159bd9dedSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namespace.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 185b825c3aSIngo Molnar #include <linux/cred.h> 1973cd49ecSMiklos Szeredi #include <linux/idr.h> 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23a07b2000SAl Viro #include <linux/file.h> 24d10577a8SAl Viro #include <linux/uaccess.h> 250bb80f24SDavid Howells #include <linux/proc_ns.h> 2620b4fb48SLinus Torvalds #include <linux/magic.h> 2757c8a661SMike Rapoport #include <linux/memblock.h> 289ea459e1SAl Viro #include <linux/task_work.h> 299164bb4aSIngo Molnar #include <linux/sched/task.h> 30e262e32dSDavid Howells #include <uapi/linux/mount.h> 319bc61ab1SDavid Howells #include <linux/fs_context.h> 32037f11b4SAl Viro #include <linux/shmem_fs.h> 339164bb4aSIngo Molnar 3407b20889SRam Pai #include "pnode.h" 35948730b0SAdrian Bunk #include "internal.h" 361da177e4SLinus Torvalds 37d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 38d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 39d2921684SEric W. Biederman 400818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 410818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 420818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 430818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 440818bf27SAl Viro 450818bf27SAl Viro static __initdata unsigned long mhash_entries; 460818bf27SAl Viro static int __init set_mhash_entries(char *str) 470818bf27SAl Viro { 480818bf27SAl Viro if (!str) 490818bf27SAl Viro return 0; 500818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 510818bf27SAl Viro return 1; 520818bf27SAl Viro } 530818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 540818bf27SAl Viro 550818bf27SAl Viro static __initdata unsigned long mphash_entries; 560818bf27SAl Viro static int __init set_mphash_entries(char *str) 570818bf27SAl Viro { 580818bf27SAl Viro if (!str) 590818bf27SAl Viro return 0; 600818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 610818bf27SAl Viro return 1; 620818bf27SAl Viro } 630818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6413f14b4dSEric Dumazet 65c7999c36SAl Viro static u64 event; 6673cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 67719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 685addc5ddSAl Viro 6938129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 700818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 71e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 7259aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 734edbe133SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 744edbe133SAl Viro static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ 751da177e4SLinus Torvalds 76f87fd4c2SMiklos Szeredi /* /sys/fs */ 7700d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7800d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 79f87fd4c2SMiklos Szeredi 8099b7db7bSNick Piggin /* 8199b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 8299b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 8399b7db7bSNick Piggin * up the tree. 8499b7db7bSNick Piggin * 8599b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8699b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8799b7db7bSNick Piggin */ 8848a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8999b7db7bSNick Piggin 9038129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 911da177e4SLinus Torvalds { 921da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 931da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 940818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 950818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 960818bf27SAl Viro } 970818bf27SAl Viro 980818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 990818bf27SAl Viro { 1000818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 1010818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 1020818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds 105b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10673cd49ecSMiklos Szeredi { 107169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 10873cd49ecSMiklos Szeredi 109169b480eSMatthew Wilcox if (res < 0) 11073cd49ecSMiklos Szeredi return res; 111169b480eSMatthew Wilcox mnt->mnt_id = res; 112169b480eSMatthew Wilcox return 0; 11373cd49ecSMiklos Szeredi } 11473cd49ecSMiklos Szeredi 115b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11673cd49ecSMiklos Szeredi { 117169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 11873cd49ecSMiklos Szeredi } 11973cd49ecSMiklos Szeredi 120719f5d7fSMiklos Szeredi /* 121719f5d7fSMiklos Szeredi * Allocate a new peer group ID 122719f5d7fSMiklos Szeredi */ 1234b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 124719f5d7fSMiklos Szeredi { 125169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 126f21f6220SAl Viro 127169b480eSMatthew Wilcox if (res < 0) 128f21f6220SAl Viro return res; 129169b480eSMatthew Wilcox mnt->mnt_group_id = res; 130169b480eSMatthew Wilcox return 0; 131719f5d7fSMiklos Szeredi } 132719f5d7fSMiklos Szeredi 133719f5d7fSMiklos Szeredi /* 134719f5d7fSMiklos Szeredi * Release a peer group ID 135719f5d7fSMiklos Szeredi */ 1364b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 137719f5d7fSMiklos Szeredi { 138169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 13915169fe7SAl Viro mnt->mnt_group_id = 0; 140719f5d7fSMiklos Szeredi } 141719f5d7fSMiklos Szeredi 142b3e19d92SNick Piggin /* 143b3e19d92SNick Piggin * vfsmount lock must be held for read 144b3e19d92SNick Piggin */ 14583adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 146b3e19d92SNick Piggin { 147b3e19d92SNick Piggin #ifdef CONFIG_SMP 14868e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 149b3e19d92SNick Piggin #else 150b3e19d92SNick Piggin preempt_disable(); 15168e8a9feSAl Viro mnt->mnt_count += n; 152b3e19d92SNick Piggin preempt_enable(); 153b3e19d92SNick Piggin #endif 154b3e19d92SNick Piggin } 155b3e19d92SNick Piggin 156b3e19d92SNick Piggin /* 157b3e19d92SNick Piggin * vfsmount lock must be held for write 158b3e19d92SNick Piggin */ 15983adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 160b3e19d92SNick Piggin { 161b3e19d92SNick Piggin #ifdef CONFIG_SMP 162f03c6599SAl Viro unsigned int count = 0; 163b3e19d92SNick Piggin int cpu; 164b3e19d92SNick Piggin 165b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 16668e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 167b3e19d92SNick Piggin } 168b3e19d92SNick Piggin 169b3e19d92SNick Piggin return count; 170b3e19d92SNick Piggin #else 17168e8a9feSAl Viro return mnt->mnt_count; 172b3e19d92SNick Piggin #endif 173b3e19d92SNick Piggin } 174b3e19d92SNick Piggin 175b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1761da177e4SLinus Torvalds { 177c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 178c63181e6SAl Viro if (mnt) { 17973cd49ecSMiklos Szeredi int err; 18073cd49ecSMiklos Szeredi 181c63181e6SAl Viro err = mnt_alloc_id(mnt); 18288b38782SLi Zefan if (err) 18388b38782SLi Zefan goto out_free_cache; 18488b38782SLi Zefan 18588b38782SLi Zefan if (name) { 186fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 187c63181e6SAl Viro if (!mnt->mnt_devname) 18888b38782SLi Zefan goto out_free_id; 18973cd49ecSMiklos Szeredi } 19073cd49ecSMiklos Szeredi 191b3e19d92SNick Piggin #ifdef CONFIG_SMP 192c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 193c63181e6SAl Viro if (!mnt->mnt_pcp) 194b3e19d92SNick Piggin goto out_free_devname; 195b3e19d92SNick Piggin 196c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 197b3e19d92SNick Piggin #else 198c63181e6SAl Viro mnt->mnt_count = 1; 199c63181e6SAl Viro mnt->mnt_writers = 0; 200b3e19d92SNick Piggin #endif 201b3e19d92SNick Piggin 20238129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 203c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 204c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 205c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 206c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 207c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 208c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 209c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2100a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 21199b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 21256cbb429SAl Viro INIT_HLIST_HEAD(&mnt->mnt_stuck_children); 2131da177e4SLinus Torvalds } 214c63181e6SAl Viro return mnt; 21588b38782SLi Zefan 216d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 217d3ef3d73Snpiggin@suse.de out_free_devname: 218fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 219d3ef3d73Snpiggin@suse.de #endif 22088b38782SLi Zefan out_free_id: 221c63181e6SAl Viro mnt_free_id(mnt); 22288b38782SLi Zefan out_free_cache: 223c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 22488b38782SLi Zefan return NULL; 2251da177e4SLinus Torvalds } 2261da177e4SLinus Torvalds 2278366025eSDave Hansen /* 2288366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2298366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2308366025eSDave Hansen * We must keep track of when those operations start 2318366025eSDave Hansen * (for permission checks) and when they end, so that 2328366025eSDave Hansen * we can determine when writes are able to occur to 2338366025eSDave Hansen * a filesystem. 2348366025eSDave Hansen */ 2353d733633SDave Hansen /* 2363d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2373d733633SDave Hansen * @mnt: the mount to check for its write status 2383d733633SDave Hansen * 2393d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2403d733633SDave Hansen * It does not guarantee that the filesystem will stay 2413d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2423d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2433d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2443d733633SDave Hansen * r/w. 2453d733633SDave Hansen */ 24643f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 2473d733633SDave Hansen { 24843f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 2493d733633SDave Hansen } 2503d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2513d733633SDave Hansen 25283adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2533d733633SDave Hansen { 254d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 25568e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 256d3ef3d73Snpiggin@suse.de #else 25768e8a9feSAl Viro mnt->mnt_writers++; 258d3ef3d73Snpiggin@suse.de #endif 2593d733633SDave Hansen } 2603d733633SDave Hansen 26183adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2623d733633SDave Hansen { 263d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 26468e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 265d3ef3d73Snpiggin@suse.de #else 26668e8a9feSAl Viro mnt->mnt_writers--; 267d3ef3d73Snpiggin@suse.de #endif 268d3ef3d73Snpiggin@suse.de } 269d3ef3d73Snpiggin@suse.de 27083adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 271d3ef3d73Snpiggin@suse.de { 272d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 273d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2743d733633SDave Hansen int cpu; 2753d733633SDave Hansen 2763d733633SDave Hansen for_each_possible_cpu(cpu) { 27768e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2783d733633SDave Hansen } 2793d733633SDave Hansen 280d3ef3d73Snpiggin@suse.de return count; 281d3ef3d73Snpiggin@suse.de #else 282d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 283d3ef3d73Snpiggin@suse.de #endif 2843d733633SDave Hansen } 2853d733633SDave Hansen 2864ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2874ed5e82fSMiklos Szeredi { 2884ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2894ed5e82fSMiklos Szeredi return 1; 2904ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2914ed5e82fSMiklos Szeredi smp_rmb(); 2924ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2934ed5e82fSMiklos Szeredi } 2944ed5e82fSMiklos Szeredi 2953d733633SDave Hansen /* 296eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 297eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 298eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 299eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3003d733633SDave Hansen */ 3018366025eSDave Hansen /** 302eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 30383adc753SAl Viro * @m: the mount on which to take a write 3048366025eSDave Hansen * 305eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 306eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 307eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 308eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 309eb04c282SJan Kara * called. This is effectively a refcount. 3108366025eSDave Hansen */ 311eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3128366025eSDave Hansen { 31383adc753SAl Viro struct mount *mnt = real_mount(m); 3143d733633SDave Hansen int ret = 0; 3153d733633SDave Hansen 316d3ef3d73Snpiggin@suse.de preempt_disable(); 317c6653a83SNick Piggin mnt_inc_writers(mnt); 318d3ef3d73Snpiggin@suse.de /* 319c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 320d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 321d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 322d3ef3d73Snpiggin@suse.de */ 323d3ef3d73Snpiggin@suse.de smp_mb(); 3246aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 325d3ef3d73Snpiggin@suse.de cpu_relax(); 326d3ef3d73Snpiggin@suse.de /* 327d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 328d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 329d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 330d3ef3d73Snpiggin@suse.de */ 331d3ef3d73Snpiggin@suse.de smp_rmb(); 3324ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 333c6653a83SNick Piggin mnt_dec_writers(mnt); 3343d733633SDave Hansen ret = -EROFS; 3353d733633SDave Hansen } 336d3ef3d73Snpiggin@suse.de preempt_enable(); 337eb04c282SJan Kara 338eb04c282SJan Kara return ret; 339eb04c282SJan Kara } 340eb04c282SJan Kara 341eb04c282SJan Kara /** 342eb04c282SJan Kara * mnt_want_write - get write access to a mount 343eb04c282SJan Kara * @m: the mount on which to take a write 344eb04c282SJan Kara * 345eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 346eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 347eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 348eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 349eb04c282SJan Kara */ 350eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 351eb04c282SJan Kara { 352eb04c282SJan Kara int ret; 353eb04c282SJan Kara 354eb04c282SJan Kara sb_start_write(m->mnt_sb); 355eb04c282SJan Kara ret = __mnt_want_write(m); 356eb04c282SJan Kara if (ret) 357eb04c282SJan Kara sb_end_write(m->mnt_sb); 3583d733633SDave Hansen return ret; 3598366025eSDave Hansen } 3608366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3618366025eSDave Hansen 3628366025eSDave Hansen /** 36396029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 36496029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 36596029c4eSnpiggin@suse.de * 36696029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 36796029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 36896029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 36996029c4eSnpiggin@suse.de * on it. This allows some optimisation. 37096029c4eSnpiggin@suse.de * 37196029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 37296029c4eSnpiggin@suse.de * drop the reference. 37396029c4eSnpiggin@suse.de */ 37496029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 37596029c4eSnpiggin@suse.de { 37696029c4eSnpiggin@suse.de /* superblock may be r/o */ 37796029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 37896029c4eSnpiggin@suse.de return -EROFS; 37996029c4eSnpiggin@suse.de preempt_disable(); 38083adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 38196029c4eSnpiggin@suse.de preempt_enable(); 38296029c4eSnpiggin@suse.de return 0; 38396029c4eSnpiggin@suse.de } 38496029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 38596029c4eSnpiggin@suse.de 38696029c4eSnpiggin@suse.de /** 387eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 388eb04c282SJan Kara * @file: the file who's mount on which to take a write 389eb04c282SJan Kara * 390eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 391eb04c282SJan Kara * do some optimisations if the file is open for write already 392eb04c282SJan Kara */ 393eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 394eb04c282SJan Kara { 39583f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 396eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 397eb04c282SJan Kara else 398eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 399eb04c282SJan Kara } 400eb04c282SJan Kara 401eb04c282SJan Kara /** 4027c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4037c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4047c6893e3SMiklos Szeredi * 4057c6893e3SMiklos Szeredi * This is like mnt_want_write, but it takes a file and can 4067c6893e3SMiklos Szeredi * do some optimisations if the file is open for write already 4077c6893e3SMiklos Szeredi */ 4087c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4097c6893e3SMiklos Szeredi { 4107c6893e3SMiklos Szeredi int ret; 4117c6893e3SMiklos Szeredi 4127c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4137c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4147c6893e3SMiklos Szeredi if (ret) 4157c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4167c6893e3SMiklos Szeredi return ret; 4177c6893e3SMiklos Szeredi } 41896029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 41996029c4eSnpiggin@suse.de 42096029c4eSnpiggin@suse.de /** 421eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4228366025eSDave Hansen * @mnt: the mount on which to give up write access 4238366025eSDave Hansen * 4248366025eSDave Hansen * Tells the low-level filesystem that we are done 4258366025eSDave Hansen * performing writes to it. Must be matched with 426eb04c282SJan Kara * __mnt_want_write() call above. 4278366025eSDave Hansen */ 428eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4298366025eSDave Hansen { 430d3ef3d73Snpiggin@suse.de preempt_disable(); 43183adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 432d3ef3d73Snpiggin@suse.de preempt_enable(); 4338366025eSDave Hansen } 434eb04c282SJan Kara 435eb04c282SJan Kara /** 436eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 437eb04c282SJan Kara * @mnt: the mount on which to give up write access 438eb04c282SJan Kara * 439eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 440eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 441eb04c282SJan Kara * mnt_want_write() call above. 442eb04c282SJan Kara */ 443eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 444eb04c282SJan Kara { 445eb04c282SJan Kara __mnt_drop_write(mnt); 446eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 447eb04c282SJan Kara } 4488366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4498366025eSDave Hansen 450eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 451eb04c282SJan Kara { 452eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 453eb04c282SJan Kara } 454eb04c282SJan Kara 4557c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4567c6893e3SMiklos Szeredi { 457a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4587c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4597c6893e3SMiklos Szeredi } 4602a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4612a79f17eSAl Viro 46283adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4638366025eSDave Hansen { 4643d733633SDave Hansen int ret = 0; 4653d733633SDave Hansen 466719ea2fbSAl Viro lock_mount_hash(); 46783adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 468d3ef3d73Snpiggin@suse.de /* 469d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 470d3ef3d73Snpiggin@suse.de * should be visible before we do. 471d3ef3d73Snpiggin@suse.de */ 472d3ef3d73Snpiggin@suse.de smp_mb(); 473d3ef3d73Snpiggin@suse.de 474d3ef3d73Snpiggin@suse.de /* 475d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 476d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 477d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 478d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 479d3ef3d73Snpiggin@suse.de * 480d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 481d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 482d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 483d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 484d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 485d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 486d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 487d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 488d3ef3d73Snpiggin@suse.de * we're counting up here. 489d3ef3d73Snpiggin@suse.de */ 490c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 491d3ef3d73Snpiggin@suse.de ret = -EBUSY; 492d3ef3d73Snpiggin@suse.de else 49383adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 494d3ef3d73Snpiggin@suse.de /* 495d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 496d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 497d3ef3d73Snpiggin@suse.de */ 498d3ef3d73Snpiggin@suse.de smp_wmb(); 49983adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 500719ea2fbSAl Viro unlock_mount_hash(); 5013d733633SDave Hansen return ret; 5023d733633SDave Hansen } 5038366025eSDave Hansen 50443f5e655SDavid Howells static int __mnt_unmake_readonly(struct mount *mnt) 5052e4b7fcdSDave Hansen { 506719ea2fbSAl Viro lock_mount_hash(); 50783adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 508719ea2fbSAl Viro unlock_mount_hash(); 50943f5e655SDavid Howells return 0; 5102e4b7fcdSDave Hansen } 5112e4b7fcdSDave Hansen 5124ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5134ed5e82fSMiklos Szeredi { 5144ed5e82fSMiklos Szeredi struct mount *mnt; 5154ed5e82fSMiklos Szeredi int err = 0; 5164ed5e82fSMiklos Szeredi 5178e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5188e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5198e8b8796SMiklos Szeredi return -EBUSY; 5208e8b8796SMiklos Szeredi 521719ea2fbSAl Viro lock_mount_hash(); 5224ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5234ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5244ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5254ed5e82fSMiklos Szeredi smp_mb(); 5264ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5274ed5e82fSMiklos Szeredi err = -EBUSY; 5284ed5e82fSMiklos Szeredi break; 5294ed5e82fSMiklos Szeredi } 5304ed5e82fSMiklos Szeredi } 5314ed5e82fSMiklos Szeredi } 5328e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5338e8b8796SMiklos Szeredi err = -EBUSY; 5348e8b8796SMiklos Szeredi 5354ed5e82fSMiklos Szeredi if (!err) { 5364ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5374ed5e82fSMiklos Szeredi smp_wmb(); 5384ed5e82fSMiklos Szeredi } 5394ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5404ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5414ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5424ed5e82fSMiklos Szeredi } 543719ea2fbSAl Viro unlock_mount_hash(); 5444ed5e82fSMiklos Szeredi 5454ed5e82fSMiklos Szeredi return err; 5464ed5e82fSMiklos Szeredi } 5474ed5e82fSMiklos Szeredi 548b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5491da177e4SLinus Torvalds { 550fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 551d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 55268e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 553d3ef3d73Snpiggin@suse.de #endif 554b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5551da177e4SLinus Torvalds } 5561da177e4SLinus Torvalds 5578ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5588ffcb32eSDavid Howells { 5598ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5608ffcb32eSDavid Howells } 5618ffcb32eSDavid Howells 56248a066e7SAl Viro /* call under rcu_read_lock */ 563294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 56448a066e7SAl Viro { 56548a066e7SAl Viro struct mount *mnt; 56648a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 567294d71ffSAl Viro return 1; 56848a066e7SAl Viro if (bastard == NULL) 569294d71ffSAl Viro return 0; 57048a066e7SAl Viro mnt = real_mount(bastard); 57148a066e7SAl Viro mnt_add_count(mnt, 1); 572119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 57348a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 574294d71ffSAl Viro return 0; 57548a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 57648a066e7SAl Viro mnt_add_count(mnt, -1); 577294d71ffSAl Viro return 1; 57848a066e7SAl Viro } 579119e1ef8SAl Viro lock_mount_hash(); 580119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 581119e1ef8SAl Viro mnt_add_count(mnt, -1); 582119e1ef8SAl Viro unlock_mount_hash(); 583119e1ef8SAl Viro return 1; 584119e1ef8SAl Viro } 585119e1ef8SAl Viro unlock_mount_hash(); 586119e1ef8SAl Viro /* caller will mntput() */ 587294d71ffSAl Viro return -1; 588294d71ffSAl Viro } 589294d71ffSAl Viro 590294d71ffSAl Viro /* call under rcu_read_lock */ 591294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 592294d71ffSAl Viro { 593294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 594294d71ffSAl Viro if (likely(!res)) 595294d71ffSAl Viro return true; 596294d71ffSAl Viro if (unlikely(res < 0)) { 59748a066e7SAl Viro rcu_read_unlock(); 59848a066e7SAl Viro mntput(bastard); 59948a066e7SAl Viro rcu_read_lock(); 600294d71ffSAl Viro } 60148a066e7SAl Viro return false; 60248a066e7SAl Viro } 60348a066e7SAl Viro 6041da177e4SLinus Torvalds /* 605474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 60648a066e7SAl Viro * call under rcu_read_lock() 6071da177e4SLinus Torvalds */ 608474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6091da177e4SLinus Torvalds { 61038129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 611474279dcSAl Viro struct mount *p; 6121da177e4SLinus Torvalds 61338129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 614474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 615474279dcSAl Viro return p; 616474279dcSAl Viro return NULL; 6171da177e4SLinus Torvalds } 618474279dcSAl Viro 619474279dcSAl Viro /* 620f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 621f015f126SDavid Howells * 622f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 623f015f126SDavid Howells * following mounts: 624f015f126SDavid Howells * 625f015f126SDavid Howells * mount /dev/sda1 /mnt 626f015f126SDavid Howells * mount /dev/sda2 /mnt 627f015f126SDavid Howells * mount /dev/sda3 /mnt 628f015f126SDavid Howells * 629f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 630f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 631f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 632f015f126SDavid Howells * 633f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 634a05964f3SRam Pai */ 635ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 636a05964f3SRam Pai { 637c7105365SAl Viro struct mount *child_mnt; 63848a066e7SAl Viro struct vfsmount *m; 63948a066e7SAl Viro unsigned seq; 64099b7db7bSNick Piggin 64148a066e7SAl Viro rcu_read_lock(); 64248a066e7SAl Viro do { 64348a066e7SAl Viro seq = read_seqbegin(&mount_lock); 644474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 64548a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 64648a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 64748a066e7SAl Viro rcu_read_unlock(); 64848a066e7SAl Viro return m; 649a05964f3SRam Pai } 650a05964f3SRam Pai 6519f6c61f9SMiklos Szeredi static inline void lock_ns_list(struct mnt_namespace *ns) 6529f6c61f9SMiklos Szeredi { 6539f6c61f9SMiklos Szeredi spin_lock(&ns->ns_lock); 6549f6c61f9SMiklos Szeredi } 6559f6c61f9SMiklos Szeredi 6569f6c61f9SMiklos Szeredi static inline void unlock_ns_list(struct mnt_namespace *ns) 6579f6c61f9SMiklos Szeredi { 6589f6c61f9SMiklos Szeredi spin_unlock(&ns->ns_lock); 6599f6c61f9SMiklos Szeredi } 6609f6c61f9SMiklos Szeredi 6619f6c61f9SMiklos Szeredi static inline bool mnt_is_cursor(struct mount *mnt) 6629f6c61f9SMiklos Szeredi { 6639f6c61f9SMiklos Szeredi return mnt->mnt.mnt_flags & MNT_CURSOR; 6649f6c61f9SMiklos Szeredi } 6659f6c61f9SMiklos Szeredi 6667af1364fSEric W. Biederman /* 6677af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6687af1364fSEric W. Biederman * current mount namespace. 6697af1364fSEric W. Biederman * 6707af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6717af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6727af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6737af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6747af1364fSEric W. Biederman * is a mountpoint. 6757af1364fSEric W. Biederman * 6767af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6777af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6787af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6797af1364fSEric W. Biederman * parent mount. 6807af1364fSEric W. Biederman */ 6817af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6827af1364fSEric W. Biederman { 6837af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 6847af1364fSEric W. Biederman struct mount *mnt; 6857af1364fSEric W. Biederman bool is_covered = false; 6867af1364fSEric W. Biederman 6877af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 6887af1364fSEric W. Biederman goto out; 6897af1364fSEric W. Biederman 6907af1364fSEric W. Biederman down_read(&namespace_sem); 6919f6c61f9SMiklos Szeredi lock_ns_list(ns); 6927af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 6939f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 6949f6c61f9SMiklos Szeredi continue; 6957af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 6967af1364fSEric W. Biederman if (is_covered) 6977af1364fSEric W. Biederman break; 6987af1364fSEric W. Biederman } 6999f6c61f9SMiklos Szeredi unlock_ns_list(ns); 7007af1364fSEric W. Biederman up_read(&namespace_sem); 7017af1364fSEric W. Biederman out: 7027af1364fSEric W. Biederman return is_covered; 7037af1364fSEric W. Biederman } 7047af1364fSEric W. Biederman 705e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 70684d17192SAl Viro { 7070818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 70884d17192SAl Viro struct mountpoint *mp; 70984d17192SAl Viro 7100818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 71184d17192SAl Viro if (mp->m_dentry == dentry) { 71284d17192SAl Viro mp->m_count++; 71384d17192SAl Viro return mp; 71484d17192SAl Viro } 71584d17192SAl Viro } 716e2dfa935SEric W. Biederman return NULL; 717e2dfa935SEric W. Biederman } 718e2dfa935SEric W. Biederman 7193895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 720e2dfa935SEric W. Biederman { 7213895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 722e2dfa935SEric W. Biederman int ret; 72384d17192SAl Viro 7243895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7251e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7261e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7271e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7283895dbf8SEric W. Biederman mountpoint: 7293895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7303895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7313895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7323895dbf8SEric W. Biederman if (mp) 7333895dbf8SEric W. Biederman goto done; 73484d17192SAl Viro } 735eed81007SMiklos Szeredi 7363895dbf8SEric W. Biederman if (!new) 7373895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7383895dbf8SEric W. Biederman if (!new) 7393895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7403895dbf8SEric W. Biederman 7413895dbf8SEric W. Biederman 7423895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7433895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7443895dbf8SEric W. Biederman 7453895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7463895dbf8SEric W. Biederman if (ret == -EBUSY) 7473895dbf8SEric W. Biederman goto mountpoint; 7483895dbf8SEric W. Biederman 7493895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7503895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7513895dbf8SEric W. Biederman if (ret) 7523895dbf8SEric W. Biederman goto done; 7533895dbf8SEric W. Biederman 7543895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7553895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7564edbe133SAl Viro new->m_dentry = dget(dentry); 7573895dbf8SEric W. Biederman new->m_count = 1; 7583895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7593895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7603895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7613895dbf8SEric W. Biederman 7623895dbf8SEric W. Biederman mp = new; 7633895dbf8SEric W. Biederman new = NULL; 7643895dbf8SEric W. Biederman done: 7653895dbf8SEric W. Biederman kfree(new); 76684d17192SAl Viro return mp; 76784d17192SAl Viro } 76884d17192SAl Viro 7694edbe133SAl Viro /* 7704edbe133SAl Viro * vfsmount lock must be held. Additionally, the caller is responsible 7714edbe133SAl Viro * for serializing calls for given disposal list. 7724edbe133SAl Viro */ 7734edbe133SAl Viro static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) 77484d17192SAl Viro { 77584d17192SAl Viro if (!--mp->m_count) { 77684d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7770a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 77884d17192SAl Viro spin_lock(&dentry->d_lock); 77984d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 78084d17192SAl Viro spin_unlock(&dentry->d_lock); 7814edbe133SAl Viro dput_to_list(dentry, list); 7820818bf27SAl Viro hlist_del(&mp->m_hash); 78384d17192SAl Viro kfree(mp); 78484d17192SAl Viro } 78584d17192SAl Viro } 78684d17192SAl Viro 7874edbe133SAl Viro /* called with namespace_lock and vfsmount lock */ 7884edbe133SAl Viro static void put_mountpoint(struct mountpoint *mp) 7894edbe133SAl Viro { 7904edbe133SAl Viro __put_mountpoint(mp, &ex_mountpoints); 7914edbe133SAl Viro } 7924edbe133SAl Viro 793143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7941da177e4SLinus Torvalds { 7956b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7961da177e4SLinus Torvalds } 7971da177e4SLinus Torvalds 79899b7db7bSNick Piggin /* 79999b7db7bSNick Piggin * vfsmount lock must be held for write 80099b7db7bSNick Piggin */ 8016b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 8025addc5ddSAl Viro { 8035addc5ddSAl Viro if (ns) { 8045addc5ddSAl Viro ns->event = ++event; 8055addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8065addc5ddSAl Viro } 8075addc5ddSAl Viro } 8085addc5ddSAl Viro 80999b7db7bSNick Piggin /* 81099b7db7bSNick Piggin * vfsmount lock must be held for write 81199b7db7bSNick Piggin */ 8126b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8135addc5ddSAl Viro { 8145addc5ddSAl Viro if (ns && ns->event != event) { 8155addc5ddSAl Viro ns->event = event; 8165addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8175addc5ddSAl Viro } 8185addc5ddSAl Viro } 8195addc5ddSAl Viro 82099b7db7bSNick Piggin /* 82199b7db7bSNick Piggin * vfsmount lock must be held for write 82299b7db7bSNick Piggin */ 823e4e59906SAl Viro static struct mountpoint *unhash_mnt(struct mount *mnt) 8241da177e4SLinus Torvalds { 825e4e59906SAl Viro struct mountpoint *mp; 8260714a533SAl Viro mnt->mnt_parent = mnt; 827a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8286b41d536SAl Viro list_del_init(&mnt->mnt_child); 82938129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8300a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 831e4e59906SAl Viro mp = mnt->mnt_mp; 83284d17192SAl Viro mnt->mnt_mp = NULL; 833e4e59906SAl Viro return mp; 8347bdb11deSEric W. Biederman } 8357bdb11deSEric W. Biederman 8367bdb11deSEric W. Biederman /* 8377bdb11deSEric W. Biederman * vfsmount lock must be held for write 8387bdb11deSEric W. Biederman */ 8396a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8406a46c573SEric W. Biederman { 841e4e59906SAl Viro put_mountpoint(unhash_mnt(mnt)); 8426a46c573SEric W. Biederman } 8436a46c573SEric W. Biederman 8446a46c573SEric W. Biederman /* 8456a46c573SEric W. Biederman * vfsmount lock must be held for write 8466a46c573SEric W. Biederman */ 84784d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 84884d17192SAl Viro struct mountpoint *mp, 84944d964d6SAl Viro struct mount *child_mnt) 850b90fa9aeSRam Pai { 85184d17192SAl Viro mp->m_count++; 8523a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 8534edbe133SAl Viro child_mnt->mnt_mountpoint = mp->m_dentry; 8543a2393d7SAl Viro child_mnt->mnt_parent = mnt; 85584d17192SAl Viro child_mnt->mnt_mp = mp; 8560a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 857b90fa9aeSRam Pai } 858b90fa9aeSRam Pai 8591064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8601064f874SEric W. Biederman { 8611064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8621064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8631064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8641064f874SEric W. Biederman } 8651064f874SEric W. Biederman 86699b7db7bSNick Piggin /* 86799b7db7bSNick Piggin * vfsmount lock must be held for write 86899b7db7bSNick Piggin */ 86984d17192SAl Viro static void attach_mnt(struct mount *mnt, 87084d17192SAl Viro struct mount *parent, 87184d17192SAl Viro struct mountpoint *mp) 8721da177e4SLinus Torvalds { 87384d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8741064f874SEric W. Biederman __attach_mnt(mnt, parent); 875b90fa9aeSRam Pai } 876b90fa9aeSRam Pai 8771064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 87812a5b529SAl Viro { 8791064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8801064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8811064f874SEric W. Biederman 8821064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8831064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8841064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8851064f874SEric W. Biederman 8861064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 8871064f874SEric W. Biederman 8881064f874SEric W. Biederman put_mountpoint(old_mp); 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 9438f291889SAl Viro /** 9448f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 9458f291889SAl Viro * @fc: The configuration context with the superblock attached 9468f291889SAl Viro * 9478f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 9488f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 9498f291889SAl Viro * 9508f291889SAl Viro * Note that this does not attach the mount to anything. 9518f291889SAl Viro */ 9528f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 9539d412a43SAl Viro { 954b105e270SAl Viro struct mount *mnt; 9559d412a43SAl Viro 9568f291889SAl Viro if (!fc->root) 9578f291889SAl Viro return ERR_PTR(-EINVAL); 9589d412a43SAl Viro 9598f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 9609d412a43SAl Viro if (!mnt) 9619d412a43SAl Viro return ERR_PTR(-ENOMEM); 9629d412a43SAl Viro 9638f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 964b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9659d412a43SAl Viro 9668f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 9678f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 9688f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 969a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9700714a533SAl Viro mnt->mnt_parent = mnt; 9718f291889SAl Viro 972719ea2fbSAl Viro lock_mount_hash(); 9738f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 974719ea2fbSAl Viro unlock_mount_hash(); 975b105e270SAl Viro return &mnt->mnt; 9769d412a43SAl Viro } 9778f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 9788f291889SAl Viro 9798f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 9808f291889SAl Viro { 9818f291889SAl Viro int err = vfs_get_tree(fc); 9828f291889SAl Viro if (!err) { 9838f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 9848f291889SAl Viro return vfs_create_mount(fc); 9858f291889SAl Viro } 9868f291889SAl Viro return ERR_PTR(err); 9878f291889SAl Viro } 9888f291889SAl Viro EXPORT_SYMBOL(fc_mount); 9898f291889SAl Viro 9909bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 9919bc61ab1SDavid Howells int flags, const char *name, 9929bc61ab1SDavid Howells void *data) 9931da177e4SLinus Torvalds { 9949bc61ab1SDavid Howells struct fs_context *fc; 9958f291889SAl Viro struct vfsmount *mnt; 9969bc61ab1SDavid Howells int ret = 0; 9979d412a43SAl Viro 9989d412a43SAl Viro if (!type) 9993e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 10009d412a43SAl Viro 10019bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 10029bc61ab1SDavid Howells if (IS_ERR(fc)) 10039bc61ab1SDavid Howells return ERR_CAST(fc); 10049bc61ab1SDavid Howells 10053e1aeb00SDavid Howells if (name) 10063e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 10073e1aeb00SDavid Howells name, strlen(name)); 10089bc61ab1SDavid Howells if (!ret) 10099bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 10109bc61ab1SDavid Howells if (!ret) 10118f291889SAl Viro mnt = fc_mount(fc); 10128f291889SAl Viro else 10138f291889SAl Viro mnt = ERR_PTR(ret); 10149d412a43SAl Viro 10159bc61ab1SDavid Howells put_fs_context(fc); 10168f291889SAl Viro return mnt; 10179d412a43SAl Viro } 10189d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10199d412a43SAl Viro 102093faccbbSEric W. Biederman struct vfsmount * 102193faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 102293faccbbSEric W. Biederman const char *name, void *data) 102393faccbbSEric W. Biederman { 102493faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 102593faccbbSEric W. Biederman * through from the parent mount to the submount don't support 102693faccbbSEric W. Biederman * unprivileged mounts with submounts. 102793faccbbSEric W. Biederman */ 102893faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 102993faccbbSEric W. Biederman return ERR_PTR(-EPERM); 103093faccbbSEric W. Biederman 1031e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 103293faccbbSEric W. Biederman } 103393faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 103493faccbbSEric W. Biederman 103587129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 103636341f64SRam Pai int flag) 10371da177e4SLinus Torvalds { 103887129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1039be34d1a3SDavid Howells struct mount *mnt; 1040be34d1a3SDavid Howells int err; 10411da177e4SLinus Torvalds 1042be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1043be34d1a3SDavid Howells if (!mnt) 1044be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1045be34d1a3SDavid Howells 10467a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 104715169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1048719f5d7fSMiklos Szeredi else 104915169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1050719f5d7fSMiklos Szeredi 105115169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1052be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1053719f5d7fSMiklos Szeredi if (err) 1054719f5d7fSMiklos Szeredi goto out_free; 1055719f5d7fSMiklos Szeredi } 1056719f5d7fSMiklos Szeredi 105716a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 105816a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 10595ff9d8a6SEric W. Biederman 10601da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1061b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1062b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1063a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10640714a533SAl Viro mnt->mnt_parent = mnt; 1065719ea2fbSAl Viro lock_mount_hash(); 106639f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1067719ea2fbSAl Viro unlock_mount_hash(); 1068b90fa9aeSRam Pai 10697a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10707a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10716776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 107232301920SAl Viro mnt->mnt_master = old; 1073fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10748aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1075fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10766776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1077d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10786776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1079d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10805235d448SAl Viro } else { 10815235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10825afe0022SRam Pai } 1083b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10840f0afb1dSAl Viro set_mnt_shared(mnt); 10851da177e4SLinus Torvalds 10861da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10871da177e4SLinus Torvalds * as the original if that was on one */ 108836341f64SRam Pai if (flag & CL_EXPIRE) { 10896776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10906776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10911da177e4SLinus Torvalds } 1092be34d1a3SDavid Howells 1093cb338d06SAl Viro return mnt; 1094719f5d7fSMiklos Szeredi 1095719f5d7fSMiklos Szeredi out_free: 10968ffcb32eSDavid Howells mnt_free_id(mnt); 1097719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1098be34d1a3SDavid Howells return ERR_PTR(err); 10991da177e4SLinus Torvalds } 11001da177e4SLinus Torvalds 11019ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11029ea459e1SAl Viro { 110356cbb429SAl Viro struct hlist_node *p; 110456cbb429SAl Viro struct mount *m; 11059ea459e1SAl Viro /* 110656cbb429SAl Viro * The warning here probably indicates that somebody messed 110756cbb429SAl Viro * up a mnt_want/drop_write() pair. If this happens, the 110856cbb429SAl Viro * filesystem was probably unable to make r/w->r/o transitions. 11099ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11109ea459e1SAl Viro * so mnt_get_writers() below is safe. 11119ea459e1SAl Viro */ 11129ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11139ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11149ea459e1SAl Viro mnt_pin_kill(mnt); 111556cbb429SAl Viro hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { 111656cbb429SAl Viro hlist_del(&m->mnt_umount); 111756cbb429SAl Viro mntput(&m->mnt); 111856cbb429SAl Viro } 11199ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11209ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11219ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11229ea459e1SAl Viro mnt_free_id(mnt); 11239ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11249ea459e1SAl Viro } 11259ea459e1SAl Viro 11269ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11279ea459e1SAl Viro { 11289ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11299ea459e1SAl Viro } 11309ea459e1SAl Viro 11319ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11329ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11339ea459e1SAl Viro { 11349ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 113529785735SByungchul Park struct mount *m, *t; 11369ea459e1SAl Viro 113729785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 113829785735SByungchul Park cleanup_mnt(m); 11399ea459e1SAl Viro } 11409ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11419ea459e1SAl Viro 1142900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11437b7b1aceSAl Viro { 11444edbe133SAl Viro LIST_HEAD(list); 11454edbe133SAl Viro 114648a066e7SAl Viro rcu_read_lock(); 11479ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11489ea0a46cSAl Viro /* 11499ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11509ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11519ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11529ea0a46cSAl Viro * be dropped until after an RCU delay done after 11539ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11549ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11559ea0a46cSAl Viro * we are dropping is not the final one. 11569ea0a46cSAl Viro */ 1157aa9c0e07SAl Viro mnt_add_count(mnt, -1); 115848a066e7SAl Viro rcu_read_unlock(); 115999b7db7bSNick Piggin return; 1160b3e19d92SNick Piggin } 1161719ea2fbSAl Viro lock_mount_hash(); 1162119e1ef8SAl Viro /* 1163119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1164119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1165119e1ef8SAl Viro */ 1166119e1ef8SAl Viro smp_mb(); 11679ea0a46cSAl Viro mnt_add_count(mnt, -1); 1168b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 116948a066e7SAl Viro rcu_read_unlock(); 1170719ea2fbSAl Viro unlock_mount_hash(); 117199b7db7bSNick Piggin return; 117299b7db7bSNick Piggin } 117348a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 117448a066e7SAl Viro rcu_read_unlock(); 117548a066e7SAl Viro unlock_mount_hash(); 117648a066e7SAl Viro return; 117748a066e7SAl Viro } 117848a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 117948a066e7SAl Viro rcu_read_unlock(); 1180962830dfSAndi Kleen 118139f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1182ce07d891SEric W. Biederman 1183ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1184ce07d891SEric W. Biederman struct mount *p, *tmp; 1185ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 11864edbe133SAl Viro __put_mountpoint(unhash_mnt(p), &list); 118756cbb429SAl Viro hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); 1188ce07d891SEric W. Biederman } 1189ce07d891SEric W. Biederman } 1190719ea2fbSAl Viro unlock_mount_hash(); 11914edbe133SAl Viro shrink_dentry_list(&list); 1192649a795aSAl Viro 11939ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11949ea459e1SAl Viro struct task_struct *task = current; 11959ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11969ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11979ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11989ea459e1SAl Viro return; 11999ea459e1SAl Viro } 12009ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 12019ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 12029ea459e1SAl Viro return; 12039ea459e1SAl Viro } 12049ea459e1SAl Viro cleanup_mnt(mnt); 1205b3e19d92SNick Piggin } 1206b3e19d92SNick Piggin 1207b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1208b3e19d92SNick Piggin { 1209b3e19d92SNick Piggin if (mnt) { 1210863d684fSAl Viro struct mount *m = real_mount(mnt); 1211b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1212863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1213863d684fSAl Viro m->mnt_expiry_mark = 0; 1214863d684fSAl Viro mntput_no_expire(m); 1215b3e19d92SNick Piggin } 1216b3e19d92SNick Piggin } 1217b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1218b3e19d92SNick Piggin 1219b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1220b3e19d92SNick Piggin { 1221b3e19d92SNick Piggin if (mnt) 122283adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1223b3e19d92SNick Piggin return mnt; 1224b3e19d92SNick Piggin } 1225b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1226b3e19d92SNick Piggin 1227c6609c0aSIan Kent /* path_is_mountpoint() - Check if path is a mount in the current 1228c6609c0aSIan Kent * namespace. 1229c6609c0aSIan Kent * 1230c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1231c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1232c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1233c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1234c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1235c6609c0aSIan Kent * alone. 1236c6609c0aSIan Kent */ 1237c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1238c6609c0aSIan Kent { 1239c6609c0aSIan Kent unsigned seq; 1240c6609c0aSIan Kent bool res; 1241c6609c0aSIan Kent 1242c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1243c6609c0aSIan Kent return false; 1244c6609c0aSIan Kent 1245c6609c0aSIan Kent rcu_read_lock(); 1246c6609c0aSIan Kent do { 1247c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1248c6609c0aSIan Kent res = __path_is_mountpoint(path); 1249c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1250c6609c0aSIan Kent rcu_read_unlock(); 1251c6609c0aSIan Kent 1252c6609c0aSIan Kent return res; 1253c6609c0aSIan Kent } 1254c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1255c6609c0aSIan Kent 1256ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12577b7b1aceSAl Viro { 12583064c356SAl Viro struct mount *p; 12593064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12603064c356SAl Viro if (IS_ERR(p)) 12613064c356SAl Viro return ERR_CAST(p); 12623064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12633064c356SAl Viro return &p->mnt; 12647b7b1aceSAl Viro } 12651da177e4SLinus Torvalds 1266a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12679f6c61f9SMiklos Szeredi static struct mount *mnt_list_next(struct mnt_namespace *ns, 12689f6c61f9SMiklos Szeredi struct list_head *p) 12699f6c61f9SMiklos Szeredi { 12709f6c61f9SMiklos Szeredi struct mount *mnt, *ret = NULL; 12719f6c61f9SMiklos Szeredi 12729f6c61f9SMiklos Szeredi lock_ns_list(ns); 12739f6c61f9SMiklos Szeredi list_for_each_continue(p, &ns->list) { 12749f6c61f9SMiklos Szeredi mnt = list_entry(p, typeof(*mnt), mnt_list); 12759f6c61f9SMiklos Szeredi if (!mnt_is_cursor(mnt)) { 12769f6c61f9SMiklos Szeredi ret = mnt; 12779f6c61f9SMiklos Szeredi break; 12789f6c61f9SMiklos Szeredi } 12799f6c61f9SMiklos Szeredi } 12809f6c61f9SMiklos Szeredi unlock_ns_list(ns); 12819f6c61f9SMiklos Szeredi 12829f6c61f9SMiklos Szeredi return ret; 12839f6c61f9SMiklos Szeredi } 12849f6c61f9SMiklos Szeredi 12850226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12861da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12871da177e4SLinus Torvalds { 1288ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12899f6c61f9SMiklos Szeredi struct list_head *prev; 12901da177e4SLinus Torvalds 1291390c6843SRam Pai down_read(&namespace_sem); 12929f6c61f9SMiklos Szeredi if (!*pos) { 12939f6c61f9SMiklos Szeredi prev = &p->ns->list; 12949f6c61f9SMiklos Szeredi } else { 12959f6c61f9SMiklos Szeredi prev = &p->cursor.mnt_list; 12969f6c61f9SMiklos Szeredi 12979f6c61f9SMiklos Szeredi /* Read after we'd reached the end? */ 12989f6c61f9SMiklos Szeredi if (list_empty(prev)) 12999f6c61f9SMiklos Szeredi return NULL; 1300c7999c36SAl Viro } 1301c7999c36SAl Viro 13029f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, prev); 13031da177e4SLinus Torvalds } 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 13061da177e4SLinus Torvalds { 1307ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13089f6c61f9SMiklos Szeredi struct mount *mnt = v; 1309b0765fb8SPavel Emelianov 13109f6c61f9SMiklos Szeredi ++*pos; 13119f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, &mnt->mnt_list); 13121da177e4SLinus Torvalds } 13131da177e4SLinus Torvalds 13141da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 13151da177e4SLinus Torvalds { 13169f6c61f9SMiklos Szeredi struct proc_mounts *p = m->private; 13179f6c61f9SMiklos Szeredi struct mount *mnt = v; 13189f6c61f9SMiklos Szeredi 13199f6c61f9SMiklos Szeredi lock_ns_list(p->ns); 13209f6c61f9SMiklos Szeredi if (mnt) 13219f6c61f9SMiklos Szeredi list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); 13229f6c61f9SMiklos Szeredi else 13239f6c61f9SMiklos Szeredi list_del_init(&p->cursor.mnt_list); 13249f6c61f9SMiklos Szeredi unlock_ns_list(p->ns); 1325390c6843SRam Pai up_read(&namespace_sem); 13261da177e4SLinus Torvalds } 13271da177e4SLinus Torvalds 13280226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13299f5596afSAl Viro { 1330ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13319f6c61f9SMiklos Szeredi struct mount *r = v; 13320226f492SAl Viro return p->show(m, &r->mnt); 13331da177e4SLinus Torvalds } 13341da177e4SLinus Torvalds 1335a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13361da177e4SLinus Torvalds .start = m_start, 13371da177e4SLinus Torvalds .next = m_next, 13381da177e4SLinus Torvalds .stop = m_stop, 13390226f492SAl Viro .show = m_show, 1340b4629fe2SChuck Lever }; 13419f6c61f9SMiklos Szeredi 13429f6c61f9SMiklos Szeredi void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) 13439f6c61f9SMiklos Szeredi { 13449f6c61f9SMiklos Szeredi down_read(&namespace_sem); 13459f6c61f9SMiklos Szeredi lock_ns_list(ns); 13469f6c61f9SMiklos Szeredi list_del(&cursor->mnt_list); 13479f6c61f9SMiklos Szeredi unlock_ns_list(ns); 13489f6c61f9SMiklos Szeredi up_read(&namespace_sem); 13499f6c61f9SMiklos Szeredi } 1350a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1351b4629fe2SChuck Lever 13521da177e4SLinus Torvalds /** 13531da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 13541da177e4SLinus Torvalds * @mnt: root of mount tree 13551da177e4SLinus Torvalds * 13561da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 13571da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 13581da177e4SLinus Torvalds * busy. 13591da177e4SLinus Torvalds */ 1360909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13611da177e4SLinus Torvalds { 1362909b0a88SAl Viro struct mount *mnt = real_mount(m); 136336341f64SRam Pai int actual_refs = 0; 136436341f64SRam Pai int minimum_refs = 0; 1365315fc83eSAl Viro struct mount *p; 1366909b0a88SAl Viro BUG_ON(!m); 13671da177e4SLinus Torvalds 1368b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1369719ea2fbSAl Viro lock_mount_hash(); 1370909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 137183adc753SAl Viro actual_refs += mnt_get_count(p); 13721da177e4SLinus Torvalds minimum_refs += 2; 13731da177e4SLinus Torvalds } 1374719ea2fbSAl Viro unlock_mount_hash(); 13751da177e4SLinus Torvalds 13761da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13771da177e4SLinus Torvalds return 0; 1378e3474a8eSIan Kent 1379e3474a8eSIan Kent return 1; 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds 13821da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13831da177e4SLinus Torvalds 13841da177e4SLinus Torvalds /** 13851da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13861da177e4SLinus Torvalds * @mnt: root of mount 13871da177e4SLinus Torvalds * 13881da177e4SLinus Torvalds * This is called to check if a mount point has any 13891da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13901da177e4SLinus Torvalds * mount has sub mounts this will return busy 13911da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13921da177e4SLinus Torvalds * 13931da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13941da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13951da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13961da177e4SLinus Torvalds */ 13971da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13981da177e4SLinus Torvalds { 1399e3474a8eSIan Kent int ret = 1; 14008ad08d8aSAl Viro down_read(&namespace_sem); 1401719ea2fbSAl Viro lock_mount_hash(); 14021ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1403e3474a8eSIan Kent ret = 0; 1404719ea2fbSAl Viro unlock_mount_hash(); 14058ad08d8aSAl Viro up_read(&namespace_sem); 1406a05964f3SRam Pai return ret; 14071da177e4SLinus Torvalds } 14081da177e4SLinus Torvalds 14091da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 14101da177e4SLinus Torvalds 141197216be0SAl Viro static void namespace_unlock(void) 14121da177e4SLinus Torvalds { 1413a3b3c562SEric W. Biederman struct hlist_head head; 141456cbb429SAl Viro struct hlist_node *p; 141556cbb429SAl Viro struct mount *m; 14164edbe133SAl Viro LIST_HEAD(list); 141797216be0SAl Viro 1418a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 14194edbe133SAl Viro list_splice_init(&ex_mountpoints, &list); 1420a3b3c562SEric W. Biederman 142197216be0SAl Viro up_write(&namespace_sem); 1422a3b3c562SEric W. Biederman 14234edbe133SAl Viro shrink_dentry_list(&list); 14244edbe133SAl Viro 1425a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 142697216be0SAl Viro return; 142797216be0SAl Viro 142822cb7405SNeilBrown synchronize_rcu_expedited(); 142948a066e7SAl Viro 143056cbb429SAl Viro hlist_for_each_entry_safe(m, p, &head, mnt_umount) { 143156cbb429SAl Viro hlist_del(&m->mnt_umount); 143256cbb429SAl Viro mntput(&m->mnt); 143356cbb429SAl Viro } 143470fbcdf4SRam Pai } 143570fbcdf4SRam Pai 143697216be0SAl Viro static inline void namespace_lock(void) 1437e3197d83SAl Viro { 143897216be0SAl Viro down_write(&namespace_sem); 1439e3197d83SAl Viro } 1440e3197d83SAl Viro 1441e819f152SEric W. Biederman enum umount_tree_flags { 1442e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1443e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1444e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1445e819f152SEric W. Biederman }; 1446f2d0a123SEric W. Biederman 1447f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1448f2d0a123SEric W. Biederman { 1449f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1450f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1451f2d0a123SEric W. Biederman return true; 1452f2d0a123SEric W. Biederman 1453f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1454f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1455f2d0a123SEric W. Biederman return true; 1456f2d0a123SEric W. Biederman 1457f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1458f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1459f2d0a123SEric W. Biederman * connected to mounted mounts. 1460f2d0a123SEric W. Biederman */ 1461f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1462f2d0a123SEric W. Biederman return true; 1463f2d0a123SEric W. Biederman 1464f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1465f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1466f2d0a123SEric W. Biederman return false; 1467f2d0a123SEric W. Biederman 1468f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1469f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1470f2d0a123SEric W. Biederman return false; 1471f2d0a123SEric W. Biederman 1472f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1473f2d0a123SEric W. Biederman return true; 1474f2d0a123SEric W. Biederman } 1475f2d0a123SEric W. Biederman 147699b7db7bSNick Piggin /* 147748a066e7SAl Viro * mount_lock must be held 147899b7db7bSNick Piggin * namespace_sem must be held for write 147999b7db7bSNick Piggin */ 1480e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 148170fbcdf4SRam Pai { 1482c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1483315fc83eSAl Viro struct mount *p; 148470fbcdf4SRam Pai 14855d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14865d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14875d88457eSEric W. Biederman 1488c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1489590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1490590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1491c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1492590ce4bcSEric W. Biederman } 1493c003b26fSEric W. Biederman 1494411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1495c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1496c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 149738129a13SAl Viro } 149870fbcdf4SRam Pai 1499c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1500e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 15017b8a53fdSAl Viro propagate_umount(&tmp_list); 1502a05964f3SRam Pai 1503c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1504d2921684SEric W. Biederman struct mnt_namespace *ns; 1505ce07d891SEric W. Biederman bool disconnect; 1506c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 15076776db3dSAl Viro list_del_init(&p->mnt_expire); 15081a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1509d2921684SEric W. Biederman ns = p->mnt_ns; 1510d2921684SEric W. Biederman if (ns) { 1511d2921684SEric W. Biederman ns->mounts--; 1512d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1513d2921684SEric W. Biederman } 151463d37a84SAl Viro p->mnt_ns = NULL; 1515e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 151648a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 151787b95ce0SAl Viro 1518f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1519676da58dSAl Viro if (mnt_has_parent(p)) { 152081b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1521ce07d891SEric W. Biederman if (!disconnect) { 1522ce07d891SEric W. Biederman /* Don't forget about p */ 1523ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1524ce07d891SEric W. Biederman } else { 15256a46c573SEric W. Biederman umount_mnt(p); 15267c4b93d8SAl Viro } 1527ce07d891SEric W. Biederman } 15280f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 152919a1c409SAl Viro if (disconnect) 153019a1c409SAl Viro hlist_add_head(&p->mnt_umount, &unmounted); 153138129a13SAl Viro } 15321da177e4SLinus Torvalds } 15331da177e4SLinus Torvalds 1534b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1535c35038beSAl Viro 15368d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 15378d0347f6SDavid Howells { 15388d0347f6SDavid Howells int ret = 0; 15398d0347f6SDavid Howells 15408d0347f6SDavid Howells down_write(&sb->s_umount); 15418d0347f6SDavid Howells if (!sb_rdonly(sb)) { 15428d0347f6SDavid Howells struct fs_context *fc; 15438d0347f6SDavid Howells 15448d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 15458d0347f6SDavid Howells SB_RDONLY); 15468d0347f6SDavid Howells if (IS_ERR(fc)) { 15478d0347f6SDavid Howells ret = PTR_ERR(fc); 15488d0347f6SDavid Howells } else { 15498d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 15508d0347f6SDavid Howells if (!ret) 15518d0347f6SDavid Howells ret = reconfigure_super(fc); 15528d0347f6SDavid Howells put_fs_context(fc); 15538d0347f6SDavid Howells } 15548d0347f6SDavid Howells } 15558d0347f6SDavid Howells up_write(&sb->s_umount); 15568d0347f6SDavid Howells return ret; 15578d0347f6SDavid Howells } 15588d0347f6SDavid Howells 15591ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 15601da177e4SLinus Torvalds { 15611ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 15621da177e4SLinus Torvalds int retval; 15631da177e4SLinus Torvalds 15641ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 15651da177e4SLinus Torvalds if (retval) 15661da177e4SLinus Torvalds return retval; 15671da177e4SLinus Torvalds 15681da177e4SLinus Torvalds /* 15691da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 15701da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 15711da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 15721da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 15731da177e4SLinus Torvalds */ 15741da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 15751ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 15761da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 15771da177e4SLinus Torvalds return -EINVAL; 15781da177e4SLinus Torvalds 1579b3e19d92SNick Piggin /* 1580b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1581b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1582b3e19d92SNick Piggin */ 1583719ea2fbSAl Viro lock_mount_hash(); 158483adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1585719ea2fbSAl Viro unlock_mount_hash(); 15861da177e4SLinus Torvalds return -EBUSY; 1587b3e19d92SNick Piggin } 1588719ea2fbSAl Viro unlock_mount_hash(); 15891da177e4SLinus Torvalds 1590863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 15911da177e4SLinus Torvalds return -EAGAIN; 15921da177e4SLinus Torvalds } 15931da177e4SLinus Torvalds 15941da177e4SLinus Torvalds /* 15951da177e4SLinus Torvalds * If we may have to abort operations to get out of this 15961da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 15971da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 15981da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 15991da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 16001da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 16011da177e4SLinus Torvalds * about for the moment. 16021da177e4SLinus Torvalds */ 16031da177e4SLinus Torvalds 160442faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 160542faad99SAl Viro sb->s_op->umount_begin(sb); 160642faad99SAl Viro } 16071da177e4SLinus Torvalds 16081da177e4SLinus Torvalds /* 16091da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 16101da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 16111da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 16121da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 16131da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 16141da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 16151da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 16161da177e4SLinus Torvalds */ 16171ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 16181da177e4SLinus Torvalds /* 16191da177e4SLinus Torvalds * Special case for "unmounting" root ... 16201da177e4SLinus Torvalds * we just try to remount it readonly. 16211da177e4SLinus Torvalds */ 1622bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1623a1480dccSAndy Lutomirski return -EPERM; 16248d0347f6SDavid Howells return do_umount_root(sb); 16251da177e4SLinus Torvalds } 16261da177e4SLinus Torvalds 162797216be0SAl Viro namespace_lock(); 1628719ea2fbSAl Viro lock_mount_hash(); 16291da177e4SLinus Torvalds 163025d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 163125d202edSEric W. Biederman retval = -EINVAL; 163225d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 163325d202edSEric W. Biederman goto out; 163425d202edSEric W. Biederman 163525d202edSEric W. Biederman event++; 163648a066e7SAl Viro if (flags & MNT_DETACH) { 163748a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1638e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 163948a066e7SAl Viro retval = 0; 164048a066e7SAl Viro } else { 1641b54b9be7SAl Viro shrink_submounts(mnt); 16421da177e4SLinus Torvalds retval = -EBUSY; 164348a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 16441a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1645e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 16461da177e4SLinus Torvalds retval = 0; 16471da177e4SLinus Torvalds } 164848a066e7SAl Viro } 164925d202edSEric W. Biederman out: 1650719ea2fbSAl Viro unlock_mount_hash(); 1651e3197d83SAl Viro namespace_unlock(); 16521da177e4SLinus Torvalds return retval; 16531da177e4SLinus Torvalds } 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds /* 165680b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 165780b5dce8SEric W. Biederman * 165880b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 165980b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 166080b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 166180b5dce8SEric W. Biederman * leaking them. 166280b5dce8SEric W. Biederman * 166380b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 166480b5dce8SEric W. Biederman */ 166580b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 166680b5dce8SEric W. Biederman { 166780b5dce8SEric W. Biederman struct mountpoint *mp; 166880b5dce8SEric W. Biederman struct mount *mnt; 166980b5dce8SEric W. Biederman 167080b5dce8SEric W. Biederman namespace_lock(); 16713895dbf8SEric W. Biederman lock_mount_hash(); 167280b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1673adc9b5c0SAl Viro if (!mp) 167480b5dce8SEric W. Biederman goto out_unlock; 167580b5dce8SEric W. Biederman 1676e06b933eSAndrey Ulanov event++; 167780b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 167880b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1679ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1680fe78fcc8SEric W. Biederman umount_mnt(mnt); 168156cbb429SAl Viro hlist_add_head(&mnt->mnt_umount, &unmounted); 1682ce07d891SEric W. Biederman } 1683e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 168480b5dce8SEric W. Biederman } 168580b5dce8SEric W. Biederman put_mountpoint(mp); 168680b5dce8SEric W. Biederman out_unlock: 16873895dbf8SEric W. Biederman unlock_mount_hash(); 168880b5dce8SEric W. Biederman namespace_unlock(); 168980b5dce8SEric W. Biederman } 169080b5dce8SEric W. Biederman 169180b5dce8SEric W. Biederman /* 16929b40bc90SAl Viro * Is the caller allowed to modify his namespace? 16939b40bc90SAl Viro */ 16949b40bc90SAl Viro static inline bool may_mount(void) 16959b40bc90SAl Viro { 16969b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 16979b40bc90SAl Viro } 16989b40bc90SAl Viro 1699df2474a2SJeff Layton #ifdef CONFIG_MANDATORY_FILE_LOCKING 17009e8925b6SJeff Layton static inline bool may_mandlock(void) 17019e8925b6SJeff Layton { 170295ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 17039e8925b6SJeff Layton } 1704df2474a2SJeff Layton #else 1705df2474a2SJeff Layton static inline bool may_mandlock(void) 1706df2474a2SJeff Layton { 1707df2474a2SJeff Layton pr_warn("VFS: \"mand\" mount option not supported"); 1708df2474a2SJeff Layton return false; 1709df2474a2SJeff Layton } 1710df2474a2SJeff Layton #endif 17119e8925b6SJeff Layton 17129b40bc90SAl Viro /* 17131da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 17141da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 17151da177e4SLinus Torvalds * 17161da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 17171da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 17181da177e4SLinus Torvalds */ 17191da177e4SLinus Torvalds 17203a18ef5cSDominik Brodowski int ksys_umount(char __user *name, int flags) 17211da177e4SLinus Torvalds { 17222d8f3038SAl Viro struct path path; 1723900148dcSAl Viro struct mount *mnt; 17241da177e4SLinus Torvalds int retval; 1725161aff1dSAl Viro int lookup_flags = LOOKUP_MOUNTPOINT; 17261da177e4SLinus Torvalds 1727db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1728db1f05bbSMiklos Szeredi return -EINVAL; 1729db1f05bbSMiklos Szeredi 17309b40bc90SAl Viro if (!may_mount()) 17319b40bc90SAl Viro return -EPERM; 17329b40bc90SAl Viro 1733db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1734db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1735db1f05bbSMiklos Szeredi 1736161aff1dSAl Viro retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); 17371da177e4SLinus Torvalds if (retval) 17381da177e4SLinus Torvalds goto out; 1739900148dcSAl Viro mnt = real_mount(path.mnt); 17401da177e4SLinus Torvalds retval = -EINVAL; 17412d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 17421da177e4SLinus Torvalds goto dput_and_out; 1743143c8c91SAl Viro if (!check_mnt(mnt)) 17441da177e4SLinus Torvalds goto dput_and_out; 174525d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 17465ff9d8a6SEric W. Biederman goto dput_and_out; 1747b2f5d4dcSEric W. Biederman retval = -EPERM; 1748b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1749b2f5d4dcSEric W. Biederman goto dput_and_out; 17501da177e4SLinus Torvalds 1751900148dcSAl Viro retval = do_umount(mnt, flags); 17521da177e4SLinus Torvalds dput_and_out: 1753429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 17542d8f3038SAl Viro dput(path.dentry); 1755900148dcSAl Viro mntput_no_expire(mnt); 17561da177e4SLinus Torvalds out: 17571da177e4SLinus Torvalds return retval; 17581da177e4SLinus Torvalds } 17591da177e4SLinus Torvalds 17603a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 17613a18ef5cSDominik Brodowski { 17623a18ef5cSDominik Brodowski return ksys_umount(name, flags); 17633a18ef5cSDominik Brodowski } 17643a18ef5cSDominik Brodowski 17651da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds /* 17681da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17691da177e4SLinus Torvalds */ 1770bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17711da177e4SLinus Torvalds { 17723a18ef5cSDominik Brodowski return ksys_umount(name, 0); 17731da177e4SLinus Torvalds } 17741da177e4SLinus Torvalds 17751da177e4SLinus Torvalds #endif 17761da177e4SLinus Torvalds 17774ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 17788823c079SEric W. Biederman { 17794ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1780e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1781e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 17824ce5d2b1SEric W. Biederman } 17834ce5d2b1SEric W. Biederman 1784213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 178558be2825SAl Viro { 178658be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 178758be2825SAl Viro } 178858be2825SAl Viro 17894ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 17904ce5d2b1SEric W. Biederman { 17914ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 17924ce5d2b1SEric W. Biederman * mount namespace loop? 17934ce5d2b1SEric W. Biederman */ 17944ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 17954ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 17964ce5d2b1SEric W. Biederman return false; 17974ce5d2b1SEric W. Biederman 1798f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 17998823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 18008823c079SEric W. Biederman } 18018823c079SEric W. Biederman 180287129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 180336341f64SRam Pai int flag) 18041da177e4SLinus Torvalds { 180584d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 18061da177e4SLinus Torvalds 18074ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 18084ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 18094ce5d2b1SEric W. Biederman 18104ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1811be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 18129676f0c6SRam Pai 181336341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1814be34d1a3SDavid Howells if (IS_ERR(q)) 1815be34d1a3SDavid Howells return q; 1816be34d1a3SDavid Howells 1817a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 18181da177e4SLinus Torvalds 18191da177e4SLinus Torvalds p = mnt; 18206b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1821315fc83eSAl Viro struct mount *s; 18227ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 18231da177e4SLinus Torvalds continue; 18241da177e4SLinus Torvalds 1825909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 18264ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 18274ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1828df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1829df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1830df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1831df7342b2SEric W. Biederman goto out; 1832df7342b2SEric W. Biederman } else { 18334ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 18344ce5d2b1SEric W. Biederman continue; 18354ce5d2b1SEric W. Biederman } 1836df7342b2SEric W. Biederman } 18374ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 18384ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 18399676f0c6SRam Pai s = skip_mnt_tree(s); 18409676f0c6SRam Pai continue; 18419676f0c6SRam Pai } 18420714a533SAl Viro while (p != s->mnt_parent) { 18430714a533SAl Viro p = p->mnt_parent; 18440714a533SAl Viro q = q->mnt_parent; 18451da177e4SLinus Torvalds } 184687129cc0SAl Viro p = s; 184784d17192SAl Viro parent = q; 184887129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1849be34d1a3SDavid Howells if (IS_ERR(q)) 1850be34d1a3SDavid Howells goto out; 1851719ea2fbSAl Viro lock_mount_hash(); 18521a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 18531064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1854719ea2fbSAl Viro unlock_mount_hash(); 18551da177e4SLinus Torvalds } 18561da177e4SLinus Torvalds } 18571da177e4SLinus Torvalds return res; 1858be34d1a3SDavid Howells out: 18591da177e4SLinus Torvalds if (res) { 1860719ea2fbSAl Viro lock_mount_hash(); 1861e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1862719ea2fbSAl Viro unlock_mount_hash(); 18631da177e4SLinus Torvalds } 1864be34d1a3SDavid Howells return q; 18651da177e4SLinus Torvalds } 18661da177e4SLinus Torvalds 1867be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1868be34d1a3SDavid Howells 1869ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 18708aec0809SAl Viro { 1871cb338d06SAl Viro struct mount *tree; 187297216be0SAl Viro namespace_lock(); 1873cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1874cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1875cd4a4017SEric W. Biederman else 187687129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 187787129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1878328e6d90SAl Viro namespace_unlock(); 1879be34d1a3SDavid Howells if (IS_ERR(tree)) 188052e220d3SDan Carpenter return ERR_CAST(tree); 1881be34d1a3SDavid Howells return &tree->mnt; 18828aec0809SAl Viro } 18838aec0809SAl Viro 1884a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1885a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1886a07b2000SAl Viro 1887a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1888a07b2000SAl Viro { 1889a07b2000SAl Viro struct mnt_namespace *ns; 1890a07b2000SAl Viro namespace_lock(); 1891a07b2000SAl Viro lock_mount_hash(); 1892a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 189344dfd84aSDavid Howells if (ns) { 189444dfd84aSDavid Howells if (is_anon_ns(ns)) 1895a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 189644dfd84aSDavid Howells else 189744dfd84aSDavid Howells ns = NULL; 189844dfd84aSDavid Howells } 1899a07b2000SAl Viro unlock_mount_hash(); 1900a07b2000SAl Viro namespace_unlock(); 190144dfd84aSDavid Howells if (ns) 1902a07b2000SAl Viro free_mnt_ns(ns); 1903a07b2000SAl Viro } 1904a07b2000SAl Viro 19058aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 19068aec0809SAl Viro { 190797216be0SAl Viro namespace_lock(); 1908719ea2fbSAl Viro lock_mount_hash(); 19099c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1910719ea2fbSAl Viro unlock_mount_hash(); 19113ab6abeeSAl Viro namespace_unlock(); 19128aec0809SAl Viro } 19138aec0809SAl Viro 1914c771d683SMiklos Szeredi /** 1915c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1916c771d683SMiklos Szeredi * 1917c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1918c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1919c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1920c771d683SMiklos Szeredi * 1921c771d683SMiklos Szeredi * Release with mntput(). 1922c771d683SMiklos Szeredi */ 1923ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1924c771d683SMiklos Szeredi { 1925c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1926c771d683SMiklos Szeredi struct mount *new_mnt; 1927c771d683SMiklos Szeredi 1928c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1929c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1930c771d683SMiklos Szeredi 1931c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1932c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1933c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1934c771d683SMiklos Szeredi 1935c771d683SMiklos Szeredi return &new_mnt->mnt; 1936c771d683SMiklos Szeredi } 1937c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1938c771d683SMiklos Szeredi 19391f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 19401f707137SAl Viro struct vfsmount *root) 19411f707137SAl Viro { 19421a4eeaf2SAl Viro struct mount *mnt; 19431f707137SAl Viro int res = f(root, arg); 19441f707137SAl Viro if (res) 19451f707137SAl Viro return res; 19461a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 19471a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 19481f707137SAl Viro if (res) 19491f707137SAl Viro return res; 19501f707137SAl Viro } 19511f707137SAl Viro return 0; 19521f707137SAl Viro } 19531f707137SAl Viro 19543bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 19553bd045ccSAl Viro { 19563bd045ccSAl Viro struct mount *p; 19573bd045ccSAl Viro 19583bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 19593bd045ccSAl Viro int flags = p->mnt.mnt_flags; 19603bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 19613bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 19623bd045ccSAl Viro 19633bd045ccSAl Viro if (flags & MNT_READONLY) 19643bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 19653bd045ccSAl Viro 19663bd045ccSAl Viro if (flags & MNT_NODEV) 19673bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 19683bd045ccSAl Viro 19693bd045ccSAl Viro if (flags & MNT_NOSUID) 19703bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 19713bd045ccSAl Viro 19723bd045ccSAl Viro if (flags & MNT_NOEXEC) 19733bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 19743bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 19753bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 19763bd045ccSAl Viro flags |= MNT_LOCKED; 19773bd045ccSAl Viro p->mnt.mnt_flags = flags; 19783bd045ccSAl Viro } 19793bd045ccSAl Viro } 19803bd045ccSAl Viro 19814b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1982719f5d7fSMiklos Szeredi { 1983315fc83eSAl Viro struct mount *p; 1984719f5d7fSMiklos Szeredi 1985909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1986fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 19874b8b21f4SAl Viro mnt_release_group_id(p); 1988719f5d7fSMiklos Szeredi } 1989719f5d7fSMiklos Szeredi } 1990719f5d7fSMiklos Szeredi 19914b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1992719f5d7fSMiklos Szeredi { 1993315fc83eSAl Viro struct mount *p; 1994719f5d7fSMiklos Szeredi 1995909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1996fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 19974b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1998719f5d7fSMiklos Szeredi if (err) { 19994b8b21f4SAl Viro cleanup_group_ids(mnt, p); 2000719f5d7fSMiklos Szeredi return err; 2001719f5d7fSMiklos Szeredi } 2002719f5d7fSMiklos Szeredi } 2003719f5d7fSMiklos Szeredi } 2004719f5d7fSMiklos Szeredi 2005719f5d7fSMiklos Szeredi return 0; 2006719f5d7fSMiklos Szeredi } 2007719f5d7fSMiklos Szeredi 2008d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 2009d2921684SEric W. Biederman { 2010d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 2011d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 2012d2921684SEric W. Biederman struct mount *p; 2013d2921684SEric W. Biederman 2014d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 2015d2921684SEric W. Biederman mounts++; 2016d2921684SEric W. Biederman 2017d2921684SEric W. Biederman old = ns->mounts; 2018d2921684SEric W. Biederman pending = ns->pending_mounts; 2019d2921684SEric W. Biederman sum = old + pending; 2020d2921684SEric W. Biederman if ((old > sum) || 2021d2921684SEric W. Biederman (pending > sum) || 2022d2921684SEric W. Biederman (max < sum) || 2023d2921684SEric W. Biederman (mounts > (max - sum))) 2024d2921684SEric W. Biederman return -ENOSPC; 2025d2921684SEric W. Biederman 2026d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 2027d2921684SEric W. Biederman return 0; 2028d2921684SEric W. Biederman } 2029d2921684SEric W. Biederman 2030b90fa9aeSRam Pai /* 2031b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 2032b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 203321444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 203421444403SRam Pai * store the parent mount and mountpoint dentry. 203521444403SRam Pai * (done when source_mnt is moved) 2036b90fa9aeSRam Pai * 2037b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 2038b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 20399676f0c6SRam Pai * --------------------------------------------------------------------------- 2040b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 20419676f0c6SRam Pai * |************************************************************************** 20429676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 20439676f0c6SRam Pai * | dest | | | | | 20449676f0c6SRam Pai * | | | | | | | 20459676f0c6SRam Pai * | v | | | | | 20469676f0c6SRam Pai * |************************************************************************** 20479676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 20485afe0022SRam Pai * | | | | | | 20499676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 20509676f0c6SRam Pai * *************************************************************************** 2051b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 2052b90fa9aeSRam Pai * destination mount. 2053b90fa9aeSRam Pai * 2054b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2055b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2056b90fa9aeSRam Pai * the peer group of the source mount. 2057b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2058b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2059b90fa9aeSRam Pai * mount. 20605afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 20615afe0022SRam Pai * of the destination mount and the cloned mount is made slave 20625afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 20635afe0022SRam Pai * is marked as 'shared and slave'. 20645afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 20655afe0022SRam Pai * source mount. 20665afe0022SRam Pai * 20679676f0c6SRam Pai * --------------------------------------------------------------------------- 206821444403SRam Pai * | MOVE MOUNT OPERATION | 20699676f0c6SRam Pai * |************************************************************************** 20709676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 20719676f0c6SRam Pai * | dest | | | | | 20729676f0c6SRam Pai * | | | | | | | 20739676f0c6SRam Pai * | v | | | | | 20749676f0c6SRam Pai * |************************************************************************** 20759676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 20765afe0022SRam Pai * | | | | | | 20779676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 20789676f0c6SRam Pai * *************************************************************************** 20795afe0022SRam Pai * 20805afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 20815afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 208221444403SRam Pai * (+*) the mount is moved to the destination. 20835afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 20845afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 20855afe0022SRam Pai * the mount is marked as 'shared and slave'. 20865afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2087b90fa9aeSRam Pai * 2088b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2089b90fa9aeSRam Pai * applied to each mount in the tree. 2090b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2091b90fa9aeSRam Pai * in allocations. 2092b90fa9aeSRam Pai */ 20930fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 209484d17192SAl Viro struct mount *dest_mnt, 209584d17192SAl Viro struct mountpoint *dest_mp, 20962763d119SAl Viro bool moving) 2097b90fa9aeSRam Pai { 20983bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 209938129a13SAl Viro HLIST_HEAD(tree_list); 2100d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 21011064f874SEric W. Biederman struct mountpoint *smp; 2102315fc83eSAl Viro struct mount *child, *p; 210338129a13SAl Viro struct hlist_node *n; 2104719f5d7fSMiklos Szeredi int err; 2105b90fa9aeSRam Pai 21061064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 21071064f874SEric W. Biederman * to be tucked under other mounts. 21081064f874SEric W. Biederman */ 21091064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 21101064f874SEric W. Biederman if (IS_ERR(smp)) 21111064f874SEric W. Biederman return PTR_ERR(smp); 21121064f874SEric W. Biederman 2113d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 21142763d119SAl Viro if (!moving) { 2115d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2116d2921684SEric W. Biederman if (err) 2117d2921684SEric W. Biederman goto out; 2118d2921684SEric W. Biederman } 2119d2921684SEric W. Biederman 2120fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 21210fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2122719f5d7fSMiklos Szeredi if (err) 2123719f5d7fSMiklos Szeredi goto out; 212484d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2125f2ebb3a9SAl Viro lock_mount_hash(); 2126719f5d7fSMiklos Szeredi if (err) 2127719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2128909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 21290f0afb1dSAl Viro set_mnt_shared(p); 21300b1b901bSAl Viro } else { 21310b1b901bSAl Viro lock_mount_hash(); 2132b90fa9aeSRam Pai } 21332763d119SAl Viro if (moving) { 21342763d119SAl Viro unhash_mnt(source_mnt); 213584d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2136143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 213721444403SRam Pai } else { 213844dfd84aSDavid Howells if (source_mnt->mnt_ns) { 213944dfd84aSDavid Howells /* move from anon - the caller will destroy */ 214044dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 214144dfd84aSDavid Howells } 214284d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 21431064f874SEric W. Biederman commit_tree(source_mnt); 214421444403SRam Pai } 2145b90fa9aeSRam Pai 214638129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 21471d6a32acSAl Viro struct mount *q; 214838129a13SAl Viro hlist_del_init(&child->mnt_hash); 21491064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 21501d6a32acSAl Viro child->mnt_mountpoint); 21511064f874SEric W. Biederman if (q) 21521064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 21533bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 21543bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 21553bd045ccSAl Viro lock_mnt_tree(child); 2156d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 21571064f874SEric W. Biederman commit_tree(child); 2158b90fa9aeSRam Pai } 21591064f874SEric W. Biederman put_mountpoint(smp); 2160719ea2fbSAl Viro unlock_mount_hash(); 216199b7db7bSNick Piggin 2162b90fa9aeSRam Pai return 0; 2163719f5d7fSMiklos Szeredi 2164719f5d7fSMiklos Szeredi out_cleanup_ids: 2165f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2166f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2167d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2168e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2169f2ebb3a9SAl Viro } 2170f2ebb3a9SAl Viro unlock_mount_hash(); 21710fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2172719f5d7fSMiklos Szeredi out: 2173d2921684SEric W. Biederman ns->pending_mounts = 0; 21741064f874SEric W. Biederman 21751064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 21761064f874SEric W. Biederman put_mountpoint(smp); 21771064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 21781064f874SEric W. Biederman 2179719f5d7fSMiklos Szeredi return err; 2180b90fa9aeSRam Pai } 2181b90fa9aeSRam Pai 218284d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2183b12cea91SAl Viro { 2184b12cea91SAl Viro struct vfsmount *mnt; 218584d17192SAl Viro struct dentry *dentry = path->dentry; 2186b12cea91SAl Viro retry: 21875955102cSAl Viro inode_lock(dentry->d_inode); 218884d17192SAl Viro if (unlikely(cant_mount(dentry))) { 21895955102cSAl Viro inode_unlock(dentry->d_inode); 219084d17192SAl Viro return ERR_PTR(-ENOENT); 2191b12cea91SAl Viro } 219297216be0SAl Viro namespace_lock(); 2193b12cea91SAl Viro mnt = lookup_mnt(path); 219484d17192SAl Viro if (likely(!mnt)) { 21953895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 219684d17192SAl Viro if (IS_ERR(mp)) { 219797216be0SAl Viro namespace_unlock(); 21985955102cSAl Viro inode_unlock(dentry->d_inode); 219984d17192SAl Viro return mp; 220084d17192SAl Viro } 220184d17192SAl Viro return mp; 220284d17192SAl Viro } 220397216be0SAl Viro namespace_unlock(); 22045955102cSAl Viro inode_unlock(path->dentry->d_inode); 2205b12cea91SAl Viro path_put(path); 2206b12cea91SAl Viro path->mnt = mnt; 220784d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2208b12cea91SAl Viro goto retry; 2209b12cea91SAl Viro } 2210b12cea91SAl Viro 221184d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2212b12cea91SAl Viro { 221384d17192SAl Viro struct dentry *dentry = where->m_dentry; 22143895dbf8SEric W. Biederman 22153895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 221684d17192SAl Viro put_mountpoint(where); 22173895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 22183895dbf8SEric W. Biederman 2219328e6d90SAl Viro namespace_unlock(); 22205955102cSAl Viro inode_unlock(dentry->d_inode); 2221b12cea91SAl Viro } 2222b12cea91SAl Viro 222384d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 22241da177e4SLinus Torvalds { 2225e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 22261da177e4SLinus Torvalds return -EINVAL; 22271da177e4SLinus Torvalds 2228e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2229e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 22301da177e4SLinus Torvalds return -ENOTDIR; 22311da177e4SLinus Torvalds 22322763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 22331da177e4SLinus Torvalds } 22341da177e4SLinus Torvalds 22351da177e4SLinus Torvalds /* 22367a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 22377a2e8a8fSValerie Aurora */ 22387a2e8a8fSValerie Aurora 2239e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 22407a2e8a8fSValerie Aurora { 2241e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 22427a2e8a8fSValerie Aurora 22437a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 22447a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 22457a2e8a8fSValerie Aurora return 0; 22467a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 22477a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 22487a2e8a8fSValerie Aurora return 0; 22497a2e8a8fSValerie Aurora return type; 22507a2e8a8fSValerie Aurora } 22517a2e8a8fSValerie Aurora 22527a2e8a8fSValerie Aurora /* 225307b20889SRam Pai * recursively change the type of the mountpoint. 225407b20889SRam Pai */ 2255e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 225607b20889SRam Pai { 2257315fc83eSAl Viro struct mount *m; 22584b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2259e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 22607a2e8a8fSValerie Aurora int type; 2261719f5d7fSMiklos Szeredi int err = 0; 226207b20889SRam Pai 22632d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 226407b20889SRam Pai return -EINVAL; 226507b20889SRam Pai 2266e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 22677a2e8a8fSValerie Aurora if (!type) 22687a2e8a8fSValerie Aurora return -EINVAL; 22697a2e8a8fSValerie Aurora 227097216be0SAl Viro namespace_lock(); 2271719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2272719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2273719f5d7fSMiklos Szeredi if (err) 2274719f5d7fSMiklos Szeredi goto out_unlock; 2275719f5d7fSMiklos Szeredi } 2276719f5d7fSMiklos Szeredi 2277719ea2fbSAl Viro lock_mount_hash(); 2278909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 22790f0afb1dSAl Viro change_mnt_propagation(m, type); 2280719ea2fbSAl Viro unlock_mount_hash(); 2281719f5d7fSMiklos Szeredi 2282719f5d7fSMiklos Szeredi out_unlock: 228397216be0SAl Viro namespace_unlock(); 2284719f5d7fSMiklos Szeredi return err; 228507b20889SRam Pai } 228607b20889SRam Pai 22875ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 22885ff9d8a6SEric W. Biederman { 22895ff9d8a6SEric W. Biederman struct mount *child; 22905ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 22915ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 22925ff9d8a6SEric W. Biederman continue; 22935ff9d8a6SEric W. Biederman 22945ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 22955ff9d8a6SEric W. Biederman return true; 22965ff9d8a6SEric W. Biederman } 22975ff9d8a6SEric W. Biederman return false; 22985ff9d8a6SEric W. Biederman } 22995ff9d8a6SEric W. Biederman 2300a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2301a07b2000SAl Viro { 2302a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2303a07b2000SAl Viro 2304a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2305a07b2000SAl Viro return mnt; 2306a07b2000SAl Viro 2307a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2308a07b2000SAl Viro return mnt; 2309a07b2000SAl Viro 2310a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2311a07b2000SAl Viro return mnt; 2312a07b2000SAl Viro 2313a07b2000SAl Viro if (recurse) 2314a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2315a07b2000SAl Viro else 2316a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2317a07b2000SAl Viro 2318a07b2000SAl Viro if (!IS_ERR(mnt)) 2319a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2320a07b2000SAl Viro 2321a07b2000SAl Viro return mnt; 2322a07b2000SAl Viro } 2323a07b2000SAl Viro 232407b20889SRam Pai /* 23251da177e4SLinus Torvalds * do loopback mount. 23261da177e4SLinus Torvalds */ 2327808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 23282dafe1c4SEric Sandeen int recurse) 23291da177e4SLinus Torvalds { 23302d92ab3cSAl Viro struct path old_path; 2331a07b2000SAl Viro struct mount *mnt = NULL, *parent; 233284d17192SAl Viro struct mountpoint *mp; 233357eccb83SAl Viro int err; 23341da177e4SLinus Torvalds if (!old_name || !*old_name) 23351da177e4SLinus Torvalds return -EINVAL; 2336815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 23371da177e4SLinus Torvalds if (err) 23381da177e4SLinus Torvalds return err; 23391da177e4SLinus Torvalds 23408823c079SEric W. Biederman err = -EINVAL; 23414ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 23428823c079SEric W. Biederman goto out; 23438823c079SEric W. Biederman 234484d17192SAl Viro mp = lock_mount(path); 2345a07b2000SAl Viro if (IS_ERR(mp)) { 234684d17192SAl Viro err = PTR_ERR(mp); 23479676f0c6SRam Pai goto out; 2348a07b2000SAl Viro } 23499676f0c6SRam Pai 235084d17192SAl Viro parent = real_mount(path->mnt); 2351e149ed2bSAl Viro if (!check_mnt(parent)) 2352e149ed2bSAl Viro goto out2; 2353e149ed2bSAl Viro 2354a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2355be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2356be34d1a3SDavid Howells err = PTR_ERR(mnt); 2357e9c5d8a5SAndrey Vagin goto out2; 2358be34d1a3SDavid Howells } 2359ccd48bc7SAl Viro 236084d17192SAl Viro err = graft_tree(mnt, parent, mp); 23611da177e4SLinus Torvalds if (err) { 2362719ea2fbSAl Viro lock_mount_hash(); 2363e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2364719ea2fbSAl Viro unlock_mount_hash(); 23655b83d2c5SRam Pai } 2366b12cea91SAl Viro out2: 236784d17192SAl Viro unlock_mount(mp); 2368ccd48bc7SAl Viro out: 23692d92ab3cSAl Viro path_put(&old_path); 23701da177e4SLinus Torvalds return err; 23711da177e4SLinus Torvalds } 23721da177e4SLinus Torvalds 2373a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2374a07b2000SAl Viro { 2375a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2376a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2377a07b2000SAl Viro struct mount *mnt, *p; 2378a07b2000SAl Viro struct file *file; 2379a07b2000SAl Viro 2380a07b2000SAl Viro if (IS_ERR(ns)) 2381a07b2000SAl Viro return ERR_CAST(ns); 2382a07b2000SAl Viro 2383a07b2000SAl Viro namespace_lock(); 2384a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2385a07b2000SAl Viro if (IS_ERR(mnt)) { 2386a07b2000SAl Viro namespace_unlock(); 2387a07b2000SAl Viro free_mnt_ns(ns); 2388a07b2000SAl Viro return ERR_CAST(mnt); 2389a07b2000SAl Viro } 2390a07b2000SAl Viro 2391a07b2000SAl Viro lock_mount_hash(); 2392a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2393a07b2000SAl Viro p->mnt_ns = ns; 2394a07b2000SAl Viro ns->mounts++; 2395a07b2000SAl Viro } 2396a07b2000SAl Viro ns->root = mnt; 2397a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2398a07b2000SAl Viro mntget(&mnt->mnt); 2399a07b2000SAl Viro unlock_mount_hash(); 2400a07b2000SAl Viro namespace_unlock(); 2401a07b2000SAl Viro 2402a07b2000SAl Viro mntput(path->mnt); 2403a07b2000SAl Viro path->mnt = &mnt->mnt; 2404a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2405a07b2000SAl Viro if (IS_ERR(file)) 2406a07b2000SAl Viro dissolve_on_fput(path->mnt); 2407a07b2000SAl Viro else 2408a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2409a07b2000SAl Viro return file; 2410a07b2000SAl Viro } 2411a07b2000SAl Viro 24122658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2413a07b2000SAl Viro { 2414a07b2000SAl Viro struct file *file; 2415a07b2000SAl Viro struct path path; 2416a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2417a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2418a07b2000SAl Viro int error; 2419a07b2000SAl Viro int fd; 2420a07b2000SAl Viro 2421a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2422a07b2000SAl Viro 2423a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2424a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2425a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2426a07b2000SAl Viro return -EINVAL; 2427a07b2000SAl Viro 2428a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2429a07b2000SAl Viro return -EINVAL; 2430a07b2000SAl Viro 2431a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2432a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2433a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2434a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2435a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2436a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2437a07b2000SAl Viro 2438a07b2000SAl Viro if (detached && !may_mount()) 2439a07b2000SAl Viro return -EPERM; 2440a07b2000SAl Viro 2441a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2442a07b2000SAl Viro if (fd < 0) 2443a07b2000SAl Viro return fd; 2444a07b2000SAl Viro 2445a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2446a07b2000SAl Viro if (unlikely(error)) { 2447a07b2000SAl Viro file = ERR_PTR(error); 2448a07b2000SAl Viro } else { 2449a07b2000SAl Viro if (detached) 2450a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2451a07b2000SAl Viro else 2452a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2453a07b2000SAl Viro path_put(&path); 2454a07b2000SAl Viro } 2455a07b2000SAl Viro if (IS_ERR(file)) { 2456a07b2000SAl Viro put_unused_fd(fd); 2457a07b2000SAl Viro return PTR_ERR(file); 2458a07b2000SAl Viro } 2459a07b2000SAl Viro fd_install(fd, file); 2460a07b2000SAl Viro return fd; 2461a07b2000SAl Viro } 2462a07b2000SAl Viro 246343f5e655SDavid Howells /* 246443f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 246543f5e655SDavid Howells * 246643f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 246743f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 246843f5e655SDavid Howells */ 246943f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 24702e4b7fcdSDave Hansen { 247143f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 24722e4b7fcdSDave Hansen 247343f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 247443f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 247543f5e655SDavid Howells return false; 247643f5e655SDavid Howells 247743f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 247843f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 247943f5e655SDavid Howells return false; 248043f5e655SDavid Howells 248143f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 248243f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 248343f5e655SDavid Howells return false; 248443f5e655SDavid Howells 248543f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 248643f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 248743f5e655SDavid Howells return false; 248843f5e655SDavid Howells 248943f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 249043f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 249143f5e655SDavid Howells return false; 249243f5e655SDavid Howells 249343f5e655SDavid Howells return true; 249443f5e655SDavid Howells } 249543f5e655SDavid Howells 249643f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 249743f5e655SDavid Howells { 249843f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 249943f5e655SDavid Howells 250043f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 25012e4b7fcdSDave Hansen return 0; 25022e4b7fcdSDave Hansen 25032e4b7fcdSDave Hansen if (readonly_request) 250443f5e655SDavid Howells return mnt_make_readonly(mnt); 250543f5e655SDavid Howells 250643f5e655SDavid Howells return __mnt_unmake_readonly(mnt); 250743f5e655SDavid Howells } 250843f5e655SDavid Howells 250943f5e655SDavid Howells /* 251043f5e655SDavid Howells * Update the user-settable attributes on a mount. The caller must hold 251143f5e655SDavid Howells * sb->s_umount for writing. 251243f5e655SDavid Howells */ 251343f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 251443f5e655SDavid Howells { 251543f5e655SDavid Howells lock_mount_hash(); 251643f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 251743f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 251843f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 251943f5e655SDavid Howells unlock_mount_hash(); 252043f5e655SDavid Howells } 252143f5e655SDavid Howells 2522f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2523f8b92ba6SDeepa Dinamani { 2524f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2525f8b92ba6SDeepa Dinamani 2526f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2527f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2528f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2529f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2530f8b92ba6SDeepa Dinamani struct tm tm; 2531f8b92ba6SDeepa Dinamani 2532f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2533f8b92ba6SDeepa Dinamani 25340ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 25350ecee669SEric Biggers sb->s_type->name, 25360ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 25370ecee669SEric Biggers mntpath, 2538f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2539f8b92ba6SDeepa Dinamani 2540f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2541f8b92ba6SDeepa Dinamani } 2542f8b92ba6SDeepa Dinamani } 2543f8b92ba6SDeepa Dinamani 254443f5e655SDavid Howells /* 254543f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 254643f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 254743f5e655SDavid Howells * to mount(2). 254843f5e655SDavid Howells */ 254943f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 255043f5e655SDavid Howells { 255143f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 255243f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 255343f5e655SDavid Howells int ret; 255443f5e655SDavid Howells 255543f5e655SDavid Howells if (!check_mnt(mnt)) 255643f5e655SDavid Howells return -EINVAL; 255743f5e655SDavid Howells 255843f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 255943f5e655SDavid Howells return -EINVAL; 256043f5e655SDavid Howells 256143f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 256243f5e655SDavid Howells return -EPERM; 256343f5e655SDavid Howells 256443f5e655SDavid Howells down_write(&sb->s_umount); 256543f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 256643f5e655SDavid Howells if (ret == 0) 256743f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 256843f5e655SDavid Howells up_write(&sb->s_umount); 2569f8b92ba6SDeepa Dinamani 2570f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2571f8b92ba6SDeepa Dinamani 257243f5e655SDavid Howells return ret; 25732e4b7fcdSDave Hansen } 25742e4b7fcdSDave Hansen 25751da177e4SLinus Torvalds /* 25761da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 25771da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 25781da177e4SLinus Torvalds * on it - tough luck. 25791da177e4SLinus Torvalds */ 2580e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2581e462ec50SDavid Howells int mnt_flags, void *data) 25821da177e4SLinus Torvalds { 25831da177e4SLinus Torvalds int err; 25842d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2585143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 25868d0347f6SDavid Howells struct fs_context *fc; 25871da177e4SLinus Torvalds 2588143c8c91SAl Viro if (!check_mnt(mnt)) 25891da177e4SLinus Torvalds return -EINVAL; 25901da177e4SLinus Torvalds 25912d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 25921da177e4SLinus Torvalds return -EINVAL; 25931da177e4SLinus Torvalds 259443f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 259507b64558SEric W. Biederman return -EPERM; 25969566d674SEric W. Biederman 25978d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 25988d0347f6SDavid Howells if (IS_ERR(fc)) 25998d0347f6SDavid Howells return PTR_ERR(fc); 2600ff36fe2cSEric Paris 26018d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 26028d0347f6SDavid Howells if (!err) { 26031da177e4SLinus Torvalds down_write(&sb->s_umount); 260457eccb83SAl Viro err = -EPERM; 260543f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 26068d0347f6SDavid Howells err = reconfigure_super(fc); 260743f5e655SDavid Howells if (!err) 260843f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 26090e55a7ccSDan Williams } 26106339dab8SAl Viro up_write(&sb->s_umount); 26118d0347f6SDavid Howells } 2612f8b92ba6SDeepa Dinamani 2613f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2614f8b92ba6SDeepa Dinamani 26158d0347f6SDavid Howells put_fs_context(fc); 26161da177e4SLinus Torvalds return err; 26171da177e4SLinus Torvalds } 26181da177e4SLinus Torvalds 2619cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 26209676f0c6SRam Pai { 2621315fc83eSAl Viro struct mount *p; 2622909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2623fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 26249676f0c6SRam Pai return 1; 26259676f0c6SRam Pai } 26269676f0c6SRam Pai return 0; 26279676f0c6SRam Pai } 26289676f0c6SRam Pai 262944dfd84aSDavid Howells /* 263044dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 263144dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 263244dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 263344dfd84aSDavid Howells * cycles to be made. 263444dfd84aSDavid Howells */ 263544dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 263644dfd84aSDavid Howells { 263744dfd84aSDavid Howells struct mount *p; 263844dfd84aSDavid Howells bool ret = false; 263944dfd84aSDavid Howells 264044dfd84aSDavid Howells lock_mount_hash(); 264144dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 264244dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 264344dfd84aSDavid Howells goto out; 264444dfd84aSDavid Howells 264544dfd84aSDavid Howells ret = true; 264644dfd84aSDavid Howells out: 264744dfd84aSDavid Howells unlock_mount_hash(); 264844dfd84aSDavid Howells return ret; 264944dfd84aSDavid Howells } 265044dfd84aSDavid Howells 26512db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 26521da177e4SLinus Torvalds { 265344dfd84aSDavid Howells struct mnt_namespace *ns; 2654676da58dSAl Viro struct mount *p; 26550fb54e50SAl Viro struct mount *old; 26562763d119SAl Viro struct mount *parent; 26572763d119SAl Viro struct mountpoint *mp, *old_mp; 265857eccb83SAl Viro int err; 265944dfd84aSDavid Howells bool attached; 26601da177e4SLinus Torvalds 26612db154b3SDavid Howells mp = lock_mount(new_path); 266284d17192SAl Viro if (IS_ERR(mp)) 26632db154b3SDavid Howells return PTR_ERR(mp); 2664cc53ce53SDavid Howells 26652db154b3SDavid Howells old = real_mount(old_path->mnt); 26662db154b3SDavid Howells p = real_mount(new_path->mnt); 26672763d119SAl Viro parent = old->mnt_parent; 266844dfd84aSDavid Howells attached = mnt_has_parent(old); 26692763d119SAl Viro old_mp = old->mnt_mp; 267044dfd84aSDavid Howells ns = old->mnt_ns; 2671143c8c91SAl Viro 26721da177e4SLinus Torvalds err = -EINVAL; 267344dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 267444dfd84aSDavid Howells if (!check_mnt(p)) 26752db154b3SDavid Howells goto out; 26761da177e4SLinus Torvalds 2677570d7a98SEric Biggers /* The thing moved must be mounted... */ 2678570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 267944dfd84aSDavid Howells goto out; 268044dfd84aSDavid Howells 2681570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2682570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 26832db154b3SDavid Howells goto out; 26841da177e4SLinus Torvalds 26852db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 26862db154b3SDavid Howells goto out; 26872db154b3SDavid Howells 26882db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 26892db154b3SDavid Howells goto out; 26902db154b3SDavid Howells 26912db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 26922db154b3SDavid Howells d_is_dir(old_path->dentry)) 26932db154b3SDavid Howells goto out; 269421444403SRam Pai /* 269521444403SRam Pai * Don't move a mount residing in a shared parent. 269621444403SRam Pai */ 26972763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 26982db154b3SDavid Howells goto out; 26999676f0c6SRam Pai /* 27009676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 27019676f0c6SRam Pai * mount which is shared. 27029676f0c6SRam Pai */ 2703fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 27042db154b3SDavid Howells goto out; 27051da177e4SLinus Torvalds err = -ELOOP; 270644dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 270744dfd84aSDavid Howells goto out; 2708fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2709676da58dSAl Viro if (p == old) 27102db154b3SDavid Howells goto out; 27111da177e4SLinus Torvalds 27122db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 27132763d119SAl Viro attached); 27144ac91378SJan Blunck if (err) 27152db154b3SDavid Howells goto out; 27161da177e4SLinus Torvalds 27171da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 27181da177e4SLinus Torvalds * automatically */ 27196776db3dSAl Viro list_del_init(&old->mnt_expire); 27202763d119SAl Viro if (attached) 27212763d119SAl Viro put_mountpoint(old_mp); 27221da177e4SLinus Torvalds out: 27232db154b3SDavid Howells unlock_mount(mp); 272444dfd84aSDavid Howells if (!err) { 27252763d119SAl Viro if (attached) 27262763d119SAl Viro mntput_no_expire(parent); 27272763d119SAl Viro else 272844dfd84aSDavid Howells free_mnt_ns(ns); 272944dfd84aSDavid Howells } 27302db154b3SDavid Howells return err; 27312db154b3SDavid Howells } 27322db154b3SDavid Howells 27332db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 27342db154b3SDavid Howells { 27352db154b3SDavid Howells struct path old_path; 27362db154b3SDavid Howells int err; 27372db154b3SDavid Howells 27382db154b3SDavid Howells if (!old_name || !*old_name) 27392db154b3SDavid Howells return -EINVAL; 27402db154b3SDavid Howells 27412db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 27422db154b3SDavid Howells if (err) 27432db154b3SDavid Howells return err; 27442db154b3SDavid Howells 27452db154b3SDavid Howells err = do_move_mount(&old_path, path); 27462d92ab3cSAl Viro path_put(&old_path); 27471da177e4SLinus Torvalds return err; 27481da177e4SLinus Torvalds } 27491da177e4SLinus Torvalds 27509d412a43SAl Viro /* 27519d412a43SAl Viro * add a mount into a namespace's mount tree 27529d412a43SAl Viro */ 27538f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 27548f11538eSAl Viro struct path *path, int mnt_flags) 27559d412a43SAl Viro { 27568f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 27579d412a43SAl Viro 2758f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 27599d412a43SAl Viro 276084d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2761156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2762156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 27638f11538eSAl Viro return -EINVAL; 2764156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 276584d17192SAl Viro if (!parent->mnt_ns) 27668f11538eSAl Viro return -EINVAL; 2767156cacb1SAl Viro } 27689d412a43SAl Viro 27699d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 277095bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 27719d412a43SAl Viro path->mnt->mnt_root == path->dentry) 27728f11538eSAl Viro return -EBUSY; 27739d412a43SAl Viro 2774e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 27758f11538eSAl Viro return -EINVAL; 27769d412a43SAl Viro 277795bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 27788f11538eSAl Viro return graft_tree(newmnt, parent, mp); 27799d412a43SAl Viro } 2780b1e75df4SAl Viro 2781132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2782132e4608SDavid Howells 2783132e4608SDavid Howells /* 2784132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2785132e4608SDavid Howells * be added to the namespace tree. 2786132e4608SDavid Howells */ 2787132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2788132e4608SDavid Howells unsigned int mnt_flags) 2789132e4608SDavid Howells { 2790132e4608SDavid Howells struct vfsmount *mnt; 27918f11538eSAl Viro struct mountpoint *mp; 2792132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2793132e4608SDavid Howells int error; 2794132e4608SDavid Howells 2795c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2796c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2797c9ce29edSAl Viro error = -EPERM; 2798c9ce29edSAl Viro 2799c9ce29edSAl Viro if (unlikely(error)) { 2800c9ce29edSAl Viro fc_drop_locked(fc); 2801c9ce29edSAl Viro return error; 2802132e4608SDavid Howells } 2803132e4608SDavid Howells 2804132e4608SDavid Howells up_write(&sb->s_umount); 2805132e4608SDavid Howells 2806132e4608SDavid Howells mnt = vfs_create_mount(fc); 2807132e4608SDavid Howells if (IS_ERR(mnt)) 2808132e4608SDavid Howells return PTR_ERR(mnt); 2809132e4608SDavid Howells 2810f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 2811f8b92ba6SDeepa Dinamani 28128f11538eSAl Viro mp = lock_mount(mountpoint); 28138f11538eSAl Viro if (IS_ERR(mp)) { 28148f11538eSAl Viro mntput(mnt); 28158f11538eSAl Viro return PTR_ERR(mp); 28168f11538eSAl Viro } 28178f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 28188f11538eSAl Viro unlock_mount(mp); 28190ecee669SEric Biggers if (error < 0) 28200ecee669SEric Biggers mntput(mnt); 2821f8b92ba6SDeepa Dinamani return error; 2822f8b92ba6SDeepa Dinamani } 2823f8b92ba6SDeepa Dinamani 28241da177e4SLinus Torvalds /* 28251da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 28261da177e4SLinus Torvalds * namespace's tree 28271da177e4SLinus Torvalds */ 2828e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2829808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 28301da177e4SLinus Torvalds { 28310c55cfc4SEric W. Biederman struct file_system_type *type; 2832a0c9a8b8SAl Viro struct fs_context *fc; 2833a0c9a8b8SAl Viro const char *subtype = NULL; 2834a0c9a8b8SAl Viro int err = 0; 28351da177e4SLinus Torvalds 28360c55cfc4SEric W. Biederman if (!fstype) 28371da177e4SLinus Torvalds return -EINVAL; 28381da177e4SLinus Torvalds 28390c55cfc4SEric W. Biederman type = get_fs_type(fstype); 28400c55cfc4SEric W. Biederman if (!type) 28410c55cfc4SEric W. Biederman return -ENODEV; 28420c55cfc4SEric W. Biederman 2843a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 2844a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 2845a0c9a8b8SAl Viro if (subtype) { 2846a0c9a8b8SAl Viro subtype++; 2847a0c9a8b8SAl Viro if (!*subtype) { 28480c55cfc4SEric W. Biederman put_filesystem(type); 2849a0c9a8b8SAl Viro return -EINVAL; 2850a0c9a8b8SAl Viro } 2851a0c9a8b8SAl Viro } 28528654df4eSEric W. Biederman } 28538654df4eSEric W. Biederman 2854a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 2855a0c9a8b8SAl Viro put_filesystem(type); 2856a0c9a8b8SAl Viro if (IS_ERR(fc)) 2857a0c9a8b8SAl Viro return PTR_ERR(fc); 2858a0c9a8b8SAl Viro 28593e1aeb00SDavid Howells if (subtype) 28603e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 28613e1aeb00SDavid Howells subtype, strlen(subtype)); 28623e1aeb00SDavid Howells if (!err && name) 28633e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 2864a0c9a8b8SAl Viro if (!err) 2865a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 2866c3aabf07SAl Viro if (!err && !mount_capable(fc)) 2867c3aabf07SAl Viro err = -EPERM; 2868a0c9a8b8SAl Viro if (!err) 2869a0c9a8b8SAl Viro err = vfs_get_tree(fc); 2870132e4608SDavid Howells if (!err) 2871132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 2872a0c9a8b8SAl Viro 2873a0c9a8b8SAl Viro put_fs_context(fc); 287415f9a3f3SAl Viro return err; 28751da177e4SLinus Torvalds } 28761da177e4SLinus Torvalds 287719a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 287819a167afSAl Viro { 287926df6034SAl Viro struct dentry *dentry = path->dentry; 28808f11538eSAl Viro struct mountpoint *mp; 288125e195aaSAl Viro struct mount *mnt; 288219a167afSAl Viro int err; 288325e195aaSAl Viro 288425e195aaSAl Viro if (!m) 288525e195aaSAl Viro return 0; 288625e195aaSAl Viro if (IS_ERR(m)) 288725e195aaSAl Viro return PTR_ERR(m); 288825e195aaSAl Viro 288925e195aaSAl Viro mnt = real_mount(m); 289019a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 289119a167afSAl Viro * expired before we get a chance to add it 289219a167afSAl Viro */ 28936776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 289419a167afSAl Viro 289519a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 289626df6034SAl Viro m->mnt_root == dentry) { 2897b1e75df4SAl Viro err = -ELOOP; 289826df6034SAl Viro goto discard; 289919a167afSAl Viro } 290019a167afSAl Viro 290126df6034SAl Viro /* 290226df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 290326df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 290426df6034SAl Viro * got", not "try to mount it on top". 290526df6034SAl Viro */ 290626df6034SAl Viro inode_lock(dentry->d_inode); 290726df6034SAl Viro namespace_lock(); 290826df6034SAl Viro if (unlikely(cant_mount(dentry))) { 290926df6034SAl Viro err = -ENOENT; 291026df6034SAl Viro goto discard_locked; 291126df6034SAl Viro } 291226df6034SAl Viro rcu_read_lock(); 291326df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 291426df6034SAl Viro rcu_read_unlock(); 291526df6034SAl Viro err = 0; 291626df6034SAl Viro goto discard_locked; 291726df6034SAl Viro } 291826df6034SAl Viro rcu_read_unlock(); 291926df6034SAl Viro mp = get_mountpoint(dentry); 29208f11538eSAl Viro if (IS_ERR(mp)) { 29218f11538eSAl Viro err = PTR_ERR(mp); 292226df6034SAl Viro goto discard_locked; 29238f11538eSAl Viro } 292426df6034SAl Viro 29258f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 29268f11538eSAl Viro unlock_mount(mp); 292726df6034SAl Viro if (unlikely(err)) 292826df6034SAl Viro goto discard; 292926df6034SAl Viro mntput(m); 2930b1e75df4SAl Viro return 0; 293126df6034SAl Viro 293226df6034SAl Viro discard_locked: 293326df6034SAl Viro namespace_unlock(); 293426df6034SAl Viro inode_unlock(dentry->d_inode); 293526df6034SAl Viro discard: 2936b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 29376776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 293897216be0SAl Viro namespace_lock(); 29396776db3dSAl Viro list_del_init(&mnt->mnt_expire); 294097216be0SAl Viro namespace_unlock(); 294119a167afSAl Viro } 2942b1e75df4SAl Viro mntput(m); 2943b1e75df4SAl Viro mntput(m); 294419a167afSAl Viro return err; 294519a167afSAl Viro } 294619a167afSAl Viro 2947ea5b778aSDavid Howells /** 2948ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2949ea5b778aSDavid Howells * @mnt: The mount to list. 2950ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2951ea5b778aSDavid Howells */ 2952ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2953ea5b778aSDavid Howells { 295497216be0SAl Viro namespace_lock(); 2955ea5b778aSDavid Howells 29566776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2957ea5b778aSDavid Howells 295897216be0SAl Viro namespace_unlock(); 2959ea5b778aSDavid Howells } 2960ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2961ea5b778aSDavid Howells 2962ea5b778aSDavid Howells /* 29631da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 29641da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 29651da177e4SLinus Torvalds * here 29661da177e4SLinus Torvalds */ 29671da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 29681da177e4SLinus Torvalds { 2969761d5c38SAl Viro struct mount *mnt, *next; 29701da177e4SLinus Torvalds LIST_HEAD(graveyard); 29711da177e4SLinus Torvalds 29721da177e4SLinus Torvalds if (list_empty(mounts)) 29731da177e4SLinus Torvalds return; 29741da177e4SLinus Torvalds 297597216be0SAl Viro namespace_lock(); 2976719ea2fbSAl Viro lock_mount_hash(); 29771da177e4SLinus Torvalds 29781da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 29791da177e4SLinus Torvalds * following criteria: 29801da177e4SLinus Torvalds * - only referenced by its parent vfsmount 29811da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 29821da177e4SLinus Torvalds * cleared by mntput()) 29831da177e4SLinus Torvalds */ 29846776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2985863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 29861ab59738SAl Viro propagate_mount_busy(mnt, 1)) 29871da177e4SLinus Torvalds continue; 29886776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 29891da177e4SLinus Torvalds } 2990bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 29916776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2992143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2993e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2994bcc5c7d2SAl Viro } 2995719ea2fbSAl Viro unlock_mount_hash(); 29963ab6abeeSAl Viro namespace_unlock(); 29971da177e4SLinus Torvalds } 29981da177e4SLinus Torvalds 29991da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 30001da177e4SLinus Torvalds 30011da177e4SLinus Torvalds /* 30025528f911STrond Myklebust * Ripoff of 'select_parent()' 30035528f911STrond Myklebust * 30045528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 30055528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 30065528f911STrond Myklebust */ 3007692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 30085528f911STrond Myklebust { 3009692afc31SAl Viro struct mount *this_parent = parent; 30105528f911STrond Myklebust struct list_head *next; 30115528f911STrond Myklebust int found = 0; 30125528f911STrond Myklebust 30135528f911STrond Myklebust repeat: 30146b41d536SAl Viro next = this_parent->mnt_mounts.next; 30155528f911STrond Myklebust resume: 30166b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 30175528f911STrond Myklebust struct list_head *tmp = next; 30186b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 30195528f911STrond Myklebust 30205528f911STrond Myklebust next = tmp->next; 3021692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 30225528f911STrond Myklebust continue; 30235528f911STrond Myklebust /* 30245528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 30255528f911STrond Myklebust */ 30266b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 30275528f911STrond Myklebust this_parent = mnt; 30285528f911STrond Myklebust goto repeat; 30295528f911STrond Myklebust } 30305528f911STrond Myklebust 30311ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 30326776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 30335528f911STrond Myklebust found++; 30345528f911STrond Myklebust } 30355528f911STrond Myklebust } 30365528f911STrond Myklebust /* 30375528f911STrond Myklebust * All done at this level ... ascend and resume the search 30385528f911STrond Myklebust */ 30395528f911STrond Myklebust if (this_parent != parent) { 30406b41d536SAl Viro next = this_parent->mnt_child.next; 30410714a533SAl Viro this_parent = this_parent->mnt_parent; 30425528f911STrond Myklebust goto resume; 30435528f911STrond Myklebust } 30445528f911STrond Myklebust return found; 30455528f911STrond Myklebust } 30465528f911STrond Myklebust 30475528f911STrond Myklebust /* 30485528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 30495528f911STrond Myklebust * submounts of a specific parent mountpoint 305099b7db7bSNick Piggin * 305148a066e7SAl Viro * mount_lock must be held for write 30525528f911STrond Myklebust */ 3053b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 30545528f911STrond Myklebust { 30555528f911STrond Myklebust LIST_HEAD(graveyard); 3056761d5c38SAl Viro struct mount *m; 30575528f911STrond Myklebust 30585528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 3059c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3060bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3061761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 30626776db3dSAl Viro mnt_expire); 3063143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3064e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3065bcc5c7d2SAl Viro } 3066bcc5c7d2SAl Viro } 30675528f911STrond Myklebust } 30685528f911STrond Myklebust 3069b40ef869SAl Viro void *copy_mount_options(const void __user * data) 30701da177e4SLinus Torvalds { 3071b40ef869SAl Viro char *copy; 307212efec56SAl Viro unsigned size; 30731da177e4SLinus Torvalds 30741da177e4SLinus Torvalds if (!data) 3075b40ef869SAl Viro return NULL; 30761da177e4SLinus Torvalds 3077b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3078b40ef869SAl Viro if (!copy) 3079b40ef869SAl Viro return ERR_PTR(-ENOMEM); 30801da177e4SLinus Torvalds 308112efec56SAl Viro size = PAGE_SIZE - offset_in_page(data); 30821da177e4SLinus Torvalds 308312efec56SAl Viro if (copy_from_user(copy, data, size)) { 3084b40ef869SAl Viro kfree(copy); 3085b40ef869SAl Viro return ERR_PTR(-EFAULT); 30861da177e4SLinus Torvalds } 308712efec56SAl Viro if (size != PAGE_SIZE) { 308812efec56SAl Viro if (copy_from_user(copy + size, data + size, PAGE_SIZE - size)) 308912efec56SAl Viro memset(copy + size, 0, PAGE_SIZE - size); 309012efec56SAl Viro } 3091b40ef869SAl Viro return copy; 30921da177e4SLinus Torvalds } 30931da177e4SLinus Torvalds 3094b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 3095eca6f534SVegard Nossum { 3096fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3097eca6f534SVegard Nossum } 3098eca6f534SVegard Nossum 30991da177e4SLinus Torvalds /* 31001da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 31011da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 31021da177e4SLinus Torvalds * 31031da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 31041da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 31051da177e4SLinus Torvalds * information (or be NULL). 31061da177e4SLinus Torvalds * 31071da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 31081da177e4SLinus Torvalds * When the flags word was introduced its top half was required 31091da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 31101da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 31111da177e4SLinus Torvalds * and must be discarded. 31121da177e4SLinus Torvalds */ 31135e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 3114808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 31151da177e4SLinus Torvalds { 31162d92ab3cSAl Viro struct path path; 3117e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 31181da177e4SLinus Torvalds int retval = 0; 31191da177e4SLinus Torvalds 31201da177e4SLinus Torvalds /* Discard magic */ 31211da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 31221da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 31231da177e4SLinus Torvalds 31241da177e4SLinus Torvalds /* Basic sanity checks */ 31251da177e4SLinus Torvalds if (data_page) 31261da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 31271da177e4SLinus Torvalds 3128e462ec50SDavid Howells if (flags & MS_NOUSER) 3129e462ec50SDavid Howells return -EINVAL; 3130e462ec50SDavid Howells 3131a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 3132ce6595a2SAl Viro retval = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3133a27ab9f2STetsuo Handa if (retval) 3134a27ab9f2STetsuo Handa return retval; 3135a27ab9f2STetsuo Handa 3136a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 3137a27ab9f2STetsuo Handa type_page, flags, data_page); 31380d5cadb8SAl Viro if (!retval && !may_mount()) 31390d5cadb8SAl Viro retval = -EPERM; 3140e462ec50SDavid Howells if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) 31419e8925b6SJeff Layton retval = -EPERM; 3142a27ab9f2STetsuo Handa if (retval) 3143a27ab9f2STetsuo Handa goto dput_out; 3144a27ab9f2STetsuo Handa 3145613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3146613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 31470a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 31480a1c01c9SMatthew Garrett 31491da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 31501da177e4SLinus Torvalds if (flags & MS_NOSUID) 31511da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 31521da177e4SLinus Torvalds if (flags & MS_NODEV) 31531da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 31541da177e4SLinus Torvalds if (flags & MS_NOEXEC) 31551da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3156fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3157fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3158fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3159fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3160d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3161d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3162a9e5b732SDavid Howells if (flags & MS_RDONLY) 31632e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3164fc33a7bbSChristoph Hellwig 3165ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3166ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3167ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3168ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3169ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3170ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 3171ffbc6f0eSEric W. Biederman } 3172ffbc6f0eSEric W. Biederman 3173e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3174e462ec50SDavid Howells SB_SYNCHRONOUS | 3175e462ec50SDavid Howells SB_MANDLOCK | 3176e462ec50SDavid Howells SB_DIRSYNC | 3177e462ec50SDavid Howells SB_SILENT | 3178917086ffSMimi Zohar SB_POSIXACL | 3179d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3180917086ffSMimi Zohar SB_I_VERSION); 31811da177e4SLinus Torvalds 318243f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 318343f5e655SDavid Howells retval = do_reconfigure_mnt(&path, mnt_flags); 318443f5e655SDavid Howells else if (flags & MS_REMOUNT) 3185e462ec50SDavid Howells retval = do_remount(&path, flags, sb_flags, mnt_flags, 31861da177e4SLinus Torvalds data_page); 31871da177e4SLinus Torvalds else if (flags & MS_BIND) 31882d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 31899676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 31902d92ab3cSAl Viro retval = do_change_type(&path, flags); 31911da177e4SLinus Torvalds else if (flags & MS_MOVE) 31922db154b3SDavid Howells retval = do_move_mount_old(&path, dev_name); 31931da177e4SLinus Torvalds else 3194e462ec50SDavid Howells retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, 31951da177e4SLinus Torvalds dev_name, data_page); 31961da177e4SLinus Torvalds dput_out: 31972d92ab3cSAl Viro path_put(&path); 31981da177e4SLinus Torvalds return retval; 31991da177e4SLinus Torvalds } 32001da177e4SLinus Torvalds 3201537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3202537f7ccbSEric W. Biederman { 3203537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3204537f7ccbSEric W. Biederman } 3205537f7ccbSEric W. Biederman 3206537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3207537f7ccbSEric W. Biederman { 3208537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3209537f7ccbSEric W. Biederman } 3210537f7ccbSEric W. Biederman 3211771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3212771b1371SEric W. Biederman { 321374e83122SAl Viro if (!is_anon_ns(ns)) 32146344c433SAl Viro ns_free_inum(&ns->ns); 3215537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3216771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3217771b1371SEric W. Biederman kfree(ns); 3218771b1371SEric W. Biederman } 3219771b1371SEric W. Biederman 32208823c079SEric W. Biederman /* 32218823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 32228823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 32238823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 32248823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 32258823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 32268823c079SEric W. Biederman */ 32278823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 32288823c079SEric W. Biederman 322974e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3230cf8d2c11STrond Myklebust { 3231cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3232537f7ccbSEric W. Biederman struct ucounts *ucounts; 323398f842e6SEric W. Biederman int ret; 3234cf8d2c11STrond Myklebust 3235537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3236537f7ccbSEric W. Biederman if (!ucounts) 3237df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3238537f7ccbSEric W. Biederman 323974e83122SAl Viro new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 3240537f7ccbSEric W. Biederman if (!new_ns) { 3241537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3242cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3243537f7ccbSEric W. Biederman } 324474e83122SAl Viro if (!anon) { 32456344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 324698f842e6SEric W. Biederman if (ret) { 324798f842e6SEric W. Biederman kfree(new_ns); 3248537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 324998f842e6SEric W. Biederman return ERR_PTR(ret); 325098f842e6SEric W. Biederman } 325174e83122SAl Viro } 325233c42940SAl Viro new_ns->ns.ops = &mntns_operations; 325374e83122SAl Viro if (!anon) 32548823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 3255cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 3256cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3257cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 32589f6c61f9SMiklos Szeredi spin_lock_init(&new_ns->ns_lock); 3259771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3260537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3261cf8d2c11STrond Myklebust return new_ns; 3262cf8d2c11STrond Myklebust } 3263cf8d2c11STrond Myklebust 32640766f788SEmese Revfy __latent_entropy 32659559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 32669559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 32671da177e4SLinus Torvalds { 32686b3286edSKirill Korotaev struct mnt_namespace *new_ns; 32697f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3270315fc83eSAl Viro struct mount *p, *q; 32719559f689SAl Viro struct mount *old; 3272cb338d06SAl Viro struct mount *new; 32737a472ef4SEric W. Biederman int copy_flags; 32741da177e4SLinus Torvalds 32759559f689SAl Viro BUG_ON(!ns); 32769559f689SAl Viro 32779559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 32789559f689SAl Viro get_mnt_ns(ns); 32799559f689SAl Viro return ns; 32809559f689SAl Viro } 32819559f689SAl Viro 32829559f689SAl Viro old = ns->root; 32839559f689SAl Viro 328474e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3285cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3286cf8d2c11STrond Myklebust return new_ns; 32871da177e4SLinus Torvalds 328897216be0SAl Viro namespace_lock(); 32891da177e4SLinus Torvalds /* First pass: copy the tree topology */ 32904ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 32919559f689SAl Viro if (user_ns != ns->user_ns) 32923bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 32937a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3294be34d1a3SDavid Howells if (IS_ERR(new)) { 3295328e6d90SAl Viro namespace_unlock(); 3296771b1371SEric W. Biederman free_mnt_ns(new_ns); 3297be34d1a3SDavid Howells return ERR_CAST(new); 32981da177e4SLinus Torvalds } 32993bd045ccSAl Viro if (user_ns != ns->user_ns) { 33003bd045ccSAl Viro lock_mount_hash(); 33013bd045ccSAl Viro lock_mnt_tree(new); 33023bd045ccSAl Viro unlock_mount_hash(); 33033bd045ccSAl Viro } 3304be08d6d2SAl Viro new_ns->root = new; 33051a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 33061da177e4SLinus Torvalds 33071da177e4SLinus Torvalds /* 33081da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 33091da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 33101da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 33111da177e4SLinus Torvalds */ 3312909b0a88SAl Viro p = old; 3313cb338d06SAl Viro q = new; 33141da177e4SLinus Torvalds while (p) { 3315143c8c91SAl Viro q->mnt_ns = new_ns; 3316d2921684SEric W. Biederman new_ns->mounts++; 33179559f689SAl Viro if (new_fs) { 33189559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 33199559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3320315fc83eSAl Viro rootmnt = &p->mnt; 33211da177e4SLinus Torvalds } 33229559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 33239559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3324315fc83eSAl Viro pwdmnt = &p->mnt; 33251da177e4SLinus Torvalds } 33261da177e4SLinus Torvalds } 3327909b0a88SAl Viro p = next_mnt(p, old); 3328909b0a88SAl Viro q = next_mnt(q, new); 33294ce5d2b1SEric W. Biederman if (!q) 33304ce5d2b1SEric W. Biederman break; 33314ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 33324ce5d2b1SEric W. Biederman p = next_mnt(p, old); 33331da177e4SLinus Torvalds } 3334328e6d90SAl Viro namespace_unlock(); 33351da177e4SLinus Torvalds 33361da177e4SLinus Torvalds if (rootmnt) 3337f03c6599SAl Viro mntput(rootmnt); 33381da177e4SLinus Torvalds if (pwdmnt) 3339f03c6599SAl Viro mntput(pwdmnt); 33401da177e4SLinus Torvalds 3341741a2951SJANAK DESAI return new_ns; 3342741a2951SJANAK DESAI } 3343741a2951SJANAK DESAI 334474e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3345cf8d2c11STrond Myklebust { 33461a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3347ea441d11SAl Viro struct mnt_namespace *ns; 3348d31da0f0SAl Viro struct super_block *s; 3349ea441d11SAl Viro struct path path; 3350ea441d11SAl Viro int err; 3351ea441d11SAl Viro 335274e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 335374e83122SAl Viro if (IS_ERR(ns)) { 335474e83122SAl Viro mntput(m); 3355ea441d11SAl Viro return ERR_CAST(ns); 335674e83122SAl Viro } 335774e83122SAl Viro mnt->mnt_ns = ns; 335874e83122SAl Viro ns->root = mnt; 335974e83122SAl Viro ns->mounts++; 336074e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3361ea441d11SAl Viro 336274e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3363ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3364ea441d11SAl Viro 3365ea441d11SAl Viro put_mnt_ns(ns); 3366ea441d11SAl Viro 3367ea441d11SAl Viro if (err) 3368ea441d11SAl Viro return ERR_PTR(err); 3369ea441d11SAl Viro 3370ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3371d31da0f0SAl Viro s = path.mnt->mnt_sb; 3372d31da0f0SAl Viro atomic_inc(&s->s_active); 3373ea441d11SAl Viro mntput(path.mnt); 3374ea441d11SAl Viro /* lock the sucker */ 3375d31da0f0SAl Viro down_write(&s->s_umount); 3376ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3377ea441d11SAl Viro return path.dentry; 3378ea441d11SAl Viro } 3379ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3380ea441d11SAl Viro 3381cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3382cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 33831da177e4SLinus Torvalds { 3384eca6f534SVegard Nossum int ret; 3385eca6f534SVegard Nossum char *kernel_type; 3386eca6f534SVegard Nossum char *kernel_dev; 3387b40ef869SAl Viro void *options; 33881da177e4SLinus Torvalds 3389b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3390b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3391b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3392eca6f534SVegard Nossum goto out_type; 33931da177e4SLinus Torvalds 3394b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3395b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3396b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3397eca6f534SVegard Nossum goto out_dev; 33981da177e4SLinus Torvalds 3399b40ef869SAl Viro options = copy_mount_options(data); 3400b40ef869SAl Viro ret = PTR_ERR(options); 3401b40ef869SAl Viro if (IS_ERR(options)) 3402eca6f534SVegard Nossum goto out_data; 34031da177e4SLinus Torvalds 3404b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3405eca6f534SVegard Nossum 3406b40ef869SAl Viro kfree(options); 3407eca6f534SVegard Nossum out_data: 3408eca6f534SVegard Nossum kfree(kernel_dev); 3409eca6f534SVegard Nossum out_dev: 3410eca6f534SVegard Nossum kfree(kernel_type); 3411eca6f534SVegard Nossum out_type: 3412eca6f534SVegard Nossum return ret; 34131da177e4SLinus Torvalds } 34141da177e4SLinus Torvalds 34151da177e4SLinus Torvalds /* 341693766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 341793766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 341893766fbdSDavid Howells */ 341993766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 342093766fbdSDavid Howells unsigned int, attr_flags) 342193766fbdSDavid Howells { 342293766fbdSDavid Howells struct mnt_namespace *ns; 342393766fbdSDavid Howells struct fs_context *fc; 342493766fbdSDavid Howells struct file *file; 342593766fbdSDavid Howells struct path newmount; 342693766fbdSDavid Howells struct mount *mnt; 342793766fbdSDavid Howells struct fd f; 342893766fbdSDavid Howells unsigned int mnt_flags = 0; 342993766fbdSDavid Howells long ret; 343093766fbdSDavid Howells 343193766fbdSDavid Howells if (!may_mount()) 343293766fbdSDavid Howells return -EPERM; 343393766fbdSDavid Howells 343493766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 343593766fbdSDavid Howells return -EINVAL; 343693766fbdSDavid Howells 343793766fbdSDavid Howells if (attr_flags & ~(MOUNT_ATTR_RDONLY | 343893766fbdSDavid Howells MOUNT_ATTR_NOSUID | 343993766fbdSDavid Howells MOUNT_ATTR_NODEV | 344093766fbdSDavid Howells MOUNT_ATTR_NOEXEC | 344193766fbdSDavid Howells MOUNT_ATTR__ATIME | 344293766fbdSDavid Howells MOUNT_ATTR_NODIRATIME)) 344393766fbdSDavid Howells return -EINVAL; 344493766fbdSDavid Howells 344593766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_RDONLY) 344693766fbdSDavid Howells mnt_flags |= MNT_READONLY; 344793766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NOSUID) 344893766fbdSDavid Howells mnt_flags |= MNT_NOSUID; 344993766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NODEV) 345093766fbdSDavid Howells mnt_flags |= MNT_NODEV; 345193766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NOEXEC) 345293766fbdSDavid Howells mnt_flags |= MNT_NOEXEC; 345393766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NODIRATIME) 345493766fbdSDavid Howells mnt_flags |= MNT_NODIRATIME; 345593766fbdSDavid Howells 345693766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 345793766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 345893766fbdSDavid Howells break; 345993766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 346093766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 346193766fbdSDavid Howells break; 346293766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 346393766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 346493766fbdSDavid Howells break; 346593766fbdSDavid Howells default: 346693766fbdSDavid Howells return -EINVAL; 346793766fbdSDavid Howells } 346893766fbdSDavid Howells 346993766fbdSDavid Howells f = fdget(fs_fd); 347093766fbdSDavid Howells if (!f.file) 347193766fbdSDavid Howells return -EBADF; 347293766fbdSDavid Howells 347393766fbdSDavid Howells ret = -EINVAL; 347493766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 347593766fbdSDavid Howells goto err_fsfd; 347693766fbdSDavid Howells 347793766fbdSDavid Howells fc = f.file->private_data; 347893766fbdSDavid Howells 347993766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 348093766fbdSDavid Howells if (ret < 0) 348193766fbdSDavid Howells goto err_fsfd; 348293766fbdSDavid Howells 348393766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 348493766fbdSDavid Howells ret = -EINVAL; 348593766fbdSDavid Howells if (!fc->root) 348693766fbdSDavid Howells goto err_unlock; 348793766fbdSDavid Howells 348893766fbdSDavid Howells ret = -EPERM; 348993766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 349093766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 349193766fbdSDavid Howells goto err_unlock; 349293766fbdSDavid Howells } 349393766fbdSDavid Howells 349493766fbdSDavid Howells ret = -EBUSY; 349593766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 349693766fbdSDavid Howells goto err_unlock; 349793766fbdSDavid Howells 349893766fbdSDavid Howells ret = -EPERM; 349993766fbdSDavid Howells if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock()) 350093766fbdSDavid Howells goto err_unlock; 350193766fbdSDavid Howells 350293766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 350393766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 350493766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 350593766fbdSDavid Howells goto err_unlock; 350693766fbdSDavid Howells } 350793766fbdSDavid Howells newmount.dentry = dget(fc->root); 350893766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 350993766fbdSDavid Howells 351093766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 351193766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 351293766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 351393766fbdSDavid Howells * don't want to have to handle any errors incurred. 351493766fbdSDavid Howells */ 351593766fbdSDavid Howells vfs_clean_context(fc); 351693766fbdSDavid Howells 351793766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 351893766fbdSDavid Howells if (IS_ERR(ns)) { 351993766fbdSDavid Howells ret = PTR_ERR(ns); 352093766fbdSDavid Howells goto err_path; 352193766fbdSDavid Howells } 352293766fbdSDavid Howells mnt = real_mount(newmount.mnt); 352393766fbdSDavid Howells mnt->mnt_ns = ns; 352493766fbdSDavid Howells ns->root = mnt; 352593766fbdSDavid Howells ns->mounts = 1; 352693766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 35271b0b9cc8SEric Biggers mntget(newmount.mnt); 352893766fbdSDavid Howells 352993766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 353093766fbdSDavid Howells * it, not just simply put it. 353193766fbdSDavid Howells */ 353293766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 353393766fbdSDavid Howells if (IS_ERR(file)) { 353493766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 353593766fbdSDavid Howells ret = PTR_ERR(file); 353693766fbdSDavid Howells goto err_path; 353793766fbdSDavid Howells } 353893766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 353993766fbdSDavid Howells 354093766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 354193766fbdSDavid Howells if (ret >= 0) 354293766fbdSDavid Howells fd_install(ret, file); 354393766fbdSDavid Howells else 354493766fbdSDavid Howells fput(file); 354593766fbdSDavid Howells 354693766fbdSDavid Howells err_path: 354793766fbdSDavid Howells path_put(&newmount); 354893766fbdSDavid Howells err_unlock: 354993766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 355093766fbdSDavid Howells err_fsfd: 355193766fbdSDavid Howells fdput(f); 355293766fbdSDavid Howells return ret; 355393766fbdSDavid Howells } 355493766fbdSDavid Howells 355593766fbdSDavid Howells /* 355693766fbdSDavid Howells * Move a mount from one place to another. In combination with 355793766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 355893766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 355993766fbdSDavid Howells * a mount subtree. 35602db154b3SDavid Howells * 35612db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 35622db154b3SDavid Howells */ 35632db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 35642658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 35652658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 35662db154b3SDavid Howells unsigned int, flags) 35672db154b3SDavid Howells { 35682db154b3SDavid Howells struct path from_path, to_path; 35692db154b3SDavid Howells unsigned int lflags; 35702db154b3SDavid Howells int ret = 0; 35712db154b3SDavid Howells 35722db154b3SDavid Howells if (!may_mount()) 35732db154b3SDavid Howells return -EPERM; 35742db154b3SDavid Howells 35752db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 35762db154b3SDavid Howells return -EINVAL; 35772db154b3SDavid Howells 35782db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 35792db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 35802db154b3SDavid Howells * to clear it afterwards. 35812db154b3SDavid Howells */ 35822db154b3SDavid Howells lflags = 0; 35832db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 35842db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 35852db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 35862db154b3SDavid Howells 35872db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 35882db154b3SDavid Howells if (ret < 0) 35892db154b3SDavid Howells return ret; 35902db154b3SDavid Howells 35912db154b3SDavid Howells lflags = 0; 35922db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 35932db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 35942db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 35952db154b3SDavid Howells 35962db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 35972db154b3SDavid Howells if (ret < 0) 35982db154b3SDavid Howells goto out_from; 35992db154b3SDavid Howells 36002db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 36012db154b3SDavid Howells if (ret < 0) 36022db154b3SDavid Howells goto out_to; 36032db154b3SDavid Howells 36042db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 36052db154b3SDavid Howells 36062db154b3SDavid Howells out_to: 36072db154b3SDavid Howells path_put(&to_path); 36082db154b3SDavid Howells out_from: 36092db154b3SDavid Howells path_put(&from_path); 36102db154b3SDavid Howells return ret; 36112db154b3SDavid Howells } 36122db154b3SDavid Howells 36132db154b3SDavid Howells /* 3614afac7cbaSAl Viro * Return true if path is reachable from root 3615afac7cbaSAl Viro * 361648a066e7SAl Viro * namespace_sem or mount_lock is held 3617afac7cbaSAl Viro */ 3618643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3619afac7cbaSAl Viro const struct path *root) 3620afac7cbaSAl Viro { 3621643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3622a73324daSAl Viro dentry = mnt->mnt_mountpoint; 36230714a533SAl Viro mnt = mnt->mnt_parent; 3624afac7cbaSAl Viro } 3625643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3626afac7cbaSAl Viro } 3627afac7cbaSAl Viro 3628640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3629afac7cbaSAl Viro { 363025ab4c9bSYaowei Bai bool res; 363148a066e7SAl Viro read_seqlock_excl(&mount_lock); 3632643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 363348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3634afac7cbaSAl Viro return res; 3635afac7cbaSAl Viro } 3636afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3637afac7cbaSAl Viro 3638afac7cbaSAl Viro /* 36391da177e4SLinus Torvalds * pivot_root Semantics: 36401da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 36411da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 36421da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 36431da177e4SLinus Torvalds * 36441da177e4SLinus Torvalds * Restrictions: 36451da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 36461da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 36471da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 36481da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 36491da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 36501da177e4SLinus Torvalds * 36514a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 36524a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 36534a0d11faSNeil Brown * in this situation. 36544a0d11faSNeil Brown * 36551da177e4SLinus Torvalds * Notes: 36561da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 36571da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 36581da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 36591da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 36601da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 36611da177e4SLinus Torvalds * first. 36621da177e4SLinus Torvalds */ 36633480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 36643480b257SHeiko Carstens const char __user *, put_old) 36651da177e4SLinus Torvalds { 36662763d119SAl Viro struct path new, old, root; 36672763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 366884d17192SAl Viro struct mountpoint *old_mp, *root_mp; 36691da177e4SLinus Torvalds int error; 36701da177e4SLinus Torvalds 36719b40bc90SAl Viro if (!may_mount()) 36721da177e4SLinus Torvalds return -EPERM; 36731da177e4SLinus Torvalds 3674ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3675ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 36761da177e4SLinus Torvalds if (error) 36771da177e4SLinus Torvalds goto out0; 36781da177e4SLinus Torvalds 3679ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 3680ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 36811da177e4SLinus Torvalds if (error) 36821da177e4SLinus Torvalds goto out1; 36831da177e4SLinus Torvalds 36842d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3685b12cea91SAl Viro if (error) 3686b12cea91SAl Viro goto out2; 36871da177e4SLinus Torvalds 3688f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 368984d17192SAl Viro old_mp = lock_mount(&old); 369084d17192SAl Viro error = PTR_ERR(old_mp); 369184d17192SAl Viro if (IS_ERR(old_mp)) 3692b12cea91SAl Viro goto out3; 3693b12cea91SAl Viro 36941da177e4SLinus Torvalds error = -EINVAL; 3695419148daSAl Viro new_mnt = real_mount(new.mnt); 3696419148daSAl Viro root_mnt = real_mount(root.mnt); 369784d17192SAl Viro old_mnt = real_mount(old.mnt); 36982763d119SAl Viro ex_parent = new_mnt->mnt_parent; 36992763d119SAl Viro root_parent = root_mnt->mnt_parent; 370084d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 37012763d119SAl Viro IS_MNT_SHARED(ex_parent) || 37022763d119SAl Viro IS_MNT_SHARED(root_parent)) 3703b12cea91SAl Viro goto out4; 3704143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3705b12cea91SAl Viro goto out4; 37065ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 37075ff9d8a6SEric W. Biederman goto out4; 37081da177e4SLinus Torvalds error = -ENOENT; 3709f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3710b12cea91SAl Viro goto out4; 37111da177e4SLinus Torvalds error = -EBUSY; 371284d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3713b12cea91SAl Viro goto out4; /* loop, on the same file system */ 37141da177e4SLinus Torvalds error = -EINVAL; 37158c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3716b12cea91SAl Viro goto out4; /* not a mountpoint */ 3717676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3718b12cea91SAl Viro goto out4; /* not attached */ 37192d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3720b12cea91SAl Viro goto out4; /* not a mountpoint */ 3721676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3722b12cea91SAl Viro goto out4; /* not attached */ 37234ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 372484d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3725b12cea91SAl Viro goto out4; 37260d082601SEric W. Biederman /* make certain new is below the root */ 37270d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 37280d082601SEric W. Biederman goto out4; 3729719ea2fbSAl Viro lock_mount_hash(); 37302763d119SAl Viro umount_mnt(new_mnt); 37312763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 37325ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 37335ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 37345ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 37355ff9d8a6SEric W. Biederman } 37364ac91378SJan Blunck /* mount old root on put_old */ 373784d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 37384ac91378SJan Blunck /* mount new_root on / */ 37392763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 37402763d119SAl Viro mnt_add_count(root_parent, -1); 37416b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 37424fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 37434fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 37443895dbf8SEric W. Biederman put_mountpoint(root_mp); 3745719ea2fbSAl Viro unlock_mount_hash(); 37462d8f3038SAl Viro chroot_fs_refs(&root, &new); 37471da177e4SLinus Torvalds error = 0; 3748b12cea91SAl Viro out4: 374984d17192SAl Viro unlock_mount(old_mp); 37502763d119SAl Viro if (!error) 37512763d119SAl Viro mntput_no_expire(ex_parent); 3752b12cea91SAl Viro out3: 37538c3ee42eSAl Viro path_put(&root); 3754b12cea91SAl Viro out2: 37552d8f3038SAl Viro path_put(&old); 37561da177e4SLinus Torvalds out1: 37572d8f3038SAl Viro path_put(&new); 37581da177e4SLinus Torvalds out0: 37591da177e4SLinus Torvalds return error; 37601da177e4SLinus Torvalds } 37611da177e4SLinus Torvalds 37621da177e4SLinus Torvalds static void __init init_mount_tree(void) 37631da177e4SLinus Torvalds { 37641da177e4SLinus Torvalds struct vfsmount *mnt; 376574e83122SAl Viro struct mount *m; 37666b3286edSKirill Korotaev struct mnt_namespace *ns; 3767ac748a09SJan Blunck struct path root; 37681da177e4SLinus Torvalds 3769fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 37701da177e4SLinus Torvalds if (IS_ERR(mnt)) 37711da177e4SLinus Torvalds panic("Can't create rootfs"); 3772b3e19d92SNick Piggin 377374e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 37743b22edc5STrond Myklebust if (IS_ERR(ns)) 37751da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 377674e83122SAl Viro m = real_mount(mnt); 377774e83122SAl Viro m->mnt_ns = ns; 377874e83122SAl Viro ns->root = m; 377974e83122SAl Viro ns->mounts = 1; 378074e83122SAl Viro list_add(&m->mnt_list, &ns->list); 37816b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 37826b3286edSKirill Korotaev get_mnt_ns(ns); 37831da177e4SLinus Torvalds 3784be08d6d2SAl Viro root.mnt = mnt; 3785be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3786da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3787ac748a09SJan Blunck 3788ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3789ac748a09SJan Blunck set_fs_root(current->fs, &root); 37901da177e4SLinus Torvalds } 37911da177e4SLinus Torvalds 379274bf17cfSDenis Cheng void __init mnt_init(void) 37931da177e4SLinus Torvalds { 379415a67dd8SRandy Dunlap int err; 37951da177e4SLinus Torvalds 37967d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 379720c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 37981da177e4SLinus Torvalds 37990818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 380038129a13SAl Viro sizeof(struct hlist_head), 38010818bf27SAl Viro mhash_entries, 19, 38023d375d78SPavel Tatashin HASH_ZERO, 38030818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 38040818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 38050818bf27SAl Viro sizeof(struct hlist_head), 38060818bf27SAl Viro mphash_entries, 19, 38073d375d78SPavel Tatashin HASH_ZERO, 38080818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 38091da177e4SLinus Torvalds 381084d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 38111da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 38121da177e4SLinus Torvalds 38134b93dc9bSTejun Heo kernfs_init(); 38144b93dc9bSTejun Heo 381515a67dd8SRandy Dunlap err = sysfs_init(); 381615a67dd8SRandy Dunlap if (err) 381715a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 38188e24eea7SHarvey Harrison __func__, err); 381900d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 382000d26666SGreg Kroah-Hartman if (!fs_kobj) 38218e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 3822037f11b4SAl Viro shmem_init(); 38231da177e4SLinus Torvalds init_rootfs(); 38241da177e4SLinus Torvalds init_mount_tree(); 38251da177e4SLinus Torvalds } 38261da177e4SLinus Torvalds 3827616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 38281da177e4SLinus Torvalds { 3829d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3830616511d0STrond Myklebust return; 38317b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3832771b1371SEric W. Biederman free_mnt_ns(ns); 38331da177e4SLinus Torvalds } 38349d412a43SAl Viro 3835d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 38369d412a43SAl Viro { 3837423e0ab0STim Chen struct vfsmount *mnt; 3838d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 3839423e0ab0STim Chen if (!IS_ERR(mnt)) { 3840423e0ab0STim Chen /* 3841423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3842423e0ab0STim Chen * we unmount before file sys is unregistered 3843423e0ab0STim Chen */ 3844f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3845423e0ab0STim Chen } 3846423e0ab0STim Chen return mnt; 38479d412a43SAl Viro } 3848d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 3849423e0ab0STim Chen 3850423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3851423e0ab0STim Chen { 3852423e0ab0STim Chen /* release long term mount so mount point can be released */ 3853423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3854f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 385548a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3856423e0ab0STim Chen mntput(mnt); 3857423e0ab0STim Chen } 3858423e0ab0STim Chen } 3859423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 386002125a82SAl Viro 386102125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 386202125a82SAl Viro { 3863143c8c91SAl Viro return check_mnt(real_mount(mnt)); 386402125a82SAl Viro } 38658823c079SEric W. Biederman 38663151527eSEric W. Biederman bool current_chrooted(void) 38673151527eSEric W. Biederman { 38683151527eSEric W. Biederman /* Does the current process have a non-standard root */ 38693151527eSEric W. Biederman struct path ns_root; 38703151527eSEric W. Biederman struct path fs_root; 38713151527eSEric W. Biederman bool chrooted; 38723151527eSEric W. Biederman 38733151527eSEric W. Biederman /* Find the namespace root */ 38743151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 38753151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 38763151527eSEric W. Biederman path_get(&ns_root); 38773151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 38783151527eSEric W. Biederman ; 38793151527eSEric W. Biederman 38803151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 38813151527eSEric W. Biederman 38823151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 38833151527eSEric W. Biederman 38843151527eSEric W. Biederman path_put(&fs_root); 38853151527eSEric W. Biederman path_put(&ns_root); 38863151527eSEric W. Biederman 38873151527eSEric W. Biederman return chrooted; 38883151527eSEric W. Biederman } 38893151527eSEric W. Biederman 3890132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 3891132e4608SDavid Howells const struct super_block *sb, 38928654df4eSEric W. Biederman int *new_mnt_flags) 389387a8ebd6SEric W. Biederman { 38948c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 389587a8ebd6SEric W. Biederman struct mount *mnt; 3896e51db735SEric W. Biederman bool visible = false; 389787a8ebd6SEric W. Biederman 389844bb4385SAl Viro down_read(&namespace_sem); 38999f6c61f9SMiklos Szeredi lock_ns_list(ns); 390087a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3901e51db735SEric W. Biederman struct mount *child; 390277b1a97dSEric W. Biederman int mnt_flags; 390377b1a97dSEric W. Biederman 39049f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 39059f6c61f9SMiklos Szeredi continue; 39069f6c61f9SMiklos Szeredi 3907132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 3908e51db735SEric W. Biederman continue; 3909e51db735SEric W. Biederman 39107e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 39117e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 39127e96c1b0SEric W. Biederman */ 39137e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 39147e96c1b0SEric W. Biederman continue; 39157e96c1b0SEric W. Biederman 3916a1935c17SEric W. Biederman /* A local view of the mount flags */ 391777b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 391877b1a97dSEric W. Biederman 3919695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3920bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 3921695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3922695e9df0SEric W. Biederman 39238c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 39248c6cf9ccSEric W. Biederman * than the proposed new mount. 39258c6cf9ccSEric W. Biederman */ 392677b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 39278c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 39288c6cf9ccSEric W. Biederman continue; 392977b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 393077b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 39318c6cf9ccSEric W. Biederman continue; 39328c6cf9ccSEric W. Biederman 3933ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3934ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3935ceeb0e5dSEric W. Biederman * empty directories. 3936e51db735SEric W. Biederman */ 3937e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3938e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3939ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3940d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3941ceeb0e5dSEric W. Biederman continue; 39427236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 39437236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3944e51db735SEric W. Biederman goto next; 394587a8ebd6SEric W. Biederman } 39468c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 394777b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 39488c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3949e51db735SEric W. Biederman visible = true; 3950e51db735SEric W. Biederman goto found; 3951e51db735SEric W. Biederman next: ; 395287a8ebd6SEric W. Biederman } 3953e51db735SEric W. Biederman found: 39549f6c61f9SMiklos Szeredi unlock_ns_list(ns); 395544bb4385SAl Viro up_read(&namespace_sem); 3956e51db735SEric W. Biederman return visible; 395787a8ebd6SEric W. Biederman } 395887a8ebd6SEric W. Biederman 3959132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 39608654df4eSEric W. Biederman { 3961a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 39628654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 39638654df4eSEric W. Biederman unsigned long s_iflags; 39648654df4eSEric W. Biederman 39658654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 39668654df4eSEric W. Biederman return false; 39678654df4eSEric W. Biederman 39688654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 3969132e4608SDavid Howells s_iflags = sb->s_iflags; 39708654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 39718654df4eSEric W. Biederman return false; 39728654df4eSEric W. Biederman 3973a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3974a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3975a1935c17SEric W. Biederman required_iflags); 3976a1935c17SEric W. Biederman return true; 3977a1935c17SEric W. Biederman } 3978a1935c17SEric W. Biederman 3979132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 39808654df4eSEric W. Biederman } 39818654df4eSEric W. Biederman 3982380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3983380cf5baSAndy Lutomirski { 3984380cf5baSAndy Lutomirski /* 3985380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3986380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3987380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3988380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3989380cf5baSAndy Lutomirski * in other namespaces. 3990380cf5baSAndy Lutomirski */ 3991380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3992380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3993380cf5baSAndy Lutomirski } 3994380cf5baSAndy Lutomirski 399564964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 39968823c079SEric W. Biederman { 399758be2825SAl Viro struct ns_common *ns = NULL; 39988823c079SEric W. Biederman struct nsproxy *nsproxy; 39998823c079SEric W. Biederman 4000728dba3aSEric W. Biederman task_lock(task); 4001728dba3aSEric W. Biederman nsproxy = task->nsproxy; 40028823c079SEric W. Biederman if (nsproxy) { 400358be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 400458be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 40058823c079SEric W. Biederman } 4006728dba3aSEric W. Biederman task_unlock(task); 40078823c079SEric W. Biederman 40088823c079SEric W. Biederman return ns; 40098823c079SEric W. Biederman } 40108823c079SEric W. Biederman 401164964528SAl Viro static void mntns_put(struct ns_common *ns) 40128823c079SEric W. Biederman { 401358be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 40148823c079SEric W. Biederman } 40158823c079SEric W. Biederman 401664964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 40178823c079SEric W. Biederman { 40188823c079SEric W. Biederman struct fs_struct *fs = current->fs; 40194f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 40208823c079SEric W. Biederman struct path root; 40214f757f3cSAl Viro int err; 40228823c079SEric W. Biederman 40230c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 4024c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 4025c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 4026ae11e0f1SZhao Hongjiang return -EPERM; 40278823c079SEric W. Biederman 402874e83122SAl Viro if (is_anon_ns(mnt_ns)) 402974e83122SAl Viro return -EINVAL; 403074e83122SAl Viro 40318823c079SEric W. Biederman if (fs->users != 1) 40328823c079SEric W. Biederman return -EINVAL; 40338823c079SEric W. Biederman 40348823c079SEric W. Biederman get_mnt_ns(mnt_ns); 40354f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 40368823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 40378823c079SEric W. Biederman 40388823c079SEric W. Biederman /* Find the root */ 40394f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 40404f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 40414f757f3cSAl Viro if (err) { 40424f757f3cSAl Viro /* revert to old namespace */ 40434f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 40444f757f3cSAl Viro put_mnt_ns(mnt_ns); 40454f757f3cSAl Viro return err; 40464f757f3cSAl Viro } 40478823c079SEric W. Biederman 40484068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 40494068367cSAndrei Vagin 40508823c079SEric W. Biederman /* Update the pwd and root */ 40518823c079SEric W. Biederman set_fs_pwd(fs, &root); 40528823c079SEric W. Biederman set_fs_root(fs, &root); 40538823c079SEric W. Biederman 40548823c079SEric W. Biederman path_put(&root); 40558823c079SEric W. Biederman return 0; 40568823c079SEric W. Biederman } 40578823c079SEric W. Biederman 4058bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 4059bcac25a5SAndrey Vagin { 4060bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 4061bcac25a5SAndrey Vagin } 4062bcac25a5SAndrey Vagin 40638823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 40648823c079SEric W. Biederman .name = "mnt", 40658823c079SEric W. Biederman .type = CLONE_NEWNS, 40668823c079SEric W. Biederman .get = mntns_get, 40678823c079SEric W. Biederman .put = mntns_put, 40688823c079SEric W. Biederman .install = mntns_install, 4069bcac25a5SAndrey Vagin .owner = mntns_owner, 40708823c079SEric W. Biederman }; 4071