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 */ 159edf7ddbfSEric Biggers int mnt_get_count(struct mount *mnt) 160b3e19d92SNick Piggin { 161b3e19d92SNick Piggin #ifdef CONFIG_SMP 162edf7ddbfSEric Biggers 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 down_read(&namespace_sem); 6889f6c61f9SMiklos Szeredi lock_ns_list(ns); 6897af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 6909f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 6919f6c61f9SMiklos Szeredi continue; 6927af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 6937af1364fSEric W. Biederman if (is_covered) 6947af1364fSEric W. Biederman break; 6957af1364fSEric W. Biederman } 6969f6c61f9SMiklos Szeredi unlock_ns_list(ns); 6977af1364fSEric W. Biederman up_read(&namespace_sem); 6985ad05cc8SNikolay Borisov 6997af1364fSEric W. Biederman return is_covered; 7007af1364fSEric W. Biederman } 7017af1364fSEric W. Biederman 702e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 70384d17192SAl Viro { 7040818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 70584d17192SAl Viro struct mountpoint *mp; 70684d17192SAl Viro 7070818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 70884d17192SAl Viro if (mp->m_dentry == dentry) { 70984d17192SAl Viro mp->m_count++; 71084d17192SAl Viro return mp; 71184d17192SAl Viro } 71284d17192SAl Viro } 713e2dfa935SEric W. Biederman return NULL; 714e2dfa935SEric W. Biederman } 715e2dfa935SEric W. Biederman 7163895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 717e2dfa935SEric W. Biederman { 7183895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 719e2dfa935SEric W. Biederman int ret; 72084d17192SAl Viro 7213895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7221e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7231e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7241e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7253895dbf8SEric W. Biederman mountpoint: 7263895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7273895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7283895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7293895dbf8SEric W. Biederman if (mp) 7303895dbf8SEric W. Biederman goto done; 73184d17192SAl Viro } 732eed81007SMiklos Szeredi 7333895dbf8SEric W. Biederman if (!new) 7343895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7353895dbf8SEric W. Biederman if (!new) 7363895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7373895dbf8SEric W. Biederman 7383895dbf8SEric W. Biederman 7393895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7403895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7413895dbf8SEric W. Biederman 7423895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7433895dbf8SEric W. Biederman if (ret == -EBUSY) 7443895dbf8SEric W. Biederman goto mountpoint; 7453895dbf8SEric W. Biederman 7463895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7473895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7483895dbf8SEric W. Biederman if (ret) 7493895dbf8SEric W. Biederman goto done; 7503895dbf8SEric W. Biederman 7513895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7523895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7534edbe133SAl Viro new->m_dentry = dget(dentry); 7543895dbf8SEric W. Biederman new->m_count = 1; 7553895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7563895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7573895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7583895dbf8SEric W. Biederman 7593895dbf8SEric W. Biederman mp = new; 7603895dbf8SEric W. Biederman new = NULL; 7613895dbf8SEric W. Biederman done: 7623895dbf8SEric W. Biederman kfree(new); 76384d17192SAl Viro return mp; 76484d17192SAl Viro } 76584d17192SAl Viro 7664edbe133SAl Viro /* 7674edbe133SAl Viro * vfsmount lock must be held. Additionally, the caller is responsible 7684edbe133SAl Viro * for serializing calls for given disposal list. 7694edbe133SAl Viro */ 7704edbe133SAl Viro static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) 77184d17192SAl Viro { 77284d17192SAl Viro if (!--mp->m_count) { 77384d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7740a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 77584d17192SAl Viro spin_lock(&dentry->d_lock); 77684d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 77784d17192SAl Viro spin_unlock(&dentry->d_lock); 7784edbe133SAl Viro dput_to_list(dentry, list); 7790818bf27SAl Viro hlist_del(&mp->m_hash); 78084d17192SAl Viro kfree(mp); 78184d17192SAl Viro } 78284d17192SAl Viro } 78384d17192SAl Viro 7844edbe133SAl Viro /* called with namespace_lock and vfsmount lock */ 7854edbe133SAl Viro static void put_mountpoint(struct mountpoint *mp) 7864edbe133SAl Viro { 7874edbe133SAl Viro __put_mountpoint(mp, &ex_mountpoints); 7884edbe133SAl Viro } 7894edbe133SAl Viro 790143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7911da177e4SLinus Torvalds { 7926b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7931da177e4SLinus Torvalds } 7941da177e4SLinus Torvalds 79599b7db7bSNick Piggin /* 79699b7db7bSNick Piggin * vfsmount lock must be held for write 79799b7db7bSNick Piggin */ 7986b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7995addc5ddSAl Viro { 8005addc5ddSAl Viro if (ns) { 8015addc5ddSAl Viro ns->event = ++event; 8025addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8035addc5ddSAl Viro } 8045addc5ddSAl Viro } 8055addc5ddSAl Viro 80699b7db7bSNick Piggin /* 80799b7db7bSNick Piggin * vfsmount lock must be held for write 80899b7db7bSNick Piggin */ 8096b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8105addc5ddSAl Viro { 8115addc5ddSAl Viro if (ns && ns->event != event) { 8125addc5ddSAl Viro ns->event = event; 8135addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8145addc5ddSAl Viro } 8155addc5ddSAl Viro } 8165addc5ddSAl Viro 81799b7db7bSNick Piggin /* 81899b7db7bSNick Piggin * vfsmount lock must be held for write 81999b7db7bSNick Piggin */ 820e4e59906SAl Viro static struct mountpoint *unhash_mnt(struct mount *mnt) 8211da177e4SLinus Torvalds { 822e4e59906SAl Viro struct mountpoint *mp; 8230714a533SAl Viro mnt->mnt_parent = mnt; 824a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8256b41d536SAl Viro list_del_init(&mnt->mnt_child); 82638129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8270a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 828e4e59906SAl Viro mp = mnt->mnt_mp; 82984d17192SAl Viro mnt->mnt_mp = NULL; 830e4e59906SAl Viro return mp; 8317bdb11deSEric W. Biederman } 8327bdb11deSEric W. Biederman 8337bdb11deSEric W. Biederman /* 8347bdb11deSEric W. Biederman * vfsmount lock must be held for write 8357bdb11deSEric W. Biederman */ 8366a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8376a46c573SEric W. Biederman { 838e4e59906SAl Viro put_mountpoint(unhash_mnt(mnt)); 8396a46c573SEric W. Biederman } 8406a46c573SEric W. Biederman 8416a46c573SEric W. Biederman /* 8426a46c573SEric W. Biederman * vfsmount lock must be held for write 8436a46c573SEric W. Biederman */ 84484d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 84584d17192SAl Viro struct mountpoint *mp, 84644d964d6SAl Viro struct mount *child_mnt) 847b90fa9aeSRam Pai { 84884d17192SAl Viro mp->m_count++; 8493a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 8504edbe133SAl Viro child_mnt->mnt_mountpoint = mp->m_dentry; 8513a2393d7SAl Viro child_mnt->mnt_parent = mnt; 85284d17192SAl Viro child_mnt->mnt_mp = mp; 8530a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 854b90fa9aeSRam Pai } 855b90fa9aeSRam Pai 8561064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8571064f874SEric W. Biederman { 8581064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8591064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8601064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8611064f874SEric W. Biederman } 8621064f874SEric W. Biederman 86399b7db7bSNick Piggin /* 86499b7db7bSNick Piggin * vfsmount lock must be held for write 86599b7db7bSNick Piggin */ 86684d17192SAl Viro static void attach_mnt(struct mount *mnt, 86784d17192SAl Viro struct mount *parent, 86884d17192SAl Viro struct mountpoint *mp) 8691da177e4SLinus Torvalds { 87084d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8711064f874SEric W. Biederman __attach_mnt(mnt, parent); 872b90fa9aeSRam Pai } 873b90fa9aeSRam Pai 8741064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 87512a5b529SAl Viro { 8761064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8771064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8781064f874SEric W. Biederman 8791064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8801064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8811064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8821064f874SEric W. Biederman 8831064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 8841064f874SEric W. Biederman 8851064f874SEric W. Biederman put_mountpoint(old_mp); 8861064f874SEric W. Biederman mnt_add_count(old_parent, -1); 88712a5b529SAl Viro } 88812a5b529SAl Viro 889b90fa9aeSRam Pai /* 89099b7db7bSNick Piggin * vfsmount lock must be held for write 891b90fa9aeSRam Pai */ 8921064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 893b90fa9aeSRam Pai { 8940714a533SAl Viro struct mount *parent = mnt->mnt_parent; 89583adc753SAl Viro struct mount *m; 896b90fa9aeSRam Pai LIST_HEAD(head); 897143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 898b90fa9aeSRam Pai 8990714a533SAl Viro BUG_ON(parent == mnt); 900b90fa9aeSRam Pai 9011a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 902f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 903143c8c91SAl Viro m->mnt_ns = n; 904f03c6599SAl Viro 905b90fa9aeSRam Pai list_splice(&head, n->list.prev); 906b90fa9aeSRam Pai 907d2921684SEric W. Biederman n->mounts += n->pending_mounts; 908d2921684SEric W. Biederman n->pending_mounts = 0; 909d2921684SEric W. Biederman 9101064f874SEric W. Biederman __attach_mnt(mnt, parent); 9116b3286edSKirill Korotaev touch_mnt_namespace(n); 9121da177e4SLinus Torvalds } 9131da177e4SLinus Torvalds 914909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9151da177e4SLinus Torvalds { 9166b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9176b41d536SAl Viro if (next == &p->mnt_mounts) { 9181da177e4SLinus Torvalds while (1) { 919909b0a88SAl Viro if (p == root) 9201da177e4SLinus Torvalds return NULL; 9216b41d536SAl Viro next = p->mnt_child.next; 9226b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9231da177e4SLinus Torvalds break; 9240714a533SAl Viro p = p->mnt_parent; 9251da177e4SLinus Torvalds } 9261da177e4SLinus Torvalds } 9276b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9281da177e4SLinus Torvalds } 9291da177e4SLinus Torvalds 930315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9319676f0c6SRam Pai { 9326b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9336b41d536SAl Viro while (prev != &p->mnt_mounts) { 9346b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9356b41d536SAl Viro prev = p->mnt_mounts.prev; 9369676f0c6SRam Pai } 9379676f0c6SRam Pai return p; 9389676f0c6SRam Pai } 9399676f0c6SRam Pai 9408f291889SAl Viro /** 9418f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 9428f291889SAl Viro * @fc: The configuration context with the superblock attached 9438f291889SAl Viro * 9448f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 9458f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 9468f291889SAl Viro * 9478f291889SAl Viro * Note that this does not attach the mount to anything. 9488f291889SAl Viro */ 9498f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 9509d412a43SAl Viro { 951b105e270SAl Viro struct mount *mnt; 9529d412a43SAl Viro 9538f291889SAl Viro if (!fc->root) 9548f291889SAl Viro return ERR_PTR(-EINVAL); 9559d412a43SAl Viro 9568f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 9579d412a43SAl Viro if (!mnt) 9589d412a43SAl Viro return ERR_PTR(-ENOMEM); 9599d412a43SAl Viro 9608f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 961b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9629d412a43SAl Viro 9638f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 9648f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 9658f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 966a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9670714a533SAl Viro mnt->mnt_parent = mnt; 9688f291889SAl Viro 969719ea2fbSAl Viro lock_mount_hash(); 9708f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 971719ea2fbSAl Viro unlock_mount_hash(); 972b105e270SAl Viro return &mnt->mnt; 9739d412a43SAl Viro } 9748f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 9758f291889SAl Viro 9768f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 9778f291889SAl Viro { 9788f291889SAl Viro int err = vfs_get_tree(fc); 9798f291889SAl Viro if (!err) { 9808f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 9818f291889SAl Viro return vfs_create_mount(fc); 9828f291889SAl Viro } 9838f291889SAl Viro return ERR_PTR(err); 9848f291889SAl Viro } 9858f291889SAl Viro EXPORT_SYMBOL(fc_mount); 9868f291889SAl Viro 9879bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 9889bc61ab1SDavid Howells int flags, const char *name, 9899bc61ab1SDavid Howells void *data) 9901da177e4SLinus Torvalds { 9919bc61ab1SDavid Howells struct fs_context *fc; 9928f291889SAl Viro struct vfsmount *mnt; 9939bc61ab1SDavid Howells int ret = 0; 9949d412a43SAl Viro 9959d412a43SAl Viro if (!type) 9963e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 9979d412a43SAl Viro 9989bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 9999bc61ab1SDavid Howells if (IS_ERR(fc)) 10009bc61ab1SDavid Howells return ERR_CAST(fc); 10019bc61ab1SDavid Howells 10023e1aeb00SDavid Howells if (name) 10033e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 10043e1aeb00SDavid Howells name, strlen(name)); 10059bc61ab1SDavid Howells if (!ret) 10069bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 10079bc61ab1SDavid Howells if (!ret) 10088f291889SAl Viro mnt = fc_mount(fc); 10098f291889SAl Viro else 10108f291889SAl Viro mnt = ERR_PTR(ret); 10119d412a43SAl Viro 10129bc61ab1SDavid Howells put_fs_context(fc); 10138f291889SAl Viro return mnt; 10149d412a43SAl Viro } 10159d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10169d412a43SAl Viro 101793faccbbSEric W. Biederman struct vfsmount * 101893faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 101993faccbbSEric W. Biederman const char *name, void *data) 102093faccbbSEric W. Biederman { 102193faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 102293faccbbSEric W. Biederman * through from the parent mount to the submount don't support 102393faccbbSEric W. Biederman * unprivileged mounts with submounts. 102493faccbbSEric W. Biederman */ 102593faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 102693faccbbSEric W. Biederman return ERR_PTR(-EPERM); 102793faccbbSEric W. Biederman 1028e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 102993faccbbSEric W. Biederman } 103093faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 103193faccbbSEric W. Biederman 103287129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 103336341f64SRam Pai int flag) 10341da177e4SLinus Torvalds { 103587129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1036be34d1a3SDavid Howells struct mount *mnt; 1037be34d1a3SDavid Howells int err; 10381da177e4SLinus Torvalds 1039be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1040be34d1a3SDavid Howells if (!mnt) 1041be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1042be34d1a3SDavid Howells 10437a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 104415169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1045719f5d7fSMiklos Szeredi else 104615169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1047719f5d7fSMiklos Szeredi 104815169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1049be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1050719f5d7fSMiklos Szeredi if (err) 1051719f5d7fSMiklos Szeredi goto out_free; 1052719f5d7fSMiklos Szeredi } 1053719f5d7fSMiklos Szeredi 105416a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 105516a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 10565ff9d8a6SEric W. Biederman 10571da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1058b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1059b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1060a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10610714a533SAl Viro mnt->mnt_parent = mnt; 1062719ea2fbSAl Viro lock_mount_hash(); 106339f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1064719ea2fbSAl Viro unlock_mount_hash(); 1065b90fa9aeSRam Pai 10667a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10677a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10686776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 106932301920SAl Viro mnt->mnt_master = old; 1070fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10718aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1072fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10736776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1074d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10756776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1076d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10775235d448SAl Viro } else { 10785235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10795afe0022SRam Pai } 1080b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10810f0afb1dSAl Viro set_mnt_shared(mnt); 10821da177e4SLinus Torvalds 10831da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10841da177e4SLinus Torvalds * as the original if that was on one */ 108536341f64SRam Pai if (flag & CL_EXPIRE) { 10866776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10876776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10881da177e4SLinus Torvalds } 1089be34d1a3SDavid Howells 1090cb338d06SAl Viro return mnt; 1091719f5d7fSMiklos Szeredi 1092719f5d7fSMiklos Szeredi out_free: 10938ffcb32eSDavid Howells mnt_free_id(mnt); 1094719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1095be34d1a3SDavid Howells return ERR_PTR(err); 10961da177e4SLinus Torvalds } 10971da177e4SLinus Torvalds 10989ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10999ea459e1SAl Viro { 110056cbb429SAl Viro struct hlist_node *p; 110156cbb429SAl Viro struct mount *m; 11029ea459e1SAl Viro /* 110356cbb429SAl Viro * The warning here probably indicates that somebody messed 110456cbb429SAl Viro * up a mnt_want/drop_write() pair. If this happens, the 110556cbb429SAl Viro * filesystem was probably unable to make r/w->r/o transitions. 11069ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11079ea459e1SAl Viro * so mnt_get_writers() below is safe. 11089ea459e1SAl Viro */ 11099ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11109ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11119ea459e1SAl Viro mnt_pin_kill(mnt); 111256cbb429SAl Viro hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { 111356cbb429SAl Viro hlist_del(&m->mnt_umount); 111456cbb429SAl Viro mntput(&m->mnt); 111556cbb429SAl Viro } 11169ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11179ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11189ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11199ea459e1SAl Viro mnt_free_id(mnt); 11209ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11219ea459e1SAl Viro } 11229ea459e1SAl Viro 11239ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11249ea459e1SAl Viro { 11259ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11269ea459e1SAl Viro } 11279ea459e1SAl Viro 11289ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11299ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11309ea459e1SAl Viro { 11319ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 113229785735SByungchul Park struct mount *m, *t; 11339ea459e1SAl Viro 113429785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 113529785735SByungchul Park cleanup_mnt(m); 11369ea459e1SAl Viro } 11379ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11389ea459e1SAl Viro 1139900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11407b7b1aceSAl Viro { 11414edbe133SAl Viro LIST_HEAD(list); 1142edf7ddbfSEric Biggers int count; 11434edbe133SAl Viro 114448a066e7SAl Viro rcu_read_lock(); 11459ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11469ea0a46cSAl Viro /* 11479ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11489ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11499ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11509ea0a46cSAl Viro * be dropped until after an RCU delay done after 11519ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11529ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11539ea0a46cSAl Viro * we are dropping is not the final one. 11549ea0a46cSAl Viro */ 1155aa9c0e07SAl Viro mnt_add_count(mnt, -1); 115648a066e7SAl Viro rcu_read_unlock(); 115799b7db7bSNick Piggin return; 1158b3e19d92SNick Piggin } 1159719ea2fbSAl Viro lock_mount_hash(); 1160119e1ef8SAl Viro /* 1161119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1162119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1163119e1ef8SAl Viro */ 1164119e1ef8SAl Viro smp_mb(); 11659ea0a46cSAl Viro mnt_add_count(mnt, -1); 1166edf7ddbfSEric Biggers count = mnt_get_count(mnt); 1167edf7ddbfSEric Biggers if (count != 0) { 1168edf7ddbfSEric Biggers WARN_ON(count < 0); 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); 119791989c70SJens Axboe if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) 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 171225ccd24fSChristoph Hellwig static int can_umount(const struct path *path, int flags) 17131da177e4SLinus Torvalds { 171425ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 17151da177e4SLinus Torvalds 17169b40bc90SAl Viro if (!may_mount()) 17179b40bc90SAl Viro return -EPERM; 171841525f56SChristoph Hellwig if (path->dentry != path->mnt->mnt_root) 171925ccd24fSChristoph Hellwig return -EINVAL; 1720143c8c91SAl Viro if (!check_mnt(mnt)) 172125ccd24fSChristoph Hellwig return -EINVAL; 172225d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 172325ccd24fSChristoph Hellwig return -EINVAL; 1724b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 172525ccd24fSChristoph Hellwig return -EPERM; 172625ccd24fSChristoph Hellwig return 0; 172725ccd24fSChristoph Hellwig } 17281da177e4SLinus Torvalds 1729*a0a6df9aSAl Viro // caller is responsible for flags being sane 173025ccd24fSChristoph Hellwig int path_umount(struct path *path, int flags) 173125ccd24fSChristoph Hellwig { 173225ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 173325ccd24fSChristoph Hellwig int ret; 173425ccd24fSChristoph Hellwig 173525ccd24fSChristoph Hellwig ret = can_umount(path, flags); 173625ccd24fSChristoph Hellwig if (!ret) 173725ccd24fSChristoph Hellwig ret = do_umount(mnt, flags); 173825ccd24fSChristoph Hellwig 1739429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 174041525f56SChristoph Hellwig dput(path->dentry); 1741900148dcSAl Viro mntput_no_expire(mnt); 174225ccd24fSChristoph Hellwig return ret; 17431da177e4SLinus Torvalds } 17441da177e4SLinus Torvalds 174509267defSChristoph Hellwig static int ksys_umount(char __user *name, int flags) 174641525f56SChristoph Hellwig { 174741525f56SChristoph Hellwig int lookup_flags = LOOKUP_MOUNTPOINT; 174841525f56SChristoph Hellwig struct path path; 174941525f56SChristoph Hellwig int ret; 175041525f56SChristoph Hellwig 1751*a0a6df9aSAl Viro // basic validity checks done first 1752*a0a6df9aSAl Viro if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1753*a0a6df9aSAl Viro return -EINVAL; 1754*a0a6df9aSAl Viro 175541525f56SChristoph Hellwig if (!(flags & UMOUNT_NOFOLLOW)) 175641525f56SChristoph Hellwig lookup_flags |= LOOKUP_FOLLOW; 175741525f56SChristoph Hellwig ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); 175841525f56SChristoph Hellwig if (ret) 175941525f56SChristoph Hellwig return ret; 176041525f56SChristoph Hellwig return path_umount(&path, flags); 176141525f56SChristoph Hellwig } 176241525f56SChristoph Hellwig 17633a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 17643a18ef5cSDominik Brodowski { 17653a18ef5cSDominik Brodowski return ksys_umount(name, flags); 17663a18ef5cSDominik Brodowski } 17673a18ef5cSDominik Brodowski 17681da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 17691da177e4SLinus Torvalds 17701da177e4SLinus Torvalds /* 17711da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17721da177e4SLinus Torvalds */ 1773bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17741da177e4SLinus Torvalds { 17753a18ef5cSDominik Brodowski return ksys_umount(name, 0); 17761da177e4SLinus Torvalds } 17771da177e4SLinus Torvalds 17781da177e4SLinus Torvalds #endif 17791da177e4SLinus Torvalds 17804ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 17818823c079SEric W. Biederman { 17824ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1783e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1784e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 17854ce5d2b1SEric W. Biederman } 17864ce5d2b1SEric W. Biederman 1787213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 178858be2825SAl Viro { 178958be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 179058be2825SAl Viro } 179158be2825SAl Viro 1792303cc571SChristian Brauner struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) 1793303cc571SChristian Brauner { 1794303cc571SChristian Brauner return &mnt->ns; 1795303cc571SChristian Brauner } 1796303cc571SChristian Brauner 17974ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 17984ce5d2b1SEric W. Biederman { 17994ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 18004ce5d2b1SEric W. Biederman * mount namespace loop? 18014ce5d2b1SEric W. Biederman */ 18024ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 18034ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 18044ce5d2b1SEric W. Biederman return false; 18054ce5d2b1SEric W. Biederman 1806f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 18078823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 18088823c079SEric W. Biederman } 18098823c079SEric W. Biederman 181087129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 181136341f64SRam Pai int flag) 18121da177e4SLinus Torvalds { 181384d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 18141da177e4SLinus Torvalds 18154ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 18164ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 18174ce5d2b1SEric W. Biederman 18184ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1819be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 18209676f0c6SRam Pai 182136341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1822be34d1a3SDavid Howells if (IS_ERR(q)) 1823be34d1a3SDavid Howells return q; 1824be34d1a3SDavid Howells 1825a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 18261da177e4SLinus Torvalds 18271da177e4SLinus Torvalds p = mnt; 18286b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1829315fc83eSAl Viro struct mount *s; 18307ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 18311da177e4SLinus Torvalds continue; 18321da177e4SLinus Torvalds 1833909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 18344ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 18354ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1836df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1837df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1838df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1839df7342b2SEric W. Biederman goto out; 1840df7342b2SEric W. Biederman } else { 18414ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 18424ce5d2b1SEric W. Biederman continue; 18434ce5d2b1SEric W. Biederman } 1844df7342b2SEric W. Biederman } 18454ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 18464ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 18479676f0c6SRam Pai s = skip_mnt_tree(s); 18489676f0c6SRam Pai continue; 18499676f0c6SRam Pai } 18500714a533SAl Viro while (p != s->mnt_parent) { 18510714a533SAl Viro p = p->mnt_parent; 18520714a533SAl Viro q = q->mnt_parent; 18531da177e4SLinus Torvalds } 185487129cc0SAl Viro p = s; 185584d17192SAl Viro parent = q; 185687129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1857be34d1a3SDavid Howells if (IS_ERR(q)) 1858be34d1a3SDavid Howells goto out; 1859719ea2fbSAl Viro lock_mount_hash(); 18601a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 18611064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1862719ea2fbSAl Viro unlock_mount_hash(); 18631da177e4SLinus Torvalds } 18641da177e4SLinus Torvalds } 18651da177e4SLinus Torvalds return res; 1866be34d1a3SDavid Howells out: 18671da177e4SLinus Torvalds if (res) { 1868719ea2fbSAl Viro lock_mount_hash(); 1869e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1870719ea2fbSAl Viro unlock_mount_hash(); 18711da177e4SLinus Torvalds } 1872be34d1a3SDavid Howells return q; 18731da177e4SLinus Torvalds } 18741da177e4SLinus Torvalds 1875be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1876be34d1a3SDavid Howells 1877ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 18788aec0809SAl Viro { 1879cb338d06SAl Viro struct mount *tree; 188097216be0SAl Viro namespace_lock(); 1881cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1882cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1883cd4a4017SEric W. Biederman else 188487129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 188587129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1886328e6d90SAl Viro namespace_unlock(); 1887be34d1a3SDavid Howells if (IS_ERR(tree)) 188852e220d3SDan Carpenter return ERR_CAST(tree); 1889be34d1a3SDavid Howells return &tree->mnt; 18908aec0809SAl Viro } 18918aec0809SAl Viro 1892a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1893a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1894a07b2000SAl Viro 1895a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1896a07b2000SAl Viro { 1897a07b2000SAl Viro struct mnt_namespace *ns; 1898a07b2000SAl Viro namespace_lock(); 1899a07b2000SAl Viro lock_mount_hash(); 1900a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 190144dfd84aSDavid Howells if (ns) { 190244dfd84aSDavid Howells if (is_anon_ns(ns)) 1903a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 190444dfd84aSDavid Howells else 190544dfd84aSDavid Howells ns = NULL; 190644dfd84aSDavid Howells } 1907a07b2000SAl Viro unlock_mount_hash(); 1908a07b2000SAl Viro namespace_unlock(); 190944dfd84aSDavid Howells if (ns) 1910a07b2000SAl Viro free_mnt_ns(ns); 1911a07b2000SAl Viro } 1912a07b2000SAl Viro 19138aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 19148aec0809SAl Viro { 191597216be0SAl Viro namespace_lock(); 1916719ea2fbSAl Viro lock_mount_hash(); 19179c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1918719ea2fbSAl Viro unlock_mount_hash(); 19193ab6abeeSAl Viro namespace_unlock(); 19208aec0809SAl Viro } 19218aec0809SAl Viro 1922c771d683SMiklos Szeredi /** 1923c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1924c771d683SMiklos Szeredi * 1925c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1926c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1927c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1928c771d683SMiklos Szeredi * 1929c771d683SMiklos Szeredi * Release with mntput(). 1930c771d683SMiklos Szeredi */ 1931ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1932c771d683SMiklos Szeredi { 1933c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1934c771d683SMiklos Szeredi struct mount *new_mnt; 1935c771d683SMiklos Szeredi 1936c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1937c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1938c771d683SMiklos Szeredi 1939c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1940c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1941c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1942c771d683SMiklos Szeredi 1943df820f8dSMiklos Szeredi /* Longterm mount to be removed by kern_unmount*() */ 1944df820f8dSMiklos Szeredi new_mnt->mnt_ns = MNT_NS_INTERNAL; 1945df820f8dSMiklos Szeredi 1946c771d683SMiklos Szeredi return &new_mnt->mnt; 1947c771d683SMiklos Szeredi } 1948c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1949c771d683SMiklos Szeredi 19501f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 19511f707137SAl Viro struct vfsmount *root) 19521f707137SAl Viro { 19531a4eeaf2SAl Viro struct mount *mnt; 19541f707137SAl Viro int res = f(root, arg); 19551f707137SAl Viro if (res) 19561f707137SAl Viro return res; 19571a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 19581a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 19591f707137SAl Viro if (res) 19601f707137SAl Viro return res; 19611f707137SAl Viro } 19621f707137SAl Viro return 0; 19631f707137SAl Viro } 19641f707137SAl Viro 19653bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 19663bd045ccSAl Viro { 19673bd045ccSAl Viro struct mount *p; 19683bd045ccSAl Viro 19693bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 19703bd045ccSAl Viro int flags = p->mnt.mnt_flags; 19713bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 19723bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 19733bd045ccSAl Viro 19743bd045ccSAl Viro if (flags & MNT_READONLY) 19753bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 19763bd045ccSAl Viro 19773bd045ccSAl Viro if (flags & MNT_NODEV) 19783bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 19793bd045ccSAl Viro 19803bd045ccSAl Viro if (flags & MNT_NOSUID) 19813bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 19823bd045ccSAl Viro 19833bd045ccSAl Viro if (flags & MNT_NOEXEC) 19843bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 19853bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 19863bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 19873bd045ccSAl Viro flags |= MNT_LOCKED; 19883bd045ccSAl Viro p->mnt.mnt_flags = flags; 19893bd045ccSAl Viro } 19903bd045ccSAl Viro } 19913bd045ccSAl Viro 19924b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1993719f5d7fSMiklos Szeredi { 1994315fc83eSAl Viro struct mount *p; 1995719f5d7fSMiklos Szeredi 1996909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1997fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 19984b8b21f4SAl Viro mnt_release_group_id(p); 1999719f5d7fSMiklos Szeredi } 2000719f5d7fSMiklos Szeredi } 2001719f5d7fSMiklos Szeredi 20024b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 2003719f5d7fSMiklos Szeredi { 2004315fc83eSAl Viro struct mount *p; 2005719f5d7fSMiklos Szeredi 2006909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 2007fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 20084b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 2009719f5d7fSMiklos Szeredi if (err) { 20104b8b21f4SAl Viro cleanup_group_ids(mnt, p); 2011719f5d7fSMiklos Szeredi return err; 2012719f5d7fSMiklos Szeredi } 2013719f5d7fSMiklos Szeredi } 2014719f5d7fSMiklos Szeredi } 2015719f5d7fSMiklos Szeredi 2016719f5d7fSMiklos Szeredi return 0; 2017719f5d7fSMiklos Szeredi } 2018719f5d7fSMiklos Szeredi 2019d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 2020d2921684SEric W. Biederman { 2021d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 2022d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 2023d2921684SEric W. Biederman struct mount *p; 2024d2921684SEric W. Biederman 2025d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 2026d2921684SEric W. Biederman mounts++; 2027d2921684SEric W. Biederman 2028d2921684SEric W. Biederman old = ns->mounts; 2029d2921684SEric W. Biederman pending = ns->pending_mounts; 2030d2921684SEric W. Biederman sum = old + pending; 2031d2921684SEric W. Biederman if ((old > sum) || 2032d2921684SEric W. Biederman (pending > sum) || 2033d2921684SEric W. Biederman (max < sum) || 2034d2921684SEric W. Biederman (mounts > (max - sum))) 2035d2921684SEric W. Biederman return -ENOSPC; 2036d2921684SEric W. Biederman 2037d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 2038d2921684SEric W. Biederman return 0; 2039d2921684SEric W. Biederman } 2040d2921684SEric W. Biederman 2041b90fa9aeSRam Pai /* 2042b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 2043b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 204421444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 204521444403SRam Pai * store the parent mount and mountpoint dentry. 204621444403SRam Pai * (done when source_mnt is moved) 2047b90fa9aeSRam Pai * 2048b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 2049b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 20509676f0c6SRam Pai * --------------------------------------------------------------------------- 2051b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 20529676f0c6SRam Pai * |************************************************************************** 20539676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 20549676f0c6SRam Pai * | dest | | | | | 20559676f0c6SRam Pai * | | | | | | | 20569676f0c6SRam Pai * | v | | | | | 20579676f0c6SRam Pai * |************************************************************************** 20589676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 20595afe0022SRam Pai * | | | | | | 20609676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 20619676f0c6SRam Pai * *************************************************************************** 2062b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 2063b90fa9aeSRam Pai * destination mount. 2064b90fa9aeSRam Pai * 2065b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2066b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2067b90fa9aeSRam Pai * the peer group of the source mount. 2068b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2069b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2070b90fa9aeSRam Pai * mount. 20715afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 20725afe0022SRam Pai * of the destination mount and the cloned mount is made slave 20735afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 20745afe0022SRam Pai * is marked as 'shared and slave'. 20755afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 20765afe0022SRam Pai * source mount. 20775afe0022SRam Pai * 20789676f0c6SRam Pai * --------------------------------------------------------------------------- 207921444403SRam Pai * | MOVE MOUNT OPERATION | 20809676f0c6SRam Pai * |************************************************************************** 20819676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 20829676f0c6SRam Pai * | dest | | | | | 20839676f0c6SRam Pai * | | | | | | | 20849676f0c6SRam Pai * | v | | | | | 20859676f0c6SRam Pai * |************************************************************************** 20869676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 20875afe0022SRam Pai * | | | | | | 20889676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 20899676f0c6SRam Pai * *************************************************************************** 20905afe0022SRam Pai * 20915afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 20925afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 209321444403SRam Pai * (+*) the mount is moved to the destination. 20945afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 20955afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 20965afe0022SRam Pai * the mount is marked as 'shared and slave'. 20975afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2098b90fa9aeSRam Pai * 2099b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2100b90fa9aeSRam Pai * applied to each mount in the tree. 2101b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2102b90fa9aeSRam Pai * in allocations. 2103b90fa9aeSRam Pai */ 21040fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 210584d17192SAl Viro struct mount *dest_mnt, 210684d17192SAl Viro struct mountpoint *dest_mp, 21072763d119SAl Viro bool moving) 2108b90fa9aeSRam Pai { 21093bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 211038129a13SAl Viro HLIST_HEAD(tree_list); 2111d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 21121064f874SEric W. Biederman struct mountpoint *smp; 2113315fc83eSAl Viro struct mount *child, *p; 211438129a13SAl Viro struct hlist_node *n; 2115719f5d7fSMiklos Szeredi int err; 2116b90fa9aeSRam Pai 21171064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 21181064f874SEric W. Biederman * to be tucked under other mounts. 21191064f874SEric W. Biederman */ 21201064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 21211064f874SEric W. Biederman if (IS_ERR(smp)) 21221064f874SEric W. Biederman return PTR_ERR(smp); 21231064f874SEric W. Biederman 2124d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 21252763d119SAl Viro if (!moving) { 2126d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2127d2921684SEric W. Biederman if (err) 2128d2921684SEric W. Biederman goto out; 2129d2921684SEric W. Biederman } 2130d2921684SEric W. Biederman 2131fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 21320fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2133719f5d7fSMiklos Szeredi if (err) 2134719f5d7fSMiklos Szeredi goto out; 213584d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2136f2ebb3a9SAl Viro lock_mount_hash(); 2137719f5d7fSMiklos Szeredi if (err) 2138719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2139909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 21400f0afb1dSAl Viro set_mnt_shared(p); 21410b1b901bSAl Viro } else { 21420b1b901bSAl Viro lock_mount_hash(); 2143b90fa9aeSRam Pai } 21442763d119SAl Viro if (moving) { 21452763d119SAl Viro unhash_mnt(source_mnt); 214684d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2147143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 214821444403SRam Pai } else { 214944dfd84aSDavid Howells if (source_mnt->mnt_ns) { 215044dfd84aSDavid Howells /* move from anon - the caller will destroy */ 215144dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 215244dfd84aSDavid Howells } 215384d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 21541064f874SEric W. Biederman commit_tree(source_mnt); 215521444403SRam Pai } 2156b90fa9aeSRam Pai 215738129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 21581d6a32acSAl Viro struct mount *q; 215938129a13SAl Viro hlist_del_init(&child->mnt_hash); 21601064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 21611d6a32acSAl Viro child->mnt_mountpoint); 21621064f874SEric W. Biederman if (q) 21631064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 21643bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 21653bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 21663bd045ccSAl Viro lock_mnt_tree(child); 2167d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 21681064f874SEric W. Biederman commit_tree(child); 2169b90fa9aeSRam Pai } 21701064f874SEric W. Biederman put_mountpoint(smp); 2171719ea2fbSAl Viro unlock_mount_hash(); 217299b7db7bSNick Piggin 2173b90fa9aeSRam Pai return 0; 2174719f5d7fSMiklos Szeredi 2175719f5d7fSMiklos Szeredi out_cleanup_ids: 2176f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2177f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2178d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2179e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2180f2ebb3a9SAl Viro } 2181f2ebb3a9SAl Viro unlock_mount_hash(); 21820fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2183719f5d7fSMiklos Szeredi out: 2184d2921684SEric W. Biederman ns->pending_mounts = 0; 21851064f874SEric W. Biederman 21861064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 21871064f874SEric W. Biederman put_mountpoint(smp); 21881064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 21891064f874SEric W. Biederman 2190719f5d7fSMiklos Szeredi return err; 2191b90fa9aeSRam Pai } 2192b90fa9aeSRam Pai 219384d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2194b12cea91SAl Viro { 2195b12cea91SAl Viro struct vfsmount *mnt; 219684d17192SAl Viro struct dentry *dentry = path->dentry; 2197b12cea91SAl Viro retry: 21985955102cSAl Viro inode_lock(dentry->d_inode); 219984d17192SAl Viro if (unlikely(cant_mount(dentry))) { 22005955102cSAl Viro inode_unlock(dentry->d_inode); 220184d17192SAl Viro return ERR_PTR(-ENOENT); 2202b12cea91SAl Viro } 220397216be0SAl Viro namespace_lock(); 2204b12cea91SAl Viro mnt = lookup_mnt(path); 220584d17192SAl Viro if (likely(!mnt)) { 22063895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 220784d17192SAl Viro if (IS_ERR(mp)) { 220897216be0SAl Viro namespace_unlock(); 22095955102cSAl Viro inode_unlock(dentry->d_inode); 221084d17192SAl Viro return mp; 221184d17192SAl Viro } 221284d17192SAl Viro return mp; 221384d17192SAl Viro } 221497216be0SAl Viro namespace_unlock(); 22155955102cSAl Viro inode_unlock(path->dentry->d_inode); 2216b12cea91SAl Viro path_put(path); 2217b12cea91SAl Viro path->mnt = mnt; 221884d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2219b12cea91SAl Viro goto retry; 2220b12cea91SAl Viro } 2221b12cea91SAl Viro 222284d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2223b12cea91SAl Viro { 222484d17192SAl Viro struct dentry *dentry = where->m_dentry; 22253895dbf8SEric W. Biederman 22263895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 222784d17192SAl Viro put_mountpoint(where); 22283895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 22293895dbf8SEric W. Biederman 2230328e6d90SAl Viro namespace_unlock(); 22315955102cSAl Viro inode_unlock(dentry->d_inode); 2232b12cea91SAl Viro } 2233b12cea91SAl Viro 223484d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 22351da177e4SLinus Torvalds { 2236e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 22371da177e4SLinus Torvalds return -EINVAL; 22381da177e4SLinus Torvalds 2239e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2240e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 22411da177e4SLinus Torvalds return -ENOTDIR; 22421da177e4SLinus Torvalds 22432763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 22441da177e4SLinus Torvalds } 22451da177e4SLinus Torvalds 22461da177e4SLinus Torvalds /* 22477a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 22487a2e8a8fSValerie Aurora */ 22497a2e8a8fSValerie Aurora 2250e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 22517a2e8a8fSValerie Aurora { 2252e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 22537a2e8a8fSValerie Aurora 22547a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 22557a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 22567a2e8a8fSValerie Aurora return 0; 22577a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 22587a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 22597a2e8a8fSValerie Aurora return 0; 22607a2e8a8fSValerie Aurora return type; 22617a2e8a8fSValerie Aurora } 22627a2e8a8fSValerie Aurora 22637a2e8a8fSValerie Aurora /* 226407b20889SRam Pai * recursively change the type of the mountpoint. 226507b20889SRam Pai */ 2266e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 226707b20889SRam Pai { 2268315fc83eSAl Viro struct mount *m; 22694b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2270e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 22717a2e8a8fSValerie Aurora int type; 2272719f5d7fSMiklos Szeredi int err = 0; 227307b20889SRam Pai 22742d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 227507b20889SRam Pai return -EINVAL; 227607b20889SRam Pai 2277e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 22787a2e8a8fSValerie Aurora if (!type) 22797a2e8a8fSValerie Aurora return -EINVAL; 22807a2e8a8fSValerie Aurora 228197216be0SAl Viro namespace_lock(); 2282719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2283719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2284719f5d7fSMiklos Szeredi if (err) 2285719f5d7fSMiklos Szeredi goto out_unlock; 2286719f5d7fSMiklos Szeredi } 2287719f5d7fSMiklos Szeredi 2288719ea2fbSAl Viro lock_mount_hash(); 2289909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 22900f0afb1dSAl Viro change_mnt_propagation(m, type); 2291719ea2fbSAl Viro unlock_mount_hash(); 2292719f5d7fSMiklos Szeredi 2293719f5d7fSMiklos Szeredi out_unlock: 229497216be0SAl Viro namespace_unlock(); 2295719f5d7fSMiklos Szeredi return err; 229607b20889SRam Pai } 229707b20889SRam Pai 22985ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 22995ff9d8a6SEric W. Biederman { 23005ff9d8a6SEric W. Biederman struct mount *child; 23015ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 23025ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 23035ff9d8a6SEric W. Biederman continue; 23045ff9d8a6SEric W. Biederman 23055ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 23065ff9d8a6SEric W. Biederman return true; 23075ff9d8a6SEric W. Biederman } 23085ff9d8a6SEric W. Biederman return false; 23095ff9d8a6SEric W. Biederman } 23105ff9d8a6SEric W. Biederman 2311a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2312a07b2000SAl Viro { 2313a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2314a07b2000SAl Viro 2315a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2316a07b2000SAl Viro return mnt; 2317a07b2000SAl Viro 2318a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2319a07b2000SAl Viro return mnt; 2320a07b2000SAl Viro 2321a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2322a07b2000SAl Viro return mnt; 2323a07b2000SAl Viro 2324a07b2000SAl Viro if (recurse) 2325a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2326a07b2000SAl Viro else 2327a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2328a07b2000SAl Viro 2329a07b2000SAl Viro if (!IS_ERR(mnt)) 2330a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2331a07b2000SAl Viro 2332a07b2000SAl Viro return mnt; 2333a07b2000SAl Viro } 2334a07b2000SAl Viro 233507b20889SRam Pai /* 23361da177e4SLinus Torvalds * do loopback mount. 23371da177e4SLinus Torvalds */ 2338808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 23392dafe1c4SEric Sandeen int recurse) 23401da177e4SLinus Torvalds { 23412d92ab3cSAl Viro struct path old_path; 2342a07b2000SAl Viro struct mount *mnt = NULL, *parent; 234384d17192SAl Viro struct mountpoint *mp; 234457eccb83SAl Viro int err; 23451da177e4SLinus Torvalds if (!old_name || !*old_name) 23461da177e4SLinus Torvalds return -EINVAL; 2347815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 23481da177e4SLinus Torvalds if (err) 23491da177e4SLinus Torvalds return err; 23501da177e4SLinus Torvalds 23518823c079SEric W. Biederman err = -EINVAL; 23524ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 23538823c079SEric W. Biederman goto out; 23548823c079SEric W. Biederman 235584d17192SAl Viro mp = lock_mount(path); 2356a07b2000SAl Viro if (IS_ERR(mp)) { 235784d17192SAl Viro err = PTR_ERR(mp); 23589676f0c6SRam Pai goto out; 2359a07b2000SAl Viro } 23609676f0c6SRam Pai 236184d17192SAl Viro parent = real_mount(path->mnt); 2362e149ed2bSAl Viro if (!check_mnt(parent)) 2363e149ed2bSAl Viro goto out2; 2364e149ed2bSAl Viro 2365a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2366be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2367be34d1a3SDavid Howells err = PTR_ERR(mnt); 2368e9c5d8a5SAndrey Vagin goto out2; 2369be34d1a3SDavid Howells } 2370ccd48bc7SAl Viro 237184d17192SAl Viro err = graft_tree(mnt, parent, mp); 23721da177e4SLinus Torvalds if (err) { 2373719ea2fbSAl Viro lock_mount_hash(); 2374e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2375719ea2fbSAl Viro unlock_mount_hash(); 23765b83d2c5SRam Pai } 2377b12cea91SAl Viro out2: 237884d17192SAl Viro unlock_mount(mp); 2379ccd48bc7SAl Viro out: 23802d92ab3cSAl Viro path_put(&old_path); 23811da177e4SLinus Torvalds return err; 23821da177e4SLinus Torvalds } 23831da177e4SLinus Torvalds 2384a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2385a07b2000SAl Viro { 2386a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2387a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2388a07b2000SAl Viro struct mount *mnt, *p; 2389a07b2000SAl Viro struct file *file; 2390a07b2000SAl Viro 2391a07b2000SAl Viro if (IS_ERR(ns)) 2392a07b2000SAl Viro return ERR_CAST(ns); 2393a07b2000SAl Viro 2394a07b2000SAl Viro namespace_lock(); 2395a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2396a07b2000SAl Viro if (IS_ERR(mnt)) { 2397a07b2000SAl Viro namespace_unlock(); 2398a07b2000SAl Viro free_mnt_ns(ns); 2399a07b2000SAl Viro return ERR_CAST(mnt); 2400a07b2000SAl Viro } 2401a07b2000SAl Viro 2402a07b2000SAl Viro lock_mount_hash(); 2403a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2404a07b2000SAl Viro p->mnt_ns = ns; 2405a07b2000SAl Viro ns->mounts++; 2406a07b2000SAl Viro } 2407a07b2000SAl Viro ns->root = mnt; 2408a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2409a07b2000SAl Viro mntget(&mnt->mnt); 2410a07b2000SAl Viro unlock_mount_hash(); 2411a07b2000SAl Viro namespace_unlock(); 2412a07b2000SAl Viro 2413a07b2000SAl Viro mntput(path->mnt); 2414a07b2000SAl Viro path->mnt = &mnt->mnt; 2415a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2416a07b2000SAl Viro if (IS_ERR(file)) 2417a07b2000SAl Viro dissolve_on_fput(path->mnt); 2418a07b2000SAl Viro else 2419a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2420a07b2000SAl Viro return file; 2421a07b2000SAl Viro } 2422a07b2000SAl Viro 24232658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2424a07b2000SAl Viro { 2425a07b2000SAl Viro struct file *file; 2426a07b2000SAl Viro struct path path; 2427a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2428a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2429a07b2000SAl Viro int error; 2430a07b2000SAl Viro int fd; 2431a07b2000SAl Viro 2432a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2433a07b2000SAl Viro 2434a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2435a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2436a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2437a07b2000SAl Viro return -EINVAL; 2438a07b2000SAl Viro 2439a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2440a07b2000SAl Viro return -EINVAL; 2441a07b2000SAl Viro 2442a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2443a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2444a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2445a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2446a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2447a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2448a07b2000SAl Viro 2449a07b2000SAl Viro if (detached && !may_mount()) 2450a07b2000SAl Viro return -EPERM; 2451a07b2000SAl Viro 2452a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2453a07b2000SAl Viro if (fd < 0) 2454a07b2000SAl Viro return fd; 2455a07b2000SAl Viro 2456a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2457a07b2000SAl Viro if (unlikely(error)) { 2458a07b2000SAl Viro file = ERR_PTR(error); 2459a07b2000SAl Viro } else { 2460a07b2000SAl Viro if (detached) 2461a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2462a07b2000SAl Viro else 2463a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2464a07b2000SAl Viro path_put(&path); 2465a07b2000SAl Viro } 2466a07b2000SAl Viro if (IS_ERR(file)) { 2467a07b2000SAl Viro put_unused_fd(fd); 2468a07b2000SAl Viro return PTR_ERR(file); 2469a07b2000SAl Viro } 2470a07b2000SAl Viro fd_install(fd, file); 2471a07b2000SAl Viro return fd; 2472a07b2000SAl Viro } 2473a07b2000SAl Viro 247443f5e655SDavid Howells /* 247543f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 247643f5e655SDavid Howells * 247743f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 247843f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 247943f5e655SDavid Howells */ 248043f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 24812e4b7fcdSDave Hansen { 248243f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 24832e4b7fcdSDave Hansen 248443f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 248543f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 248643f5e655SDavid Howells return false; 248743f5e655SDavid Howells 248843f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 248943f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 249043f5e655SDavid Howells return false; 249143f5e655SDavid Howells 249243f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 249343f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 249443f5e655SDavid Howells return false; 249543f5e655SDavid Howells 249643f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 249743f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 249843f5e655SDavid Howells return false; 249943f5e655SDavid Howells 250043f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 250143f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 250243f5e655SDavid Howells return false; 250343f5e655SDavid Howells 250443f5e655SDavid Howells return true; 250543f5e655SDavid Howells } 250643f5e655SDavid Howells 250743f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 250843f5e655SDavid Howells { 250943f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 251043f5e655SDavid Howells 251143f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 25122e4b7fcdSDave Hansen return 0; 25132e4b7fcdSDave Hansen 25142e4b7fcdSDave Hansen if (readonly_request) 251543f5e655SDavid Howells return mnt_make_readonly(mnt); 251643f5e655SDavid Howells 251743f5e655SDavid Howells return __mnt_unmake_readonly(mnt); 251843f5e655SDavid Howells } 251943f5e655SDavid Howells 252043f5e655SDavid Howells /* 252143f5e655SDavid Howells * Update the user-settable attributes on a mount. The caller must hold 252243f5e655SDavid Howells * sb->s_umount for writing. 252343f5e655SDavid Howells */ 252443f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 252543f5e655SDavid Howells { 252643f5e655SDavid Howells lock_mount_hash(); 252743f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 252843f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 252943f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 253043f5e655SDavid Howells unlock_mount_hash(); 253143f5e655SDavid Howells } 253243f5e655SDavid Howells 2533f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2534f8b92ba6SDeepa Dinamani { 2535f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2536f8b92ba6SDeepa Dinamani 2537f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2538f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2539f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2540f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2541f8b92ba6SDeepa Dinamani struct tm tm; 2542f8b92ba6SDeepa Dinamani 2543f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2544f8b92ba6SDeepa Dinamani 25450ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 25460ecee669SEric Biggers sb->s_type->name, 25470ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 25480ecee669SEric Biggers mntpath, 2549f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2550f8b92ba6SDeepa Dinamani 2551f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2552f8b92ba6SDeepa Dinamani } 2553f8b92ba6SDeepa Dinamani } 2554f8b92ba6SDeepa Dinamani 255543f5e655SDavid Howells /* 255643f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 255743f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 255843f5e655SDavid Howells * to mount(2). 255943f5e655SDavid Howells */ 256043f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 256143f5e655SDavid Howells { 256243f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 256343f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 256443f5e655SDavid Howells int ret; 256543f5e655SDavid Howells 256643f5e655SDavid Howells if (!check_mnt(mnt)) 256743f5e655SDavid Howells return -EINVAL; 256843f5e655SDavid Howells 256943f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 257043f5e655SDavid Howells return -EINVAL; 257143f5e655SDavid Howells 257243f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 257343f5e655SDavid Howells return -EPERM; 257443f5e655SDavid Howells 257543f5e655SDavid Howells down_write(&sb->s_umount); 257643f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 257743f5e655SDavid Howells if (ret == 0) 257843f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 257943f5e655SDavid Howells up_write(&sb->s_umount); 2580f8b92ba6SDeepa Dinamani 2581f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2582f8b92ba6SDeepa Dinamani 258343f5e655SDavid Howells return ret; 25842e4b7fcdSDave Hansen } 25852e4b7fcdSDave Hansen 25861da177e4SLinus Torvalds /* 25871da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 25881da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 25891da177e4SLinus Torvalds * on it - tough luck. 25901da177e4SLinus Torvalds */ 2591e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2592e462ec50SDavid Howells int mnt_flags, void *data) 25931da177e4SLinus Torvalds { 25941da177e4SLinus Torvalds int err; 25952d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2596143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 25978d0347f6SDavid Howells struct fs_context *fc; 25981da177e4SLinus Torvalds 2599143c8c91SAl Viro if (!check_mnt(mnt)) 26001da177e4SLinus Torvalds return -EINVAL; 26011da177e4SLinus Torvalds 26022d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 26031da177e4SLinus Torvalds return -EINVAL; 26041da177e4SLinus Torvalds 260543f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 260607b64558SEric W. Biederman return -EPERM; 26079566d674SEric W. Biederman 26088d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 26098d0347f6SDavid Howells if (IS_ERR(fc)) 26108d0347f6SDavid Howells return PTR_ERR(fc); 2611ff36fe2cSEric Paris 2612b330966fSMiklos Szeredi fc->oldapi = true; 26138d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 26148d0347f6SDavid Howells if (!err) { 26151da177e4SLinus Torvalds down_write(&sb->s_umount); 261657eccb83SAl Viro err = -EPERM; 261743f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 26188d0347f6SDavid Howells err = reconfigure_super(fc); 261943f5e655SDavid Howells if (!err) 262043f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 26210e55a7ccSDan Williams } 26226339dab8SAl Viro up_write(&sb->s_umount); 26238d0347f6SDavid Howells } 2624f8b92ba6SDeepa Dinamani 2625f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2626f8b92ba6SDeepa Dinamani 26278d0347f6SDavid Howells put_fs_context(fc); 26281da177e4SLinus Torvalds return err; 26291da177e4SLinus Torvalds } 26301da177e4SLinus Torvalds 2631cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 26329676f0c6SRam Pai { 2633315fc83eSAl Viro struct mount *p; 2634909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2635fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 26369676f0c6SRam Pai return 1; 26379676f0c6SRam Pai } 26389676f0c6SRam Pai return 0; 26399676f0c6SRam Pai } 26409676f0c6SRam Pai 264144dfd84aSDavid Howells /* 264244dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 264344dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 264444dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 264544dfd84aSDavid Howells * cycles to be made. 264644dfd84aSDavid Howells */ 264744dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 264844dfd84aSDavid Howells { 264944dfd84aSDavid Howells struct mount *p; 265044dfd84aSDavid Howells bool ret = false; 265144dfd84aSDavid Howells 265244dfd84aSDavid Howells lock_mount_hash(); 265344dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 265444dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 265544dfd84aSDavid Howells goto out; 265644dfd84aSDavid Howells 265744dfd84aSDavid Howells ret = true; 265844dfd84aSDavid Howells out: 265944dfd84aSDavid Howells unlock_mount_hash(); 266044dfd84aSDavid Howells return ret; 266144dfd84aSDavid Howells } 266244dfd84aSDavid Howells 26632db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 26641da177e4SLinus Torvalds { 266544dfd84aSDavid Howells struct mnt_namespace *ns; 2666676da58dSAl Viro struct mount *p; 26670fb54e50SAl Viro struct mount *old; 26682763d119SAl Viro struct mount *parent; 26692763d119SAl Viro struct mountpoint *mp, *old_mp; 267057eccb83SAl Viro int err; 267144dfd84aSDavid Howells bool attached; 26721da177e4SLinus Torvalds 26732db154b3SDavid Howells mp = lock_mount(new_path); 267484d17192SAl Viro if (IS_ERR(mp)) 26752db154b3SDavid Howells return PTR_ERR(mp); 2676cc53ce53SDavid Howells 26772db154b3SDavid Howells old = real_mount(old_path->mnt); 26782db154b3SDavid Howells p = real_mount(new_path->mnt); 26792763d119SAl Viro parent = old->mnt_parent; 268044dfd84aSDavid Howells attached = mnt_has_parent(old); 26812763d119SAl Viro old_mp = old->mnt_mp; 268244dfd84aSDavid Howells ns = old->mnt_ns; 2683143c8c91SAl Viro 26841da177e4SLinus Torvalds err = -EINVAL; 268544dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 268644dfd84aSDavid Howells if (!check_mnt(p)) 26872db154b3SDavid Howells goto out; 26881da177e4SLinus Torvalds 2689570d7a98SEric Biggers /* The thing moved must be mounted... */ 2690570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 269144dfd84aSDavid Howells goto out; 269244dfd84aSDavid Howells 2693570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2694570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 26952db154b3SDavid Howells goto out; 26961da177e4SLinus Torvalds 26972db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 26982db154b3SDavid Howells goto out; 26992db154b3SDavid Howells 27002db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 27012db154b3SDavid Howells goto out; 27022db154b3SDavid Howells 27032db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 27042db154b3SDavid Howells d_is_dir(old_path->dentry)) 27052db154b3SDavid Howells goto out; 270621444403SRam Pai /* 270721444403SRam Pai * Don't move a mount residing in a shared parent. 270821444403SRam Pai */ 27092763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 27102db154b3SDavid Howells goto out; 27119676f0c6SRam Pai /* 27129676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 27139676f0c6SRam Pai * mount which is shared. 27149676f0c6SRam Pai */ 2715fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 27162db154b3SDavid Howells goto out; 27171da177e4SLinus Torvalds err = -ELOOP; 271844dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 271944dfd84aSDavid Howells goto out; 2720fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2721676da58dSAl Viro if (p == old) 27222db154b3SDavid Howells goto out; 27231da177e4SLinus Torvalds 27242db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 27252763d119SAl Viro attached); 27264ac91378SJan Blunck if (err) 27272db154b3SDavid Howells goto out; 27281da177e4SLinus Torvalds 27291da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 27301da177e4SLinus Torvalds * automatically */ 27316776db3dSAl Viro list_del_init(&old->mnt_expire); 27322763d119SAl Viro if (attached) 27332763d119SAl Viro put_mountpoint(old_mp); 27341da177e4SLinus Torvalds out: 27352db154b3SDavid Howells unlock_mount(mp); 273644dfd84aSDavid Howells if (!err) { 27372763d119SAl Viro if (attached) 27382763d119SAl Viro mntput_no_expire(parent); 27392763d119SAl Viro else 274044dfd84aSDavid Howells free_mnt_ns(ns); 274144dfd84aSDavid Howells } 27422db154b3SDavid Howells return err; 27432db154b3SDavid Howells } 27442db154b3SDavid Howells 27452db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 27462db154b3SDavid Howells { 27472db154b3SDavid Howells struct path old_path; 27482db154b3SDavid Howells int err; 27492db154b3SDavid Howells 27502db154b3SDavid Howells if (!old_name || !*old_name) 27512db154b3SDavid Howells return -EINVAL; 27522db154b3SDavid Howells 27532db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 27542db154b3SDavid Howells if (err) 27552db154b3SDavid Howells return err; 27562db154b3SDavid Howells 27572db154b3SDavid Howells err = do_move_mount(&old_path, path); 27582d92ab3cSAl Viro path_put(&old_path); 27591da177e4SLinus Torvalds return err; 27601da177e4SLinus Torvalds } 27611da177e4SLinus Torvalds 27629d412a43SAl Viro /* 27639d412a43SAl Viro * add a mount into a namespace's mount tree 27649d412a43SAl Viro */ 27658f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 27668f11538eSAl Viro struct path *path, int mnt_flags) 27679d412a43SAl Viro { 27688f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 27699d412a43SAl Viro 2770f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 27719d412a43SAl Viro 277284d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2773156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2774156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 27758f11538eSAl Viro return -EINVAL; 2776156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 277784d17192SAl Viro if (!parent->mnt_ns) 27788f11538eSAl Viro return -EINVAL; 2779156cacb1SAl Viro } 27809d412a43SAl Viro 27819d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 278295bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 27839d412a43SAl Viro path->mnt->mnt_root == path->dentry) 27848f11538eSAl Viro return -EBUSY; 27859d412a43SAl Viro 2786e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 27878f11538eSAl Viro return -EINVAL; 27889d412a43SAl Viro 278995bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 27908f11538eSAl Viro return graft_tree(newmnt, parent, mp); 27919d412a43SAl Viro } 2792b1e75df4SAl Viro 2793132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2794132e4608SDavid Howells 2795132e4608SDavid Howells /* 2796132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2797132e4608SDavid Howells * be added to the namespace tree. 2798132e4608SDavid Howells */ 2799132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2800132e4608SDavid Howells unsigned int mnt_flags) 2801132e4608SDavid Howells { 2802132e4608SDavid Howells struct vfsmount *mnt; 28038f11538eSAl Viro struct mountpoint *mp; 2804132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2805132e4608SDavid Howells int error; 2806132e4608SDavid Howells 2807c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2808c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2809c9ce29edSAl Viro error = -EPERM; 2810c9ce29edSAl Viro 2811c9ce29edSAl Viro if (unlikely(error)) { 2812c9ce29edSAl Viro fc_drop_locked(fc); 2813c9ce29edSAl Viro return error; 2814132e4608SDavid Howells } 2815132e4608SDavid Howells 2816132e4608SDavid Howells up_write(&sb->s_umount); 2817132e4608SDavid Howells 2818132e4608SDavid Howells mnt = vfs_create_mount(fc); 2819132e4608SDavid Howells if (IS_ERR(mnt)) 2820132e4608SDavid Howells return PTR_ERR(mnt); 2821132e4608SDavid Howells 2822f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 2823f8b92ba6SDeepa Dinamani 28248f11538eSAl Viro mp = lock_mount(mountpoint); 28258f11538eSAl Viro if (IS_ERR(mp)) { 28268f11538eSAl Viro mntput(mnt); 28278f11538eSAl Viro return PTR_ERR(mp); 28288f11538eSAl Viro } 28298f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 28308f11538eSAl Viro unlock_mount(mp); 28310ecee669SEric Biggers if (error < 0) 28320ecee669SEric Biggers mntput(mnt); 2833f8b92ba6SDeepa Dinamani return error; 2834f8b92ba6SDeepa Dinamani } 2835f8b92ba6SDeepa Dinamani 28361da177e4SLinus Torvalds /* 28371da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 28381da177e4SLinus Torvalds * namespace's tree 28391da177e4SLinus Torvalds */ 2840e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2841808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 28421da177e4SLinus Torvalds { 28430c55cfc4SEric W. Biederman struct file_system_type *type; 2844a0c9a8b8SAl Viro struct fs_context *fc; 2845a0c9a8b8SAl Viro const char *subtype = NULL; 2846a0c9a8b8SAl Viro int err = 0; 28471da177e4SLinus Torvalds 28480c55cfc4SEric W. Biederman if (!fstype) 28491da177e4SLinus Torvalds return -EINVAL; 28501da177e4SLinus Torvalds 28510c55cfc4SEric W. Biederman type = get_fs_type(fstype); 28520c55cfc4SEric W. Biederman if (!type) 28530c55cfc4SEric W. Biederman return -ENODEV; 28540c55cfc4SEric W. Biederman 2855a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 2856a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 2857a0c9a8b8SAl Viro if (subtype) { 2858a0c9a8b8SAl Viro subtype++; 2859a0c9a8b8SAl Viro if (!*subtype) { 28600c55cfc4SEric W. Biederman put_filesystem(type); 2861a0c9a8b8SAl Viro return -EINVAL; 2862a0c9a8b8SAl Viro } 2863a0c9a8b8SAl Viro } 28648654df4eSEric W. Biederman } 28658654df4eSEric W. Biederman 2866a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 2867a0c9a8b8SAl Viro put_filesystem(type); 2868a0c9a8b8SAl Viro if (IS_ERR(fc)) 2869a0c9a8b8SAl Viro return PTR_ERR(fc); 2870a0c9a8b8SAl Viro 28713e1aeb00SDavid Howells if (subtype) 28723e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 28733e1aeb00SDavid Howells subtype, strlen(subtype)); 28743e1aeb00SDavid Howells if (!err && name) 28753e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 2876a0c9a8b8SAl Viro if (!err) 2877a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 2878c3aabf07SAl Viro if (!err && !mount_capable(fc)) 2879c3aabf07SAl Viro err = -EPERM; 2880a0c9a8b8SAl Viro if (!err) 2881a0c9a8b8SAl Viro err = vfs_get_tree(fc); 2882132e4608SDavid Howells if (!err) 2883132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 2884a0c9a8b8SAl Viro 2885a0c9a8b8SAl Viro put_fs_context(fc); 288615f9a3f3SAl Viro return err; 28871da177e4SLinus Torvalds } 28881da177e4SLinus Torvalds 288919a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 289019a167afSAl Viro { 289126df6034SAl Viro struct dentry *dentry = path->dentry; 28928f11538eSAl Viro struct mountpoint *mp; 289325e195aaSAl Viro struct mount *mnt; 289419a167afSAl Viro int err; 289525e195aaSAl Viro 289625e195aaSAl Viro if (!m) 289725e195aaSAl Viro return 0; 289825e195aaSAl Viro if (IS_ERR(m)) 289925e195aaSAl Viro return PTR_ERR(m); 290025e195aaSAl Viro 290125e195aaSAl Viro mnt = real_mount(m); 290219a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 290319a167afSAl Viro * expired before we get a chance to add it 290419a167afSAl Viro */ 29056776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 290619a167afSAl Viro 290719a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 290826df6034SAl Viro m->mnt_root == dentry) { 2909b1e75df4SAl Viro err = -ELOOP; 291026df6034SAl Viro goto discard; 291119a167afSAl Viro } 291219a167afSAl Viro 291326df6034SAl Viro /* 291426df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 291526df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 291626df6034SAl Viro * got", not "try to mount it on top". 291726df6034SAl Viro */ 291826df6034SAl Viro inode_lock(dentry->d_inode); 291926df6034SAl Viro namespace_lock(); 292026df6034SAl Viro if (unlikely(cant_mount(dentry))) { 292126df6034SAl Viro err = -ENOENT; 292226df6034SAl Viro goto discard_locked; 292326df6034SAl Viro } 292426df6034SAl Viro rcu_read_lock(); 292526df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 292626df6034SAl Viro rcu_read_unlock(); 292726df6034SAl Viro err = 0; 292826df6034SAl Viro goto discard_locked; 292926df6034SAl Viro } 293026df6034SAl Viro rcu_read_unlock(); 293126df6034SAl Viro mp = get_mountpoint(dentry); 29328f11538eSAl Viro if (IS_ERR(mp)) { 29338f11538eSAl Viro err = PTR_ERR(mp); 293426df6034SAl Viro goto discard_locked; 29358f11538eSAl Viro } 293626df6034SAl Viro 29378f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 29388f11538eSAl Viro unlock_mount(mp); 293926df6034SAl Viro if (unlikely(err)) 294026df6034SAl Viro goto discard; 294126df6034SAl Viro mntput(m); 2942b1e75df4SAl Viro return 0; 294326df6034SAl Viro 294426df6034SAl Viro discard_locked: 294526df6034SAl Viro namespace_unlock(); 294626df6034SAl Viro inode_unlock(dentry->d_inode); 294726df6034SAl Viro discard: 2948b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 29496776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 295097216be0SAl Viro namespace_lock(); 29516776db3dSAl Viro list_del_init(&mnt->mnt_expire); 295297216be0SAl Viro namespace_unlock(); 295319a167afSAl Viro } 2954b1e75df4SAl Viro mntput(m); 2955b1e75df4SAl Viro mntput(m); 295619a167afSAl Viro return err; 295719a167afSAl Viro } 295819a167afSAl Viro 2959ea5b778aSDavid Howells /** 2960ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2961ea5b778aSDavid Howells * @mnt: The mount to list. 2962ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2963ea5b778aSDavid Howells */ 2964ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2965ea5b778aSDavid Howells { 296697216be0SAl Viro namespace_lock(); 2967ea5b778aSDavid Howells 29686776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2969ea5b778aSDavid Howells 297097216be0SAl Viro namespace_unlock(); 2971ea5b778aSDavid Howells } 2972ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2973ea5b778aSDavid Howells 2974ea5b778aSDavid Howells /* 29751da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 29761da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 29771da177e4SLinus Torvalds * here 29781da177e4SLinus Torvalds */ 29791da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 29801da177e4SLinus Torvalds { 2981761d5c38SAl Viro struct mount *mnt, *next; 29821da177e4SLinus Torvalds LIST_HEAD(graveyard); 29831da177e4SLinus Torvalds 29841da177e4SLinus Torvalds if (list_empty(mounts)) 29851da177e4SLinus Torvalds return; 29861da177e4SLinus Torvalds 298797216be0SAl Viro namespace_lock(); 2988719ea2fbSAl Viro lock_mount_hash(); 29891da177e4SLinus Torvalds 29901da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 29911da177e4SLinus Torvalds * following criteria: 29921da177e4SLinus Torvalds * - only referenced by its parent vfsmount 29931da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 29941da177e4SLinus Torvalds * cleared by mntput()) 29951da177e4SLinus Torvalds */ 29966776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2997863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 29981ab59738SAl Viro propagate_mount_busy(mnt, 1)) 29991da177e4SLinus Torvalds continue; 30006776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 30011da177e4SLinus Torvalds } 3002bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 30036776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 3004143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 3005e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3006bcc5c7d2SAl Viro } 3007719ea2fbSAl Viro unlock_mount_hash(); 30083ab6abeeSAl Viro namespace_unlock(); 30091da177e4SLinus Torvalds } 30101da177e4SLinus Torvalds 30111da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 30121da177e4SLinus Torvalds 30131da177e4SLinus Torvalds /* 30145528f911STrond Myklebust * Ripoff of 'select_parent()' 30155528f911STrond Myklebust * 30165528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 30175528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 30185528f911STrond Myklebust */ 3019692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 30205528f911STrond Myklebust { 3021692afc31SAl Viro struct mount *this_parent = parent; 30225528f911STrond Myklebust struct list_head *next; 30235528f911STrond Myklebust int found = 0; 30245528f911STrond Myklebust 30255528f911STrond Myklebust repeat: 30266b41d536SAl Viro next = this_parent->mnt_mounts.next; 30275528f911STrond Myklebust resume: 30286b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 30295528f911STrond Myklebust struct list_head *tmp = next; 30306b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 30315528f911STrond Myklebust 30325528f911STrond Myklebust next = tmp->next; 3033692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 30345528f911STrond Myklebust continue; 30355528f911STrond Myklebust /* 30365528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 30375528f911STrond Myklebust */ 30386b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 30395528f911STrond Myklebust this_parent = mnt; 30405528f911STrond Myklebust goto repeat; 30415528f911STrond Myklebust } 30425528f911STrond Myklebust 30431ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 30446776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 30455528f911STrond Myklebust found++; 30465528f911STrond Myklebust } 30475528f911STrond Myklebust } 30485528f911STrond Myklebust /* 30495528f911STrond Myklebust * All done at this level ... ascend and resume the search 30505528f911STrond Myklebust */ 30515528f911STrond Myklebust if (this_parent != parent) { 30526b41d536SAl Viro next = this_parent->mnt_child.next; 30530714a533SAl Viro this_parent = this_parent->mnt_parent; 30545528f911STrond Myklebust goto resume; 30555528f911STrond Myklebust } 30565528f911STrond Myklebust return found; 30575528f911STrond Myklebust } 30585528f911STrond Myklebust 30595528f911STrond Myklebust /* 30605528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 30615528f911STrond Myklebust * submounts of a specific parent mountpoint 306299b7db7bSNick Piggin * 306348a066e7SAl Viro * mount_lock must be held for write 30645528f911STrond Myklebust */ 3065b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 30665528f911STrond Myklebust { 30675528f911STrond Myklebust LIST_HEAD(graveyard); 3068761d5c38SAl Viro struct mount *m; 30695528f911STrond Myklebust 30705528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 3071c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3072bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3073761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 30746776db3dSAl Viro mnt_expire); 3075143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3076e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3077bcc5c7d2SAl Viro } 3078bcc5c7d2SAl Viro } 30795528f911STrond Myklebust } 30805528f911STrond Myklebust 3081028abd92SChristoph Hellwig static void *copy_mount_options(const void __user * data) 30821da177e4SLinus Torvalds { 3083b40ef869SAl Viro char *copy; 3084d563d678SCatalin Marinas unsigned left, offset; 30851da177e4SLinus Torvalds 30861da177e4SLinus Torvalds if (!data) 3087b40ef869SAl Viro return NULL; 30881da177e4SLinus Torvalds 3089b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3090b40ef869SAl Viro if (!copy) 3091b40ef869SAl Viro return ERR_PTR(-ENOMEM); 30921da177e4SLinus Torvalds 3093d563d678SCatalin Marinas left = copy_from_user(copy, data, PAGE_SIZE); 30941da177e4SLinus Torvalds 3095d563d678SCatalin Marinas /* 3096d563d678SCatalin Marinas * Not all architectures have an exact copy_from_user(). Resort to 3097d563d678SCatalin Marinas * byte at a time. 3098d563d678SCatalin Marinas */ 3099d563d678SCatalin Marinas offset = PAGE_SIZE - left; 3100d563d678SCatalin Marinas while (left) { 3101d563d678SCatalin Marinas char c; 3102d563d678SCatalin Marinas if (get_user(c, (const char __user *)data + offset)) 3103d563d678SCatalin Marinas break; 3104d563d678SCatalin Marinas copy[offset] = c; 3105d563d678SCatalin Marinas left--; 3106d563d678SCatalin Marinas offset++; 3107d563d678SCatalin Marinas } 3108d563d678SCatalin Marinas 3109d563d678SCatalin Marinas if (left == PAGE_SIZE) { 3110b40ef869SAl Viro kfree(copy); 3111b40ef869SAl Viro return ERR_PTR(-EFAULT); 31121da177e4SLinus Torvalds } 3113d563d678SCatalin Marinas 3114b40ef869SAl Viro return copy; 31151da177e4SLinus Torvalds } 31161da177e4SLinus Torvalds 3117028abd92SChristoph Hellwig static char *copy_mount_string(const void __user *data) 3118eca6f534SVegard Nossum { 3119fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3120eca6f534SVegard Nossum } 3121eca6f534SVegard Nossum 31221da177e4SLinus Torvalds /* 31231da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 31241da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 31251da177e4SLinus Torvalds * 31261da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 31271da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 31281da177e4SLinus Torvalds * information (or be NULL). 31291da177e4SLinus Torvalds * 31301da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 31311da177e4SLinus Torvalds * When the flags word was introduced its top half was required 31321da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 31331da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 31341da177e4SLinus Torvalds * and must be discarded. 31351da177e4SLinus Torvalds */ 3136c60166f0SChristoph Hellwig int path_mount(const char *dev_name, struct path *path, 3137808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 31381da177e4SLinus Torvalds { 3139e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 3140a1e6aaa3SChristoph Hellwig int ret; 31411da177e4SLinus Torvalds 31421da177e4SLinus Torvalds /* Discard magic */ 31431da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 31441da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 31451da177e4SLinus Torvalds 31461da177e4SLinus Torvalds /* Basic sanity checks */ 31471da177e4SLinus Torvalds if (data_page) 31481da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 31491da177e4SLinus Torvalds 3150e462ec50SDavid Howells if (flags & MS_NOUSER) 3151e462ec50SDavid Howells return -EINVAL; 3152e462ec50SDavid Howells 3153a1e6aaa3SChristoph Hellwig ret = security_sb_mount(dev_name, path, type_page, flags, data_page); 3154a1e6aaa3SChristoph Hellwig if (ret) 3155a1e6aaa3SChristoph Hellwig return ret; 3156a1e6aaa3SChristoph Hellwig if (!may_mount()) 3157a1e6aaa3SChristoph Hellwig return -EPERM; 3158a1e6aaa3SChristoph Hellwig if ((flags & SB_MANDLOCK) && !may_mandlock()) 3159a1e6aaa3SChristoph Hellwig return -EPERM; 3160a27ab9f2STetsuo Handa 3161613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3162613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 31630a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 31640a1c01c9SMatthew Garrett 31651da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 31661da177e4SLinus Torvalds if (flags & MS_NOSUID) 31671da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 31681da177e4SLinus Torvalds if (flags & MS_NODEV) 31691da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 31701da177e4SLinus Torvalds if (flags & MS_NOEXEC) 31711da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3172fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3173fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3174fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3175fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3176d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3177d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3178a9e5b732SDavid Howells if (flags & MS_RDONLY) 31792e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3180dab741e0SMattias Nissler if (flags & MS_NOSYMFOLLOW) 3181dab741e0SMattias Nissler mnt_flags |= MNT_NOSYMFOLLOW; 3182fc33a7bbSChristoph Hellwig 3183ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3184ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3185ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3186ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3187ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3188a1e6aaa3SChristoph Hellwig mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; 3189ffbc6f0eSEric W. Biederman } 3190ffbc6f0eSEric W. Biederman 3191e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3192e462ec50SDavid Howells SB_SYNCHRONOUS | 3193e462ec50SDavid Howells SB_MANDLOCK | 3194e462ec50SDavid Howells SB_DIRSYNC | 3195e462ec50SDavid Howells SB_SILENT | 3196917086ffSMimi Zohar SB_POSIXACL | 3197d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3198917086ffSMimi Zohar SB_I_VERSION); 31991da177e4SLinus Torvalds 320043f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 3201a1e6aaa3SChristoph Hellwig return do_reconfigure_mnt(path, mnt_flags); 3202a1e6aaa3SChristoph Hellwig if (flags & MS_REMOUNT) 3203a1e6aaa3SChristoph Hellwig return do_remount(path, flags, sb_flags, mnt_flags, data_page); 3204a1e6aaa3SChristoph Hellwig if (flags & MS_BIND) 3205a1e6aaa3SChristoph Hellwig return do_loopback(path, dev_name, flags & MS_REC); 3206a1e6aaa3SChristoph Hellwig if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 3207a1e6aaa3SChristoph Hellwig return do_change_type(path, flags); 3208a1e6aaa3SChristoph Hellwig if (flags & MS_MOVE) 3209a1e6aaa3SChristoph Hellwig return do_move_mount_old(path, dev_name); 3210a1e6aaa3SChristoph Hellwig 3211a1e6aaa3SChristoph Hellwig return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, 32121da177e4SLinus Torvalds data_page); 3213a1e6aaa3SChristoph Hellwig } 3214a1e6aaa3SChristoph Hellwig 3215a1e6aaa3SChristoph Hellwig long do_mount(const char *dev_name, const char __user *dir_name, 3216a1e6aaa3SChristoph Hellwig const char *type_page, unsigned long flags, void *data_page) 3217a1e6aaa3SChristoph Hellwig { 3218a1e6aaa3SChristoph Hellwig struct path path; 3219a1e6aaa3SChristoph Hellwig int ret; 3220a1e6aaa3SChristoph Hellwig 3221a1e6aaa3SChristoph Hellwig ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3222a1e6aaa3SChristoph Hellwig if (ret) 3223a1e6aaa3SChristoph Hellwig return ret; 3224a1e6aaa3SChristoph Hellwig ret = path_mount(dev_name, &path, type_page, flags, data_page); 32252d92ab3cSAl Viro path_put(&path); 3226a1e6aaa3SChristoph Hellwig return ret; 32271da177e4SLinus Torvalds } 32281da177e4SLinus Torvalds 3229537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3230537f7ccbSEric W. Biederman { 3231537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3232537f7ccbSEric W. Biederman } 3233537f7ccbSEric W. Biederman 3234537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3235537f7ccbSEric W. Biederman { 3236537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3237537f7ccbSEric W. Biederman } 3238537f7ccbSEric W. Biederman 3239771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3240771b1371SEric W. Biederman { 324174e83122SAl Viro if (!is_anon_ns(ns)) 32426344c433SAl Viro ns_free_inum(&ns->ns); 3243537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3244771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3245771b1371SEric W. Biederman kfree(ns); 3246771b1371SEric W. Biederman } 3247771b1371SEric W. Biederman 32488823c079SEric W. Biederman /* 32498823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 32508823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 32518823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 32528823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 32538823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 32548823c079SEric W. Biederman */ 32558823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 32568823c079SEric W. Biederman 325774e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3258cf8d2c11STrond Myklebust { 3259cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3260537f7ccbSEric W. Biederman struct ucounts *ucounts; 326198f842e6SEric W. Biederman int ret; 3262cf8d2c11STrond Myklebust 3263537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3264537f7ccbSEric W. Biederman if (!ucounts) 3265df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3266537f7ccbSEric W. Biederman 326774e83122SAl Viro new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 3268537f7ccbSEric W. Biederman if (!new_ns) { 3269537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3270cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3271537f7ccbSEric W. Biederman } 327274e83122SAl Viro if (!anon) { 32736344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 327498f842e6SEric W. Biederman if (ret) { 327598f842e6SEric W. Biederman kfree(new_ns); 3276537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 327798f842e6SEric W. Biederman return ERR_PTR(ret); 327898f842e6SEric W. Biederman } 327974e83122SAl Viro } 328033c42940SAl Viro new_ns->ns.ops = &mntns_operations; 328174e83122SAl Viro if (!anon) 32828823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 32831a7b8969SKirill Tkhai refcount_set(&new_ns->ns.count, 1); 3284cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3285cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 32869f6c61f9SMiklos Szeredi spin_lock_init(&new_ns->ns_lock); 3287771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3288537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3289cf8d2c11STrond Myklebust return new_ns; 3290cf8d2c11STrond Myklebust } 3291cf8d2c11STrond Myklebust 32920766f788SEmese Revfy __latent_entropy 32939559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 32949559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 32951da177e4SLinus Torvalds { 32966b3286edSKirill Korotaev struct mnt_namespace *new_ns; 32977f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3298315fc83eSAl Viro struct mount *p, *q; 32999559f689SAl Viro struct mount *old; 3300cb338d06SAl Viro struct mount *new; 33017a472ef4SEric W. Biederman int copy_flags; 33021da177e4SLinus Torvalds 33039559f689SAl Viro BUG_ON(!ns); 33049559f689SAl Viro 33059559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 33069559f689SAl Viro get_mnt_ns(ns); 33079559f689SAl Viro return ns; 33089559f689SAl Viro } 33099559f689SAl Viro 33109559f689SAl Viro old = ns->root; 33119559f689SAl Viro 331274e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3313cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3314cf8d2c11STrond Myklebust return new_ns; 33151da177e4SLinus Torvalds 331697216be0SAl Viro namespace_lock(); 33171da177e4SLinus Torvalds /* First pass: copy the tree topology */ 33184ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 33199559f689SAl Viro if (user_ns != ns->user_ns) 33203bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 33217a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3322be34d1a3SDavid Howells if (IS_ERR(new)) { 3323328e6d90SAl Viro namespace_unlock(); 3324771b1371SEric W. Biederman free_mnt_ns(new_ns); 3325be34d1a3SDavid Howells return ERR_CAST(new); 33261da177e4SLinus Torvalds } 33273bd045ccSAl Viro if (user_ns != ns->user_ns) { 33283bd045ccSAl Viro lock_mount_hash(); 33293bd045ccSAl Viro lock_mnt_tree(new); 33303bd045ccSAl Viro unlock_mount_hash(); 33313bd045ccSAl Viro } 3332be08d6d2SAl Viro new_ns->root = new; 33331a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 33341da177e4SLinus Torvalds 33351da177e4SLinus Torvalds /* 33361da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 33371da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 33381da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 33391da177e4SLinus Torvalds */ 3340909b0a88SAl Viro p = old; 3341cb338d06SAl Viro q = new; 33421da177e4SLinus Torvalds while (p) { 3343143c8c91SAl Viro q->mnt_ns = new_ns; 3344d2921684SEric W. Biederman new_ns->mounts++; 33459559f689SAl Viro if (new_fs) { 33469559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 33479559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3348315fc83eSAl Viro rootmnt = &p->mnt; 33491da177e4SLinus Torvalds } 33509559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 33519559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3352315fc83eSAl Viro pwdmnt = &p->mnt; 33531da177e4SLinus Torvalds } 33541da177e4SLinus Torvalds } 3355909b0a88SAl Viro p = next_mnt(p, old); 3356909b0a88SAl Viro q = next_mnt(q, new); 33574ce5d2b1SEric W. Biederman if (!q) 33584ce5d2b1SEric W. Biederman break; 33594ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 33604ce5d2b1SEric W. Biederman p = next_mnt(p, old); 33611da177e4SLinus Torvalds } 3362328e6d90SAl Viro namespace_unlock(); 33631da177e4SLinus Torvalds 33641da177e4SLinus Torvalds if (rootmnt) 3365f03c6599SAl Viro mntput(rootmnt); 33661da177e4SLinus Torvalds if (pwdmnt) 3367f03c6599SAl Viro mntput(pwdmnt); 33681da177e4SLinus Torvalds 3369741a2951SJANAK DESAI return new_ns; 3370741a2951SJANAK DESAI } 3371741a2951SJANAK DESAI 337274e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3373cf8d2c11STrond Myklebust { 33741a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3375ea441d11SAl Viro struct mnt_namespace *ns; 3376d31da0f0SAl Viro struct super_block *s; 3377ea441d11SAl Viro struct path path; 3378ea441d11SAl Viro int err; 3379ea441d11SAl Viro 338074e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 338174e83122SAl Viro if (IS_ERR(ns)) { 338274e83122SAl Viro mntput(m); 3383ea441d11SAl Viro return ERR_CAST(ns); 338474e83122SAl Viro } 338574e83122SAl Viro mnt->mnt_ns = ns; 338674e83122SAl Viro ns->root = mnt; 338774e83122SAl Viro ns->mounts++; 338874e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3389ea441d11SAl Viro 339074e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3391ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3392ea441d11SAl Viro 3393ea441d11SAl Viro put_mnt_ns(ns); 3394ea441d11SAl Viro 3395ea441d11SAl Viro if (err) 3396ea441d11SAl Viro return ERR_PTR(err); 3397ea441d11SAl Viro 3398ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3399d31da0f0SAl Viro s = path.mnt->mnt_sb; 3400d31da0f0SAl Viro atomic_inc(&s->s_active); 3401ea441d11SAl Viro mntput(path.mnt); 3402ea441d11SAl Viro /* lock the sucker */ 3403d31da0f0SAl Viro down_write(&s->s_umount); 3404ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3405ea441d11SAl Viro return path.dentry; 3406ea441d11SAl Viro } 3407ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3408ea441d11SAl Viro 3409cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3410cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 34111da177e4SLinus Torvalds { 3412eca6f534SVegard Nossum int ret; 3413eca6f534SVegard Nossum char *kernel_type; 3414eca6f534SVegard Nossum char *kernel_dev; 3415b40ef869SAl Viro void *options; 34161da177e4SLinus Torvalds 3417b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3418b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3419b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3420eca6f534SVegard Nossum goto out_type; 34211da177e4SLinus Torvalds 3422b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3423b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3424b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3425eca6f534SVegard Nossum goto out_dev; 34261da177e4SLinus Torvalds 3427b40ef869SAl Viro options = copy_mount_options(data); 3428b40ef869SAl Viro ret = PTR_ERR(options); 3429b40ef869SAl Viro if (IS_ERR(options)) 3430eca6f534SVegard Nossum goto out_data; 34311da177e4SLinus Torvalds 3432b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3433eca6f534SVegard Nossum 3434b40ef869SAl Viro kfree(options); 3435eca6f534SVegard Nossum out_data: 3436eca6f534SVegard Nossum kfree(kernel_dev); 3437eca6f534SVegard Nossum out_dev: 3438eca6f534SVegard Nossum kfree(kernel_type); 3439eca6f534SVegard Nossum out_type: 3440eca6f534SVegard Nossum return ret; 34411da177e4SLinus Torvalds } 34421da177e4SLinus Torvalds 34431da177e4SLinus Torvalds /* 344493766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 344593766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 344693766fbdSDavid Howells */ 344793766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 344893766fbdSDavid Howells unsigned int, attr_flags) 344993766fbdSDavid Howells { 345093766fbdSDavid Howells struct mnt_namespace *ns; 345193766fbdSDavid Howells struct fs_context *fc; 345293766fbdSDavid Howells struct file *file; 345393766fbdSDavid Howells struct path newmount; 345493766fbdSDavid Howells struct mount *mnt; 345593766fbdSDavid Howells struct fd f; 345693766fbdSDavid Howells unsigned int mnt_flags = 0; 345793766fbdSDavid Howells long ret; 345893766fbdSDavid Howells 345993766fbdSDavid Howells if (!may_mount()) 346093766fbdSDavid Howells return -EPERM; 346193766fbdSDavid Howells 346293766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 346393766fbdSDavid Howells return -EINVAL; 346493766fbdSDavid Howells 346593766fbdSDavid Howells if (attr_flags & ~(MOUNT_ATTR_RDONLY | 346693766fbdSDavid Howells MOUNT_ATTR_NOSUID | 346793766fbdSDavid Howells MOUNT_ATTR_NODEV | 346893766fbdSDavid Howells MOUNT_ATTR_NOEXEC | 346993766fbdSDavid Howells MOUNT_ATTR__ATIME | 347093766fbdSDavid Howells MOUNT_ATTR_NODIRATIME)) 347193766fbdSDavid Howells return -EINVAL; 347293766fbdSDavid Howells 347393766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_RDONLY) 347493766fbdSDavid Howells mnt_flags |= MNT_READONLY; 347593766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NOSUID) 347693766fbdSDavid Howells mnt_flags |= MNT_NOSUID; 347793766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NODEV) 347893766fbdSDavid Howells mnt_flags |= MNT_NODEV; 347993766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NOEXEC) 348093766fbdSDavid Howells mnt_flags |= MNT_NOEXEC; 348193766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NODIRATIME) 348293766fbdSDavid Howells mnt_flags |= MNT_NODIRATIME; 348393766fbdSDavid Howells 348493766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 348593766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 348693766fbdSDavid Howells break; 348793766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 348893766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 348993766fbdSDavid Howells break; 349093766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 349193766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 349293766fbdSDavid Howells break; 349393766fbdSDavid Howells default: 349493766fbdSDavid Howells return -EINVAL; 349593766fbdSDavid Howells } 349693766fbdSDavid Howells 349793766fbdSDavid Howells f = fdget(fs_fd); 349893766fbdSDavid Howells if (!f.file) 349993766fbdSDavid Howells return -EBADF; 350093766fbdSDavid Howells 350193766fbdSDavid Howells ret = -EINVAL; 350293766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 350393766fbdSDavid Howells goto err_fsfd; 350493766fbdSDavid Howells 350593766fbdSDavid Howells fc = f.file->private_data; 350693766fbdSDavid Howells 350793766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 350893766fbdSDavid Howells if (ret < 0) 350993766fbdSDavid Howells goto err_fsfd; 351093766fbdSDavid Howells 351193766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 351293766fbdSDavid Howells ret = -EINVAL; 351393766fbdSDavid Howells if (!fc->root) 351493766fbdSDavid Howells goto err_unlock; 351593766fbdSDavid Howells 351693766fbdSDavid Howells ret = -EPERM; 351793766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 351893766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 351993766fbdSDavid Howells goto err_unlock; 352093766fbdSDavid Howells } 352193766fbdSDavid Howells 352293766fbdSDavid Howells ret = -EBUSY; 352393766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 352493766fbdSDavid Howells goto err_unlock; 352593766fbdSDavid Howells 352693766fbdSDavid Howells ret = -EPERM; 352793766fbdSDavid Howells if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock()) 352893766fbdSDavid Howells goto err_unlock; 352993766fbdSDavid Howells 353093766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 353193766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 353293766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 353393766fbdSDavid Howells goto err_unlock; 353493766fbdSDavid Howells } 353593766fbdSDavid Howells newmount.dentry = dget(fc->root); 353693766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 353793766fbdSDavid Howells 353893766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 353993766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 354093766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 354193766fbdSDavid Howells * don't want to have to handle any errors incurred. 354293766fbdSDavid Howells */ 354393766fbdSDavid Howells vfs_clean_context(fc); 354493766fbdSDavid Howells 354593766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 354693766fbdSDavid Howells if (IS_ERR(ns)) { 354793766fbdSDavid Howells ret = PTR_ERR(ns); 354893766fbdSDavid Howells goto err_path; 354993766fbdSDavid Howells } 355093766fbdSDavid Howells mnt = real_mount(newmount.mnt); 355193766fbdSDavid Howells mnt->mnt_ns = ns; 355293766fbdSDavid Howells ns->root = mnt; 355393766fbdSDavid Howells ns->mounts = 1; 355493766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 35551b0b9cc8SEric Biggers mntget(newmount.mnt); 355693766fbdSDavid Howells 355793766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 355893766fbdSDavid Howells * it, not just simply put it. 355993766fbdSDavid Howells */ 356093766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 356193766fbdSDavid Howells if (IS_ERR(file)) { 356293766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 356393766fbdSDavid Howells ret = PTR_ERR(file); 356493766fbdSDavid Howells goto err_path; 356593766fbdSDavid Howells } 356693766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 356793766fbdSDavid Howells 356893766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 356993766fbdSDavid Howells if (ret >= 0) 357093766fbdSDavid Howells fd_install(ret, file); 357193766fbdSDavid Howells else 357293766fbdSDavid Howells fput(file); 357393766fbdSDavid Howells 357493766fbdSDavid Howells err_path: 357593766fbdSDavid Howells path_put(&newmount); 357693766fbdSDavid Howells err_unlock: 357793766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 357893766fbdSDavid Howells err_fsfd: 357993766fbdSDavid Howells fdput(f); 358093766fbdSDavid Howells return ret; 358193766fbdSDavid Howells } 358293766fbdSDavid Howells 358393766fbdSDavid Howells /* 358493766fbdSDavid Howells * Move a mount from one place to another. In combination with 358593766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 358693766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 358793766fbdSDavid Howells * a mount subtree. 35882db154b3SDavid Howells * 35892db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 35902db154b3SDavid Howells */ 35912db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 35922658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 35932658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 35942db154b3SDavid Howells unsigned int, flags) 35952db154b3SDavid Howells { 35962db154b3SDavid Howells struct path from_path, to_path; 35972db154b3SDavid Howells unsigned int lflags; 35982db154b3SDavid Howells int ret = 0; 35992db154b3SDavid Howells 36002db154b3SDavid Howells if (!may_mount()) 36012db154b3SDavid Howells return -EPERM; 36022db154b3SDavid Howells 36032db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 36042db154b3SDavid Howells return -EINVAL; 36052db154b3SDavid Howells 36062db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 36072db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 36082db154b3SDavid Howells * to clear it afterwards. 36092db154b3SDavid Howells */ 36102db154b3SDavid Howells lflags = 0; 36112db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 36122db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 36132db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 36142db154b3SDavid Howells 36152db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 36162db154b3SDavid Howells if (ret < 0) 36172db154b3SDavid Howells return ret; 36182db154b3SDavid Howells 36192db154b3SDavid Howells lflags = 0; 36202db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 36212db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 36222db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 36232db154b3SDavid Howells 36242db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 36252db154b3SDavid Howells if (ret < 0) 36262db154b3SDavid Howells goto out_from; 36272db154b3SDavid Howells 36282db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 36292db154b3SDavid Howells if (ret < 0) 36302db154b3SDavid Howells goto out_to; 36312db154b3SDavid Howells 36322db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 36332db154b3SDavid Howells 36342db154b3SDavid Howells out_to: 36352db154b3SDavid Howells path_put(&to_path); 36362db154b3SDavid Howells out_from: 36372db154b3SDavid Howells path_put(&from_path); 36382db154b3SDavid Howells return ret; 36392db154b3SDavid Howells } 36402db154b3SDavid Howells 36412db154b3SDavid Howells /* 3642afac7cbaSAl Viro * Return true if path is reachable from root 3643afac7cbaSAl Viro * 364448a066e7SAl Viro * namespace_sem or mount_lock is held 3645afac7cbaSAl Viro */ 3646643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3647afac7cbaSAl Viro const struct path *root) 3648afac7cbaSAl Viro { 3649643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3650a73324daSAl Viro dentry = mnt->mnt_mountpoint; 36510714a533SAl Viro mnt = mnt->mnt_parent; 3652afac7cbaSAl Viro } 3653643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3654afac7cbaSAl Viro } 3655afac7cbaSAl Viro 3656640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3657afac7cbaSAl Viro { 365825ab4c9bSYaowei Bai bool res; 365948a066e7SAl Viro read_seqlock_excl(&mount_lock); 3660643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 366148a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3662afac7cbaSAl Viro return res; 3663afac7cbaSAl Viro } 3664afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3665afac7cbaSAl Viro 3666afac7cbaSAl Viro /* 36671da177e4SLinus Torvalds * pivot_root Semantics: 36681da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 36691da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 36701da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 36711da177e4SLinus Torvalds * 36721da177e4SLinus Torvalds * Restrictions: 36731da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 36741da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 36751da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 36761da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 36771da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 36781da177e4SLinus Torvalds * 36794a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 36800c1bc6b8SMauro Carvalho Chehab * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives 36814a0d11faSNeil Brown * in this situation. 36824a0d11faSNeil Brown * 36831da177e4SLinus Torvalds * Notes: 36841da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 36851da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 36861da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 36871da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 36881da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 36891da177e4SLinus Torvalds * first. 36901da177e4SLinus Torvalds */ 36913480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 36923480b257SHeiko Carstens const char __user *, put_old) 36931da177e4SLinus Torvalds { 36942763d119SAl Viro struct path new, old, root; 36952763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 369684d17192SAl Viro struct mountpoint *old_mp, *root_mp; 36971da177e4SLinus Torvalds int error; 36981da177e4SLinus Torvalds 36999b40bc90SAl Viro if (!may_mount()) 37001da177e4SLinus Torvalds return -EPERM; 37011da177e4SLinus Torvalds 3702ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3703ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 37041da177e4SLinus Torvalds if (error) 37051da177e4SLinus Torvalds goto out0; 37061da177e4SLinus Torvalds 3707ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 3708ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 37091da177e4SLinus Torvalds if (error) 37101da177e4SLinus Torvalds goto out1; 37111da177e4SLinus Torvalds 37122d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3713b12cea91SAl Viro if (error) 3714b12cea91SAl Viro goto out2; 37151da177e4SLinus Torvalds 3716f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 371784d17192SAl Viro old_mp = lock_mount(&old); 371884d17192SAl Viro error = PTR_ERR(old_mp); 371984d17192SAl Viro if (IS_ERR(old_mp)) 3720b12cea91SAl Viro goto out3; 3721b12cea91SAl Viro 37221da177e4SLinus Torvalds error = -EINVAL; 3723419148daSAl Viro new_mnt = real_mount(new.mnt); 3724419148daSAl Viro root_mnt = real_mount(root.mnt); 372584d17192SAl Viro old_mnt = real_mount(old.mnt); 37262763d119SAl Viro ex_parent = new_mnt->mnt_parent; 37272763d119SAl Viro root_parent = root_mnt->mnt_parent; 372884d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 37292763d119SAl Viro IS_MNT_SHARED(ex_parent) || 37302763d119SAl Viro IS_MNT_SHARED(root_parent)) 3731b12cea91SAl Viro goto out4; 3732143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3733b12cea91SAl Viro goto out4; 37345ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 37355ff9d8a6SEric W. Biederman goto out4; 37361da177e4SLinus Torvalds error = -ENOENT; 3737f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3738b12cea91SAl Viro goto out4; 37391da177e4SLinus Torvalds error = -EBUSY; 374084d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3741b12cea91SAl Viro goto out4; /* loop, on the same file system */ 37421da177e4SLinus Torvalds error = -EINVAL; 37438c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3744b12cea91SAl Viro goto out4; /* not a mountpoint */ 3745676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3746b12cea91SAl Viro goto out4; /* not attached */ 37472d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3748b12cea91SAl Viro goto out4; /* not a mountpoint */ 3749676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3750b12cea91SAl Viro goto out4; /* not attached */ 37514ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 375284d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3753b12cea91SAl Viro goto out4; 37540d082601SEric W. Biederman /* make certain new is below the root */ 37550d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 37560d082601SEric W. Biederman goto out4; 3757719ea2fbSAl Viro lock_mount_hash(); 37582763d119SAl Viro umount_mnt(new_mnt); 37592763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 37605ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 37615ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 37625ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 37635ff9d8a6SEric W. Biederman } 37644ac91378SJan Blunck /* mount old root on put_old */ 376584d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 37664ac91378SJan Blunck /* mount new_root on / */ 37672763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 37682763d119SAl Viro mnt_add_count(root_parent, -1); 37696b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 37704fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 37714fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 37723895dbf8SEric W. Biederman put_mountpoint(root_mp); 3773719ea2fbSAl Viro unlock_mount_hash(); 37742d8f3038SAl Viro chroot_fs_refs(&root, &new); 37751da177e4SLinus Torvalds error = 0; 3776b12cea91SAl Viro out4: 377784d17192SAl Viro unlock_mount(old_mp); 37782763d119SAl Viro if (!error) 37792763d119SAl Viro mntput_no_expire(ex_parent); 3780b12cea91SAl Viro out3: 37818c3ee42eSAl Viro path_put(&root); 3782b12cea91SAl Viro out2: 37832d8f3038SAl Viro path_put(&old); 37841da177e4SLinus Torvalds out1: 37852d8f3038SAl Viro path_put(&new); 37861da177e4SLinus Torvalds out0: 37871da177e4SLinus Torvalds return error; 37881da177e4SLinus Torvalds } 37891da177e4SLinus Torvalds 37901da177e4SLinus Torvalds static void __init init_mount_tree(void) 37911da177e4SLinus Torvalds { 37921da177e4SLinus Torvalds struct vfsmount *mnt; 379374e83122SAl Viro struct mount *m; 37946b3286edSKirill Korotaev struct mnt_namespace *ns; 3795ac748a09SJan Blunck struct path root; 37961da177e4SLinus Torvalds 3797fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 37981da177e4SLinus Torvalds if (IS_ERR(mnt)) 37991da177e4SLinus Torvalds panic("Can't create rootfs"); 3800b3e19d92SNick Piggin 380174e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 38023b22edc5STrond Myklebust if (IS_ERR(ns)) 38031da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 380474e83122SAl Viro m = real_mount(mnt); 380574e83122SAl Viro m->mnt_ns = ns; 380674e83122SAl Viro ns->root = m; 380774e83122SAl Viro ns->mounts = 1; 380874e83122SAl Viro list_add(&m->mnt_list, &ns->list); 38096b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 38106b3286edSKirill Korotaev get_mnt_ns(ns); 38111da177e4SLinus Torvalds 3812be08d6d2SAl Viro root.mnt = mnt; 3813be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3814da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3815ac748a09SJan Blunck 3816ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3817ac748a09SJan Blunck set_fs_root(current->fs, &root); 38181da177e4SLinus Torvalds } 38191da177e4SLinus Torvalds 382074bf17cfSDenis Cheng void __init mnt_init(void) 38211da177e4SLinus Torvalds { 382215a67dd8SRandy Dunlap int err; 38231da177e4SLinus Torvalds 38247d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 382520c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 38261da177e4SLinus Torvalds 38270818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 382838129a13SAl Viro sizeof(struct hlist_head), 38290818bf27SAl Viro mhash_entries, 19, 38303d375d78SPavel Tatashin HASH_ZERO, 38310818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 38320818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 38330818bf27SAl Viro sizeof(struct hlist_head), 38340818bf27SAl Viro mphash_entries, 19, 38353d375d78SPavel Tatashin HASH_ZERO, 38360818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 38371da177e4SLinus Torvalds 383884d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 38391da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 38401da177e4SLinus Torvalds 38414b93dc9bSTejun Heo kernfs_init(); 38424b93dc9bSTejun Heo 384315a67dd8SRandy Dunlap err = sysfs_init(); 384415a67dd8SRandy Dunlap if (err) 384515a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 38468e24eea7SHarvey Harrison __func__, err); 384700d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 384800d26666SGreg Kroah-Hartman if (!fs_kobj) 38498e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 3850037f11b4SAl Viro shmem_init(); 38511da177e4SLinus Torvalds init_rootfs(); 38521da177e4SLinus Torvalds init_mount_tree(); 38531da177e4SLinus Torvalds } 38541da177e4SLinus Torvalds 3855616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 38561da177e4SLinus Torvalds { 38571a7b8969SKirill Tkhai if (!refcount_dec_and_test(&ns->ns.count)) 3858616511d0STrond Myklebust return; 38597b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3860771b1371SEric W. Biederman free_mnt_ns(ns); 38611da177e4SLinus Torvalds } 38629d412a43SAl Viro 3863d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 38649d412a43SAl Viro { 3865423e0ab0STim Chen struct vfsmount *mnt; 3866d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 3867423e0ab0STim Chen if (!IS_ERR(mnt)) { 3868423e0ab0STim Chen /* 3869423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3870423e0ab0STim Chen * we unmount before file sys is unregistered 3871423e0ab0STim Chen */ 3872f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3873423e0ab0STim Chen } 3874423e0ab0STim Chen return mnt; 38759d412a43SAl Viro } 3876d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 3877423e0ab0STim Chen 3878423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3879423e0ab0STim Chen { 3880423e0ab0STim Chen /* release long term mount so mount point can be released */ 3881423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3882f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 388348a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3884423e0ab0STim Chen mntput(mnt); 3885423e0ab0STim Chen } 3886423e0ab0STim Chen } 3887423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 388802125a82SAl Viro 3889df820f8dSMiklos Szeredi void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) 3890df820f8dSMiklos Szeredi { 3891df820f8dSMiklos Szeredi unsigned int i; 3892df820f8dSMiklos Szeredi 3893df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 3894df820f8dSMiklos Szeredi if (mnt[i]) 3895df820f8dSMiklos Szeredi real_mount(mnt[i])->mnt_ns = NULL; 3896df820f8dSMiklos Szeredi synchronize_rcu_expedited(); 3897df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 3898df820f8dSMiklos Szeredi mntput(mnt[i]); 3899df820f8dSMiklos Szeredi } 3900df820f8dSMiklos Szeredi EXPORT_SYMBOL(kern_unmount_array); 3901df820f8dSMiklos Szeredi 390202125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 390302125a82SAl Viro { 3904143c8c91SAl Viro return check_mnt(real_mount(mnt)); 390502125a82SAl Viro } 39068823c079SEric W. Biederman 39073151527eSEric W. Biederman bool current_chrooted(void) 39083151527eSEric W. Biederman { 39093151527eSEric W. Biederman /* Does the current process have a non-standard root */ 39103151527eSEric W. Biederman struct path ns_root; 39113151527eSEric W. Biederman struct path fs_root; 39123151527eSEric W. Biederman bool chrooted; 39133151527eSEric W. Biederman 39143151527eSEric W. Biederman /* Find the namespace root */ 39153151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 39163151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 39173151527eSEric W. Biederman path_get(&ns_root); 39183151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 39193151527eSEric W. Biederman ; 39203151527eSEric W. Biederman 39213151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 39223151527eSEric W. Biederman 39233151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 39243151527eSEric W. Biederman 39253151527eSEric W. Biederman path_put(&fs_root); 39263151527eSEric W. Biederman path_put(&ns_root); 39273151527eSEric W. Biederman 39283151527eSEric W. Biederman return chrooted; 39293151527eSEric W. Biederman } 39303151527eSEric W. Biederman 3931132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 3932132e4608SDavid Howells const struct super_block *sb, 39338654df4eSEric W. Biederman int *new_mnt_flags) 393487a8ebd6SEric W. Biederman { 39358c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 393687a8ebd6SEric W. Biederman struct mount *mnt; 3937e51db735SEric W. Biederman bool visible = false; 393887a8ebd6SEric W. Biederman 393944bb4385SAl Viro down_read(&namespace_sem); 39409f6c61f9SMiklos Szeredi lock_ns_list(ns); 394187a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3942e51db735SEric W. Biederman struct mount *child; 394377b1a97dSEric W. Biederman int mnt_flags; 394477b1a97dSEric W. Biederman 39459f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 39469f6c61f9SMiklos Szeredi continue; 39479f6c61f9SMiklos Szeredi 3948132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 3949e51db735SEric W. Biederman continue; 3950e51db735SEric W. Biederman 39517e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 39527e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 39537e96c1b0SEric W. Biederman */ 39547e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 39557e96c1b0SEric W. Biederman continue; 39567e96c1b0SEric W. Biederman 3957a1935c17SEric W. Biederman /* A local view of the mount flags */ 395877b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 395977b1a97dSEric W. Biederman 3960695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3961bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 3962695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3963695e9df0SEric W. Biederman 39648c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 39658c6cf9ccSEric W. Biederman * than the proposed new mount. 39668c6cf9ccSEric W. Biederman */ 396777b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 39688c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 39698c6cf9ccSEric W. Biederman continue; 397077b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 397177b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 39728c6cf9ccSEric W. Biederman continue; 39738c6cf9ccSEric W. Biederman 3974ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3975ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3976ceeb0e5dSEric W. Biederman * empty directories. 3977e51db735SEric W. Biederman */ 3978e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3979e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3980ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3981d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3982ceeb0e5dSEric W. Biederman continue; 39837236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 39847236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3985e51db735SEric W. Biederman goto next; 398687a8ebd6SEric W. Biederman } 39878c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 398877b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 39898c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3990e51db735SEric W. Biederman visible = true; 3991e51db735SEric W. Biederman goto found; 3992e51db735SEric W. Biederman next: ; 399387a8ebd6SEric W. Biederman } 3994e51db735SEric W. Biederman found: 39959f6c61f9SMiklos Szeredi unlock_ns_list(ns); 399644bb4385SAl Viro up_read(&namespace_sem); 3997e51db735SEric W. Biederman return visible; 399887a8ebd6SEric W. Biederman } 399987a8ebd6SEric W. Biederman 4000132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 40018654df4eSEric W. Biederman { 4002a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 40038654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 40048654df4eSEric W. Biederman unsigned long s_iflags; 40058654df4eSEric W. Biederman 40068654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 40078654df4eSEric W. Biederman return false; 40088654df4eSEric W. Biederman 40098654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 4010132e4608SDavid Howells s_iflags = sb->s_iflags; 40118654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 40128654df4eSEric W. Biederman return false; 40138654df4eSEric W. Biederman 4014a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 4015a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 4016a1935c17SEric W. Biederman required_iflags); 4017a1935c17SEric W. Biederman return true; 4018a1935c17SEric W. Biederman } 4019a1935c17SEric W. Biederman 4020132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 40218654df4eSEric W. Biederman } 40228654df4eSEric W. Biederman 4023380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 4024380cf5baSAndy Lutomirski { 4025380cf5baSAndy Lutomirski /* 4026380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 4027380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 4028380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 4029380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 4030380cf5baSAndy Lutomirski * in other namespaces. 4031380cf5baSAndy Lutomirski */ 4032380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 4033380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 4034380cf5baSAndy Lutomirski } 4035380cf5baSAndy Lutomirski 403664964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 40378823c079SEric W. Biederman { 403858be2825SAl Viro struct ns_common *ns = NULL; 40398823c079SEric W. Biederman struct nsproxy *nsproxy; 40408823c079SEric W. Biederman 4041728dba3aSEric W. Biederman task_lock(task); 4042728dba3aSEric W. Biederman nsproxy = task->nsproxy; 40438823c079SEric W. Biederman if (nsproxy) { 404458be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 404558be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 40468823c079SEric W. Biederman } 4047728dba3aSEric W. Biederman task_unlock(task); 40488823c079SEric W. Biederman 40498823c079SEric W. Biederman return ns; 40508823c079SEric W. Biederman } 40518823c079SEric W. Biederman 405264964528SAl Viro static void mntns_put(struct ns_common *ns) 40538823c079SEric W. Biederman { 405458be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 40558823c079SEric W. Biederman } 40568823c079SEric W. Biederman 4057f2a8d52eSChristian Brauner static int mntns_install(struct nsset *nsset, struct ns_common *ns) 40588823c079SEric W. Biederman { 4059f2a8d52eSChristian Brauner struct nsproxy *nsproxy = nsset->nsproxy; 4060f2a8d52eSChristian Brauner struct fs_struct *fs = nsset->fs; 40614f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 4062f2a8d52eSChristian Brauner struct user_namespace *user_ns = nsset->cred->user_ns; 40638823c079SEric W. Biederman struct path root; 40644f757f3cSAl Viro int err; 40658823c079SEric W. Biederman 40660c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 4067f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_CHROOT) || 4068f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_ADMIN)) 4069ae11e0f1SZhao Hongjiang return -EPERM; 40708823c079SEric W. Biederman 407174e83122SAl Viro if (is_anon_ns(mnt_ns)) 407274e83122SAl Viro return -EINVAL; 407374e83122SAl Viro 40748823c079SEric W. Biederman if (fs->users != 1) 40758823c079SEric W. Biederman return -EINVAL; 40768823c079SEric W. Biederman 40778823c079SEric W. Biederman get_mnt_ns(mnt_ns); 40784f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 40798823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 40808823c079SEric W. Biederman 40818823c079SEric W. Biederman /* Find the root */ 40824f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 40834f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 40844f757f3cSAl Viro if (err) { 40854f757f3cSAl Viro /* revert to old namespace */ 40864f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 40874f757f3cSAl Viro put_mnt_ns(mnt_ns); 40884f757f3cSAl Viro return err; 40894f757f3cSAl Viro } 40908823c079SEric W. Biederman 40914068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 40924068367cSAndrei Vagin 40938823c079SEric W. Biederman /* Update the pwd and root */ 40948823c079SEric W. Biederman set_fs_pwd(fs, &root); 40958823c079SEric W. Biederman set_fs_root(fs, &root); 40968823c079SEric W. Biederman 40978823c079SEric W. Biederman path_put(&root); 40988823c079SEric W. Biederman return 0; 40998823c079SEric W. Biederman } 41008823c079SEric W. Biederman 4101bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 4102bcac25a5SAndrey Vagin { 4103bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 4104bcac25a5SAndrey Vagin } 4105bcac25a5SAndrey Vagin 41068823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 41078823c079SEric W. Biederman .name = "mnt", 41088823c079SEric W. Biederman .type = CLONE_NEWNS, 41098823c079SEric W. Biederman .get = mntns_get, 41108823c079SEric W. Biederman .put = mntns_put, 41118823c079SEric W. Biederman .install = mntns_install, 4112bcac25a5SAndrey Vagin .owner = mntns_owner, 41138823c079SEric W. Biederman }; 4114