159bd9dedSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namespace.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 185b825c3aSIngo Molnar #include <linux/cred.h> 1973cd49ecSMiklos Szeredi #include <linux/idr.h> 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23a07b2000SAl Viro #include <linux/file.h> 24d10577a8SAl Viro #include <linux/uaccess.h> 250bb80f24SDavid Howells #include <linux/proc_ns.h> 2620b4fb48SLinus Torvalds #include <linux/magic.h> 2757c8a661SMike Rapoport #include <linux/memblock.h> 289ea459e1SAl Viro #include <linux/task_work.h> 299164bb4aSIngo Molnar #include <linux/sched/task.h> 30e262e32dSDavid Howells #include <uapi/linux/mount.h> 319bc61ab1SDavid Howells #include <linux/fs_context.h> 32037f11b4SAl Viro #include <linux/shmem_fs.h> 339164bb4aSIngo Molnar 3407b20889SRam Pai #include "pnode.h" 35948730b0SAdrian Bunk #include "internal.h" 361da177e4SLinus Torvalds 37d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 38d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 39d2921684SEric W. Biederman 400818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 410818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 420818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 430818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 440818bf27SAl Viro 450818bf27SAl Viro static __initdata unsigned long mhash_entries; 460818bf27SAl Viro static int __init set_mhash_entries(char *str) 470818bf27SAl Viro { 480818bf27SAl Viro if (!str) 490818bf27SAl Viro return 0; 500818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 510818bf27SAl Viro return 1; 520818bf27SAl Viro } 530818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 540818bf27SAl Viro 550818bf27SAl Viro static __initdata unsigned long mphash_entries; 560818bf27SAl Viro static int __init set_mphash_entries(char *str) 570818bf27SAl Viro { 580818bf27SAl Viro if (!str) 590818bf27SAl Viro return 0; 600818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 610818bf27SAl Viro return 1; 620818bf27SAl Viro } 630818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6413f14b4dSEric Dumazet 65c7999c36SAl Viro static u64 event; 6673cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 67719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 685addc5ddSAl Viro 6938129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 700818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 71e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 7259aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 734edbe133SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 744edbe133SAl Viro static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ 751da177e4SLinus Torvalds 76f87fd4c2SMiklos Szeredi /* /sys/fs */ 7700d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7800d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 79f87fd4c2SMiklos Szeredi 8099b7db7bSNick Piggin /* 8199b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 8299b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 8399b7db7bSNick Piggin * up the tree. 8499b7db7bSNick Piggin * 8599b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8699b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8799b7db7bSNick Piggin */ 8848a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8999b7db7bSNick Piggin 9038129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 911da177e4SLinus Torvalds { 921da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 931da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 940818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 950818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 960818bf27SAl Viro } 970818bf27SAl Viro 980818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 990818bf27SAl Viro { 1000818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 1010818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 1020818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 1031da177e4SLinus Torvalds } 1041da177e4SLinus Torvalds 105b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10673cd49ecSMiklos Szeredi { 107169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 10873cd49ecSMiklos Szeredi 109169b480eSMatthew Wilcox if (res < 0) 11073cd49ecSMiklos Szeredi return res; 111169b480eSMatthew Wilcox mnt->mnt_id = res; 112169b480eSMatthew Wilcox return 0; 11373cd49ecSMiklos Szeredi } 11473cd49ecSMiklos Szeredi 115b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11673cd49ecSMiklos Szeredi { 117169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 11873cd49ecSMiklos Szeredi } 11973cd49ecSMiklos Szeredi 120719f5d7fSMiklos Szeredi /* 121719f5d7fSMiklos Szeredi * Allocate a new peer group ID 122719f5d7fSMiklos Szeredi */ 1234b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 124719f5d7fSMiklos Szeredi { 125169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 126f21f6220SAl Viro 127169b480eSMatthew Wilcox if (res < 0) 128f21f6220SAl Viro return res; 129169b480eSMatthew Wilcox mnt->mnt_group_id = res; 130169b480eSMatthew Wilcox return 0; 131719f5d7fSMiklos Szeredi } 132719f5d7fSMiklos Szeredi 133719f5d7fSMiklos Szeredi /* 134719f5d7fSMiklos Szeredi * Release a peer group ID 135719f5d7fSMiklos Szeredi */ 1364b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 137719f5d7fSMiklos Szeredi { 138169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 13915169fe7SAl Viro mnt->mnt_group_id = 0; 140719f5d7fSMiklos Szeredi } 141719f5d7fSMiklos Szeredi 142b3e19d92SNick Piggin /* 143b3e19d92SNick Piggin * vfsmount lock must be held for read 144b3e19d92SNick Piggin */ 14583adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 146b3e19d92SNick Piggin { 147b3e19d92SNick Piggin #ifdef CONFIG_SMP 14868e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 149b3e19d92SNick Piggin #else 150b3e19d92SNick Piggin preempt_disable(); 15168e8a9feSAl Viro mnt->mnt_count += n; 152b3e19d92SNick Piggin preempt_enable(); 153b3e19d92SNick Piggin #endif 154b3e19d92SNick Piggin } 155b3e19d92SNick Piggin 156b3e19d92SNick Piggin /* 157b3e19d92SNick Piggin * vfsmount lock must be held for write 158b3e19d92SNick Piggin */ 15983adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 160b3e19d92SNick Piggin { 161b3e19d92SNick Piggin #ifdef CONFIG_SMP 162f03c6599SAl Viro unsigned int count = 0; 163b3e19d92SNick Piggin int cpu; 164b3e19d92SNick Piggin 165b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 16668e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 167b3e19d92SNick Piggin } 168b3e19d92SNick Piggin 169b3e19d92SNick Piggin return count; 170b3e19d92SNick Piggin #else 17168e8a9feSAl Viro return mnt->mnt_count; 172b3e19d92SNick Piggin #endif 173b3e19d92SNick Piggin } 174b3e19d92SNick Piggin 175b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1761da177e4SLinus Torvalds { 177c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 178c63181e6SAl Viro if (mnt) { 17973cd49ecSMiklos Szeredi int err; 18073cd49ecSMiklos Szeredi 181c63181e6SAl Viro err = mnt_alloc_id(mnt); 18288b38782SLi Zefan if (err) 18388b38782SLi Zefan goto out_free_cache; 18488b38782SLi Zefan 18588b38782SLi Zefan if (name) { 186fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 187c63181e6SAl Viro if (!mnt->mnt_devname) 18888b38782SLi Zefan goto out_free_id; 18973cd49ecSMiklos Szeredi } 19073cd49ecSMiklos Szeredi 191b3e19d92SNick Piggin #ifdef CONFIG_SMP 192c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 193c63181e6SAl Viro if (!mnt->mnt_pcp) 194b3e19d92SNick Piggin goto out_free_devname; 195b3e19d92SNick Piggin 196c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 197b3e19d92SNick Piggin #else 198c63181e6SAl Viro mnt->mnt_count = 1; 199c63181e6SAl Viro mnt->mnt_writers = 0; 200b3e19d92SNick Piggin #endif 201b3e19d92SNick Piggin 20238129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 203c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 204c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 205c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 206c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 207c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 208c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 209c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2100a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 21199b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 21256cbb429SAl Viro INIT_HLIST_HEAD(&mnt->mnt_stuck_children); 2131da177e4SLinus Torvalds } 214c63181e6SAl Viro return mnt; 21588b38782SLi Zefan 216d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 217d3ef3d73Snpiggin@suse.de out_free_devname: 218fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 219d3ef3d73Snpiggin@suse.de #endif 22088b38782SLi Zefan out_free_id: 221c63181e6SAl Viro mnt_free_id(mnt); 22288b38782SLi Zefan out_free_cache: 223c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 22488b38782SLi Zefan return NULL; 2251da177e4SLinus Torvalds } 2261da177e4SLinus Torvalds 2278366025eSDave Hansen /* 2288366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2298366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2308366025eSDave Hansen * We must keep track of when those operations start 2318366025eSDave Hansen * (for permission checks) and when they end, so that 2328366025eSDave Hansen * we can determine when writes are able to occur to 2338366025eSDave Hansen * a filesystem. 2348366025eSDave Hansen */ 2353d733633SDave Hansen /* 2363d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2373d733633SDave Hansen * @mnt: the mount to check for its write status 2383d733633SDave Hansen * 2393d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2403d733633SDave Hansen * It does not guarantee that the filesystem will stay 2413d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2423d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2433d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2443d733633SDave Hansen * r/w. 2453d733633SDave Hansen */ 24643f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 2473d733633SDave Hansen { 24843f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 2493d733633SDave Hansen } 2503d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2513d733633SDave Hansen 25283adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2533d733633SDave Hansen { 254d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 25568e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 256d3ef3d73Snpiggin@suse.de #else 25768e8a9feSAl Viro mnt->mnt_writers++; 258d3ef3d73Snpiggin@suse.de #endif 2593d733633SDave Hansen } 2603d733633SDave Hansen 26183adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2623d733633SDave Hansen { 263d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 26468e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 265d3ef3d73Snpiggin@suse.de #else 26668e8a9feSAl Viro mnt->mnt_writers--; 267d3ef3d73Snpiggin@suse.de #endif 268d3ef3d73Snpiggin@suse.de } 269d3ef3d73Snpiggin@suse.de 27083adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 271d3ef3d73Snpiggin@suse.de { 272d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 273d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2743d733633SDave Hansen int cpu; 2753d733633SDave Hansen 2763d733633SDave Hansen for_each_possible_cpu(cpu) { 27768e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2783d733633SDave Hansen } 2793d733633SDave Hansen 280d3ef3d73Snpiggin@suse.de return count; 281d3ef3d73Snpiggin@suse.de #else 282d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 283d3ef3d73Snpiggin@suse.de #endif 2843d733633SDave Hansen } 2853d733633SDave Hansen 2864ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2874ed5e82fSMiklos Szeredi { 2884ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2894ed5e82fSMiklos Szeredi return 1; 2904ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2914ed5e82fSMiklos Szeredi smp_rmb(); 2924ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2934ed5e82fSMiklos Szeredi } 2944ed5e82fSMiklos Szeredi 2953d733633SDave Hansen /* 296eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 297eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 298eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 299eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3003d733633SDave Hansen */ 3018366025eSDave Hansen /** 302eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 30383adc753SAl Viro * @m: the mount on which to take a write 3048366025eSDave Hansen * 305eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 306eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 307eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 308eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 309eb04c282SJan Kara * called. This is effectively a refcount. 3108366025eSDave Hansen */ 311eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3128366025eSDave Hansen { 31383adc753SAl Viro struct mount *mnt = real_mount(m); 3143d733633SDave Hansen int ret = 0; 3153d733633SDave Hansen 316d3ef3d73Snpiggin@suse.de preempt_disable(); 317c6653a83SNick Piggin mnt_inc_writers(mnt); 318d3ef3d73Snpiggin@suse.de /* 319c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 320d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 321d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 322d3ef3d73Snpiggin@suse.de */ 323d3ef3d73Snpiggin@suse.de smp_mb(); 3246aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 325d3ef3d73Snpiggin@suse.de cpu_relax(); 326d3ef3d73Snpiggin@suse.de /* 327d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 328d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 329d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 330d3ef3d73Snpiggin@suse.de */ 331d3ef3d73Snpiggin@suse.de smp_rmb(); 3324ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 333c6653a83SNick Piggin mnt_dec_writers(mnt); 3343d733633SDave Hansen ret = -EROFS; 3353d733633SDave Hansen } 336d3ef3d73Snpiggin@suse.de preempt_enable(); 337eb04c282SJan Kara 338eb04c282SJan Kara return ret; 339eb04c282SJan Kara } 340eb04c282SJan Kara 341eb04c282SJan Kara /** 342eb04c282SJan Kara * mnt_want_write - get write access to a mount 343eb04c282SJan Kara * @m: the mount on which to take a write 344eb04c282SJan Kara * 345eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 346eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 347eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 348eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 349eb04c282SJan Kara */ 350eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 351eb04c282SJan Kara { 352eb04c282SJan Kara int ret; 353eb04c282SJan Kara 354eb04c282SJan Kara sb_start_write(m->mnt_sb); 355eb04c282SJan Kara ret = __mnt_want_write(m); 356eb04c282SJan Kara if (ret) 357eb04c282SJan Kara sb_end_write(m->mnt_sb); 3583d733633SDave Hansen return ret; 3598366025eSDave Hansen } 3608366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3618366025eSDave Hansen 3628366025eSDave Hansen /** 36396029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 36496029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 36596029c4eSnpiggin@suse.de * 36696029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 36796029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 36896029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 36996029c4eSnpiggin@suse.de * on it. This allows some optimisation. 37096029c4eSnpiggin@suse.de * 37196029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 37296029c4eSnpiggin@suse.de * drop the reference. 37396029c4eSnpiggin@suse.de */ 37496029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 37596029c4eSnpiggin@suse.de { 37696029c4eSnpiggin@suse.de /* superblock may be r/o */ 37796029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 37896029c4eSnpiggin@suse.de return -EROFS; 37996029c4eSnpiggin@suse.de preempt_disable(); 38083adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 38196029c4eSnpiggin@suse.de preempt_enable(); 38296029c4eSnpiggin@suse.de return 0; 38396029c4eSnpiggin@suse.de } 38496029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 38596029c4eSnpiggin@suse.de 38696029c4eSnpiggin@suse.de /** 387eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 388eb04c282SJan Kara * @file: the file who's mount on which to take a write 389eb04c282SJan Kara * 390eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 391eb04c282SJan Kara * do some optimisations if the file is open for write already 392eb04c282SJan Kara */ 393eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 394eb04c282SJan Kara { 39583f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 396eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 397eb04c282SJan Kara else 398eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 399eb04c282SJan Kara } 400eb04c282SJan Kara 401eb04c282SJan Kara /** 4027c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4037c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4047c6893e3SMiklos Szeredi * 4057c6893e3SMiklos Szeredi * This is like mnt_want_write, but it takes a file and can 4067c6893e3SMiklos Szeredi * do some optimisations if the file is open for write already 4077c6893e3SMiklos Szeredi */ 4087c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4097c6893e3SMiklos Szeredi { 4107c6893e3SMiklos Szeredi int ret; 4117c6893e3SMiklos Szeredi 4127c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4137c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4147c6893e3SMiklos Szeredi if (ret) 4157c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4167c6893e3SMiklos Szeredi return ret; 4177c6893e3SMiklos Szeredi } 41896029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 41996029c4eSnpiggin@suse.de 42096029c4eSnpiggin@suse.de /** 421eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4228366025eSDave Hansen * @mnt: the mount on which to give up write access 4238366025eSDave Hansen * 4248366025eSDave Hansen * Tells the low-level filesystem that we are done 4258366025eSDave Hansen * performing writes to it. Must be matched with 426eb04c282SJan Kara * __mnt_want_write() call above. 4278366025eSDave Hansen */ 428eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4298366025eSDave Hansen { 430d3ef3d73Snpiggin@suse.de preempt_disable(); 43183adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 432d3ef3d73Snpiggin@suse.de preempt_enable(); 4338366025eSDave Hansen } 434eb04c282SJan Kara 435eb04c282SJan Kara /** 436eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 437eb04c282SJan Kara * @mnt: the mount on which to give up write access 438eb04c282SJan Kara * 439eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 440eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 441eb04c282SJan Kara * mnt_want_write() call above. 442eb04c282SJan Kara */ 443eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 444eb04c282SJan Kara { 445eb04c282SJan Kara __mnt_drop_write(mnt); 446eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 447eb04c282SJan Kara } 4488366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4498366025eSDave Hansen 450eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 451eb04c282SJan Kara { 452eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 453eb04c282SJan Kara } 454eb04c282SJan Kara 4557c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4567c6893e3SMiklos Szeredi { 457a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4587c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4597c6893e3SMiklos Szeredi } 4602a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4612a79f17eSAl Viro 46283adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4638366025eSDave Hansen { 4643d733633SDave Hansen int ret = 0; 4653d733633SDave Hansen 466719ea2fbSAl Viro lock_mount_hash(); 46783adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 468d3ef3d73Snpiggin@suse.de /* 469d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 470d3ef3d73Snpiggin@suse.de * should be visible before we do. 471d3ef3d73Snpiggin@suse.de */ 472d3ef3d73Snpiggin@suse.de smp_mb(); 473d3ef3d73Snpiggin@suse.de 474d3ef3d73Snpiggin@suse.de /* 475d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 476d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 477d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 478d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 479d3ef3d73Snpiggin@suse.de * 480d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 481d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 482d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 483d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 484d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 485d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 486d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 487d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 488d3ef3d73Snpiggin@suse.de * we're counting up here. 489d3ef3d73Snpiggin@suse.de */ 490c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 491d3ef3d73Snpiggin@suse.de ret = -EBUSY; 492d3ef3d73Snpiggin@suse.de else 49383adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 494d3ef3d73Snpiggin@suse.de /* 495d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 496d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 497d3ef3d73Snpiggin@suse.de */ 498d3ef3d73Snpiggin@suse.de smp_wmb(); 49983adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 500719ea2fbSAl Viro unlock_mount_hash(); 5013d733633SDave Hansen return ret; 5023d733633SDave Hansen } 5038366025eSDave Hansen 50443f5e655SDavid Howells static int __mnt_unmake_readonly(struct mount *mnt) 5052e4b7fcdSDave Hansen { 506719ea2fbSAl Viro lock_mount_hash(); 50783adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 508719ea2fbSAl Viro unlock_mount_hash(); 50943f5e655SDavid Howells return 0; 5102e4b7fcdSDave Hansen } 5112e4b7fcdSDave Hansen 5124ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5134ed5e82fSMiklos Szeredi { 5144ed5e82fSMiklos Szeredi struct mount *mnt; 5154ed5e82fSMiklos Szeredi int err = 0; 5164ed5e82fSMiklos Szeredi 5178e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5188e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5198e8b8796SMiklos Szeredi return -EBUSY; 5208e8b8796SMiklos Szeredi 521719ea2fbSAl Viro lock_mount_hash(); 5224ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5234ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5244ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5254ed5e82fSMiklos Szeredi smp_mb(); 5264ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5274ed5e82fSMiklos Szeredi err = -EBUSY; 5284ed5e82fSMiklos Szeredi break; 5294ed5e82fSMiklos Szeredi } 5304ed5e82fSMiklos Szeredi } 5314ed5e82fSMiklos Szeredi } 5328e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5338e8b8796SMiklos Szeredi err = -EBUSY; 5348e8b8796SMiklos Szeredi 5354ed5e82fSMiklos Szeredi if (!err) { 5364ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5374ed5e82fSMiklos Szeredi smp_wmb(); 5384ed5e82fSMiklos Szeredi } 5394ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5404ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5414ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5424ed5e82fSMiklos Szeredi } 543719ea2fbSAl Viro unlock_mount_hash(); 5444ed5e82fSMiklos Szeredi 5454ed5e82fSMiklos Szeredi return err; 5464ed5e82fSMiklos Szeredi } 5474ed5e82fSMiklos Szeredi 548b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5491da177e4SLinus Torvalds { 550fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 551d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 55268e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 553d3ef3d73Snpiggin@suse.de #endif 554b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5551da177e4SLinus Torvalds } 5561da177e4SLinus Torvalds 5578ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5588ffcb32eSDavid Howells { 5598ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5608ffcb32eSDavid Howells } 5618ffcb32eSDavid Howells 56248a066e7SAl Viro /* call under rcu_read_lock */ 563294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 56448a066e7SAl Viro { 56548a066e7SAl Viro struct mount *mnt; 56648a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 567294d71ffSAl Viro return 1; 56848a066e7SAl Viro if (bastard == NULL) 569294d71ffSAl Viro return 0; 57048a066e7SAl Viro mnt = real_mount(bastard); 57148a066e7SAl Viro mnt_add_count(mnt, 1); 572119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 57348a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 574294d71ffSAl Viro return 0; 57548a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 57648a066e7SAl Viro mnt_add_count(mnt, -1); 577294d71ffSAl Viro return 1; 57848a066e7SAl Viro } 579119e1ef8SAl Viro lock_mount_hash(); 580119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 581119e1ef8SAl Viro mnt_add_count(mnt, -1); 582119e1ef8SAl Viro unlock_mount_hash(); 583119e1ef8SAl Viro return 1; 584119e1ef8SAl Viro } 585119e1ef8SAl Viro unlock_mount_hash(); 586119e1ef8SAl Viro /* caller will mntput() */ 587294d71ffSAl Viro return -1; 588294d71ffSAl Viro } 589294d71ffSAl Viro 590294d71ffSAl Viro /* call under rcu_read_lock */ 591294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 592294d71ffSAl Viro { 593294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 594294d71ffSAl Viro if (likely(!res)) 595294d71ffSAl Viro return true; 596294d71ffSAl Viro if (unlikely(res < 0)) { 59748a066e7SAl Viro rcu_read_unlock(); 59848a066e7SAl Viro mntput(bastard); 59948a066e7SAl Viro rcu_read_lock(); 600294d71ffSAl Viro } 60148a066e7SAl Viro return false; 60248a066e7SAl Viro } 60348a066e7SAl Viro 6041da177e4SLinus Torvalds /* 605474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 60648a066e7SAl Viro * call under rcu_read_lock() 6071da177e4SLinus Torvalds */ 608474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6091da177e4SLinus Torvalds { 61038129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 611474279dcSAl Viro struct mount *p; 6121da177e4SLinus Torvalds 61338129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 614474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 615474279dcSAl Viro return p; 616474279dcSAl Viro return NULL; 6171da177e4SLinus Torvalds } 618474279dcSAl Viro 619474279dcSAl Viro /* 620f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 621f015f126SDavid Howells * 622f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 623f015f126SDavid Howells * following mounts: 624f015f126SDavid Howells * 625f015f126SDavid Howells * mount /dev/sda1 /mnt 626f015f126SDavid Howells * mount /dev/sda2 /mnt 627f015f126SDavid Howells * mount /dev/sda3 /mnt 628f015f126SDavid Howells * 629f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 630f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 631f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 632f015f126SDavid Howells * 633f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 634a05964f3SRam Pai */ 635ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 636a05964f3SRam Pai { 637c7105365SAl Viro struct mount *child_mnt; 63848a066e7SAl Viro struct vfsmount *m; 63948a066e7SAl Viro unsigned seq; 64099b7db7bSNick Piggin 64148a066e7SAl Viro rcu_read_lock(); 64248a066e7SAl Viro do { 64348a066e7SAl Viro seq = read_seqbegin(&mount_lock); 644474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 64548a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 64648a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 64748a066e7SAl Viro rcu_read_unlock(); 64848a066e7SAl Viro return m; 649a05964f3SRam Pai } 650a05964f3SRam Pai 6517af1364fSEric W. Biederman /* 6527af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6537af1364fSEric W. Biederman * current mount namespace. 6547af1364fSEric W. Biederman * 6557af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6567af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6577af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6587af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6597af1364fSEric W. Biederman * is a mountpoint. 6607af1364fSEric W. Biederman * 6617af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6627af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6637af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6647af1364fSEric W. Biederman * parent mount. 6657af1364fSEric W. Biederman */ 6667af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6677af1364fSEric W. Biederman { 6687af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 6697af1364fSEric W. Biederman struct mount *mnt; 6707af1364fSEric W. Biederman bool is_covered = false; 6717af1364fSEric W. Biederman 6727af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 6737af1364fSEric W. Biederman goto out; 6747af1364fSEric W. Biederman 6757af1364fSEric W. Biederman down_read(&namespace_sem); 6767af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 6777af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 6787af1364fSEric W. Biederman if (is_covered) 6797af1364fSEric W. Biederman break; 6807af1364fSEric W. Biederman } 6817af1364fSEric W. Biederman up_read(&namespace_sem); 6827af1364fSEric W. Biederman out: 6837af1364fSEric W. Biederman return is_covered; 6847af1364fSEric W. Biederman } 6857af1364fSEric W. Biederman 686e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 68784d17192SAl Viro { 6880818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 68984d17192SAl Viro struct mountpoint *mp; 69084d17192SAl Viro 6910818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 69284d17192SAl Viro if (mp->m_dentry == dentry) { 69384d17192SAl Viro mp->m_count++; 69484d17192SAl Viro return mp; 69584d17192SAl Viro } 69684d17192SAl Viro } 697e2dfa935SEric W. Biederman return NULL; 698e2dfa935SEric W. Biederman } 699e2dfa935SEric W. Biederman 7003895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 701e2dfa935SEric W. Biederman { 7023895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 703e2dfa935SEric W. Biederman int ret; 70484d17192SAl Viro 7053895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7061e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7071e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7081e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7093895dbf8SEric W. Biederman mountpoint: 7103895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7113895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7123895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7133895dbf8SEric W. Biederman if (mp) 7143895dbf8SEric W. Biederman goto done; 71584d17192SAl Viro } 716eed81007SMiklos Szeredi 7173895dbf8SEric W. Biederman if (!new) 7183895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7193895dbf8SEric W. Biederman if (!new) 7203895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7213895dbf8SEric W. Biederman 7223895dbf8SEric W. Biederman 7233895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7243895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7253895dbf8SEric W. Biederman 7263895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7273895dbf8SEric W. Biederman if (ret == -EBUSY) 7283895dbf8SEric W. Biederman goto mountpoint; 7293895dbf8SEric W. Biederman 7303895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7313895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7323895dbf8SEric W. Biederman if (ret) 7333895dbf8SEric W. Biederman goto done; 7343895dbf8SEric W. Biederman 7353895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7363895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7374edbe133SAl Viro new->m_dentry = dget(dentry); 7383895dbf8SEric W. Biederman new->m_count = 1; 7393895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7403895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7413895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7423895dbf8SEric W. Biederman 7433895dbf8SEric W. Biederman mp = new; 7443895dbf8SEric W. Biederman new = NULL; 7453895dbf8SEric W. Biederman done: 7463895dbf8SEric W. Biederman kfree(new); 74784d17192SAl Viro return mp; 74884d17192SAl Viro } 74984d17192SAl Viro 7504edbe133SAl Viro /* 7514edbe133SAl Viro * vfsmount lock must be held. Additionally, the caller is responsible 7524edbe133SAl Viro * for serializing calls for given disposal list. 7534edbe133SAl Viro */ 7544edbe133SAl Viro static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) 75584d17192SAl Viro { 75684d17192SAl Viro if (!--mp->m_count) { 75784d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7580a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 75984d17192SAl Viro spin_lock(&dentry->d_lock); 76084d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 76184d17192SAl Viro spin_unlock(&dentry->d_lock); 7624edbe133SAl Viro dput_to_list(dentry, list); 7630818bf27SAl Viro hlist_del(&mp->m_hash); 76484d17192SAl Viro kfree(mp); 76584d17192SAl Viro } 76684d17192SAl Viro } 76784d17192SAl Viro 7684edbe133SAl Viro /* called with namespace_lock and vfsmount lock */ 7694edbe133SAl Viro static void put_mountpoint(struct mountpoint *mp) 7704edbe133SAl Viro { 7714edbe133SAl Viro __put_mountpoint(mp, &ex_mountpoints); 7724edbe133SAl Viro } 7734edbe133SAl Viro 774143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7751da177e4SLinus Torvalds { 7766b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7771da177e4SLinus Torvalds } 7781da177e4SLinus Torvalds 77999b7db7bSNick Piggin /* 78099b7db7bSNick Piggin * vfsmount lock must be held for write 78199b7db7bSNick Piggin */ 7826b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7835addc5ddSAl Viro { 7845addc5ddSAl Viro if (ns) { 7855addc5ddSAl Viro ns->event = ++event; 7865addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7875addc5ddSAl Viro } 7885addc5ddSAl Viro } 7895addc5ddSAl Viro 79099b7db7bSNick Piggin /* 79199b7db7bSNick Piggin * vfsmount lock must be held for write 79299b7db7bSNick Piggin */ 7936b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 7945addc5ddSAl Viro { 7955addc5ddSAl Viro if (ns && ns->event != event) { 7965addc5ddSAl Viro ns->event = event; 7975addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7985addc5ddSAl Viro } 7995addc5ddSAl Viro } 8005addc5ddSAl Viro 80199b7db7bSNick Piggin /* 80299b7db7bSNick Piggin * vfsmount lock must be held for write 80399b7db7bSNick Piggin */ 804e4e59906SAl Viro static struct mountpoint *unhash_mnt(struct mount *mnt) 8051da177e4SLinus Torvalds { 806e4e59906SAl Viro struct mountpoint *mp; 8070714a533SAl Viro mnt->mnt_parent = mnt; 808a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8096b41d536SAl Viro list_del_init(&mnt->mnt_child); 81038129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8110a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 812e4e59906SAl Viro mp = mnt->mnt_mp; 81384d17192SAl Viro mnt->mnt_mp = NULL; 814e4e59906SAl Viro return mp; 8157bdb11deSEric W. Biederman } 8167bdb11deSEric W. Biederman 8177bdb11deSEric W. Biederman /* 8187bdb11deSEric W. Biederman * vfsmount lock must be held for write 8197bdb11deSEric W. Biederman */ 8206a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8216a46c573SEric W. Biederman { 822e4e59906SAl Viro put_mountpoint(unhash_mnt(mnt)); 8236a46c573SEric W. Biederman } 8246a46c573SEric W. Biederman 8256a46c573SEric W. Biederman /* 8266a46c573SEric W. Biederman * vfsmount lock must be held for write 8276a46c573SEric W. Biederman */ 82884d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 82984d17192SAl Viro struct mountpoint *mp, 83044d964d6SAl Viro struct mount *child_mnt) 831b90fa9aeSRam Pai { 83284d17192SAl Viro mp->m_count++; 8333a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 8344edbe133SAl Viro child_mnt->mnt_mountpoint = mp->m_dentry; 8353a2393d7SAl Viro child_mnt->mnt_parent = mnt; 83684d17192SAl Viro child_mnt->mnt_mp = mp; 8370a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 838b90fa9aeSRam Pai } 839b90fa9aeSRam Pai 8401064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8411064f874SEric W. Biederman { 8421064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8431064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8441064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8451064f874SEric W. Biederman } 8461064f874SEric W. Biederman 84799b7db7bSNick Piggin /* 84899b7db7bSNick Piggin * vfsmount lock must be held for write 84999b7db7bSNick Piggin */ 85084d17192SAl Viro static void attach_mnt(struct mount *mnt, 85184d17192SAl Viro struct mount *parent, 85284d17192SAl Viro struct mountpoint *mp) 8531da177e4SLinus Torvalds { 85484d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8551064f874SEric W. Biederman __attach_mnt(mnt, parent); 856b90fa9aeSRam Pai } 857b90fa9aeSRam Pai 8581064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 85912a5b529SAl Viro { 8601064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8611064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8621064f874SEric W. Biederman 8631064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8641064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8651064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8661064f874SEric W. Biederman 8671064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 8681064f874SEric W. Biederman 8691064f874SEric W. Biederman put_mountpoint(old_mp); 8701064f874SEric W. Biederman mnt_add_count(old_parent, -1); 87112a5b529SAl Viro } 87212a5b529SAl Viro 873b90fa9aeSRam Pai /* 87499b7db7bSNick Piggin * vfsmount lock must be held for write 875b90fa9aeSRam Pai */ 8761064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 877b90fa9aeSRam Pai { 8780714a533SAl Viro struct mount *parent = mnt->mnt_parent; 87983adc753SAl Viro struct mount *m; 880b90fa9aeSRam Pai LIST_HEAD(head); 881143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 882b90fa9aeSRam Pai 8830714a533SAl Viro BUG_ON(parent == mnt); 884b90fa9aeSRam Pai 8851a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 886f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 887143c8c91SAl Viro m->mnt_ns = n; 888f03c6599SAl Viro 889b90fa9aeSRam Pai list_splice(&head, n->list.prev); 890b90fa9aeSRam Pai 891d2921684SEric W. Biederman n->mounts += n->pending_mounts; 892d2921684SEric W. Biederman n->pending_mounts = 0; 893d2921684SEric W. Biederman 8941064f874SEric W. Biederman __attach_mnt(mnt, parent); 8956b3286edSKirill Korotaev touch_mnt_namespace(n); 8961da177e4SLinus Torvalds } 8971da177e4SLinus Torvalds 898909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 8991da177e4SLinus Torvalds { 9006b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9016b41d536SAl Viro if (next == &p->mnt_mounts) { 9021da177e4SLinus Torvalds while (1) { 903909b0a88SAl Viro if (p == root) 9041da177e4SLinus Torvalds return NULL; 9056b41d536SAl Viro next = p->mnt_child.next; 9066b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9071da177e4SLinus Torvalds break; 9080714a533SAl Viro p = p->mnt_parent; 9091da177e4SLinus Torvalds } 9101da177e4SLinus Torvalds } 9116b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9121da177e4SLinus Torvalds } 9131da177e4SLinus Torvalds 914315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9159676f0c6SRam Pai { 9166b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9176b41d536SAl Viro while (prev != &p->mnt_mounts) { 9186b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9196b41d536SAl Viro prev = p->mnt_mounts.prev; 9209676f0c6SRam Pai } 9219676f0c6SRam Pai return p; 9229676f0c6SRam Pai } 9239676f0c6SRam Pai 9248f291889SAl Viro /** 9258f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 9268f291889SAl Viro * @fc: The configuration context with the superblock attached 9278f291889SAl Viro * 9288f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 9298f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 9308f291889SAl Viro * 9318f291889SAl Viro * Note that this does not attach the mount to anything. 9328f291889SAl Viro */ 9338f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 9349d412a43SAl Viro { 935b105e270SAl Viro struct mount *mnt; 9369d412a43SAl Viro 9378f291889SAl Viro if (!fc->root) 9388f291889SAl Viro return ERR_PTR(-EINVAL); 9399d412a43SAl Viro 9408f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 9419d412a43SAl Viro if (!mnt) 9429d412a43SAl Viro return ERR_PTR(-ENOMEM); 9439d412a43SAl Viro 9448f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 945b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9469d412a43SAl Viro 9478f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 9488f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 9498f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 950a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9510714a533SAl Viro mnt->mnt_parent = mnt; 9528f291889SAl Viro 953719ea2fbSAl Viro lock_mount_hash(); 9548f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 955719ea2fbSAl Viro unlock_mount_hash(); 956b105e270SAl Viro return &mnt->mnt; 9579d412a43SAl Viro } 9588f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 9598f291889SAl Viro 9608f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 9618f291889SAl Viro { 9628f291889SAl Viro int err = vfs_get_tree(fc); 9638f291889SAl Viro if (!err) { 9648f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 9658f291889SAl Viro return vfs_create_mount(fc); 9668f291889SAl Viro } 9678f291889SAl Viro return ERR_PTR(err); 9688f291889SAl Viro } 9698f291889SAl Viro EXPORT_SYMBOL(fc_mount); 9708f291889SAl Viro 9719bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 9729bc61ab1SDavid Howells int flags, const char *name, 9739bc61ab1SDavid Howells void *data) 9741da177e4SLinus Torvalds { 9759bc61ab1SDavid Howells struct fs_context *fc; 9768f291889SAl Viro struct vfsmount *mnt; 9779bc61ab1SDavid Howells int ret = 0; 9789d412a43SAl Viro 9799d412a43SAl Viro if (!type) 9803e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 9819d412a43SAl Viro 9829bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 9839bc61ab1SDavid Howells if (IS_ERR(fc)) 9849bc61ab1SDavid Howells return ERR_CAST(fc); 9859bc61ab1SDavid Howells 9863e1aeb00SDavid Howells if (name) 9873e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 9883e1aeb00SDavid Howells name, strlen(name)); 9899bc61ab1SDavid Howells if (!ret) 9909bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 9919bc61ab1SDavid Howells if (!ret) 9928f291889SAl Viro mnt = fc_mount(fc); 9938f291889SAl Viro else 9948f291889SAl Viro mnt = ERR_PTR(ret); 9959d412a43SAl Viro 9969bc61ab1SDavid Howells put_fs_context(fc); 9978f291889SAl Viro return mnt; 9989d412a43SAl Viro } 9999d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10009d412a43SAl Viro 100193faccbbSEric W. Biederman struct vfsmount * 100293faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 100393faccbbSEric W. Biederman const char *name, void *data) 100493faccbbSEric W. Biederman { 100593faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 100693faccbbSEric W. Biederman * through from the parent mount to the submount don't support 100793faccbbSEric W. Biederman * unprivileged mounts with submounts. 100893faccbbSEric W. Biederman */ 100993faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 101093faccbbSEric W. Biederman return ERR_PTR(-EPERM); 101193faccbbSEric W. Biederman 1012e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 101393faccbbSEric W. Biederman } 101493faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 101593faccbbSEric W. Biederman 101687129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 101736341f64SRam Pai int flag) 10181da177e4SLinus Torvalds { 101987129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1020be34d1a3SDavid Howells struct mount *mnt; 1021be34d1a3SDavid Howells int err; 10221da177e4SLinus Torvalds 1023be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1024be34d1a3SDavid Howells if (!mnt) 1025be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1026be34d1a3SDavid Howells 10277a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 102815169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1029719f5d7fSMiklos Szeredi else 103015169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1031719f5d7fSMiklos Szeredi 103215169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1033be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1034719f5d7fSMiklos Szeredi if (err) 1035719f5d7fSMiklos Szeredi goto out_free; 1036719f5d7fSMiklos Szeredi } 1037719f5d7fSMiklos Szeredi 103816a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 103916a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 10405ff9d8a6SEric W. Biederman 10411da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1042b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1043b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1044a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10450714a533SAl Viro mnt->mnt_parent = mnt; 1046719ea2fbSAl Viro lock_mount_hash(); 104739f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1048719ea2fbSAl Viro unlock_mount_hash(); 1049b90fa9aeSRam Pai 10507a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10517a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10526776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 105332301920SAl Viro mnt->mnt_master = old; 1054fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10558aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1056fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10576776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1058d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10596776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1060d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10615235d448SAl Viro } else { 10625235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10635afe0022SRam Pai } 1064b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10650f0afb1dSAl Viro set_mnt_shared(mnt); 10661da177e4SLinus Torvalds 10671da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10681da177e4SLinus Torvalds * as the original if that was on one */ 106936341f64SRam Pai if (flag & CL_EXPIRE) { 10706776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10716776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10721da177e4SLinus Torvalds } 1073be34d1a3SDavid Howells 1074cb338d06SAl Viro return mnt; 1075719f5d7fSMiklos Szeredi 1076719f5d7fSMiklos Szeredi out_free: 10778ffcb32eSDavid Howells mnt_free_id(mnt); 1078719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1079be34d1a3SDavid Howells return ERR_PTR(err); 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds 10829ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10839ea459e1SAl Viro { 108456cbb429SAl Viro struct hlist_node *p; 108556cbb429SAl Viro struct mount *m; 10869ea459e1SAl Viro /* 108756cbb429SAl Viro * The warning here probably indicates that somebody messed 108856cbb429SAl Viro * up a mnt_want/drop_write() pair. If this happens, the 108956cbb429SAl Viro * filesystem was probably unable to make r/w->r/o transitions. 10909ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 10919ea459e1SAl Viro * so mnt_get_writers() below is safe. 10929ea459e1SAl Viro */ 10939ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 10949ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 10959ea459e1SAl Viro mnt_pin_kill(mnt); 109656cbb429SAl Viro hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { 109756cbb429SAl Viro hlist_del(&m->mnt_umount); 109856cbb429SAl Viro mntput(&m->mnt); 109956cbb429SAl Viro } 11009ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11019ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11029ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11039ea459e1SAl Viro mnt_free_id(mnt); 11049ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11059ea459e1SAl Viro } 11069ea459e1SAl Viro 11079ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11089ea459e1SAl Viro { 11099ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11109ea459e1SAl Viro } 11119ea459e1SAl Viro 11129ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11139ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11149ea459e1SAl Viro { 11159ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 111629785735SByungchul Park struct mount *m, *t; 11179ea459e1SAl Viro 111829785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 111929785735SByungchul Park cleanup_mnt(m); 11209ea459e1SAl Viro } 11219ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11229ea459e1SAl Viro 1123900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11247b7b1aceSAl Viro { 11254edbe133SAl Viro LIST_HEAD(list); 11264edbe133SAl Viro 112748a066e7SAl Viro rcu_read_lock(); 11289ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11299ea0a46cSAl Viro /* 11309ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11319ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11329ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11339ea0a46cSAl Viro * be dropped until after an RCU delay done after 11349ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11359ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11369ea0a46cSAl Viro * we are dropping is not the final one. 11379ea0a46cSAl Viro */ 1138aa9c0e07SAl Viro mnt_add_count(mnt, -1); 113948a066e7SAl Viro rcu_read_unlock(); 114099b7db7bSNick Piggin return; 1141b3e19d92SNick Piggin } 1142719ea2fbSAl Viro lock_mount_hash(); 1143119e1ef8SAl Viro /* 1144119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1145119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1146119e1ef8SAl Viro */ 1147119e1ef8SAl Viro smp_mb(); 11489ea0a46cSAl Viro mnt_add_count(mnt, -1); 1149b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 115048a066e7SAl Viro rcu_read_unlock(); 1151719ea2fbSAl Viro unlock_mount_hash(); 115299b7db7bSNick Piggin return; 115399b7db7bSNick Piggin } 115448a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 115548a066e7SAl Viro rcu_read_unlock(); 115648a066e7SAl Viro unlock_mount_hash(); 115748a066e7SAl Viro return; 115848a066e7SAl Viro } 115948a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 116048a066e7SAl Viro rcu_read_unlock(); 1161962830dfSAndi Kleen 116239f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1163ce07d891SEric W. Biederman 1164ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1165ce07d891SEric W. Biederman struct mount *p, *tmp; 1166ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 11674edbe133SAl Viro __put_mountpoint(unhash_mnt(p), &list); 116856cbb429SAl Viro hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); 1169ce07d891SEric W. Biederman } 1170ce07d891SEric W. Biederman } 1171719ea2fbSAl Viro unlock_mount_hash(); 11724edbe133SAl Viro shrink_dentry_list(&list); 1173649a795aSAl Viro 11749ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11759ea459e1SAl Viro struct task_struct *task = current; 11769ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11779ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11789ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11799ea459e1SAl Viro return; 11809ea459e1SAl Viro } 11819ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11829ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11839ea459e1SAl Viro return; 11849ea459e1SAl Viro } 11859ea459e1SAl Viro cleanup_mnt(mnt); 1186b3e19d92SNick Piggin } 1187b3e19d92SNick Piggin 1188b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1189b3e19d92SNick Piggin { 1190b3e19d92SNick Piggin if (mnt) { 1191863d684fSAl Viro struct mount *m = real_mount(mnt); 1192b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1193863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1194863d684fSAl Viro m->mnt_expiry_mark = 0; 1195863d684fSAl Viro mntput_no_expire(m); 1196b3e19d92SNick Piggin } 1197b3e19d92SNick Piggin } 1198b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1199b3e19d92SNick Piggin 1200b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1201b3e19d92SNick Piggin { 1202b3e19d92SNick Piggin if (mnt) 120383adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1204b3e19d92SNick Piggin return mnt; 1205b3e19d92SNick Piggin } 1206b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1207b3e19d92SNick Piggin 1208c6609c0aSIan Kent /* path_is_mountpoint() - Check if path is a mount in the current 1209c6609c0aSIan Kent * namespace. 1210c6609c0aSIan Kent * 1211c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1212c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1213c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1214c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1215c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1216c6609c0aSIan Kent * alone. 1217c6609c0aSIan Kent */ 1218c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1219c6609c0aSIan Kent { 1220c6609c0aSIan Kent unsigned seq; 1221c6609c0aSIan Kent bool res; 1222c6609c0aSIan Kent 1223c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1224c6609c0aSIan Kent return false; 1225c6609c0aSIan Kent 1226c6609c0aSIan Kent rcu_read_lock(); 1227c6609c0aSIan Kent do { 1228c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1229c6609c0aSIan Kent res = __path_is_mountpoint(path); 1230c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1231c6609c0aSIan Kent rcu_read_unlock(); 1232c6609c0aSIan Kent 1233c6609c0aSIan Kent return res; 1234c6609c0aSIan Kent } 1235c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1236c6609c0aSIan Kent 1237ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12387b7b1aceSAl Viro { 12393064c356SAl Viro struct mount *p; 12403064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12413064c356SAl Viro if (IS_ERR(p)) 12423064c356SAl Viro return ERR_CAST(p); 12433064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12443064c356SAl Viro return &p->mnt; 12457b7b1aceSAl Viro } 12461da177e4SLinus Torvalds 1247a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12480226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12491da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12501da177e4SLinus Torvalds { 1251ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12521da177e4SLinus Torvalds 1253390c6843SRam Pai down_read(&namespace_sem); 1254c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1255c7999c36SAl Viro void *v = p->cached_mount; 1256c7999c36SAl Viro if (*pos == p->cached_index) 1257c7999c36SAl Viro return v; 1258c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1259c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1260c7999c36SAl Viro return p->cached_mount = v; 1261c7999c36SAl Viro } 1262c7999c36SAl Viro } 1263c7999c36SAl Viro 1264c7999c36SAl Viro p->cached_event = p->ns->event; 1265c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1266c7999c36SAl Viro p->cached_index = *pos; 1267c7999c36SAl Viro return p->cached_mount; 12681da177e4SLinus Torvalds } 12691da177e4SLinus Torvalds 12701da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12711da177e4SLinus Torvalds { 1272ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1273b0765fb8SPavel Emelianov 1274c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1275c7999c36SAl Viro p->cached_index = *pos; 1276c7999c36SAl Viro return p->cached_mount; 12771da177e4SLinus Torvalds } 12781da177e4SLinus Torvalds 12791da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12801da177e4SLinus Torvalds { 1281390c6843SRam Pai up_read(&namespace_sem); 12821da177e4SLinus Torvalds } 12831da177e4SLinus Torvalds 12840226f492SAl Viro static int m_show(struct seq_file *m, void *v) 12859f5596afSAl Viro { 1286ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12871a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 12880226f492SAl Viro return p->show(m, &r->mnt); 12891da177e4SLinus Torvalds } 12901da177e4SLinus Torvalds 1291a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 12921da177e4SLinus Torvalds .start = m_start, 12931da177e4SLinus Torvalds .next = m_next, 12941da177e4SLinus Torvalds .stop = m_stop, 12950226f492SAl Viro .show = m_show, 1296b4629fe2SChuck Lever }; 1297a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1298b4629fe2SChuck Lever 12991da177e4SLinus Torvalds /** 13001da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 13011da177e4SLinus Torvalds * @mnt: root of mount tree 13021da177e4SLinus Torvalds * 13031da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 13041da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 13051da177e4SLinus Torvalds * busy. 13061da177e4SLinus Torvalds */ 1307909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13081da177e4SLinus Torvalds { 1309909b0a88SAl Viro struct mount *mnt = real_mount(m); 131036341f64SRam Pai int actual_refs = 0; 131136341f64SRam Pai int minimum_refs = 0; 1312315fc83eSAl Viro struct mount *p; 1313909b0a88SAl Viro BUG_ON(!m); 13141da177e4SLinus Torvalds 1315b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1316719ea2fbSAl Viro lock_mount_hash(); 1317909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 131883adc753SAl Viro actual_refs += mnt_get_count(p); 13191da177e4SLinus Torvalds minimum_refs += 2; 13201da177e4SLinus Torvalds } 1321719ea2fbSAl Viro unlock_mount_hash(); 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13241da177e4SLinus Torvalds return 0; 1325e3474a8eSIan Kent 1326e3474a8eSIan Kent return 1; 13271da177e4SLinus Torvalds } 13281da177e4SLinus Torvalds 13291da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13301da177e4SLinus Torvalds 13311da177e4SLinus Torvalds /** 13321da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13331da177e4SLinus Torvalds * @mnt: root of mount 13341da177e4SLinus Torvalds * 13351da177e4SLinus Torvalds * This is called to check if a mount point has any 13361da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13371da177e4SLinus Torvalds * mount has sub mounts this will return busy 13381da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13391da177e4SLinus Torvalds * 13401da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13411da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13421da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13431da177e4SLinus Torvalds */ 13441da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13451da177e4SLinus Torvalds { 1346e3474a8eSIan Kent int ret = 1; 13478ad08d8aSAl Viro down_read(&namespace_sem); 1348719ea2fbSAl Viro lock_mount_hash(); 13491ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1350e3474a8eSIan Kent ret = 0; 1351719ea2fbSAl Viro unlock_mount_hash(); 13528ad08d8aSAl Viro up_read(&namespace_sem); 1353a05964f3SRam Pai return ret; 13541da177e4SLinus Torvalds } 13551da177e4SLinus Torvalds 13561da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13571da177e4SLinus Torvalds 135897216be0SAl Viro static void namespace_unlock(void) 13591da177e4SLinus Torvalds { 1360a3b3c562SEric W. Biederman struct hlist_head head; 136156cbb429SAl Viro struct hlist_node *p; 136256cbb429SAl Viro struct mount *m; 13634edbe133SAl Viro LIST_HEAD(list); 136497216be0SAl Viro 1365a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 13664edbe133SAl Viro list_splice_init(&ex_mountpoints, &list); 1367a3b3c562SEric W. Biederman 136897216be0SAl Viro up_write(&namespace_sem); 1369a3b3c562SEric W. Biederman 13704edbe133SAl Viro shrink_dentry_list(&list); 13714edbe133SAl Viro 1372a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 137397216be0SAl Viro return; 137497216be0SAl Viro 137522cb7405SNeilBrown synchronize_rcu_expedited(); 137648a066e7SAl Viro 137756cbb429SAl Viro hlist_for_each_entry_safe(m, p, &head, mnt_umount) { 137856cbb429SAl Viro hlist_del(&m->mnt_umount); 137956cbb429SAl Viro mntput(&m->mnt); 138056cbb429SAl Viro } 138170fbcdf4SRam Pai } 138270fbcdf4SRam Pai 138397216be0SAl Viro static inline void namespace_lock(void) 1384e3197d83SAl Viro { 138597216be0SAl Viro down_write(&namespace_sem); 1386e3197d83SAl Viro } 1387e3197d83SAl Viro 1388e819f152SEric W. Biederman enum umount_tree_flags { 1389e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1390e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1391e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1392e819f152SEric W. Biederman }; 1393f2d0a123SEric W. Biederman 1394f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1395f2d0a123SEric W. Biederman { 1396f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1397f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1398f2d0a123SEric W. Biederman return true; 1399f2d0a123SEric W. Biederman 1400f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1401f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1402f2d0a123SEric W. Biederman return true; 1403f2d0a123SEric W. Biederman 1404f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1405f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1406f2d0a123SEric W. Biederman * connected to mounted mounts. 1407f2d0a123SEric W. Biederman */ 1408f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1409f2d0a123SEric W. Biederman return true; 1410f2d0a123SEric W. Biederman 1411f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1412f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1413f2d0a123SEric W. Biederman return false; 1414f2d0a123SEric W. Biederman 1415f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1416f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1417f2d0a123SEric W. Biederman return false; 1418f2d0a123SEric W. Biederman 1419f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1420f2d0a123SEric W. Biederman return true; 1421f2d0a123SEric W. Biederman } 1422f2d0a123SEric W. Biederman 142399b7db7bSNick Piggin /* 142448a066e7SAl Viro * mount_lock must be held 142599b7db7bSNick Piggin * namespace_sem must be held for write 142699b7db7bSNick Piggin */ 1427e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 142870fbcdf4SRam Pai { 1429c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1430315fc83eSAl Viro struct mount *p; 143170fbcdf4SRam Pai 14325d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14335d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14345d88457eSEric W. Biederman 1435c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1436590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1437590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1438c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1439590ce4bcSEric W. Biederman } 1440c003b26fSEric W. Biederman 1441411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1442c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1443c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 144438129a13SAl Viro } 144570fbcdf4SRam Pai 1446c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1447e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14487b8a53fdSAl Viro propagate_umount(&tmp_list); 1449a05964f3SRam Pai 1450c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1451d2921684SEric W. Biederman struct mnt_namespace *ns; 1452ce07d891SEric W. Biederman bool disconnect; 1453c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 14546776db3dSAl Viro list_del_init(&p->mnt_expire); 14551a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1456d2921684SEric W. Biederman ns = p->mnt_ns; 1457d2921684SEric W. Biederman if (ns) { 1458d2921684SEric W. Biederman ns->mounts--; 1459d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1460d2921684SEric W. Biederman } 146163d37a84SAl Viro p->mnt_ns = NULL; 1462e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 146348a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 146487b95ce0SAl Viro 1465f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1466676da58dSAl Viro if (mnt_has_parent(p)) { 146781b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1468ce07d891SEric W. Biederman if (!disconnect) { 1469ce07d891SEric W. Biederman /* Don't forget about p */ 1470ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1471ce07d891SEric W. Biederman } else { 14726a46c573SEric W. Biederman umount_mnt(p); 14737c4b93d8SAl Viro } 1474ce07d891SEric W. Biederman } 14750f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 147619a1c409SAl Viro if (disconnect) 147719a1c409SAl Viro hlist_add_head(&p->mnt_umount, &unmounted); 147838129a13SAl Viro } 14791da177e4SLinus Torvalds } 14801da177e4SLinus Torvalds 1481b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1482c35038beSAl Viro 14838d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 14848d0347f6SDavid Howells { 14858d0347f6SDavid Howells int ret = 0; 14868d0347f6SDavid Howells 14878d0347f6SDavid Howells down_write(&sb->s_umount); 14888d0347f6SDavid Howells if (!sb_rdonly(sb)) { 14898d0347f6SDavid Howells struct fs_context *fc; 14908d0347f6SDavid Howells 14918d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 14928d0347f6SDavid Howells SB_RDONLY); 14938d0347f6SDavid Howells if (IS_ERR(fc)) { 14948d0347f6SDavid Howells ret = PTR_ERR(fc); 14958d0347f6SDavid Howells } else { 14968d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 14978d0347f6SDavid Howells if (!ret) 14988d0347f6SDavid Howells ret = reconfigure_super(fc); 14998d0347f6SDavid Howells put_fs_context(fc); 15008d0347f6SDavid Howells } 15018d0347f6SDavid Howells } 15028d0347f6SDavid Howells up_write(&sb->s_umount); 15038d0347f6SDavid Howells return ret; 15048d0347f6SDavid Howells } 15058d0347f6SDavid Howells 15061ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 15071da177e4SLinus Torvalds { 15081ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 15091da177e4SLinus Torvalds int retval; 15101da177e4SLinus Torvalds 15111ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 15121da177e4SLinus Torvalds if (retval) 15131da177e4SLinus Torvalds return retval; 15141da177e4SLinus Torvalds 15151da177e4SLinus Torvalds /* 15161da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 15171da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 15181da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 15191da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 15201da177e4SLinus Torvalds */ 15211da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 15221ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 15231da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 15241da177e4SLinus Torvalds return -EINVAL; 15251da177e4SLinus Torvalds 1526b3e19d92SNick Piggin /* 1527b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1528b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1529b3e19d92SNick Piggin */ 1530719ea2fbSAl Viro lock_mount_hash(); 153183adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1532719ea2fbSAl Viro unlock_mount_hash(); 15331da177e4SLinus Torvalds return -EBUSY; 1534b3e19d92SNick Piggin } 1535719ea2fbSAl Viro unlock_mount_hash(); 15361da177e4SLinus Torvalds 1537863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 15381da177e4SLinus Torvalds return -EAGAIN; 15391da177e4SLinus Torvalds } 15401da177e4SLinus Torvalds 15411da177e4SLinus Torvalds /* 15421da177e4SLinus Torvalds * If we may have to abort operations to get out of this 15431da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 15441da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 15451da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 15461da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 15471da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 15481da177e4SLinus Torvalds * about for the moment. 15491da177e4SLinus Torvalds */ 15501da177e4SLinus Torvalds 155142faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 155242faad99SAl Viro sb->s_op->umount_begin(sb); 155342faad99SAl Viro } 15541da177e4SLinus Torvalds 15551da177e4SLinus Torvalds /* 15561da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15571da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15581da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15591da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15601da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15611da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15621da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15631da177e4SLinus Torvalds */ 15641ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15651da177e4SLinus Torvalds /* 15661da177e4SLinus Torvalds * Special case for "unmounting" root ... 15671da177e4SLinus Torvalds * we just try to remount it readonly. 15681da177e4SLinus Torvalds */ 1569bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1570a1480dccSAndy Lutomirski return -EPERM; 15718d0347f6SDavid Howells return do_umount_root(sb); 15721da177e4SLinus Torvalds } 15731da177e4SLinus Torvalds 157497216be0SAl Viro namespace_lock(); 1575719ea2fbSAl Viro lock_mount_hash(); 15761da177e4SLinus Torvalds 157725d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 157825d202edSEric W. Biederman retval = -EINVAL; 157925d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 158025d202edSEric W. Biederman goto out; 158125d202edSEric W. Biederman 158225d202edSEric W. Biederman event++; 158348a066e7SAl Viro if (flags & MNT_DETACH) { 158448a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1585e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 158648a066e7SAl Viro retval = 0; 158748a066e7SAl Viro } else { 1588b54b9be7SAl Viro shrink_submounts(mnt); 15891da177e4SLinus Torvalds retval = -EBUSY; 159048a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 15911a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1592e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 15931da177e4SLinus Torvalds retval = 0; 15941da177e4SLinus Torvalds } 159548a066e7SAl Viro } 159625d202edSEric W. Biederman out: 1597719ea2fbSAl Viro unlock_mount_hash(); 1598e3197d83SAl Viro namespace_unlock(); 15991da177e4SLinus Torvalds return retval; 16001da177e4SLinus Torvalds } 16011da177e4SLinus Torvalds 16021da177e4SLinus Torvalds /* 160380b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 160480b5dce8SEric W. Biederman * 160580b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 160680b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 160780b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 160880b5dce8SEric W. Biederman * leaking them. 160980b5dce8SEric W. Biederman * 161080b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 161180b5dce8SEric W. Biederman */ 161280b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 161380b5dce8SEric W. Biederman { 161480b5dce8SEric W. Biederman struct mountpoint *mp; 161580b5dce8SEric W. Biederman struct mount *mnt; 161680b5dce8SEric W. Biederman 161780b5dce8SEric W. Biederman namespace_lock(); 16183895dbf8SEric W. Biederman lock_mount_hash(); 161980b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1620adc9b5c0SAl Viro if (!mp) 162180b5dce8SEric W. Biederman goto out_unlock; 162280b5dce8SEric W. Biederman 1623e06b933eSAndrey Ulanov event++; 162480b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 162580b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1626ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1627fe78fcc8SEric W. Biederman umount_mnt(mnt); 162856cbb429SAl Viro hlist_add_head(&mnt->mnt_umount, &unmounted); 1629ce07d891SEric W. Biederman } 1630e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 163180b5dce8SEric W. Biederman } 163280b5dce8SEric W. Biederman put_mountpoint(mp); 163380b5dce8SEric W. Biederman out_unlock: 16343895dbf8SEric W. Biederman unlock_mount_hash(); 163580b5dce8SEric W. Biederman namespace_unlock(); 163680b5dce8SEric W. Biederman } 163780b5dce8SEric W. Biederman 163880b5dce8SEric W. Biederman /* 16399b40bc90SAl Viro * Is the caller allowed to modify his namespace? 16409b40bc90SAl Viro */ 16419b40bc90SAl Viro static inline bool may_mount(void) 16429b40bc90SAl Viro { 16439b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 16449b40bc90SAl Viro } 16459b40bc90SAl Viro 1646df2474a2SJeff Layton #ifdef CONFIG_MANDATORY_FILE_LOCKING 16479e8925b6SJeff Layton static inline bool may_mandlock(void) 16489e8925b6SJeff Layton { 164995ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 16509e8925b6SJeff Layton } 1651df2474a2SJeff Layton #else 1652df2474a2SJeff Layton static inline bool may_mandlock(void) 1653df2474a2SJeff Layton { 1654df2474a2SJeff Layton pr_warn("VFS: \"mand\" mount option not supported"); 1655df2474a2SJeff Layton return false; 1656df2474a2SJeff Layton } 1657df2474a2SJeff Layton #endif 16589e8925b6SJeff Layton 16599b40bc90SAl Viro /* 16601da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16611da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16621da177e4SLinus Torvalds * 16631da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16641da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16651da177e4SLinus Torvalds */ 16661da177e4SLinus Torvalds 16673a18ef5cSDominik Brodowski int ksys_umount(char __user *name, int flags) 16681da177e4SLinus Torvalds { 16692d8f3038SAl Viro struct path path; 1670900148dcSAl Viro struct mount *mnt; 16711da177e4SLinus Torvalds int retval; 1672db1f05bbSMiklos Szeredi int lookup_flags = 0; 16731da177e4SLinus Torvalds 1674db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1675db1f05bbSMiklos Szeredi return -EINVAL; 1676db1f05bbSMiklos Szeredi 16779b40bc90SAl Viro if (!may_mount()) 16789b40bc90SAl Viro return -EPERM; 16799b40bc90SAl Viro 1680db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1681db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1682db1f05bbSMiklos Szeredi 1683197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 16841da177e4SLinus Torvalds if (retval) 16851da177e4SLinus Torvalds goto out; 1686900148dcSAl Viro mnt = real_mount(path.mnt); 16871da177e4SLinus Torvalds retval = -EINVAL; 16882d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 16891da177e4SLinus Torvalds goto dput_and_out; 1690143c8c91SAl Viro if (!check_mnt(mnt)) 16911da177e4SLinus Torvalds goto dput_and_out; 169225d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 16935ff9d8a6SEric W. Biederman goto dput_and_out; 1694b2f5d4dcSEric W. Biederman retval = -EPERM; 1695b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1696b2f5d4dcSEric W. Biederman goto dput_and_out; 16971da177e4SLinus Torvalds 1698900148dcSAl Viro retval = do_umount(mnt, flags); 16991da177e4SLinus Torvalds dput_and_out: 1700429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 17012d8f3038SAl Viro dput(path.dentry); 1702900148dcSAl Viro mntput_no_expire(mnt); 17031da177e4SLinus Torvalds out: 17041da177e4SLinus Torvalds return retval; 17051da177e4SLinus Torvalds } 17061da177e4SLinus Torvalds 17073a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 17083a18ef5cSDominik Brodowski { 17093a18ef5cSDominik Brodowski return ksys_umount(name, flags); 17103a18ef5cSDominik Brodowski } 17113a18ef5cSDominik Brodowski 17121da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 17131da177e4SLinus Torvalds 17141da177e4SLinus Torvalds /* 17151da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17161da177e4SLinus Torvalds */ 1717bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17181da177e4SLinus Torvalds { 17193a18ef5cSDominik Brodowski return ksys_umount(name, 0); 17201da177e4SLinus Torvalds } 17211da177e4SLinus Torvalds 17221da177e4SLinus Torvalds #endif 17231da177e4SLinus Torvalds 17244ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 17258823c079SEric W. Biederman { 17264ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1727e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1728e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 17294ce5d2b1SEric W. Biederman } 17304ce5d2b1SEric W. Biederman 1731213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 173258be2825SAl Viro { 173358be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 173458be2825SAl Viro } 173558be2825SAl Viro 17364ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 17374ce5d2b1SEric W. Biederman { 17384ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 17394ce5d2b1SEric W. Biederman * mount namespace loop? 17404ce5d2b1SEric W. Biederman */ 17414ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 17424ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 17434ce5d2b1SEric W. Biederman return false; 17444ce5d2b1SEric W. Biederman 1745f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 17468823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 17478823c079SEric W. Biederman } 17488823c079SEric W. Biederman 174987129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 175036341f64SRam Pai int flag) 17511da177e4SLinus Torvalds { 175284d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 17531da177e4SLinus Torvalds 17544ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 17554ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 17564ce5d2b1SEric W. Biederman 17574ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1758be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 17599676f0c6SRam Pai 176036341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1761be34d1a3SDavid Howells if (IS_ERR(q)) 1762be34d1a3SDavid Howells return q; 1763be34d1a3SDavid Howells 1764a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17651da177e4SLinus Torvalds 17661da177e4SLinus Torvalds p = mnt; 17676b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1768315fc83eSAl Viro struct mount *s; 17697ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17701da177e4SLinus Torvalds continue; 17711da177e4SLinus Torvalds 1772909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 17734ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17744ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1775df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1776df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1777df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1778df7342b2SEric W. Biederman goto out; 1779df7342b2SEric W. Biederman } else { 17804ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17814ce5d2b1SEric W. Biederman continue; 17824ce5d2b1SEric W. Biederman } 1783df7342b2SEric W. Biederman } 17844ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17854ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17869676f0c6SRam Pai s = skip_mnt_tree(s); 17879676f0c6SRam Pai continue; 17889676f0c6SRam Pai } 17890714a533SAl Viro while (p != s->mnt_parent) { 17900714a533SAl Viro p = p->mnt_parent; 17910714a533SAl Viro q = q->mnt_parent; 17921da177e4SLinus Torvalds } 179387129cc0SAl Viro p = s; 179484d17192SAl Viro parent = q; 179587129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1796be34d1a3SDavid Howells if (IS_ERR(q)) 1797be34d1a3SDavid Howells goto out; 1798719ea2fbSAl Viro lock_mount_hash(); 17991a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 18001064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1801719ea2fbSAl Viro unlock_mount_hash(); 18021da177e4SLinus Torvalds } 18031da177e4SLinus Torvalds } 18041da177e4SLinus Torvalds return res; 1805be34d1a3SDavid Howells out: 18061da177e4SLinus Torvalds if (res) { 1807719ea2fbSAl Viro lock_mount_hash(); 1808e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1809719ea2fbSAl Viro unlock_mount_hash(); 18101da177e4SLinus Torvalds } 1811be34d1a3SDavid Howells return q; 18121da177e4SLinus Torvalds } 18131da177e4SLinus Torvalds 1814be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1815be34d1a3SDavid Howells 1816ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 18178aec0809SAl Viro { 1818cb338d06SAl Viro struct mount *tree; 181997216be0SAl Viro namespace_lock(); 1820cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1821cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1822cd4a4017SEric W. Biederman else 182387129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 182487129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1825328e6d90SAl Viro namespace_unlock(); 1826be34d1a3SDavid Howells if (IS_ERR(tree)) 182752e220d3SDan Carpenter return ERR_CAST(tree); 1828be34d1a3SDavid Howells return &tree->mnt; 18298aec0809SAl Viro } 18308aec0809SAl Viro 1831a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1832a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1833a07b2000SAl Viro 1834a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1835a07b2000SAl Viro { 1836a07b2000SAl Viro struct mnt_namespace *ns; 1837a07b2000SAl Viro namespace_lock(); 1838a07b2000SAl Viro lock_mount_hash(); 1839a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 184044dfd84aSDavid Howells if (ns) { 184144dfd84aSDavid Howells if (is_anon_ns(ns)) 1842a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 184344dfd84aSDavid Howells else 184444dfd84aSDavid Howells ns = NULL; 184544dfd84aSDavid Howells } 1846a07b2000SAl Viro unlock_mount_hash(); 1847a07b2000SAl Viro namespace_unlock(); 184844dfd84aSDavid Howells if (ns) 1849a07b2000SAl Viro free_mnt_ns(ns); 1850a07b2000SAl Viro } 1851a07b2000SAl Viro 18528aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 18538aec0809SAl Viro { 185497216be0SAl Viro namespace_lock(); 1855719ea2fbSAl Viro lock_mount_hash(); 18569c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1857719ea2fbSAl Viro unlock_mount_hash(); 18583ab6abeeSAl Viro namespace_unlock(); 18598aec0809SAl Viro } 18608aec0809SAl Viro 1861c771d683SMiklos Szeredi /** 1862c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1863c771d683SMiklos Szeredi * 1864c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1865c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1866c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1867c771d683SMiklos Szeredi * 1868c771d683SMiklos Szeredi * Release with mntput(). 1869c771d683SMiklos Szeredi */ 1870ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1871c771d683SMiklos Szeredi { 1872c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1873c771d683SMiklos Szeredi struct mount *new_mnt; 1874c771d683SMiklos Szeredi 1875c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1876c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1877c771d683SMiklos Szeredi 1878c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1879c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1880c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1881c771d683SMiklos Szeredi 1882c771d683SMiklos Szeredi return &new_mnt->mnt; 1883c771d683SMiklos Szeredi } 1884c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1885c771d683SMiklos Szeredi 18861f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18871f707137SAl Viro struct vfsmount *root) 18881f707137SAl Viro { 18891a4eeaf2SAl Viro struct mount *mnt; 18901f707137SAl Viro int res = f(root, arg); 18911f707137SAl Viro if (res) 18921f707137SAl Viro return res; 18931a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18941a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18951f707137SAl Viro if (res) 18961f707137SAl Viro return res; 18971f707137SAl Viro } 18981f707137SAl Viro return 0; 18991f707137SAl Viro } 19001f707137SAl Viro 19013bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 19023bd045ccSAl Viro { 19033bd045ccSAl Viro struct mount *p; 19043bd045ccSAl Viro 19053bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 19063bd045ccSAl Viro int flags = p->mnt.mnt_flags; 19073bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 19083bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 19093bd045ccSAl Viro 19103bd045ccSAl Viro if (flags & MNT_READONLY) 19113bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 19123bd045ccSAl Viro 19133bd045ccSAl Viro if (flags & MNT_NODEV) 19143bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 19153bd045ccSAl Viro 19163bd045ccSAl Viro if (flags & MNT_NOSUID) 19173bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 19183bd045ccSAl Viro 19193bd045ccSAl Viro if (flags & MNT_NOEXEC) 19203bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 19213bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 19223bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 19233bd045ccSAl Viro flags |= MNT_LOCKED; 19243bd045ccSAl Viro p->mnt.mnt_flags = flags; 19253bd045ccSAl Viro } 19263bd045ccSAl Viro } 19273bd045ccSAl Viro 19284b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1929719f5d7fSMiklos Szeredi { 1930315fc83eSAl Viro struct mount *p; 1931719f5d7fSMiklos Szeredi 1932909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1933fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 19344b8b21f4SAl Viro mnt_release_group_id(p); 1935719f5d7fSMiklos Szeredi } 1936719f5d7fSMiklos Szeredi } 1937719f5d7fSMiklos Szeredi 19384b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1939719f5d7fSMiklos Szeredi { 1940315fc83eSAl Viro struct mount *p; 1941719f5d7fSMiklos Szeredi 1942909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1943fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 19444b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1945719f5d7fSMiklos Szeredi if (err) { 19464b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1947719f5d7fSMiklos Szeredi return err; 1948719f5d7fSMiklos Szeredi } 1949719f5d7fSMiklos Szeredi } 1950719f5d7fSMiklos Szeredi } 1951719f5d7fSMiklos Szeredi 1952719f5d7fSMiklos Szeredi return 0; 1953719f5d7fSMiklos Szeredi } 1954719f5d7fSMiklos Szeredi 1955d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1956d2921684SEric W. Biederman { 1957d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1958d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1959d2921684SEric W. Biederman struct mount *p; 1960d2921684SEric W. Biederman 1961d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1962d2921684SEric W. Biederman mounts++; 1963d2921684SEric W. Biederman 1964d2921684SEric W. Biederman old = ns->mounts; 1965d2921684SEric W. Biederman pending = ns->pending_mounts; 1966d2921684SEric W. Biederman sum = old + pending; 1967d2921684SEric W. Biederman if ((old > sum) || 1968d2921684SEric W. Biederman (pending > sum) || 1969d2921684SEric W. Biederman (max < sum) || 1970d2921684SEric W. Biederman (mounts > (max - sum))) 1971d2921684SEric W. Biederman return -ENOSPC; 1972d2921684SEric W. Biederman 1973d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1974d2921684SEric W. Biederman return 0; 1975d2921684SEric W. Biederman } 1976d2921684SEric W. Biederman 1977b90fa9aeSRam Pai /* 1978b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1979b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 198021444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 198121444403SRam Pai * store the parent mount and mountpoint dentry. 198221444403SRam Pai * (done when source_mnt is moved) 1983b90fa9aeSRam Pai * 1984b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1985b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 19869676f0c6SRam Pai * --------------------------------------------------------------------------- 1987b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 19889676f0c6SRam Pai * |************************************************************************** 19899676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19909676f0c6SRam Pai * | dest | | | | | 19919676f0c6SRam Pai * | | | | | | | 19929676f0c6SRam Pai * | v | | | | | 19939676f0c6SRam Pai * |************************************************************************** 19949676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 19955afe0022SRam Pai * | | | | | | 19969676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 19979676f0c6SRam Pai * *************************************************************************** 1998b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1999b90fa9aeSRam Pai * destination mount. 2000b90fa9aeSRam Pai * 2001b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2002b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2003b90fa9aeSRam Pai * the peer group of the source mount. 2004b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2005b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2006b90fa9aeSRam Pai * mount. 20075afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 20085afe0022SRam Pai * of the destination mount and the cloned mount is made slave 20095afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 20105afe0022SRam Pai * is marked as 'shared and slave'. 20115afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 20125afe0022SRam Pai * source mount. 20135afe0022SRam Pai * 20149676f0c6SRam Pai * --------------------------------------------------------------------------- 201521444403SRam Pai * | MOVE MOUNT OPERATION | 20169676f0c6SRam Pai * |************************************************************************** 20179676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 20189676f0c6SRam Pai * | dest | | | | | 20199676f0c6SRam Pai * | | | | | | | 20209676f0c6SRam Pai * | v | | | | | 20219676f0c6SRam Pai * |************************************************************************** 20229676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 20235afe0022SRam Pai * | | | | | | 20249676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 20259676f0c6SRam Pai * *************************************************************************** 20265afe0022SRam Pai * 20275afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 20285afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 202921444403SRam Pai * (+*) the mount is moved to the destination. 20305afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 20315afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 20325afe0022SRam Pai * the mount is marked as 'shared and slave'. 20335afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2034b90fa9aeSRam Pai * 2035b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2036b90fa9aeSRam Pai * applied to each mount in the tree. 2037b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2038b90fa9aeSRam Pai * in allocations. 2039b90fa9aeSRam Pai */ 20400fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 204184d17192SAl Viro struct mount *dest_mnt, 204284d17192SAl Viro struct mountpoint *dest_mp, 20432763d119SAl Viro bool moving) 2044b90fa9aeSRam Pai { 20453bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 204638129a13SAl Viro HLIST_HEAD(tree_list); 2047d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 20481064f874SEric W. Biederman struct mountpoint *smp; 2049315fc83eSAl Viro struct mount *child, *p; 205038129a13SAl Viro struct hlist_node *n; 2051719f5d7fSMiklos Szeredi int err; 2052b90fa9aeSRam Pai 20531064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 20541064f874SEric W. Biederman * to be tucked under other mounts. 20551064f874SEric W. Biederman */ 20561064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 20571064f874SEric W. Biederman if (IS_ERR(smp)) 20581064f874SEric W. Biederman return PTR_ERR(smp); 20591064f874SEric W. Biederman 2060d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 20612763d119SAl Viro if (!moving) { 2062d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2063d2921684SEric W. Biederman if (err) 2064d2921684SEric W. Biederman goto out; 2065d2921684SEric W. Biederman } 2066d2921684SEric W. Biederman 2067fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 20680fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2069719f5d7fSMiklos Szeredi if (err) 2070719f5d7fSMiklos Szeredi goto out; 207184d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2072f2ebb3a9SAl Viro lock_mount_hash(); 2073719f5d7fSMiklos Szeredi if (err) 2074719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2075909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 20760f0afb1dSAl Viro set_mnt_shared(p); 20770b1b901bSAl Viro } else { 20780b1b901bSAl Viro lock_mount_hash(); 2079b90fa9aeSRam Pai } 20802763d119SAl Viro if (moving) { 20812763d119SAl Viro unhash_mnt(source_mnt); 208284d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2083143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 208421444403SRam Pai } else { 208544dfd84aSDavid Howells if (source_mnt->mnt_ns) { 208644dfd84aSDavid Howells /* move from anon - the caller will destroy */ 208744dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 208844dfd84aSDavid Howells } 208984d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 20901064f874SEric W. Biederman commit_tree(source_mnt); 209121444403SRam Pai } 2092b90fa9aeSRam Pai 209338129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 20941d6a32acSAl Viro struct mount *q; 209538129a13SAl Viro hlist_del_init(&child->mnt_hash); 20961064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 20971d6a32acSAl Viro child->mnt_mountpoint); 20981064f874SEric W. Biederman if (q) 20991064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 21003bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 21013bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 21023bd045ccSAl Viro lock_mnt_tree(child); 2103d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 21041064f874SEric W. Biederman commit_tree(child); 2105b90fa9aeSRam Pai } 21061064f874SEric W. Biederman put_mountpoint(smp); 2107719ea2fbSAl Viro unlock_mount_hash(); 210899b7db7bSNick Piggin 2109b90fa9aeSRam Pai return 0; 2110719f5d7fSMiklos Szeredi 2111719f5d7fSMiklos Szeredi out_cleanup_ids: 2112f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2113f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2114d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2115e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2116f2ebb3a9SAl Viro } 2117f2ebb3a9SAl Viro unlock_mount_hash(); 21180fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2119719f5d7fSMiklos Szeredi out: 2120d2921684SEric W. Biederman ns->pending_mounts = 0; 21211064f874SEric W. Biederman 21221064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 21231064f874SEric W. Biederman put_mountpoint(smp); 21241064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 21251064f874SEric W. Biederman 2126719f5d7fSMiklos Szeredi return err; 2127b90fa9aeSRam Pai } 2128b90fa9aeSRam Pai 212984d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2130b12cea91SAl Viro { 2131b12cea91SAl Viro struct vfsmount *mnt; 213284d17192SAl Viro struct dentry *dentry = path->dentry; 2133b12cea91SAl Viro retry: 21345955102cSAl Viro inode_lock(dentry->d_inode); 213584d17192SAl Viro if (unlikely(cant_mount(dentry))) { 21365955102cSAl Viro inode_unlock(dentry->d_inode); 213784d17192SAl Viro return ERR_PTR(-ENOENT); 2138b12cea91SAl Viro } 213997216be0SAl Viro namespace_lock(); 2140b12cea91SAl Viro mnt = lookup_mnt(path); 214184d17192SAl Viro if (likely(!mnt)) { 21423895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 214384d17192SAl Viro if (IS_ERR(mp)) { 214497216be0SAl Viro namespace_unlock(); 21455955102cSAl Viro inode_unlock(dentry->d_inode); 214684d17192SAl Viro return mp; 214784d17192SAl Viro } 214884d17192SAl Viro return mp; 214984d17192SAl Viro } 215097216be0SAl Viro namespace_unlock(); 21515955102cSAl Viro inode_unlock(path->dentry->d_inode); 2152b12cea91SAl Viro path_put(path); 2153b12cea91SAl Viro path->mnt = mnt; 215484d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2155b12cea91SAl Viro goto retry; 2156b12cea91SAl Viro } 2157b12cea91SAl Viro 215884d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2159b12cea91SAl Viro { 216084d17192SAl Viro struct dentry *dentry = where->m_dentry; 21613895dbf8SEric W. Biederman 21623895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 216384d17192SAl Viro put_mountpoint(where); 21643895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 21653895dbf8SEric W. Biederman 2166328e6d90SAl Viro namespace_unlock(); 21675955102cSAl Viro inode_unlock(dentry->d_inode); 2168b12cea91SAl Viro } 2169b12cea91SAl Viro 217084d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 21711da177e4SLinus Torvalds { 2172e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 21731da177e4SLinus Torvalds return -EINVAL; 21741da177e4SLinus Torvalds 2175e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2176e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 21771da177e4SLinus Torvalds return -ENOTDIR; 21781da177e4SLinus Torvalds 21792763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 21801da177e4SLinus Torvalds } 21811da177e4SLinus Torvalds 21821da177e4SLinus Torvalds /* 21837a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 21847a2e8a8fSValerie Aurora */ 21857a2e8a8fSValerie Aurora 2186e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 21877a2e8a8fSValerie Aurora { 2188e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 21897a2e8a8fSValerie Aurora 21907a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 21917a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 21927a2e8a8fSValerie Aurora return 0; 21937a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 21947a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 21957a2e8a8fSValerie Aurora return 0; 21967a2e8a8fSValerie Aurora return type; 21977a2e8a8fSValerie Aurora } 21987a2e8a8fSValerie Aurora 21997a2e8a8fSValerie Aurora /* 220007b20889SRam Pai * recursively change the type of the mountpoint. 220107b20889SRam Pai */ 2202e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 220307b20889SRam Pai { 2204315fc83eSAl Viro struct mount *m; 22054b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2206e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 22077a2e8a8fSValerie Aurora int type; 2208719f5d7fSMiklos Szeredi int err = 0; 220907b20889SRam Pai 22102d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 221107b20889SRam Pai return -EINVAL; 221207b20889SRam Pai 2213e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 22147a2e8a8fSValerie Aurora if (!type) 22157a2e8a8fSValerie Aurora return -EINVAL; 22167a2e8a8fSValerie Aurora 221797216be0SAl Viro namespace_lock(); 2218719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2219719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2220719f5d7fSMiklos Szeredi if (err) 2221719f5d7fSMiklos Szeredi goto out_unlock; 2222719f5d7fSMiklos Szeredi } 2223719f5d7fSMiklos Szeredi 2224719ea2fbSAl Viro lock_mount_hash(); 2225909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 22260f0afb1dSAl Viro change_mnt_propagation(m, type); 2227719ea2fbSAl Viro unlock_mount_hash(); 2228719f5d7fSMiklos Szeredi 2229719f5d7fSMiklos Szeredi out_unlock: 223097216be0SAl Viro namespace_unlock(); 2231719f5d7fSMiklos Szeredi return err; 223207b20889SRam Pai } 223307b20889SRam Pai 22345ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 22355ff9d8a6SEric W. Biederman { 22365ff9d8a6SEric W. Biederman struct mount *child; 22375ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 22385ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 22395ff9d8a6SEric W. Biederman continue; 22405ff9d8a6SEric W. Biederman 22415ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 22425ff9d8a6SEric W. Biederman return true; 22435ff9d8a6SEric W. Biederman } 22445ff9d8a6SEric W. Biederman return false; 22455ff9d8a6SEric W. Biederman } 22465ff9d8a6SEric W. Biederman 2247a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2248a07b2000SAl Viro { 2249a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2250a07b2000SAl Viro 2251a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2252a07b2000SAl Viro return mnt; 2253a07b2000SAl Viro 2254a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2255a07b2000SAl Viro return mnt; 2256a07b2000SAl Viro 2257a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2258a07b2000SAl Viro return mnt; 2259a07b2000SAl Viro 2260a07b2000SAl Viro if (recurse) 2261a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2262a07b2000SAl Viro else 2263a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2264a07b2000SAl Viro 2265a07b2000SAl Viro if (!IS_ERR(mnt)) 2266a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2267a07b2000SAl Viro 2268a07b2000SAl Viro return mnt; 2269a07b2000SAl Viro } 2270a07b2000SAl Viro 227107b20889SRam Pai /* 22721da177e4SLinus Torvalds * do loopback mount. 22731da177e4SLinus Torvalds */ 2274808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 22752dafe1c4SEric Sandeen int recurse) 22761da177e4SLinus Torvalds { 22772d92ab3cSAl Viro struct path old_path; 2278a07b2000SAl Viro struct mount *mnt = NULL, *parent; 227984d17192SAl Viro struct mountpoint *mp; 228057eccb83SAl Viro int err; 22811da177e4SLinus Torvalds if (!old_name || !*old_name) 22821da177e4SLinus Torvalds return -EINVAL; 2283815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 22841da177e4SLinus Torvalds if (err) 22851da177e4SLinus Torvalds return err; 22861da177e4SLinus Torvalds 22878823c079SEric W. Biederman err = -EINVAL; 22884ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 22898823c079SEric W. Biederman goto out; 22908823c079SEric W. Biederman 229184d17192SAl Viro mp = lock_mount(path); 2292a07b2000SAl Viro if (IS_ERR(mp)) { 229384d17192SAl Viro err = PTR_ERR(mp); 22949676f0c6SRam Pai goto out; 2295a07b2000SAl Viro } 22969676f0c6SRam Pai 229784d17192SAl Viro parent = real_mount(path->mnt); 2298e149ed2bSAl Viro if (!check_mnt(parent)) 2299e149ed2bSAl Viro goto out2; 2300e149ed2bSAl Viro 2301a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2302be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2303be34d1a3SDavid Howells err = PTR_ERR(mnt); 2304e9c5d8a5SAndrey Vagin goto out2; 2305be34d1a3SDavid Howells } 2306ccd48bc7SAl Viro 230784d17192SAl Viro err = graft_tree(mnt, parent, mp); 23081da177e4SLinus Torvalds if (err) { 2309719ea2fbSAl Viro lock_mount_hash(); 2310e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2311719ea2fbSAl Viro unlock_mount_hash(); 23125b83d2c5SRam Pai } 2313b12cea91SAl Viro out2: 231484d17192SAl Viro unlock_mount(mp); 2315ccd48bc7SAl Viro out: 23162d92ab3cSAl Viro path_put(&old_path); 23171da177e4SLinus Torvalds return err; 23181da177e4SLinus Torvalds } 23191da177e4SLinus Torvalds 2320a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2321a07b2000SAl Viro { 2322a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2323a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2324a07b2000SAl Viro struct mount *mnt, *p; 2325a07b2000SAl Viro struct file *file; 2326a07b2000SAl Viro 2327a07b2000SAl Viro if (IS_ERR(ns)) 2328a07b2000SAl Viro return ERR_CAST(ns); 2329a07b2000SAl Viro 2330a07b2000SAl Viro namespace_lock(); 2331a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2332a07b2000SAl Viro if (IS_ERR(mnt)) { 2333a07b2000SAl Viro namespace_unlock(); 2334a07b2000SAl Viro free_mnt_ns(ns); 2335a07b2000SAl Viro return ERR_CAST(mnt); 2336a07b2000SAl Viro } 2337a07b2000SAl Viro 2338a07b2000SAl Viro lock_mount_hash(); 2339a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2340a07b2000SAl Viro p->mnt_ns = ns; 2341a07b2000SAl Viro ns->mounts++; 2342a07b2000SAl Viro } 2343a07b2000SAl Viro ns->root = mnt; 2344a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2345a07b2000SAl Viro mntget(&mnt->mnt); 2346a07b2000SAl Viro unlock_mount_hash(); 2347a07b2000SAl Viro namespace_unlock(); 2348a07b2000SAl Viro 2349a07b2000SAl Viro mntput(path->mnt); 2350a07b2000SAl Viro path->mnt = &mnt->mnt; 2351a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2352a07b2000SAl Viro if (IS_ERR(file)) 2353a07b2000SAl Viro dissolve_on_fput(path->mnt); 2354a07b2000SAl Viro else 2355a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2356a07b2000SAl Viro return file; 2357a07b2000SAl Viro } 2358a07b2000SAl Viro 23592658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2360a07b2000SAl Viro { 2361a07b2000SAl Viro struct file *file; 2362a07b2000SAl Viro struct path path; 2363a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2364a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2365a07b2000SAl Viro int error; 2366a07b2000SAl Viro int fd; 2367a07b2000SAl Viro 2368a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2369a07b2000SAl Viro 2370a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2371a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2372a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2373a07b2000SAl Viro return -EINVAL; 2374a07b2000SAl Viro 2375a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2376a07b2000SAl Viro return -EINVAL; 2377a07b2000SAl Viro 2378a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2379a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2380a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2381a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2382a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2383a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2384a07b2000SAl Viro 2385a07b2000SAl Viro if (detached && !may_mount()) 2386a07b2000SAl Viro return -EPERM; 2387a07b2000SAl Viro 2388a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2389a07b2000SAl Viro if (fd < 0) 2390a07b2000SAl Viro return fd; 2391a07b2000SAl Viro 2392a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2393a07b2000SAl Viro if (unlikely(error)) { 2394a07b2000SAl Viro file = ERR_PTR(error); 2395a07b2000SAl Viro } else { 2396a07b2000SAl Viro if (detached) 2397a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2398a07b2000SAl Viro else 2399a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2400a07b2000SAl Viro path_put(&path); 2401a07b2000SAl Viro } 2402a07b2000SAl Viro if (IS_ERR(file)) { 2403a07b2000SAl Viro put_unused_fd(fd); 2404a07b2000SAl Viro return PTR_ERR(file); 2405a07b2000SAl Viro } 2406a07b2000SAl Viro fd_install(fd, file); 2407a07b2000SAl Viro return fd; 2408a07b2000SAl Viro } 2409a07b2000SAl Viro 241043f5e655SDavid Howells /* 241143f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 241243f5e655SDavid Howells * 241343f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 241443f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 241543f5e655SDavid Howells */ 241643f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 24172e4b7fcdSDave Hansen { 241843f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 24192e4b7fcdSDave Hansen 242043f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 242143f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 242243f5e655SDavid Howells return false; 242343f5e655SDavid Howells 242443f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 242543f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 242643f5e655SDavid Howells return false; 242743f5e655SDavid Howells 242843f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 242943f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 243043f5e655SDavid Howells return false; 243143f5e655SDavid Howells 243243f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 243343f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 243443f5e655SDavid Howells return false; 243543f5e655SDavid Howells 243643f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 243743f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 243843f5e655SDavid Howells return false; 243943f5e655SDavid Howells 244043f5e655SDavid Howells return true; 244143f5e655SDavid Howells } 244243f5e655SDavid Howells 244343f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 244443f5e655SDavid Howells { 244543f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 244643f5e655SDavid Howells 244743f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 24482e4b7fcdSDave Hansen return 0; 24492e4b7fcdSDave Hansen 24502e4b7fcdSDave Hansen if (readonly_request) 245143f5e655SDavid Howells return mnt_make_readonly(mnt); 245243f5e655SDavid Howells 245343f5e655SDavid Howells return __mnt_unmake_readonly(mnt); 245443f5e655SDavid Howells } 245543f5e655SDavid Howells 245643f5e655SDavid Howells /* 245743f5e655SDavid Howells * Update the user-settable attributes on a mount. The caller must hold 245843f5e655SDavid Howells * sb->s_umount for writing. 245943f5e655SDavid Howells */ 246043f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 246143f5e655SDavid Howells { 246243f5e655SDavid Howells lock_mount_hash(); 246343f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 246443f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 246543f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 246643f5e655SDavid Howells unlock_mount_hash(); 246743f5e655SDavid Howells } 246843f5e655SDavid Howells 2469f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2470f8b92ba6SDeepa Dinamani { 2471f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2472f8b92ba6SDeepa Dinamani 2473f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2474f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2475f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2476f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2477f8b92ba6SDeepa Dinamani struct tm tm; 2478f8b92ba6SDeepa Dinamani 2479f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2480f8b92ba6SDeepa Dinamani 24810ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 24820ecee669SEric Biggers sb->s_type->name, 24830ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 24840ecee669SEric Biggers mntpath, 2485f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2486f8b92ba6SDeepa Dinamani 2487f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2488f8b92ba6SDeepa Dinamani } 2489f8b92ba6SDeepa Dinamani } 2490f8b92ba6SDeepa Dinamani 249143f5e655SDavid Howells /* 249243f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 249343f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 249443f5e655SDavid Howells * to mount(2). 249543f5e655SDavid Howells */ 249643f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 249743f5e655SDavid Howells { 249843f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 249943f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 250043f5e655SDavid Howells int ret; 250143f5e655SDavid Howells 250243f5e655SDavid Howells if (!check_mnt(mnt)) 250343f5e655SDavid Howells return -EINVAL; 250443f5e655SDavid Howells 250543f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 250643f5e655SDavid Howells return -EINVAL; 250743f5e655SDavid Howells 250843f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 250943f5e655SDavid Howells return -EPERM; 251043f5e655SDavid Howells 251143f5e655SDavid Howells down_write(&sb->s_umount); 251243f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 251343f5e655SDavid Howells if (ret == 0) 251443f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 251543f5e655SDavid Howells up_write(&sb->s_umount); 2516f8b92ba6SDeepa Dinamani 2517f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2518f8b92ba6SDeepa Dinamani 251943f5e655SDavid Howells return ret; 25202e4b7fcdSDave Hansen } 25212e4b7fcdSDave Hansen 25221da177e4SLinus Torvalds /* 25231da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 25241da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 25251da177e4SLinus Torvalds * on it - tough luck. 25261da177e4SLinus Torvalds */ 2527e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2528e462ec50SDavid Howells int mnt_flags, void *data) 25291da177e4SLinus Torvalds { 25301da177e4SLinus Torvalds int err; 25312d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2532143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 25338d0347f6SDavid Howells struct fs_context *fc; 25341da177e4SLinus Torvalds 2535143c8c91SAl Viro if (!check_mnt(mnt)) 25361da177e4SLinus Torvalds return -EINVAL; 25371da177e4SLinus Torvalds 25382d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 25391da177e4SLinus Torvalds return -EINVAL; 25401da177e4SLinus Torvalds 254143f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 254207b64558SEric W. Biederman return -EPERM; 25439566d674SEric W. Biederman 25448d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 25458d0347f6SDavid Howells if (IS_ERR(fc)) 25468d0347f6SDavid Howells return PTR_ERR(fc); 2547ff36fe2cSEric Paris 25488d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 25498d0347f6SDavid Howells if (!err) { 25501da177e4SLinus Torvalds down_write(&sb->s_umount); 255157eccb83SAl Viro err = -EPERM; 255243f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 25538d0347f6SDavid Howells err = reconfigure_super(fc); 255443f5e655SDavid Howells if (!err) 255543f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 25560e55a7ccSDan Williams } 25576339dab8SAl Viro up_write(&sb->s_umount); 25588d0347f6SDavid Howells } 2559f8b92ba6SDeepa Dinamani 2560f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2561f8b92ba6SDeepa Dinamani 25628d0347f6SDavid Howells put_fs_context(fc); 25631da177e4SLinus Torvalds return err; 25641da177e4SLinus Torvalds } 25651da177e4SLinus Torvalds 2566cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 25679676f0c6SRam Pai { 2568315fc83eSAl Viro struct mount *p; 2569909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2570fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 25719676f0c6SRam Pai return 1; 25729676f0c6SRam Pai } 25739676f0c6SRam Pai return 0; 25749676f0c6SRam Pai } 25759676f0c6SRam Pai 257644dfd84aSDavid Howells /* 257744dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 257844dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 257944dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 258044dfd84aSDavid Howells * cycles to be made. 258144dfd84aSDavid Howells */ 258244dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 258344dfd84aSDavid Howells { 258444dfd84aSDavid Howells struct mount *p; 258544dfd84aSDavid Howells bool ret = false; 258644dfd84aSDavid Howells 258744dfd84aSDavid Howells lock_mount_hash(); 258844dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 258944dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 259044dfd84aSDavid Howells goto out; 259144dfd84aSDavid Howells 259244dfd84aSDavid Howells ret = true; 259344dfd84aSDavid Howells out: 259444dfd84aSDavid Howells unlock_mount_hash(); 259544dfd84aSDavid Howells return ret; 259644dfd84aSDavid Howells } 259744dfd84aSDavid Howells 25982db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 25991da177e4SLinus Torvalds { 260044dfd84aSDavid Howells struct mnt_namespace *ns; 2601676da58dSAl Viro struct mount *p; 26020fb54e50SAl Viro struct mount *old; 26032763d119SAl Viro struct mount *parent; 26042763d119SAl Viro struct mountpoint *mp, *old_mp; 260557eccb83SAl Viro int err; 260644dfd84aSDavid Howells bool attached; 26071da177e4SLinus Torvalds 26082db154b3SDavid Howells mp = lock_mount(new_path); 260984d17192SAl Viro if (IS_ERR(mp)) 26102db154b3SDavid Howells return PTR_ERR(mp); 2611cc53ce53SDavid Howells 26122db154b3SDavid Howells old = real_mount(old_path->mnt); 26132db154b3SDavid Howells p = real_mount(new_path->mnt); 26142763d119SAl Viro parent = old->mnt_parent; 261544dfd84aSDavid Howells attached = mnt_has_parent(old); 26162763d119SAl Viro old_mp = old->mnt_mp; 261744dfd84aSDavid Howells ns = old->mnt_ns; 2618143c8c91SAl Viro 26191da177e4SLinus Torvalds err = -EINVAL; 262044dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 262144dfd84aSDavid Howells if (!check_mnt(p)) 26222db154b3SDavid Howells goto out; 26231da177e4SLinus Torvalds 2624570d7a98SEric Biggers /* The thing moved must be mounted... */ 2625570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 262644dfd84aSDavid Howells goto out; 262744dfd84aSDavid Howells 2628570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2629570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 26302db154b3SDavid Howells goto out; 26311da177e4SLinus Torvalds 26322db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 26332db154b3SDavid Howells goto out; 26342db154b3SDavid Howells 26352db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 26362db154b3SDavid Howells goto out; 26372db154b3SDavid Howells 26382db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 26392db154b3SDavid Howells d_is_dir(old_path->dentry)) 26402db154b3SDavid Howells goto out; 264121444403SRam Pai /* 264221444403SRam Pai * Don't move a mount residing in a shared parent. 264321444403SRam Pai */ 26442763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 26452db154b3SDavid Howells goto out; 26469676f0c6SRam Pai /* 26479676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 26489676f0c6SRam Pai * mount which is shared. 26499676f0c6SRam Pai */ 2650fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 26512db154b3SDavid Howells goto out; 26521da177e4SLinus Torvalds err = -ELOOP; 265344dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 265444dfd84aSDavid Howells goto out; 2655fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2656676da58dSAl Viro if (p == old) 26572db154b3SDavid Howells goto out; 26581da177e4SLinus Torvalds 26592db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 26602763d119SAl Viro attached); 26614ac91378SJan Blunck if (err) 26622db154b3SDavid Howells goto out; 26631da177e4SLinus Torvalds 26641da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 26651da177e4SLinus Torvalds * automatically */ 26666776db3dSAl Viro list_del_init(&old->mnt_expire); 26672763d119SAl Viro if (attached) 26682763d119SAl Viro put_mountpoint(old_mp); 26691da177e4SLinus Torvalds out: 26702db154b3SDavid Howells unlock_mount(mp); 267144dfd84aSDavid Howells if (!err) { 26722763d119SAl Viro if (attached) 26732763d119SAl Viro mntput_no_expire(parent); 26742763d119SAl Viro else 267544dfd84aSDavid Howells free_mnt_ns(ns); 267644dfd84aSDavid Howells } 26772db154b3SDavid Howells return err; 26782db154b3SDavid Howells } 26792db154b3SDavid Howells 26802db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 26812db154b3SDavid Howells { 26822db154b3SDavid Howells struct path old_path; 26832db154b3SDavid Howells int err; 26842db154b3SDavid Howells 26852db154b3SDavid Howells if (!old_name || !*old_name) 26862db154b3SDavid Howells return -EINVAL; 26872db154b3SDavid Howells 26882db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 26892db154b3SDavid Howells if (err) 26902db154b3SDavid Howells return err; 26912db154b3SDavid Howells 26922db154b3SDavid Howells err = do_move_mount(&old_path, path); 26932d92ab3cSAl Viro path_put(&old_path); 26941da177e4SLinus Torvalds return err; 26951da177e4SLinus Torvalds } 26961da177e4SLinus Torvalds 26979d412a43SAl Viro /* 26989d412a43SAl Viro * add a mount into a namespace's mount tree 26999d412a43SAl Viro */ 27008f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 27018f11538eSAl Viro struct path *path, int mnt_flags) 27029d412a43SAl Viro { 27038f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 27049d412a43SAl Viro 2705f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 27069d412a43SAl Viro 270784d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2708156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2709156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 27108f11538eSAl Viro return -EINVAL; 2711156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 271284d17192SAl Viro if (!parent->mnt_ns) 27138f11538eSAl Viro return -EINVAL; 2714156cacb1SAl Viro } 27159d412a43SAl Viro 27169d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 271795bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 27189d412a43SAl Viro path->mnt->mnt_root == path->dentry) 27198f11538eSAl Viro return -EBUSY; 27209d412a43SAl Viro 2721e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 27228f11538eSAl Viro return -EINVAL; 27239d412a43SAl Viro 272495bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 27258f11538eSAl Viro return graft_tree(newmnt, parent, mp); 27269d412a43SAl Viro } 2727b1e75df4SAl Viro 2728132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2729132e4608SDavid Howells 2730132e4608SDavid Howells /* 2731132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2732132e4608SDavid Howells * be added to the namespace tree. 2733132e4608SDavid Howells */ 2734132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2735132e4608SDavid Howells unsigned int mnt_flags) 2736132e4608SDavid Howells { 2737132e4608SDavid Howells struct vfsmount *mnt; 27388f11538eSAl Viro struct mountpoint *mp; 2739132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2740132e4608SDavid Howells int error; 2741132e4608SDavid Howells 2742c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2743c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2744c9ce29edSAl Viro error = -EPERM; 2745c9ce29edSAl Viro 2746c9ce29edSAl Viro if (unlikely(error)) { 2747c9ce29edSAl Viro fc_drop_locked(fc); 2748c9ce29edSAl Viro return error; 2749132e4608SDavid Howells } 2750132e4608SDavid Howells 2751132e4608SDavid Howells up_write(&sb->s_umount); 2752132e4608SDavid Howells 2753132e4608SDavid Howells mnt = vfs_create_mount(fc); 2754132e4608SDavid Howells if (IS_ERR(mnt)) 2755132e4608SDavid Howells return PTR_ERR(mnt); 2756132e4608SDavid Howells 2757f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 2758f8b92ba6SDeepa Dinamani 27598f11538eSAl Viro mp = lock_mount(mountpoint); 27608f11538eSAl Viro if (IS_ERR(mp)) { 27618f11538eSAl Viro mntput(mnt); 27628f11538eSAl Viro return PTR_ERR(mp); 27638f11538eSAl Viro } 27648f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 27658f11538eSAl Viro unlock_mount(mp); 27660ecee669SEric Biggers if (error < 0) 27670ecee669SEric Biggers mntput(mnt); 2768f8b92ba6SDeepa Dinamani return error; 2769f8b92ba6SDeepa Dinamani } 2770f8b92ba6SDeepa Dinamani 27711da177e4SLinus Torvalds /* 27721da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 27731da177e4SLinus Torvalds * namespace's tree 27741da177e4SLinus Torvalds */ 2775e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2776808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 27771da177e4SLinus Torvalds { 27780c55cfc4SEric W. Biederman struct file_system_type *type; 2779a0c9a8b8SAl Viro struct fs_context *fc; 2780a0c9a8b8SAl Viro const char *subtype = NULL; 2781a0c9a8b8SAl Viro int err = 0; 27821da177e4SLinus Torvalds 27830c55cfc4SEric W. Biederman if (!fstype) 27841da177e4SLinus Torvalds return -EINVAL; 27851da177e4SLinus Torvalds 27860c55cfc4SEric W. Biederman type = get_fs_type(fstype); 27870c55cfc4SEric W. Biederman if (!type) 27880c55cfc4SEric W. Biederman return -ENODEV; 27890c55cfc4SEric W. Biederman 2790a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 2791a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 2792a0c9a8b8SAl Viro if (subtype) { 2793a0c9a8b8SAl Viro subtype++; 2794a0c9a8b8SAl Viro if (!*subtype) { 27950c55cfc4SEric W. Biederman put_filesystem(type); 2796a0c9a8b8SAl Viro return -EINVAL; 2797a0c9a8b8SAl Viro } 2798a0c9a8b8SAl Viro } 27998654df4eSEric W. Biederman } 28008654df4eSEric W. Biederman 2801a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 2802a0c9a8b8SAl Viro put_filesystem(type); 2803a0c9a8b8SAl Viro if (IS_ERR(fc)) 2804a0c9a8b8SAl Viro return PTR_ERR(fc); 2805a0c9a8b8SAl Viro 28063e1aeb00SDavid Howells if (subtype) 28073e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 28083e1aeb00SDavid Howells subtype, strlen(subtype)); 28093e1aeb00SDavid Howells if (!err && name) 28103e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 2811a0c9a8b8SAl Viro if (!err) 2812a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 2813c3aabf07SAl Viro if (!err && !mount_capable(fc)) 2814c3aabf07SAl Viro err = -EPERM; 2815a0c9a8b8SAl Viro if (!err) 2816a0c9a8b8SAl Viro err = vfs_get_tree(fc); 2817132e4608SDavid Howells if (!err) 2818132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 2819a0c9a8b8SAl Viro 2820a0c9a8b8SAl Viro put_fs_context(fc); 282115f9a3f3SAl Viro return err; 28221da177e4SLinus Torvalds } 28231da177e4SLinus Torvalds 282419a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 282519a167afSAl Viro { 282626df6034SAl Viro struct dentry *dentry = path->dentry; 28276776db3dSAl Viro struct mount *mnt = real_mount(m); 28288f11538eSAl Viro struct mountpoint *mp; 282919a167afSAl Viro int err; 283019a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 283119a167afSAl Viro * expired before we get a chance to add it 283219a167afSAl Viro */ 28336776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 283419a167afSAl Viro 283519a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 283626df6034SAl Viro m->mnt_root == dentry) { 2837b1e75df4SAl Viro err = -ELOOP; 283826df6034SAl Viro goto discard; 283919a167afSAl Viro } 284019a167afSAl Viro 284126df6034SAl Viro /* 284226df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 284326df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 284426df6034SAl Viro * got", not "try to mount it on top". 284526df6034SAl Viro */ 284626df6034SAl Viro inode_lock(dentry->d_inode); 284726df6034SAl Viro namespace_lock(); 284826df6034SAl Viro if (unlikely(cant_mount(dentry))) { 284926df6034SAl Viro err = -ENOENT; 285026df6034SAl Viro goto discard_locked; 285126df6034SAl Viro } 285226df6034SAl Viro rcu_read_lock(); 285326df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 285426df6034SAl Viro rcu_read_unlock(); 285526df6034SAl Viro err = 0; 285626df6034SAl Viro goto discard_locked; 285726df6034SAl Viro } 285826df6034SAl Viro rcu_read_unlock(); 285926df6034SAl Viro mp = get_mountpoint(dentry); 28608f11538eSAl Viro if (IS_ERR(mp)) { 28618f11538eSAl Viro err = PTR_ERR(mp); 286226df6034SAl Viro goto discard_locked; 28638f11538eSAl Viro } 286426df6034SAl Viro 28658f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 28668f11538eSAl Viro unlock_mount(mp); 286726df6034SAl Viro if (unlikely(err)) 286826df6034SAl Viro goto discard; 286926df6034SAl Viro mntput(m); 2870b1e75df4SAl Viro return 0; 287126df6034SAl Viro 287226df6034SAl Viro discard_locked: 287326df6034SAl Viro namespace_unlock(); 287426df6034SAl Viro inode_unlock(dentry->d_inode); 287526df6034SAl Viro discard: 2876b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 28776776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 287897216be0SAl Viro namespace_lock(); 28796776db3dSAl Viro list_del_init(&mnt->mnt_expire); 288097216be0SAl Viro namespace_unlock(); 288119a167afSAl Viro } 2882b1e75df4SAl Viro mntput(m); 2883b1e75df4SAl Viro mntput(m); 288419a167afSAl Viro return err; 288519a167afSAl Viro } 288619a167afSAl Viro 2887ea5b778aSDavid Howells /** 2888ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2889ea5b778aSDavid Howells * @mnt: The mount to list. 2890ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2891ea5b778aSDavid Howells */ 2892ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2893ea5b778aSDavid Howells { 289497216be0SAl Viro namespace_lock(); 2895ea5b778aSDavid Howells 28966776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2897ea5b778aSDavid Howells 289897216be0SAl Viro namespace_unlock(); 2899ea5b778aSDavid Howells } 2900ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2901ea5b778aSDavid Howells 2902ea5b778aSDavid Howells /* 29031da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 29041da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 29051da177e4SLinus Torvalds * here 29061da177e4SLinus Torvalds */ 29071da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 29081da177e4SLinus Torvalds { 2909761d5c38SAl Viro struct mount *mnt, *next; 29101da177e4SLinus Torvalds LIST_HEAD(graveyard); 29111da177e4SLinus Torvalds 29121da177e4SLinus Torvalds if (list_empty(mounts)) 29131da177e4SLinus Torvalds return; 29141da177e4SLinus Torvalds 291597216be0SAl Viro namespace_lock(); 2916719ea2fbSAl Viro lock_mount_hash(); 29171da177e4SLinus Torvalds 29181da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 29191da177e4SLinus Torvalds * following criteria: 29201da177e4SLinus Torvalds * - only referenced by its parent vfsmount 29211da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 29221da177e4SLinus Torvalds * cleared by mntput()) 29231da177e4SLinus Torvalds */ 29246776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2925863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 29261ab59738SAl Viro propagate_mount_busy(mnt, 1)) 29271da177e4SLinus Torvalds continue; 29286776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 29291da177e4SLinus Torvalds } 2930bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 29316776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2932143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2933e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2934bcc5c7d2SAl Viro } 2935719ea2fbSAl Viro unlock_mount_hash(); 29363ab6abeeSAl Viro namespace_unlock(); 29371da177e4SLinus Torvalds } 29381da177e4SLinus Torvalds 29391da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 29401da177e4SLinus Torvalds 29411da177e4SLinus Torvalds /* 29425528f911STrond Myklebust * Ripoff of 'select_parent()' 29435528f911STrond Myklebust * 29445528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 29455528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 29465528f911STrond Myklebust */ 2947692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 29485528f911STrond Myklebust { 2949692afc31SAl Viro struct mount *this_parent = parent; 29505528f911STrond Myklebust struct list_head *next; 29515528f911STrond Myklebust int found = 0; 29525528f911STrond Myklebust 29535528f911STrond Myklebust repeat: 29546b41d536SAl Viro next = this_parent->mnt_mounts.next; 29555528f911STrond Myklebust resume: 29566b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 29575528f911STrond Myklebust struct list_head *tmp = next; 29586b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 29595528f911STrond Myklebust 29605528f911STrond Myklebust next = tmp->next; 2961692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 29625528f911STrond Myklebust continue; 29635528f911STrond Myklebust /* 29645528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 29655528f911STrond Myklebust */ 29666b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 29675528f911STrond Myklebust this_parent = mnt; 29685528f911STrond Myklebust goto repeat; 29695528f911STrond Myklebust } 29705528f911STrond Myklebust 29711ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 29726776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 29735528f911STrond Myklebust found++; 29745528f911STrond Myklebust } 29755528f911STrond Myklebust } 29765528f911STrond Myklebust /* 29775528f911STrond Myklebust * All done at this level ... ascend and resume the search 29785528f911STrond Myklebust */ 29795528f911STrond Myklebust if (this_parent != parent) { 29806b41d536SAl Viro next = this_parent->mnt_child.next; 29810714a533SAl Viro this_parent = this_parent->mnt_parent; 29825528f911STrond Myklebust goto resume; 29835528f911STrond Myklebust } 29845528f911STrond Myklebust return found; 29855528f911STrond Myklebust } 29865528f911STrond Myklebust 29875528f911STrond Myklebust /* 29885528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 29895528f911STrond Myklebust * submounts of a specific parent mountpoint 299099b7db7bSNick Piggin * 299148a066e7SAl Viro * mount_lock must be held for write 29925528f911STrond Myklebust */ 2993b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 29945528f911STrond Myklebust { 29955528f911STrond Myklebust LIST_HEAD(graveyard); 2996761d5c38SAl Viro struct mount *m; 29975528f911STrond Myklebust 29985528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2999c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3000bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3001761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 30026776db3dSAl Viro mnt_expire); 3003143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3004e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3005bcc5c7d2SAl Viro } 3006bcc5c7d2SAl Viro } 30075528f911STrond Myklebust } 30085528f911STrond Myklebust 3009b40ef869SAl Viro void *copy_mount_options(const void __user * data) 30101da177e4SLinus Torvalds { 3011b40ef869SAl Viro char *copy; 301212efec56SAl Viro unsigned size; 30131da177e4SLinus Torvalds 30141da177e4SLinus Torvalds if (!data) 3015b40ef869SAl Viro return NULL; 30161da177e4SLinus Torvalds 3017b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3018b40ef869SAl Viro if (!copy) 3019b40ef869SAl Viro return ERR_PTR(-ENOMEM); 30201da177e4SLinus Torvalds 302112efec56SAl Viro size = PAGE_SIZE - offset_in_page(data); 30221da177e4SLinus Torvalds 302312efec56SAl Viro if (copy_from_user(copy, data, size)) { 3024b40ef869SAl Viro kfree(copy); 3025b40ef869SAl Viro return ERR_PTR(-EFAULT); 30261da177e4SLinus Torvalds } 302712efec56SAl Viro if (size != PAGE_SIZE) { 302812efec56SAl Viro if (copy_from_user(copy + size, data + size, PAGE_SIZE - size)) 302912efec56SAl Viro memset(copy + size, 0, PAGE_SIZE - size); 303012efec56SAl Viro } 3031b40ef869SAl Viro return copy; 30321da177e4SLinus Torvalds } 30331da177e4SLinus Torvalds 3034b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 3035eca6f534SVegard Nossum { 3036fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3037eca6f534SVegard Nossum } 3038eca6f534SVegard Nossum 30391da177e4SLinus Torvalds /* 30401da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 30411da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 30421da177e4SLinus Torvalds * 30431da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 30441da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 30451da177e4SLinus Torvalds * information (or be NULL). 30461da177e4SLinus Torvalds * 30471da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 30481da177e4SLinus Torvalds * When the flags word was introduced its top half was required 30491da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 30501da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 30511da177e4SLinus Torvalds * and must be discarded. 30521da177e4SLinus Torvalds */ 30535e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 3054808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 30551da177e4SLinus Torvalds { 30562d92ab3cSAl Viro struct path path; 3057e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 30581da177e4SLinus Torvalds int retval = 0; 30591da177e4SLinus Torvalds 30601da177e4SLinus Torvalds /* Discard magic */ 30611da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 30621da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 30631da177e4SLinus Torvalds 30641da177e4SLinus Torvalds /* Basic sanity checks */ 30651da177e4SLinus Torvalds if (data_page) 30661da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 30671da177e4SLinus Torvalds 3068e462ec50SDavid Howells if (flags & MS_NOUSER) 3069e462ec50SDavid Howells return -EINVAL; 3070e462ec50SDavid Howells 3071a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 3072ce6595a2SAl Viro retval = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3073a27ab9f2STetsuo Handa if (retval) 3074a27ab9f2STetsuo Handa return retval; 3075a27ab9f2STetsuo Handa 3076a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 3077a27ab9f2STetsuo Handa type_page, flags, data_page); 30780d5cadb8SAl Viro if (!retval && !may_mount()) 30790d5cadb8SAl Viro retval = -EPERM; 3080e462ec50SDavid Howells if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) 30819e8925b6SJeff Layton retval = -EPERM; 3082a27ab9f2STetsuo Handa if (retval) 3083a27ab9f2STetsuo Handa goto dput_out; 3084a27ab9f2STetsuo Handa 3085613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3086613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 30870a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 30880a1c01c9SMatthew Garrett 30891da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 30901da177e4SLinus Torvalds if (flags & MS_NOSUID) 30911da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 30921da177e4SLinus Torvalds if (flags & MS_NODEV) 30931da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 30941da177e4SLinus Torvalds if (flags & MS_NOEXEC) 30951da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3096fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3097fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3098fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3099fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3100d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3101d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3102a9e5b732SDavid Howells if (flags & MS_RDONLY) 31032e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3104fc33a7bbSChristoph Hellwig 3105ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3106ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3107ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3108ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3109ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3110ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 3111ffbc6f0eSEric W. Biederman } 3112ffbc6f0eSEric W. Biederman 3113e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3114e462ec50SDavid Howells SB_SYNCHRONOUS | 3115e462ec50SDavid Howells SB_MANDLOCK | 3116e462ec50SDavid Howells SB_DIRSYNC | 3117e462ec50SDavid Howells SB_SILENT | 3118917086ffSMimi Zohar SB_POSIXACL | 3119d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3120917086ffSMimi Zohar SB_I_VERSION); 31211da177e4SLinus Torvalds 312243f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 312343f5e655SDavid Howells retval = do_reconfigure_mnt(&path, mnt_flags); 312443f5e655SDavid Howells else if (flags & MS_REMOUNT) 3125e462ec50SDavid Howells retval = do_remount(&path, flags, sb_flags, mnt_flags, 31261da177e4SLinus Torvalds data_page); 31271da177e4SLinus Torvalds else if (flags & MS_BIND) 31282d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 31299676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 31302d92ab3cSAl Viro retval = do_change_type(&path, flags); 31311da177e4SLinus Torvalds else if (flags & MS_MOVE) 31322db154b3SDavid Howells retval = do_move_mount_old(&path, dev_name); 31331da177e4SLinus Torvalds else 3134e462ec50SDavid Howells retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, 31351da177e4SLinus Torvalds dev_name, data_page); 31361da177e4SLinus Torvalds dput_out: 31372d92ab3cSAl Viro path_put(&path); 31381da177e4SLinus Torvalds return retval; 31391da177e4SLinus Torvalds } 31401da177e4SLinus Torvalds 3141537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3142537f7ccbSEric W. Biederman { 3143537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3144537f7ccbSEric W. Biederman } 3145537f7ccbSEric W. Biederman 3146537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3147537f7ccbSEric W. Biederman { 3148537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3149537f7ccbSEric W. Biederman } 3150537f7ccbSEric W. Biederman 3151771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3152771b1371SEric W. Biederman { 315374e83122SAl Viro if (!is_anon_ns(ns)) 31546344c433SAl Viro ns_free_inum(&ns->ns); 3155537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3156771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3157771b1371SEric W. Biederman kfree(ns); 3158771b1371SEric W. Biederman } 3159771b1371SEric W. Biederman 31608823c079SEric W. Biederman /* 31618823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 31628823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 31638823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 31648823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 31658823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 31668823c079SEric W. Biederman */ 31678823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 31688823c079SEric W. Biederman 316974e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3170cf8d2c11STrond Myklebust { 3171cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3172537f7ccbSEric W. Biederman struct ucounts *ucounts; 317398f842e6SEric W. Biederman int ret; 3174cf8d2c11STrond Myklebust 3175537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3176537f7ccbSEric W. Biederman if (!ucounts) 3177df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3178537f7ccbSEric W. Biederman 317974e83122SAl Viro new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 3180537f7ccbSEric W. Biederman if (!new_ns) { 3181537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3182cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3183537f7ccbSEric W. Biederman } 318474e83122SAl Viro if (!anon) { 31856344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 318698f842e6SEric W. Biederman if (ret) { 318798f842e6SEric W. Biederman kfree(new_ns); 3188537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 318998f842e6SEric W. Biederman return ERR_PTR(ret); 319098f842e6SEric W. Biederman } 319174e83122SAl Viro } 319233c42940SAl Viro new_ns->ns.ops = &mntns_operations; 319374e83122SAl Viro if (!anon) 31948823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 3195cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 3196cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3197cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 3198771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3199537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3200cf8d2c11STrond Myklebust return new_ns; 3201cf8d2c11STrond Myklebust } 3202cf8d2c11STrond Myklebust 32030766f788SEmese Revfy __latent_entropy 32049559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 32059559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 32061da177e4SLinus Torvalds { 32076b3286edSKirill Korotaev struct mnt_namespace *new_ns; 32087f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3209315fc83eSAl Viro struct mount *p, *q; 32109559f689SAl Viro struct mount *old; 3211cb338d06SAl Viro struct mount *new; 32127a472ef4SEric W. Biederman int copy_flags; 32131da177e4SLinus Torvalds 32149559f689SAl Viro BUG_ON(!ns); 32159559f689SAl Viro 32169559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 32179559f689SAl Viro get_mnt_ns(ns); 32189559f689SAl Viro return ns; 32199559f689SAl Viro } 32209559f689SAl Viro 32219559f689SAl Viro old = ns->root; 32229559f689SAl Viro 322374e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3224cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3225cf8d2c11STrond Myklebust return new_ns; 32261da177e4SLinus Torvalds 322797216be0SAl Viro namespace_lock(); 32281da177e4SLinus Torvalds /* First pass: copy the tree topology */ 32294ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 32309559f689SAl Viro if (user_ns != ns->user_ns) 32313bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 32327a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3233be34d1a3SDavid Howells if (IS_ERR(new)) { 3234328e6d90SAl Viro namespace_unlock(); 3235771b1371SEric W. Biederman free_mnt_ns(new_ns); 3236be34d1a3SDavid Howells return ERR_CAST(new); 32371da177e4SLinus Torvalds } 32383bd045ccSAl Viro if (user_ns != ns->user_ns) { 32393bd045ccSAl Viro lock_mount_hash(); 32403bd045ccSAl Viro lock_mnt_tree(new); 32413bd045ccSAl Viro unlock_mount_hash(); 32423bd045ccSAl Viro } 3243be08d6d2SAl Viro new_ns->root = new; 32441a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 32451da177e4SLinus Torvalds 32461da177e4SLinus Torvalds /* 32471da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 32481da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 32491da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 32501da177e4SLinus Torvalds */ 3251909b0a88SAl Viro p = old; 3252cb338d06SAl Viro q = new; 32531da177e4SLinus Torvalds while (p) { 3254143c8c91SAl Viro q->mnt_ns = new_ns; 3255d2921684SEric W. Biederman new_ns->mounts++; 32569559f689SAl Viro if (new_fs) { 32579559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 32589559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3259315fc83eSAl Viro rootmnt = &p->mnt; 32601da177e4SLinus Torvalds } 32619559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 32629559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3263315fc83eSAl Viro pwdmnt = &p->mnt; 32641da177e4SLinus Torvalds } 32651da177e4SLinus Torvalds } 3266909b0a88SAl Viro p = next_mnt(p, old); 3267909b0a88SAl Viro q = next_mnt(q, new); 32684ce5d2b1SEric W. Biederman if (!q) 32694ce5d2b1SEric W. Biederman break; 32704ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 32714ce5d2b1SEric W. Biederman p = next_mnt(p, old); 32721da177e4SLinus Torvalds } 3273328e6d90SAl Viro namespace_unlock(); 32741da177e4SLinus Torvalds 32751da177e4SLinus Torvalds if (rootmnt) 3276f03c6599SAl Viro mntput(rootmnt); 32771da177e4SLinus Torvalds if (pwdmnt) 3278f03c6599SAl Viro mntput(pwdmnt); 32791da177e4SLinus Torvalds 3280741a2951SJANAK DESAI return new_ns; 3281741a2951SJANAK DESAI } 3282741a2951SJANAK DESAI 328374e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3284cf8d2c11STrond Myklebust { 32851a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3286ea441d11SAl Viro struct mnt_namespace *ns; 3287d31da0f0SAl Viro struct super_block *s; 3288ea441d11SAl Viro struct path path; 3289ea441d11SAl Viro int err; 3290ea441d11SAl Viro 329174e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 329274e83122SAl Viro if (IS_ERR(ns)) { 329374e83122SAl Viro mntput(m); 3294ea441d11SAl Viro return ERR_CAST(ns); 329574e83122SAl Viro } 329674e83122SAl Viro mnt->mnt_ns = ns; 329774e83122SAl Viro ns->root = mnt; 329874e83122SAl Viro ns->mounts++; 329974e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3300ea441d11SAl Viro 330174e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3302ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3303ea441d11SAl Viro 3304ea441d11SAl Viro put_mnt_ns(ns); 3305ea441d11SAl Viro 3306ea441d11SAl Viro if (err) 3307ea441d11SAl Viro return ERR_PTR(err); 3308ea441d11SAl Viro 3309ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3310d31da0f0SAl Viro s = path.mnt->mnt_sb; 3311d31da0f0SAl Viro atomic_inc(&s->s_active); 3312ea441d11SAl Viro mntput(path.mnt); 3313ea441d11SAl Viro /* lock the sucker */ 3314d31da0f0SAl Viro down_write(&s->s_umount); 3315ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3316ea441d11SAl Viro return path.dentry; 3317ea441d11SAl Viro } 3318ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3319ea441d11SAl Viro 3320cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3321cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 33221da177e4SLinus Torvalds { 3323eca6f534SVegard Nossum int ret; 3324eca6f534SVegard Nossum char *kernel_type; 3325eca6f534SVegard Nossum char *kernel_dev; 3326b40ef869SAl Viro void *options; 33271da177e4SLinus Torvalds 3328b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3329b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3330b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3331eca6f534SVegard Nossum goto out_type; 33321da177e4SLinus Torvalds 3333b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3334b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3335b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3336eca6f534SVegard Nossum goto out_dev; 33371da177e4SLinus Torvalds 3338b40ef869SAl Viro options = copy_mount_options(data); 3339b40ef869SAl Viro ret = PTR_ERR(options); 3340b40ef869SAl Viro if (IS_ERR(options)) 3341eca6f534SVegard Nossum goto out_data; 33421da177e4SLinus Torvalds 3343b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3344eca6f534SVegard Nossum 3345b40ef869SAl Viro kfree(options); 3346eca6f534SVegard Nossum out_data: 3347eca6f534SVegard Nossum kfree(kernel_dev); 3348eca6f534SVegard Nossum out_dev: 3349eca6f534SVegard Nossum kfree(kernel_type); 3350eca6f534SVegard Nossum out_type: 3351eca6f534SVegard Nossum return ret; 33521da177e4SLinus Torvalds } 33531da177e4SLinus Torvalds 33541da177e4SLinus Torvalds /* 335593766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 335693766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 335793766fbdSDavid Howells */ 335893766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 335993766fbdSDavid Howells unsigned int, attr_flags) 336093766fbdSDavid Howells { 336193766fbdSDavid Howells struct mnt_namespace *ns; 336293766fbdSDavid Howells struct fs_context *fc; 336393766fbdSDavid Howells struct file *file; 336493766fbdSDavid Howells struct path newmount; 336593766fbdSDavid Howells struct mount *mnt; 336693766fbdSDavid Howells struct fd f; 336793766fbdSDavid Howells unsigned int mnt_flags = 0; 336893766fbdSDavid Howells long ret; 336993766fbdSDavid Howells 337093766fbdSDavid Howells if (!may_mount()) 337193766fbdSDavid Howells return -EPERM; 337293766fbdSDavid Howells 337393766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 337493766fbdSDavid Howells return -EINVAL; 337593766fbdSDavid Howells 337693766fbdSDavid Howells if (attr_flags & ~(MOUNT_ATTR_RDONLY | 337793766fbdSDavid Howells MOUNT_ATTR_NOSUID | 337893766fbdSDavid Howells MOUNT_ATTR_NODEV | 337993766fbdSDavid Howells MOUNT_ATTR_NOEXEC | 338093766fbdSDavid Howells MOUNT_ATTR__ATIME | 338193766fbdSDavid Howells MOUNT_ATTR_NODIRATIME)) 338293766fbdSDavid Howells return -EINVAL; 338393766fbdSDavid Howells 338493766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_RDONLY) 338593766fbdSDavid Howells mnt_flags |= MNT_READONLY; 338693766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NOSUID) 338793766fbdSDavid Howells mnt_flags |= MNT_NOSUID; 338893766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NODEV) 338993766fbdSDavid Howells mnt_flags |= MNT_NODEV; 339093766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NOEXEC) 339193766fbdSDavid Howells mnt_flags |= MNT_NOEXEC; 339293766fbdSDavid Howells if (attr_flags & MOUNT_ATTR_NODIRATIME) 339393766fbdSDavid Howells mnt_flags |= MNT_NODIRATIME; 339493766fbdSDavid Howells 339593766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 339693766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 339793766fbdSDavid Howells break; 339893766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 339993766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 340093766fbdSDavid Howells break; 340193766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 340293766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 340393766fbdSDavid Howells break; 340493766fbdSDavid Howells default: 340593766fbdSDavid Howells return -EINVAL; 340693766fbdSDavid Howells } 340793766fbdSDavid Howells 340893766fbdSDavid Howells f = fdget(fs_fd); 340993766fbdSDavid Howells if (!f.file) 341093766fbdSDavid Howells return -EBADF; 341193766fbdSDavid Howells 341293766fbdSDavid Howells ret = -EINVAL; 341393766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 341493766fbdSDavid Howells goto err_fsfd; 341593766fbdSDavid Howells 341693766fbdSDavid Howells fc = f.file->private_data; 341793766fbdSDavid Howells 341893766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 341993766fbdSDavid Howells if (ret < 0) 342093766fbdSDavid Howells goto err_fsfd; 342193766fbdSDavid Howells 342293766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 342393766fbdSDavid Howells ret = -EINVAL; 342493766fbdSDavid Howells if (!fc->root) 342593766fbdSDavid Howells goto err_unlock; 342693766fbdSDavid Howells 342793766fbdSDavid Howells ret = -EPERM; 342893766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 342993766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 343093766fbdSDavid Howells goto err_unlock; 343193766fbdSDavid Howells } 343293766fbdSDavid Howells 343393766fbdSDavid Howells ret = -EBUSY; 343493766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 343593766fbdSDavid Howells goto err_unlock; 343693766fbdSDavid Howells 343793766fbdSDavid Howells ret = -EPERM; 343893766fbdSDavid Howells if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock()) 343993766fbdSDavid Howells goto err_unlock; 344093766fbdSDavid Howells 344193766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 344293766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 344393766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 344493766fbdSDavid Howells goto err_unlock; 344593766fbdSDavid Howells } 344693766fbdSDavid Howells newmount.dentry = dget(fc->root); 344793766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 344893766fbdSDavid Howells 344993766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 345093766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 345193766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 345293766fbdSDavid Howells * don't want to have to handle any errors incurred. 345393766fbdSDavid Howells */ 345493766fbdSDavid Howells vfs_clean_context(fc); 345593766fbdSDavid Howells 345693766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 345793766fbdSDavid Howells if (IS_ERR(ns)) { 345893766fbdSDavid Howells ret = PTR_ERR(ns); 345993766fbdSDavid Howells goto err_path; 346093766fbdSDavid Howells } 346193766fbdSDavid Howells mnt = real_mount(newmount.mnt); 346293766fbdSDavid Howells mnt->mnt_ns = ns; 346393766fbdSDavid Howells ns->root = mnt; 346493766fbdSDavid Howells ns->mounts = 1; 346593766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 34661b0b9cc8SEric Biggers mntget(newmount.mnt); 346793766fbdSDavid Howells 346893766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 346993766fbdSDavid Howells * it, not just simply put it. 347093766fbdSDavid Howells */ 347193766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 347293766fbdSDavid Howells if (IS_ERR(file)) { 347393766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 347493766fbdSDavid Howells ret = PTR_ERR(file); 347593766fbdSDavid Howells goto err_path; 347693766fbdSDavid Howells } 347793766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 347893766fbdSDavid Howells 347993766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 348093766fbdSDavid Howells if (ret >= 0) 348193766fbdSDavid Howells fd_install(ret, file); 348293766fbdSDavid Howells else 348393766fbdSDavid Howells fput(file); 348493766fbdSDavid Howells 348593766fbdSDavid Howells err_path: 348693766fbdSDavid Howells path_put(&newmount); 348793766fbdSDavid Howells err_unlock: 348893766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 348993766fbdSDavid Howells err_fsfd: 349093766fbdSDavid Howells fdput(f); 349193766fbdSDavid Howells return ret; 349293766fbdSDavid Howells } 349393766fbdSDavid Howells 349493766fbdSDavid Howells /* 349593766fbdSDavid Howells * Move a mount from one place to another. In combination with 349693766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 349793766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 349893766fbdSDavid Howells * a mount subtree. 34992db154b3SDavid Howells * 35002db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 35012db154b3SDavid Howells */ 35022db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 35032658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 35042658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 35052db154b3SDavid Howells unsigned int, flags) 35062db154b3SDavid Howells { 35072db154b3SDavid Howells struct path from_path, to_path; 35082db154b3SDavid Howells unsigned int lflags; 35092db154b3SDavid Howells int ret = 0; 35102db154b3SDavid Howells 35112db154b3SDavid Howells if (!may_mount()) 35122db154b3SDavid Howells return -EPERM; 35132db154b3SDavid Howells 35142db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 35152db154b3SDavid Howells return -EINVAL; 35162db154b3SDavid Howells 35172db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 35182db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 35192db154b3SDavid Howells * to clear it afterwards. 35202db154b3SDavid Howells */ 35212db154b3SDavid Howells lflags = 0; 35222db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 35232db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 35242db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 35252db154b3SDavid Howells 35262db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 35272db154b3SDavid Howells if (ret < 0) 35282db154b3SDavid Howells return ret; 35292db154b3SDavid Howells 35302db154b3SDavid Howells lflags = 0; 35312db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 35322db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 35332db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 35342db154b3SDavid Howells 35352db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 35362db154b3SDavid Howells if (ret < 0) 35372db154b3SDavid Howells goto out_from; 35382db154b3SDavid Howells 35392db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 35402db154b3SDavid Howells if (ret < 0) 35412db154b3SDavid Howells goto out_to; 35422db154b3SDavid Howells 35432db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 35442db154b3SDavid Howells 35452db154b3SDavid Howells out_to: 35462db154b3SDavid Howells path_put(&to_path); 35472db154b3SDavid Howells out_from: 35482db154b3SDavid Howells path_put(&from_path); 35492db154b3SDavid Howells return ret; 35502db154b3SDavid Howells } 35512db154b3SDavid Howells 35522db154b3SDavid Howells /* 3553afac7cbaSAl Viro * Return true if path is reachable from root 3554afac7cbaSAl Viro * 355548a066e7SAl Viro * namespace_sem or mount_lock is held 3556afac7cbaSAl Viro */ 3557643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3558afac7cbaSAl Viro const struct path *root) 3559afac7cbaSAl Viro { 3560643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3561a73324daSAl Viro dentry = mnt->mnt_mountpoint; 35620714a533SAl Viro mnt = mnt->mnt_parent; 3563afac7cbaSAl Viro } 3564643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3565afac7cbaSAl Viro } 3566afac7cbaSAl Viro 3567640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3568afac7cbaSAl Viro { 356925ab4c9bSYaowei Bai bool res; 357048a066e7SAl Viro read_seqlock_excl(&mount_lock); 3571643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 357248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3573afac7cbaSAl Viro return res; 3574afac7cbaSAl Viro } 3575afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3576afac7cbaSAl Viro 3577afac7cbaSAl Viro /* 35781da177e4SLinus Torvalds * pivot_root Semantics: 35791da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 35801da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 35811da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 35821da177e4SLinus Torvalds * 35831da177e4SLinus Torvalds * Restrictions: 35841da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 35851da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 35861da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 35871da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 35881da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 35891da177e4SLinus Torvalds * 35904a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 35914a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 35924a0d11faSNeil Brown * in this situation. 35934a0d11faSNeil Brown * 35941da177e4SLinus Torvalds * Notes: 35951da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 35961da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 35971da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 35981da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 35991da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 36001da177e4SLinus Torvalds * first. 36011da177e4SLinus Torvalds */ 36023480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 36033480b257SHeiko Carstens const char __user *, put_old) 36041da177e4SLinus Torvalds { 36052763d119SAl Viro struct path new, old, root; 36062763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 360784d17192SAl Viro struct mountpoint *old_mp, *root_mp; 36081da177e4SLinus Torvalds int error; 36091da177e4SLinus Torvalds 36109b40bc90SAl Viro if (!may_mount()) 36111da177e4SLinus Torvalds return -EPERM; 36121da177e4SLinus Torvalds 3613ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3614ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 36151da177e4SLinus Torvalds if (error) 36161da177e4SLinus Torvalds goto out0; 36171da177e4SLinus Torvalds 3618ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 3619ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 36201da177e4SLinus Torvalds if (error) 36211da177e4SLinus Torvalds goto out1; 36221da177e4SLinus Torvalds 36232d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3624b12cea91SAl Viro if (error) 3625b12cea91SAl Viro goto out2; 36261da177e4SLinus Torvalds 3627f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 362884d17192SAl Viro old_mp = lock_mount(&old); 362984d17192SAl Viro error = PTR_ERR(old_mp); 363084d17192SAl Viro if (IS_ERR(old_mp)) 3631b12cea91SAl Viro goto out3; 3632b12cea91SAl Viro 36331da177e4SLinus Torvalds error = -EINVAL; 3634419148daSAl Viro new_mnt = real_mount(new.mnt); 3635419148daSAl Viro root_mnt = real_mount(root.mnt); 363684d17192SAl Viro old_mnt = real_mount(old.mnt); 36372763d119SAl Viro ex_parent = new_mnt->mnt_parent; 36382763d119SAl Viro root_parent = root_mnt->mnt_parent; 363984d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 36402763d119SAl Viro IS_MNT_SHARED(ex_parent) || 36412763d119SAl Viro IS_MNT_SHARED(root_parent)) 3642b12cea91SAl Viro goto out4; 3643143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3644b12cea91SAl Viro goto out4; 36455ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 36465ff9d8a6SEric W. Biederman goto out4; 36471da177e4SLinus Torvalds error = -ENOENT; 3648f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3649b12cea91SAl Viro goto out4; 36501da177e4SLinus Torvalds error = -EBUSY; 365184d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3652b12cea91SAl Viro goto out4; /* loop, on the same file system */ 36531da177e4SLinus Torvalds error = -EINVAL; 36548c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3655b12cea91SAl Viro goto out4; /* not a mountpoint */ 3656676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3657b12cea91SAl Viro goto out4; /* not attached */ 36582d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3659b12cea91SAl Viro goto out4; /* not a mountpoint */ 3660676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3661b12cea91SAl Viro goto out4; /* not attached */ 36624ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 366384d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3664b12cea91SAl Viro goto out4; 36650d082601SEric W. Biederman /* make certain new is below the root */ 36660d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 36670d082601SEric W. Biederman goto out4; 3668719ea2fbSAl Viro lock_mount_hash(); 36692763d119SAl Viro umount_mnt(new_mnt); 36702763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 36715ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 36725ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 36735ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 36745ff9d8a6SEric W. Biederman } 36754ac91378SJan Blunck /* mount old root on put_old */ 367684d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 36774ac91378SJan Blunck /* mount new_root on / */ 36782763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 36792763d119SAl Viro mnt_add_count(root_parent, -1); 36806b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 36814fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 36824fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 36833895dbf8SEric W. Biederman put_mountpoint(root_mp); 3684719ea2fbSAl Viro unlock_mount_hash(); 36852d8f3038SAl Viro chroot_fs_refs(&root, &new); 36861da177e4SLinus Torvalds error = 0; 3687b12cea91SAl Viro out4: 368884d17192SAl Viro unlock_mount(old_mp); 36892763d119SAl Viro if (!error) 36902763d119SAl Viro mntput_no_expire(ex_parent); 3691b12cea91SAl Viro out3: 36928c3ee42eSAl Viro path_put(&root); 3693b12cea91SAl Viro out2: 36942d8f3038SAl Viro path_put(&old); 36951da177e4SLinus Torvalds out1: 36962d8f3038SAl Viro path_put(&new); 36971da177e4SLinus Torvalds out0: 36981da177e4SLinus Torvalds return error; 36991da177e4SLinus Torvalds } 37001da177e4SLinus Torvalds 37011da177e4SLinus Torvalds static void __init init_mount_tree(void) 37021da177e4SLinus Torvalds { 37031da177e4SLinus Torvalds struct vfsmount *mnt; 370474e83122SAl Viro struct mount *m; 37056b3286edSKirill Korotaev struct mnt_namespace *ns; 3706ac748a09SJan Blunck struct path root; 37071da177e4SLinus Torvalds 3708fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 37091da177e4SLinus Torvalds if (IS_ERR(mnt)) 37101da177e4SLinus Torvalds panic("Can't create rootfs"); 3711b3e19d92SNick Piggin 371274e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 37133b22edc5STrond Myklebust if (IS_ERR(ns)) 37141da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 371574e83122SAl Viro m = real_mount(mnt); 371674e83122SAl Viro m->mnt_ns = ns; 371774e83122SAl Viro ns->root = m; 371874e83122SAl Viro ns->mounts = 1; 371974e83122SAl Viro list_add(&m->mnt_list, &ns->list); 37206b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 37216b3286edSKirill Korotaev get_mnt_ns(ns); 37221da177e4SLinus Torvalds 3723be08d6d2SAl Viro root.mnt = mnt; 3724be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3725da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3726ac748a09SJan Blunck 3727ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3728ac748a09SJan Blunck set_fs_root(current->fs, &root); 37291da177e4SLinus Torvalds } 37301da177e4SLinus Torvalds 373174bf17cfSDenis Cheng void __init mnt_init(void) 37321da177e4SLinus Torvalds { 373315a67dd8SRandy Dunlap int err; 37341da177e4SLinus Torvalds 37357d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 373620c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 37371da177e4SLinus Torvalds 37380818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 373938129a13SAl Viro sizeof(struct hlist_head), 37400818bf27SAl Viro mhash_entries, 19, 37413d375d78SPavel Tatashin HASH_ZERO, 37420818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 37430818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 37440818bf27SAl Viro sizeof(struct hlist_head), 37450818bf27SAl Viro mphash_entries, 19, 37463d375d78SPavel Tatashin HASH_ZERO, 37470818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 37481da177e4SLinus Torvalds 374984d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 37501da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 37511da177e4SLinus Torvalds 37524b93dc9bSTejun Heo kernfs_init(); 37534b93dc9bSTejun Heo 375415a67dd8SRandy Dunlap err = sysfs_init(); 375515a67dd8SRandy Dunlap if (err) 375615a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 37578e24eea7SHarvey Harrison __func__, err); 375800d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 375900d26666SGreg Kroah-Hartman if (!fs_kobj) 37608e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 3761037f11b4SAl Viro shmem_init(); 37621da177e4SLinus Torvalds init_rootfs(); 37631da177e4SLinus Torvalds init_mount_tree(); 37641da177e4SLinus Torvalds } 37651da177e4SLinus Torvalds 3766616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 37671da177e4SLinus Torvalds { 3768d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3769616511d0STrond Myklebust return; 37707b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3771771b1371SEric W. Biederman free_mnt_ns(ns); 37721da177e4SLinus Torvalds } 37739d412a43SAl Viro 3774d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 37759d412a43SAl Viro { 3776423e0ab0STim Chen struct vfsmount *mnt; 3777d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 3778423e0ab0STim Chen if (!IS_ERR(mnt)) { 3779423e0ab0STim Chen /* 3780423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3781423e0ab0STim Chen * we unmount before file sys is unregistered 3782423e0ab0STim Chen */ 3783f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3784423e0ab0STim Chen } 3785423e0ab0STim Chen return mnt; 37869d412a43SAl Viro } 3787d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 3788423e0ab0STim Chen 3789423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3790423e0ab0STim Chen { 3791423e0ab0STim Chen /* release long term mount so mount point can be released */ 3792423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3793f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 379448a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3795423e0ab0STim Chen mntput(mnt); 3796423e0ab0STim Chen } 3797423e0ab0STim Chen } 3798423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 379902125a82SAl Viro 380002125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 380102125a82SAl Viro { 3802143c8c91SAl Viro return check_mnt(real_mount(mnt)); 380302125a82SAl Viro } 38048823c079SEric W. Biederman 38053151527eSEric W. Biederman bool current_chrooted(void) 38063151527eSEric W. Biederman { 38073151527eSEric W. Biederman /* Does the current process have a non-standard root */ 38083151527eSEric W. Biederman struct path ns_root; 38093151527eSEric W. Biederman struct path fs_root; 38103151527eSEric W. Biederman bool chrooted; 38113151527eSEric W. Biederman 38123151527eSEric W. Biederman /* Find the namespace root */ 38133151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 38143151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 38153151527eSEric W. Biederman path_get(&ns_root); 38163151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 38173151527eSEric W. Biederman ; 38183151527eSEric W. Biederman 38193151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 38203151527eSEric W. Biederman 38213151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 38223151527eSEric W. Biederman 38233151527eSEric W. Biederman path_put(&fs_root); 38243151527eSEric W. Biederman path_put(&ns_root); 38253151527eSEric W. Biederman 38263151527eSEric W. Biederman return chrooted; 38273151527eSEric W. Biederman } 38283151527eSEric W. Biederman 3829132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 3830132e4608SDavid Howells const struct super_block *sb, 38318654df4eSEric W. Biederman int *new_mnt_flags) 383287a8ebd6SEric W. Biederman { 38338c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 383487a8ebd6SEric W. Biederman struct mount *mnt; 3835e51db735SEric W. Biederman bool visible = false; 383687a8ebd6SEric W. Biederman 383744bb4385SAl Viro down_read(&namespace_sem); 383887a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3839e51db735SEric W. Biederman struct mount *child; 384077b1a97dSEric W. Biederman int mnt_flags; 384177b1a97dSEric W. Biederman 3842132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 3843e51db735SEric W. Biederman continue; 3844e51db735SEric W. Biederman 38457e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 38467e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 38477e96c1b0SEric W. Biederman */ 38487e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 38497e96c1b0SEric W. Biederman continue; 38507e96c1b0SEric W. Biederman 3851a1935c17SEric W. Biederman /* A local view of the mount flags */ 385277b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 385377b1a97dSEric W. Biederman 3854695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3855bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 3856695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3857695e9df0SEric W. Biederman 38588c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 38598c6cf9ccSEric W. Biederman * than the proposed new mount. 38608c6cf9ccSEric W. Biederman */ 386177b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 38628c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 38638c6cf9ccSEric W. Biederman continue; 386477b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 386577b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 38668c6cf9ccSEric W. Biederman continue; 38678c6cf9ccSEric W. Biederman 3868ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3869ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3870ceeb0e5dSEric W. Biederman * empty directories. 3871e51db735SEric W. Biederman */ 3872e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3873e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3874ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3875d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3876ceeb0e5dSEric W. Biederman continue; 38777236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 38787236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3879e51db735SEric W. Biederman goto next; 388087a8ebd6SEric W. Biederman } 38818c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 388277b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 38838c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3884e51db735SEric W. Biederman visible = true; 3885e51db735SEric W. Biederman goto found; 3886e51db735SEric W. Biederman next: ; 388787a8ebd6SEric W. Biederman } 3888e51db735SEric W. Biederman found: 388944bb4385SAl Viro up_read(&namespace_sem); 3890e51db735SEric W. Biederman return visible; 389187a8ebd6SEric W. Biederman } 389287a8ebd6SEric W. Biederman 3893132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 38948654df4eSEric W. Biederman { 3895a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 38968654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 38978654df4eSEric W. Biederman unsigned long s_iflags; 38988654df4eSEric W. Biederman 38998654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 39008654df4eSEric W. Biederman return false; 39018654df4eSEric W. Biederman 39028654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 3903132e4608SDavid Howells s_iflags = sb->s_iflags; 39048654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 39058654df4eSEric W. Biederman return false; 39068654df4eSEric W. Biederman 3907a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3908a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3909a1935c17SEric W. Biederman required_iflags); 3910a1935c17SEric W. Biederman return true; 3911a1935c17SEric W. Biederman } 3912a1935c17SEric W. Biederman 3913132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 39148654df4eSEric W. Biederman } 39158654df4eSEric W. Biederman 3916380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3917380cf5baSAndy Lutomirski { 3918380cf5baSAndy Lutomirski /* 3919380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3920380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3921380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3922380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3923380cf5baSAndy Lutomirski * in other namespaces. 3924380cf5baSAndy Lutomirski */ 3925380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3926380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3927380cf5baSAndy Lutomirski } 3928380cf5baSAndy Lutomirski 392964964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 39308823c079SEric W. Biederman { 393158be2825SAl Viro struct ns_common *ns = NULL; 39328823c079SEric W. Biederman struct nsproxy *nsproxy; 39338823c079SEric W. Biederman 3934728dba3aSEric W. Biederman task_lock(task); 3935728dba3aSEric W. Biederman nsproxy = task->nsproxy; 39368823c079SEric W. Biederman if (nsproxy) { 393758be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 393858be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 39398823c079SEric W. Biederman } 3940728dba3aSEric W. Biederman task_unlock(task); 39418823c079SEric W. Biederman 39428823c079SEric W. Biederman return ns; 39438823c079SEric W. Biederman } 39448823c079SEric W. Biederman 394564964528SAl Viro static void mntns_put(struct ns_common *ns) 39468823c079SEric W. Biederman { 394758be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 39488823c079SEric W. Biederman } 39498823c079SEric W. Biederman 395064964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 39518823c079SEric W. Biederman { 39528823c079SEric W. Biederman struct fs_struct *fs = current->fs; 39534f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 39548823c079SEric W. Biederman struct path root; 39554f757f3cSAl Viro int err; 39568823c079SEric W. Biederman 39570c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3958c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3959c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3960ae11e0f1SZhao Hongjiang return -EPERM; 39618823c079SEric W. Biederman 396274e83122SAl Viro if (is_anon_ns(mnt_ns)) 396374e83122SAl Viro return -EINVAL; 396474e83122SAl Viro 39658823c079SEric W. Biederman if (fs->users != 1) 39668823c079SEric W. Biederman return -EINVAL; 39678823c079SEric W. Biederman 39688823c079SEric W. Biederman get_mnt_ns(mnt_ns); 39694f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 39708823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 39718823c079SEric W. Biederman 39728823c079SEric W. Biederman /* Find the root */ 39734f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 39744f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 39754f757f3cSAl Viro if (err) { 39764f757f3cSAl Viro /* revert to old namespace */ 39774f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 39784f757f3cSAl Viro put_mnt_ns(mnt_ns); 39794f757f3cSAl Viro return err; 39804f757f3cSAl Viro } 39818823c079SEric W. Biederman 39824068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 39834068367cSAndrei Vagin 39848823c079SEric W. Biederman /* Update the pwd and root */ 39858823c079SEric W. Biederman set_fs_pwd(fs, &root); 39868823c079SEric W. Biederman set_fs_root(fs, &root); 39878823c079SEric W. Biederman 39888823c079SEric W. Biederman path_put(&root); 39898823c079SEric W. Biederman return 0; 39908823c079SEric W. Biederman } 39918823c079SEric W. Biederman 3992bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3993bcac25a5SAndrey Vagin { 3994bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3995bcac25a5SAndrey Vagin } 3996bcac25a5SAndrey Vagin 39978823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 39988823c079SEric W. Biederman .name = "mnt", 39998823c079SEric W. Biederman .type = CLONE_NEWNS, 40008823c079SEric W. Biederman .get = mntns_get, 40018823c079SEric W. Biederman .put = mntns_put, 40028823c079SEric W. Biederman .install = mntns_install, 4003bcac25a5SAndrey Vagin .owner = mntns_owner, 40048823c079SEric W. Biederman }; 4005