount new_root on / */ 320784d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 32086b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 32094fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 32104fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 32113895dbf8SEric W. Biederman put_mountpoint(root_mp); 3212719ea2fbSAl Viro unlock_mount_hash(); 32132d8f3038SAl Viro chroot_fs_refs(&root, &new); 32141da177e4SLinus Torvalds error = 0; 3215b12cea91SAl Viro out4: 321684d17192SAl Viro unlock_mount(old_mp); 3217b12cea91SAl Viro if (!error) { 32181a390689SAl Viro path_put(&root_parent); 32191a390689SAl Viro path_put(&parent_path); 3220b12cea91SAl Viro } 3221b12cea91SAl Viro out3: 32228c3ee42eSAl Viro path_put(&root); 3223b12cea91SAl Viro out2: 32242d8f3038SAl Viro path_put(&old); 32251da177e4SLinus Torvalds out1: 32262d8f3038SAl Viro path_put(&new); 32271da177e4SLinus Torvalds out0: 32281da177e4SLinus Torvalds return error; 32291da177e4SLinus Torvalds } 32301da177e4SLinus Torvalds 32311da177e4SLinus Torvalds static void __init init_mount_tree(void) 32321da177e4SLinus Torvalds { 32331da177e4SLinus Torvalds struct vfsmount *mnt; 323474e83122SAl Viro struct mount *m; 32356b3286edSKirill Korotaev struct mnt_namespace *ns; 3236ac748a09SJan Blunck struct path root; 32370c55cfc4SEric W. Biederman struct file_system_type *type; 32381da177e4SLinus Torvalds 32390c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 32400c55cfc4SEric W. Biederman if (!type) 32410c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 32420c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 32430c55cfc4SEric W. Biederman put_filesystem(type); 32441da177e4SLinus Torvalds if (IS_ERR(mnt)) 32451da177e4SLinus Torvalds panic("Can't create rootfs"); 3246b3e19d92SNick Piggin 324774e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 32483b22edc5STrond Myklebust if (IS_ERR(ns)) 32491da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 325074e83122SAl Viro m = real_mount(mnt); 325174e83122SAl Viro m->mnt_ns = ns; 325274e83122SAl Viro ns->root = m; 325374e83122SAl Viro ns->mounts = 1; 325474e83122SAl Viro list_add(&m->mnt_list, &ns->list); 32556b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 32566b3286edSKirill Korotaev get_mnt_ns(ns); 32571da177e4SLinus Torvalds 3258be08d6d2SAl Viro root.mnt = mnt; 3259be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3260da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3261ac748a09SJan Blunck 3262ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3263ac748a09SJan Blunck set_fs_root(current->fs, &root); 32641da177e4SLinus Torvalds } 32651da177e4SLinus Torvalds 326674bf17cfSDenis Cheng void __init mnt_init(void) 32671da177e4SLinus Torvalds { 326815a67dd8SRandy Dunlap int err; 32691da177e4SLinus Torvalds 32707d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 327120c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 32721da177e4SLinus Torvalds 32730818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 327438129a13SAl Viro sizeof(struct hlist_head), 32750818bf27SAl Viro mhash_entries, 19, 32763d375d78SPavel Tatashin HASH_ZERO, 32770818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 32780818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 32790818bf27SAl Viro sizeof(struct hlist_head), 32800818bf27SAl Viro mphash_entries, 19, 32813d375d78SPavel Tatashin HASH_ZERO, 32820818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 32831da177e4SLinus Torvalds 328484d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 32851da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 32861da177e4SLinus Torvalds 32874b93dc9bSTejun Heo kernfs_init(); 32884b93dc9bSTejun Heo 328915a67dd8SRandy Dunlap err = sysfs_init(); 329015a67dd8SRandy Dunlap if (err) 329115a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 32928e24eea7SHarvey Harrison __func__, err); 329300d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 329400d26666SGreg Kroah-Hartman if (!fs_kobj) 32958e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 32961da177e4SLinus Torvalds init_rootfs(); 32971da177e4SLinus Torvalds init_mount_tree(); 32981da177e4SLinus Torvalds } 32991da177e4SLinus Torvalds 3300616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 33011da177e4SLinus Torvalds { 3302d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3303616511d0STrond Myklebust return; 33047b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3305771b1371SEric W. Biederman free_mnt_ns(ns); 33061da177e4SLinus Torvalds } 33079d412a43SAl Viro 33089d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 33099d412a43SAl Viro { 3310423e0ab0STim Chen struct vfsmount *mnt; 3311e462ec50SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); 3312423e0ab0STim Chen if (!IS_ERR(mnt)) { 3313423e0ab0STim Chen /* 3314423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3315423e0ab0STim Chen * we unmount before file sys is unregistered 3316423e0ab0STim Chen */ 3317f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3318423e0ab0STim Chen } 3319423e0ab0STim Chen return mnt; 33209d412a43SAl Viro } 33219d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3322423e0ab0STim Chen 3323423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3324423e0ab0STim Chen { 3325423e0ab0STim Chen /* release long term mount so mount point can be released */ 3326423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3327f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 332848a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3329423e0ab0STim Chen mntput(mnt); 3330423e0ab0STim Chen } 3331423e0ab0STim Chen } 3332423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 333302125a82SAl Viro 333402125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 333502125a82SAl Viro { 3336143c8c91SAl Viro return check_mnt(real_mount(mnt)); 333702125a82SAl Viro } 33388823c079SEric W. Biederman 33393151527eSEric W. Biederman bool current_chrooted(void) 33403151527eSEric W. Biederman { 33413151527eSEric W. Biederman /* Does the current process have a non-standard root */ 33423151527eSEric W. Biederman struct path ns_root; 33433151527eSEric W. Biederman struct path fs_root; 33443151527eSEric W. Biederman bool chrooted; 33453151527eSEric W. Biederman 33463151527eSEric W. Biederman /* Find the namespace root */ 33473151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 33483151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 33493151527eSEric W. Biederman path_get(&ns_root); 33503151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 33513151527eSEric W. Biederman ; 33523151527eSEric W. Biederman 33533151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 33543151527eSEric W. Biederman 33553151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 33563151527eSEric W. Biederman 33573151527eSEric W. Biederman path_put(&fs_root); 33583151527eSEric W. Biederman path_put(&ns_root); 33593151527eSEric W. Biederman 33603151527eSEric W. Biederman return chrooted; 33613151527eSEric W. Biederman } 33623151527eSEric W. Biederman 33638654df4eSEric W. Biederman static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, 33648654df4eSEric W. Biederman int *new_mnt_flags) 336587a8ebd6SEric W. Biederman { 33668c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 336787a8ebd6SEric W. Biederman struct mount *mnt; 3368e51db735SEric W. Biederman bool visible = false; 336987a8ebd6SEric W. Biederman 337044bb4385SAl Viro down_read(&namespace_sem); 337187a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3372e51db735SEric W. Biederman struct mount *child; 337377b1a97dSEric W. Biederman int mnt_flags; 337477b1a97dSEric W. Biederman 33758654df4eSEric W. Biederman if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) 3376e51db735SEric W. Biederman continue; 3377e51db735SEric W. Biederman 33787e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 33797e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 33807e96c1b0SEric W. Biederman */ 33817e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 33827e96c1b0SEric W. Biederman continue; 33837e96c1b0SEric W. Biederman 3384a1935c17SEric W. Biederman /* A local view of the mount flags */ 338577b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 338677b1a97dSEric W. Biederman 3387695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3388bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 3389695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3390695e9df0SEric W. Biederman 33918c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 33928c6cf9ccSEric W. Biederman * than the proposed new mount. 33938c6cf9ccSEric W. Biederman */ 339477b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 33958c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 33968c6cf9ccSEric W. Biederman continue; 339777b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 339877b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 33998c6cf9ccSEric W. Biederman continue; 34008c6cf9ccSEric W. Biederman 3401ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3402ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3403ceeb0e5dSEric W. Biederman * empty directories. 3404e51db735SEric W. Biederman */ 3405e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3406e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3407ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3408d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3409ceeb0e5dSEric W. Biederman continue; 34107236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 34117236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3412e51db735SEric W. Biederman goto next; 341387a8ebd6SEric W. Biederman } 34148c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 341577b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 34168c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3417e51db735SEric W. Biederman visible = true; 3418e51db735SEric W. Biederman goto found; 3419e51db735SEric W. Biederman next: ; 342087a8ebd6SEric W. Biederman } 3421e51db735SEric W. Biederman found: 342244bb4385SAl Viro up_read(&namespace_sem); 3423e51db735SEric W. Biederman return visible; 342487a8ebd6SEric W. Biederman } 342587a8ebd6SEric W. Biederman 34268654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) 34278654df4eSEric W. Biederman { 3428a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 34298654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 34308654df4eSEric W. Biederman unsigned long s_iflags; 34318654df4eSEric W. Biederman 34328654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 34338654df4eSEric W. Biederman return false; 34348654df4eSEric W. Biederman 34358654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 34368654df4eSEric W. Biederman s_iflags = mnt->mnt_sb->s_iflags; 34378654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 34388654df4eSEric W. Biederman return false; 34398654df4eSEric W. Biederman 3440a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3441a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3442a1935c17SEric W. Biederman required_iflags); 3443a1935c17SEric W. Biederman return true; 3444a1935c17SEric W. Biederman } 3445a1935c17SEric W. Biederman 34468654df4eSEric W. Biederman return !mnt_already_visible(ns, mnt, new_mnt_flags); 34478654df4eSEric W. Biederman } 34488654df4eSEric W. Biederman 3449380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3450380cf5baSAndy Lutomirski { 3451380cf5baSAndy Lutomirski /* 3452380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3453380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3454380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3455380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3456380cf5baSAndy Lutomirski * in other namespaces. 3457380cf5baSAndy Lutomirski */ 3458380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3459380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3460380cf5baSAndy Lutomirski } 3461380cf5baSAndy Lutomirski 346264964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 34638823c079SEric W. Biederman { 346458be2825SAl Viro struct ns_common *ns = NULL; 34658823c079SEric W. Biederman struct nsproxy *nsproxy; 34668823c079SEric W. Biederman 3467728dba3aSEric W. Biederman task_lock(task); 3468728dba3aSEric W. Biederman nsproxy = task->nsproxy; 34698823c079SEric W. Biederman if (nsproxy) { 347058be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 347158be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 34728823c079SEric W. Biederman } 3473728dba3aSEric W. Biederman task_unlock(task); 34748823c079SEric W. Biederman 34758823c079SEric W. Biederman return ns; 34768823c079SEric W. Biederman } 34778823c079SEric W. Biederman 347864964528SAl Viro static void mntns_put(struct ns_common *ns) 34798823c079SEric W. Biederman { 348058be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 34818823c079SEric W. Biederman } 34828823c079SEric W. Biederman 348364964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 34848823c079SEric W. Biederman { 34858823c079SEric W. Biederman struct fs_struct *fs = current->fs; 34864f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 34878823c079SEric W. Biederman struct path root; 34884f757f3cSAl Viro int err; 34898823c079SEric W. Biederman 34900c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3491c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3492c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3493ae11e0f1SZhao Hongjiang return -EPERM; 34948823c079SEric W. Biederman 349574e83122SAl Viro if (is_anon_ns(mnt_ns)) 349674e83122SAl Viro return -EINVAL; 349774e83122SAl Viro 34988823c079SEric W. Biederman if (fs->users != 1) 34998823c079SEric W. Biederman return -EINVAL; 35008823c079SEric W. Biederman 35018823c079SEric W. Biederman get_mnt_ns(mnt_ns); 35024f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 35038823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 35048823c079SEric W. Biederman 35058823c079SEric W. Biederman /* Find the root */ 35064f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 35074f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 35084f757f3cSAl Viro if (err) { 35094f757f3cSAl Viro /* revert to old namespace */ 35104f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 35114f757f3cSAl Viro put_mnt_ns(mnt_ns); 35124f757f3cSAl Viro return err; 35134f757f3cSAl Viro } 35148823c079SEric W. Biederman 35154068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 35164068367cSAndrei Vagin 35178823c079SEric W. Biederman /* Update the pwd and root */ 35188823c079SEric W. Biederman set_fs_pwd(fs, &root); 35198823c079SEric W. Biederman set_fs_root(fs, &root); 35208823c079SEric W. Biederman 35218823c079SEric W. Biederman path_put(&root); 35228823c079SEric W. Biederman return 0; 35238823c079SEric W. Biederman } 35248823c079SEric W. Biederman 3525bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3526bcac25a5SAndrey Vagin { 3527bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3528bcac25a5SAndrey Vagin } 3529bcac25a5SAndrey Vagin 35308823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 35318823c079SEric W. Biederman .name = "mnt", 35328823c079SEric W. Biederman .type = CLONE_NEWNS, 35338823c079SEric W. Biederman .get = mntns_get, 35348823c079SEric W. Biederman .put = mntns_put, 35358823c079SEric W. Biederman .install = mntns_install, 3536bcac25a5SAndrey Vagin .owner = mntns_owner, 35378823c079SEric W. Biederman }; 3538