11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 1873cd49ecSMiklos Szeredi #include <linux/idr.h> 1957f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 20d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 21d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 22d10577a8SAl Viro #include <linux/uaccess.h> 230bb80f24SDavid Howells #include <linux/proc_ns.h> 2420b4fb48SLinus Torvalds #include <linux/magic.h> 250818bf27SAl Viro #include <linux/bootmem.h> 269ea459e1SAl Viro #include <linux/task_work.h> 2707b20889SRam Pai #include "pnode.h" 28948730b0SAdrian Bunk #include "internal.h" 291da177e4SLinus Torvalds 30d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 31d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 32d2921684SEric W. Biederman 330818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 340818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 350818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 360818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 370818bf27SAl Viro 380818bf27SAl Viro static __initdata unsigned long mhash_entries; 390818bf27SAl Viro static int __init set_mhash_entries(char *str) 400818bf27SAl Viro { 410818bf27SAl Viro if (!str) 420818bf27SAl Viro return 0; 430818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 440818bf27SAl Viro return 1; 450818bf27SAl Viro } 460818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 470818bf27SAl Viro 480818bf27SAl Viro static __initdata unsigned long mphash_entries; 490818bf27SAl Viro static int __init set_mphash_entries(char *str) 500818bf27SAl Viro { 510818bf27SAl Viro if (!str) 520818bf27SAl Viro return 0; 530818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 540818bf27SAl Viro return 1; 550818bf27SAl Viro } 560818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 5713f14b4dSEric Dumazet 58c7999c36SAl Viro static u64 event; 5973cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 60719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 6199b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 62f21f6220SAl Viro static int mnt_id_start = 0; 63f21f6220SAl Viro static int mnt_group_start = 1; 645addc5ddSAl Viro 6538129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 660818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 67e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 6859aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 691da177e4SLinus Torvalds 70f87fd4c2SMiklos Szeredi /* /sys/fs */ 7100d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7200d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 73f87fd4c2SMiklos Szeredi 7499b7db7bSNick Piggin /* 7599b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7699b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 7799b7db7bSNick Piggin * up the tree. 7899b7db7bSNick Piggin * 7999b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8099b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8199b7db7bSNick Piggin */ 8248a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8399b7db7bSNick Piggin 8438129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 851da177e4SLinus Torvalds { 861da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 871da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 880818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 890818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 900818bf27SAl Viro } 910818bf27SAl Viro 920818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 930818bf27SAl Viro { 940818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 950818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 960818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 971da177e4SLinus Torvalds } 981da177e4SLinus Torvalds 9999b7db7bSNick Piggin /* 10099b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 10199b7db7bSNick Piggin * serialize with freeing. 10299b7db7bSNick Piggin */ 103b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10473cd49ecSMiklos Szeredi { 10573cd49ecSMiklos Szeredi int res; 10673cd49ecSMiklos Szeredi 10773cd49ecSMiklos Szeredi retry: 10873cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 10999b7db7bSNick Piggin spin_lock(&mnt_id_lock); 11015169fe7SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 111f21f6220SAl Viro if (!res) 11215169fe7SAl Viro mnt_id_start = mnt->mnt_id + 1; 11399b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 11473cd49ecSMiklos Szeredi if (res == -EAGAIN) 11573cd49ecSMiklos Szeredi goto retry; 11673cd49ecSMiklos Szeredi 11773cd49ecSMiklos Szeredi return res; 11873cd49ecSMiklos Szeredi } 11973cd49ecSMiklos Szeredi 120b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 12173cd49ecSMiklos Szeredi { 12215169fe7SAl Viro int id = mnt->mnt_id; 12399b7db7bSNick Piggin spin_lock(&mnt_id_lock); 124f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 125f21f6220SAl Viro if (mnt_id_start > id) 126f21f6220SAl Viro mnt_id_start = id; 12799b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 12873cd49ecSMiklos Szeredi } 12973cd49ecSMiklos Szeredi 130719f5d7fSMiklos Szeredi /* 131719f5d7fSMiklos Szeredi * Allocate a new peer group ID 132719f5d7fSMiklos Szeredi * 133719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 134719f5d7fSMiklos Szeredi */ 1354b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 136719f5d7fSMiklos Szeredi { 137f21f6220SAl Viro int res; 138f21f6220SAl Viro 139719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 140719f5d7fSMiklos Szeredi return -ENOMEM; 141719f5d7fSMiklos Szeredi 142f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 143f21f6220SAl Viro mnt_group_start, 14415169fe7SAl Viro &mnt->mnt_group_id); 145f21f6220SAl Viro if (!res) 14615169fe7SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 147f21f6220SAl Viro 148f21f6220SAl Viro return res; 149719f5d7fSMiklos Szeredi } 150719f5d7fSMiklos Szeredi 151719f5d7fSMiklos Szeredi /* 152719f5d7fSMiklos Szeredi * Release a peer group ID 153719f5d7fSMiklos Szeredi */ 1544b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 155719f5d7fSMiklos Szeredi { 15615169fe7SAl Viro int id = mnt->mnt_group_id; 157f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 158f21f6220SAl Viro if (mnt_group_start > id) 159f21f6220SAl Viro mnt_group_start = id; 16015169fe7SAl Viro mnt->mnt_group_id = 0; 161719f5d7fSMiklos Szeredi } 162719f5d7fSMiklos Szeredi 163b3e19d92SNick Piggin /* 164b3e19d92SNick Piggin * vfsmount lock must be held for read 165b3e19d92SNick Piggin */ 16683adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 167b3e19d92SNick Piggin { 168b3e19d92SNick Piggin #ifdef CONFIG_SMP 16968e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 170b3e19d92SNick Piggin #else 171b3e19d92SNick Piggin preempt_disable(); 17268e8a9feSAl Viro mnt->mnt_count += n; 173b3e19d92SNick Piggin preempt_enable(); 174b3e19d92SNick Piggin #endif 175b3e19d92SNick Piggin } 176b3e19d92SNick Piggin 177b3e19d92SNick Piggin /* 178b3e19d92SNick Piggin * vfsmount lock must be held for write 179b3e19d92SNick Piggin */ 18083adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 181b3e19d92SNick Piggin { 182b3e19d92SNick Piggin #ifdef CONFIG_SMP 183f03c6599SAl Viro unsigned int count = 0; 184b3e19d92SNick Piggin int cpu; 185b3e19d92SNick Piggin 186b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 18768e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 188b3e19d92SNick Piggin } 189b3e19d92SNick Piggin 190b3e19d92SNick Piggin return count; 191b3e19d92SNick Piggin #else 19268e8a9feSAl Viro return mnt->mnt_count; 193b3e19d92SNick Piggin #endif 194b3e19d92SNick Piggin } 195b3e19d92SNick Piggin 19687b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 19787b95ce0SAl Viro { 19887b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 19987b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 20087b95ce0SAl Viro pin_remove(p); 20187b95ce0SAl Viro mntput(&m->mnt); 20287b95ce0SAl Viro } 20387b95ce0SAl Viro 204b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 2051da177e4SLinus Torvalds { 206c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 207c63181e6SAl Viro if (mnt) { 20873cd49ecSMiklos Szeredi int err; 20973cd49ecSMiklos Szeredi 210c63181e6SAl Viro err = mnt_alloc_id(mnt); 21188b38782SLi Zefan if (err) 21288b38782SLi Zefan goto out_free_cache; 21388b38782SLi Zefan 21488b38782SLi Zefan if (name) { 215fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 216c63181e6SAl Viro if (!mnt->mnt_devname) 21788b38782SLi Zefan goto out_free_id; 21873cd49ecSMiklos Szeredi } 21973cd49ecSMiklos Szeredi 220b3e19d92SNick Piggin #ifdef CONFIG_SMP 221c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 222c63181e6SAl Viro if (!mnt->mnt_pcp) 223b3e19d92SNick Piggin goto out_free_devname; 224b3e19d92SNick Piggin 225c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 226b3e19d92SNick Piggin #else 227c63181e6SAl Viro mnt->mnt_count = 1; 228c63181e6SAl Viro mnt->mnt_writers = 0; 229b3e19d92SNick Piggin #endif 230b3e19d92SNick Piggin 23138129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 232c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 233c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 234c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 235c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 236c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 237c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 238c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2390a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 2402504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 2412504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 2422504c5d6SAndreas Gruenbacher #endif 24387b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2441da177e4SLinus Torvalds } 245c63181e6SAl Viro return mnt; 24688b38782SLi Zefan 247d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 248d3ef3d73Snpiggin@suse.de out_free_devname: 249fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 250d3ef3d73Snpiggin@suse.de #endif 25188b38782SLi Zefan out_free_id: 252c63181e6SAl Viro mnt_free_id(mnt); 25388b38782SLi Zefan out_free_cache: 254c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 25588b38782SLi Zefan return NULL; 2561da177e4SLinus Torvalds } 2571da177e4SLinus Torvalds 2588366025eSDave Hansen /* 2598366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2608366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2618366025eSDave Hansen * We must keep track of when those operations start 2628366025eSDave Hansen * (for permission checks) and when they end, so that 2638366025eSDave Hansen * we can determine when writes are able to occur to 2648366025eSDave Hansen * a filesystem. 2658366025eSDave Hansen */ 2663d733633SDave Hansen /* 2673d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2683d733633SDave Hansen * @mnt: the mount to check for its write status 2693d733633SDave Hansen * 2703d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2713d733633SDave Hansen * It does not guarantee that the filesystem will stay 2723d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2733d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2743d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2753d733633SDave Hansen * r/w. 2763d733633SDave Hansen */ 2773d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2783d733633SDave Hansen { 2792e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2802e4b7fcdSDave Hansen return 1; 2812e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2822e4b7fcdSDave Hansen return 1; 2832e4b7fcdSDave Hansen return 0; 2843d733633SDave Hansen } 2853d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2863d733633SDave Hansen 28783adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2883d733633SDave Hansen { 289d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 29068e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 291d3ef3d73Snpiggin@suse.de #else 29268e8a9feSAl Viro mnt->mnt_writers++; 293d3ef3d73Snpiggin@suse.de #endif 2943d733633SDave Hansen } 2953d733633SDave Hansen 29683adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2973d733633SDave Hansen { 298d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 29968e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 300d3ef3d73Snpiggin@suse.de #else 30168e8a9feSAl Viro mnt->mnt_writers--; 302d3ef3d73Snpiggin@suse.de #endif 303d3ef3d73Snpiggin@suse.de } 304d3ef3d73Snpiggin@suse.de 30583adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 306d3ef3d73Snpiggin@suse.de { 307d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 308d3ef3d73Snpiggin@suse.de unsigned int count = 0; 3093d733633SDave Hansen int cpu; 3103d733633SDave Hansen 3113d733633SDave Hansen for_each_possible_cpu(cpu) { 31268e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 3133d733633SDave Hansen } 3143d733633SDave Hansen 315d3ef3d73Snpiggin@suse.de return count; 316d3ef3d73Snpiggin@suse.de #else 317d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 318d3ef3d73Snpiggin@suse.de #endif 3193d733633SDave Hansen } 3203d733633SDave Hansen 3214ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3224ed5e82fSMiklos Szeredi { 3234ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3244ed5e82fSMiklos Szeredi return 1; 3254ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3264ed5e82fSMiklos Szeredi smp_rmb(); 3274ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3284ed5e82fSMiklos Szeredi } 3294ed5e82fSMiklos Szeredi 3303d733633SDave Hansen /* 331eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 332eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 333eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 334eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3353d733633SDave Hansen */ 3368366025eSDave Hansen /** 337eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 33883adc753SAl Viro * @m: the mount on which to take a write 3398366025eSDave Hansen * 340eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 341eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 342eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 343eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 344eb04c282SJan Kara * called. This is effectively a refcount. 3458366025eSDave Hansen */ 346eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3478366025eSDave Hansen { 34883adc753SAl Viro struct mount *mnt = real_mount(m); 3493d733633SDave Hansen int ret = 0; 3503d733633SDave Hansen 351d3ef3d73Snpiggin@suse.de preempt_disable(); 352c6653a83SNick Piggin mnt_inc_writers(mnt); 353d3ef3d73Snpiggin@suse.de /* 354c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 355d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 356d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 357d3ef3d73Snpiggin@suse.de */ 358d3ef3d73Snpiggin@suse.de smp_mb(); 3591e75529eSMiao Xie while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 360d3ef3d73Snpiggin@suse.de cpu_relax(); 361d3ef3d73Snpiggin@suse.de /* 362d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 363d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 364d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 365d3ef3d73Snpiggin@suse.de */ 366d3ef3d73Snpiggin@suse.de smp_rmb(); 3674ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 368c6653a83SNick Piggin mnt_dec_writers(mnt); 3693d733633SDave Hansen ret = -EROFS; 3703d733633SDave Hansen } 371d3ef3d73Snpiggin@suse.de preempt_enable(); 372eb04c282SJan Kara 373eb04c282SJan Kara return ret; 374eb04c282SJan Kara } 375eb04c282SJan Kara 376eb04c282SJan Kara /** 377eb04c282SJan Kara * mnt_want_write - get write access to a mount 378eb04c282SJan Kara * @m: the mount on which to take a write 379eb04c282SJan Kara * 380eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 381eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 382eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 383eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 384eb04c282SJan Kara */ 385eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 386eb04c282SJan Kara { 387eb04c282SJan Kara int ret; 388eb04c282SJan Kara 389eb04c282SJan Kara sb_start_write(m->mnt_sb); 390eb04c282SJan Kara ret = __mnt_want_write(m); 391eb04c282SJan Kara if (ret) 392eb04c282SJan Kara sb_end_write(m->mnt_sb); 3933d733633SDave Hansen return ret; 3948366025eSDave Hansen } 3958366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3968366025eSDave Hansen 3978366025eSDave Hansen /** 39896029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 39996029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 40096029c4eSnpiggin@suse.de * 40196029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 40296029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 40396029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 40496029c4eSnpiggin@suse.de * on it. This allows some optimisation. 40596029c4eSnpiggin@suse.de * 40696029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 40796029c4eSnpiggin@suse.de * drop the reference. 40896029c4eSnpiggin@suse.de */ 40996029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 41096029c4eSnpiggin@suse.de { 41196029c4eSnpiggin@suse.de /* superblock may be r/o */ 41296029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 41396029c4eSnpiggin@suse.de return -EROFS; 41496029c4eSnpiggin@suse.de preempt_disable(); 41583adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 41696029c4eSnpiggin@suse.de preempt_enable(); 41796029c4eSnpiggin@suse.de return 0; 41896029c4eSnpiggin@suse.de } 41996029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 42096029c4eSnpiggin@suse.de 42196029c4eSnpiggin@suse.de /** 422eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 423eb04c282SJan Kara * @file: the file who's mount on which to take a write 424eb04c282SJan Kara * 425eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 426eb04c282SJan Kara * do some optimisations if the file is open for write already 427eb04c282SJan Kara */ 428eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 429eb04c282SJan Kara { 43083f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 431eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 432eb04c282SJan Kara else 433eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 434eb04c282SJan Kara } 435eb04c282SJan Kara 436eb04c282SJan Kara /** 43796029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 43896029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 43996029c4eSnpiggin@suse.de * 44096029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 44196029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 44296029c4eSnpiggin@suse.de */ 44396029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 44496029c4eSnpiggin@suse.de { 445eb04c282SJan Kara int ret; 446eb04c282SJan Kara 447eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 448eb04c282SJan Kara ret = __mnt_want_write_file(file); 449eb04c282SJan Kara if (ret) 450eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 451eb04c282SJan Kara return ret; 45296029c4eSnpiggin@suse.de } 45396029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 45496029c4eSnpiggin@suse.de 45596029c4eSnpiggin@suse.de /** 456eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4578366025eSDave Hansen * @mnt: the mount on which to give up write access 4588366025eSDave Hansen * 4598366025eSDave Hansen * Tells the low-level filesystem that we are done 4608366025eSDave Hansen * performing writes to it. Must be matched with 461eb04c282SJan Kara * __mnt_want_write() call above. 4628366025eSDave Hansen */ 463eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4648366025eSDave Hansen { 465d3ef3d73Snpiggin@suse.de preempt_disable(); 46683adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 467d3ef3d73Snpiggin@suse.de preempt_enable(); 4688366025eSDave Hansen } 469eb04c282SJan Kara 470eb04c282SJan Kara /** 471eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 472eb04c282SJan Kara * @mnt: the mount on which to give up write access 473eb04c282SJan Kara * 474eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 475eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 476eb04c282SJan Kara * mnt_want_write() call above. 477eb04c282SJan Kara */ 478eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 479eb04c282SJan Kara { 480eb04c282SJan Kara __mnt_drop_write(mnt); 481eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 482eb04c282SJan Kara } 4838366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4848366025eSDave Hansen 485eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 486eb04c282SJan Kara { 487eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 488eb04c282SJan Kara } 489eb04c282SJan Kara 4902a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4912a79f17eSAl Viro { 4922a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4932a79f17eSAl Viro } 4942a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4952a79f17eSAl Viro 49683adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4978366025eSDave Hansen { 4983d733633SDave Hansen int ret = 0; 4993d733633SDave Hansen 500719ea2fbSAl Viro lock_mount_hash(); 50183adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 502d3ef3d73Snpiggin@suse.de /* 503d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 504d3ef3d73Snpiggin@suse.de * should be visible before we do. 505d3ef3d73Snpiggin@suse.de */ 506d3ef3d73Snpiggin@suse.de smp_mb(); 507d3ef3d73Snpiggin@suse.de 508d3ef3d73Snpiggin@suse.de /* 509d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 510d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 511d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 512d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 513d3ef3d73Snpiggin@suse.de * 514d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 515d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 516d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 517d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 518d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 519d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 520d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 521d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 522d3ef3d73Snpiggin@suse.de * we're counting up here. 523d3ef3d73Snpiggin@suse.de */ 524c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 525d3ef3d73Snpiggin@suse.de ret = -EBUSY; 526d3ef3d73Snpiggin@suse.de else 52783adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 528d3ef3d73Snpiggin@suse.de /* 529d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 530d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 531d3ef3d73Snpiggin@suse.de */ 532d3ef3d73Snpiggin@suse.de smp_wmb(); 53383adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 534719ea2fbSAl Viro unlock_mount_hash(); 5353d733633SDave Hansen return ret; 5363d733633SDave Hansen } 5378366025eSDave Hansen 53883adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5392e4b7fcdSDave Hansen { 540719ea2fbSAl Viro lock_mount_hash(); 54183adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 542719ea2fbSAl Viro unlock_mount_hash(); 5432e4b7fcdSDave Hansen } 5442e4b7fcdSDave Hansen 5454ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5464ed5e82fSMiklos Szeredi { 5474ed5e82fSMiklos Szeredi struct mount *mnt; 5484ed5e82fSMiklos Szeredi int err = 0; 5494ed5e82fSMiklos Szeredi 5508e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5518e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5528e8b8796SMiklos Szeredi return -EBUSY; 5538e8b8796SMiklos Szeredi 554719ea2fbSAl Viro lock_mount_hash(); 5554ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5564ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5574ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5584ed5e82fSMiklos Szeredi smp_mb(); 5594ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5604ed5e82fSMiklos Szeredi err = -EBUSY; 5614ed5e82fSMiklos Szeredi break; 5624ed5e82fSMiklos Szeredi } 5634ed5e82fSMiklos Szeredi } 5644ed5e82fSMiklos Szeredi } 5658e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5668e8b8796SMiklos Szeredi err = -EBUSY; 5678e8b8796SMiklos Szeredi 5684ed5e82fSMiklos Szeredi if (!err) { 5694ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5704ed5e82fSMiklos Szeredi smp_wmb(); 5714ed5e82fSMiklos Szeredi } 5724ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5734ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5744ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5754ed5e82fSMiklos Szeredi } 576719ea2fbSAl Viro unlock_mount_hash(); 5774ed5e82fSMiklos Szeredi 5784ed5e82fSMiklos Szeredi return err; 5794ed5e82fSMiklos Szeredi } 5804ed5e82fSMiklos Szeredi 581b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5821da177e4SLinus Torvalds { 583fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 584d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 58568e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 586d3ef3d73Snpiggin@suse.de #endif 587b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5881da177e4SLinus Torvalds } 5891da177e4SLinus Torvalds 5908ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5918ffcb32eSDavid Howells { 5928ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5938ffcb32eSDavid Howells } 5948ffcb32eSDavid Howells 59548a066e7SAl Viro /* call under rcu_read_lock */ 596294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 59748a066e7SAl Viro { 59848a066e7SAl Viro struct mount *mnt; 59948a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 600294d71ffSAl Viro return 1; 60148a066e7SAl Viro if (bastard == NULL) 602294d71ffSAl Viro return 0; 60348a066e7SAl Viro mnt = real_mount(bastard); 60448a066e7SAl Viro mnt_add_count(mnt, 1); 60548a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 606294d71ffSAl Viro return 0; 60748a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 60848a066e7SAl Viro mnt_add_count(mnt, -1); 609294d71ffSAl Viro return 1; 61048a066e7SAl Viro } 611294d71ffSAl Viro return -1; 612294d71ffSAl Viro } 613294d71ffSAl Viro 614294d71ffSAl Viro /* call under rcu_read_lock */ 615294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 616294d71ffSAl Viro { 617294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 618294d71ffSAl Viro if (likely(!res)) 619294d71ffSAl Viro return true; 620294d71ffSAl Viro if (unlikely(res < 0)) { 62148a066e7SAl Viro rcu_read_unlock(); 62248a066e7SAl Viro mntput(bastard); 62348a066e7SAl Viro rcu_read_lock(); 624294d71ffSAl Viro } 62548a066e7SAl Viro return false; 62648a066e7SAl Viro } 62748a066e7SAl Viro 6281da177e4SLinus Torvalds /* 629474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 63048a066e7SAl Viro * call under rcu_read_lock() 6311da177e4SLinus Torvalds */ 632474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6331da177e4SLinus Torvalds { 63438129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 635474279dcSAl Viro struct mount *p; 6361da177e4SLinus Torvalds 63738129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 638474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 639474279dcSAl Viro return p; 640474279dcSAl Viro return NULL; 6411da177e4SLinus Torvalds } 642474279dcSAl Viro 643474279dcSAl Viro /* 644474279dcSAl Viro * find the last mount at @dentry on vfsmount @mnt. 64548a066e7SAl Viro * mount_lock must be held. 646474279dcSAl Viro */ 647474279dcSAl Viro struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) 648474279dcSAl Viro { 649411a938bSEric W. Biederman struct mount *p, *res = NULL; 650411a938bSEric W. Biederman p = __lookup_mnt(mnt, dentry); 65138129a13SAl Viro if (!p) 65238129a13SAl Viro goto out; 653411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 654411a938bSEric W. Biederman res = p; 65538129a13SAl Viro hlist_for_each_entry_continue(p, mnt_hash) { 6561d6a32acSAl Viro if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) 6571d6a32acSAl Viro break; 658411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 6591d6a32acSAl Viro res = p; 6601d6a32acSAl Viro } 66138129a13SAl Viro out: 6621d6a32acSAl Viro return res; 6631da177e4SLinus Torvalds } 6641da177e4SLinus Torvalds 665a05964f3SRam Pai /* 666f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 667f015f126SDavid Howells * 668f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 669f015f126SDavid Howells * following mounts: 670f015f126SDavid Howells * 671f015f126SDavid Howells * mount /dev/sda1 /mnt 672f015f126SDavid Howells * mount /dev/sda2 /mnt 673f015f126SDavid Howells * mount /dev/sda3 /mnt 674f015f126SDavid Howells * 675f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 676f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 677f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 678f015f126SDavid Howells * 679f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 680a05964f3SRam Pai */ 681ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 682a05964f3SRam Pai { 683c7105365SAl Viro struct mount *child_mnt; 68448a066e7SAl Viro struct vfsmount *m; 68548a066e7SAl Viro unsigned seq; 68699b7db7bSNick Piggin 68748a066e7SAl Viro rcu_read_lock(); 68848a066e7SAl Viro do { 68948a066e7SAl Viro seq = read_seqbegin(&mount_lock); 690474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 69148a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 69248a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 69348a066e7SAl Viro rcu_read_unlock(); 69448a066e7SAl Viro return m; 695a05964f3SRam Pai } 696a05964f3SRam Pai 6977af1364fSEric W. Biederman /* 6987af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6997af1364fSEric W. Biederman * current mount namespace. 7007af1364fSEric W. Biederman * 7017af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 7027af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 7037af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 7047af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 7057af1364fSEric W. Biederman * is a mountpoint. 7067af1364fSEric W. Biederman * 7077af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 7087af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 7097af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 7107af1364fSEric W. Biederman * parent mount. 7117af1364fSEric W. Biederman */ 7127af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 7137af1364fSEric W. Biederman { 7147af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7157af1364fSEric W. Biederman struct mount *mnt; 7167af1364fSEric W. Biederman bool is_covered = false; 7177af1364fSEric W. Biederman 7187af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 7197af1364fSEric W. Biederman goto out; 7207af1364fSEric W. Biederman 7217af1364fSEric W. Biederman down_read(&namespace_sem); 7227af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7237af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7247af1364fSEric W. Biederman if (is_covered) 7257af1364fSEric W. Biederman break; 7267af1364fSEric W. Biederman } 7277af1364fSEric W. Biederman up_read(&namespace_sem); 7287af1364fSEric W. Biederman out: 7297af1364fSEric W. Biederman return is_covered; 7307af1364fSEric W. Biederman } 7317af1364fSEric W. Biederman 732e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 73384d17192SAl Viro { 7340818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 73584d17192SAl Viro struct mountpoint *mp; 73684d17192SAl Viro 7370818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 73884d17192SAl Viro if (mp->m_dentry == dentry) { 73984d17192SAl Viro /* might be worth a WARN_ON() */ 74084d17192SAl Viro if (d_unlinked(dentry)) 74184d17192SAl Viro return ERR_PTR(-ENOENT); 74284d17192SAl Viro mp->m_count++; 74384d17192SAl Viro return mp; 74484d17192SAl Viro } 74584d17192SAl Viro } 746e2dfa935SEric W. Biederman return NULL; 747e2dfa935SEric W. Biederman } 748e2dfa935SEric W. Biederman 749e2dfa935SEric W. Biederman static struct mountpoint *new_mountpoint(struct dentry *dentry) 750e2dfa935SEric W. Biederman { 751e2dfa935SEric W. Biederman struct hlist_head *chain = mp_hash(dentry); 752e2dfa935SEric W. Biederman struct mountpoint *mp; 753e2dfa935SEric W. Biederman int ret; 75484d17192SAl Viro 75584d17192SAl Viro mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 75684d17192SAl Viro if (!mp) 75784d17192SAl Viro return ERR_PTR(-ENOMEM); 75884d17192SAl Viro 759eed81007SMiklos Szeredi ret = d_set_mounted(dentry); 760eed81007SMiklos Szeredi if (ret) { 76184d17192SAl Viro kfree(mp); 762eed81007SMiklos Szeredi return ERR_PTR(ret); 76384d17192SAl Viro } 764eed81007SMiklos Szeredi 76584d17192SAl Viro mp->m_dentry = dentry; 76684d17192SAl Viro mp->m_count = 1; 7670818bf27SAl Viro hlist_add_head(&mp->m_hash, chain); 7680a5eb7c8SEric W. Biederman INIT_HLIST_HEAD(&mp->m_list); 76984d17192SAl Viro return mp; 77084d17192SAl Viro } 77184d17192SAl Viro 77284d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 77384d17192SAl Viro { 77484d17192SAl Viro if (!--mp->m_count) { 77584d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7760a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 77784d17192SAl Viro spin_lock(&dentry->d_lock); 77884d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 77984d17192SAl Viro spin_unlock(&dentry->d_lock); 7800818bf27SAl Viro hlist_del(&mp->m_hash); 78184d17192SAl Viro kfree(mp); 78284d17192SAl Viro } 78384d17192SAl Viro } 78484d17192SAl Viro 785143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7861da177e4SLinus Torvalds { 7876b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7881da177e4SLinus Torvalds } 7891da177e4SLinus Torvalds 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) { 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 */ 8046b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8055addc5ddSAl Viro { 8065addc5ddSAl Viro if (ns && ns->event != event) { 8075addc5ddSAl Viro ns->event = event; 8085addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8095addc5ddSAl Viro } 8105addc5ddSAl Viro } 8115addc5ddSAl Viro 81299b7db7bSNick Piggin /* 81399b7db7bSNick Piggin * vfsmount lock must be held for write 81499b7db7bSNick Piggin */ 8157bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 8161da177e4SLinus Torvalds { 8170714a533SAl Viro mnt->mnt_parent = mnt; 818a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8196b41d536SAl Viro list_del_init(&mnt->mnt_child); 82038129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8210a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 82284d17192SAl Viro put_mountpoint(mnt->mnt_mp); 82384d17192SAl Viro mnt->mnt_mp = NULL; 8241da177e4SLinus Torvalds } 8251da177e4SLinus Torvalds 82699b7db7bSNick Piggin /* 82799b7db7bSNick Piggin * vfsmount lock must be held for write 82899b7db7bSNick Piggin */ 8297bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8307bdb11deSEric W. Biederman { 8317bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8327bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8337bdb11deSEric W. Biederman unhash_mnt(mnt); 8347bdb11deSEric W. Biederman } 8357bdb11deSEric W. Biederman 8367bdb11deSEric W. Biederman /* 8377bdb11deSEric W. Biederman * vfsmount lock must be held for write 8387bdb11deSEric W. Biederman */ 8396a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8406a46c573SEric W. Biederman { 8416a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8426a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8436a46c573SEric W. Biederman unhash_mnt(mnt); 8446a46c573SEric W. Biederman } 8456a46c573SEric W. Biederman 8466a46c573SEric W. Biederman /* 8476a46c573SEric W. Biederman * vfsmount lock must be held for write 8486a46c573SEric W. Biederman */ 84984d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 85084d17192SAl Viro struct mountpoint *mp, 85144d964d6SAl Viro struct mount *child_mnt) 852b90fa9aeSRam Pai { 85384d17192SAl Viro mp->m_count++; 8543a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 85584d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8563a2393d7SAl Viro child_mnt->mnt_parent = mnt; 85784d17192SAl Viro child_mnt->mnt_mp = mp; 8580a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 859b90fa9aeSRam Pai } 860b90fa9aeSRam Pai 86199b7db7bSNick Piggin /* 86299b7db7bSNick Piggin * vfsmount lock must be held for write 86399b7db7bSNick Piggin */ 86484d17192SAl Viro static void attach_mnt(struct mount *mnt, 86584d17192SAl Viro struct mount *parent, 86684d17192SAl Viro struct mountpoint *mp) 8671da177e4SLinus Torvalds { 86884d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 86938129a13SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); 87084d17192SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 871b90fa9aeSRam Pai } 872b90fa9aeSRam Pai 87312a5b529SAl Viro static void attach_shadowed(struct mount *mnt, 87412a5b529SAl Viro struct mount *parent, 87512a5b529SAl Viro struct mount *shadows) 87612a5b529SAl Viro { 87712a5b529SAl Viro if (shadows) { 878f6f99332SLinus Torvalds hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); 87912a5b529SAl Viro list_add(&mnt->mnt_child, &shadows->mnt_child); 88012a5b529SAl Viro } else { 88112a5b529SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, 88212a5b529SAl Viro m_hash(&parent->mnt, mnt->mnt_mountpoint)); 88312a5b529SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 88412a5b529SAl Viro } 88512a5b529SAl Viro } 88612a5b529SAl Viro 887b90fa9aeSRam Pai /* 88899b7db7bSNick Piggin * vfsmount lock must be held for write 889b90fa9aeSRam Pai */ 8901d6a32acSAl Viro static void commit_tree(struct mount *mnt, struct mount *shadows) 891b90fa9aeSRam Pai { 8920714a533SAl Viro struct mount *parent = mnt->mnt_parent; 89383adc753SAl Viro struct mount *m; 894b90fa9aeSRam Pai LIST_HEAD(head); 895143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 896b90fa9aeSRam Pai 8970714a533SAl Viro BUG_ON(parent == mnt); 898b90fa9aeSRam Pai 8991a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 900f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 901143c8c91SAl Viro m->mnt_ns = n; 902f03c6599SAl Viro 903b90fa9aeSRam Pai list_splice(&head, n->list.prev); 904b90fa9aeSRam Pai 905d2921684SEric W. Biederman n->mounts += n->pending_mounts; 906d2921684SEric W. Biederman n->pending_mounts = 0; 907d2921684SEric W. Biederman 90812a5b529SAl Viro attach_shadowed(mnt, parent, shadows); 9096b3286edSKirill Korotaev touch_mnt_namespace(n); 9101da177e4SLinus Torvalds } 9111da177e4SLinus Torvalds 912909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9131da177e4SLinus Torvalds { 9146b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9156b41d536SAl Viro if (next == &p->mnt_mounts) { 9161da177e4SLinus Torvalds while (1) { 917909b0a88SAl Viro if (p == root) 9181da177e4SLinus Torvalds return NULL; 9196b41d536SAl Viro next = p->mnt_child.next; 9206b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9211da177e4SLinus Torvalds break; 9220714a533SAl Viro p = p->mnt_parent; 9231da177e4SLinus Torvalds } 9241da177e4SLinus Torvalds } 9256b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9261da177e4SLinus Torvalds } 9271da177e4SLinus Torvalds 928315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9299676f0c6SRam Pai { 9306b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9316b41d536SAl Viro while (prev != &p->mnt_mounts) { 9326b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9336b41d536SAl Viro prev = p->mnt_mounts.prev; 9349676f0c6SRam Pai } 9359676f0c6SRam Pai return p; 9369676f0c6SRam Pai } 9379676f0c6SRam Pai 9389d412a43SAl Viro struct vfsmount * 9399d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 9409d412a43SAl Viro { 941b105e270SAl Viro struct mount *mnt; 9429d412a43SAl Viro struct dentry *root; 9439d412a43SAl Viro 9449d412a43SAl Viro if (!type) 9459d412a43SAl Viro return ERR_PTR(-ENODEV); 9469d412a43SAl Viro 9479d412a43SAl Viro mnt = alloc_vfsmnt(name); 9489d412a43SAl Viro if (!mnt) 9499d412a43SAl Viro return ERR_PTR(-ENOMEM); 9509d412a43SAl Viro 9519d412a43SAl Viro if (flags & MS_KERNMOUNT) 952b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9539d412a43SAl Viro 9549d412a43SAl Viro root = mount_fs(type, flags, name, data); 9559d412a43SAl Viro if (IS_ERR(root)) { 9568ffcb32eSDavid Howells mnt_free_id(mnt); 9579d412a43SAl Viro free_vfsmnt(mnt); 9589d412a43SAl Viro return ERR_CAST(root); 9599d412a43SAl Viro } 9609d412a43SAl Viro 961b105e270SAl Viro mnt->mnt.mnt_root = root; 962b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 963a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9640714a533SAl Viro mnt->mnt_parent = mnt; 965719ea2fbSAl Viro lock_mount_hash(); 96639f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 967719ea2fbSAl Viro unlock_mount_hash(); 968b105e270SAl Viro return &mnt->mnt; 9699d412a43SAl Viro } 9709d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 9719d412a43SAl Viro 97287129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 97336341f64SRam Pai int flag) 9741da177e4SLinus Torvalds { 97587129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 976be34d1a3SDavid Howells struct mount *mnt; 977be34d1a3SDavid Howells int err; 9781da177e4SLinus Torvalds 979be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 980be34d1a3SDavid Howells if (!mnt) 981be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 982be34d1a3SDavid Howells 9837a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 98415169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 985719f5d7fSMiklos Szeredi else 98615169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 987719f5d7fSMiklos Szeredi 98815169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 989be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 990719f5d7fSMiklos Szeredi if (err) 991719f5d7fSMiklos Szeredi goto out_free; 992719f5d7fSMiklos Szeredi } 993719f5d7fSMiklos Szeredi 994f2ebb3a9SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); 995132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 9969566d674SEric W. Biederman if (flag & CL_UNPRIVILEGED) { 9979566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; 9989566d674SEric W. Biederman 9999566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_READONLY) 1000132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 1001132c94e3SEric W. Biederman 10029566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NODEV) 10039566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; 10049566d674SEric W. Biederman 10059566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOSUID) 10069566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; 10079566d674SEric W. Biederman 10089566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOEXEC) 10099566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; 10109566d674SEric W. Biederman } 10119566d674SEric W. Biederman 10125ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 1013381cacb1SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && 1014381cacb1SEric W. Biederman (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) 10155ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 10165ff9d8a6SEric W. Biederman 10171da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1018b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1019b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1020a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10210714a533SAl Viro mnt->mnt_parent = mnt; 1022719ea2fbSAl Viro lock_mount_hash(); 102339f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1024719ea2fbSAl Viro unlock_mount_hash(); 1025b90fa9aeSRam Pai 10267a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10277a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10286776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 102932301920SAl Viro mnt->mnt_master = old; 1030fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10318aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1032fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10336776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1034d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10356776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1036d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10375afe0022SRam Pai } 1038b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10390f0afb1dSAl Viro set_mnt_shared(mnt); 10401da177e4SLinus Torvalds 10411da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10421da177e4SLinus Torvalds * as the original if that was on one */ 104336341f64SRam Pai if (flag & CL_EXPIRE) { 10446776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10456776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10461da177e4SLinus Torvalds } 1047be34d1a3SDavid Howells 1048cb338d06SAl Viro return mnt; 1049719f5d7fSMiklos Szeredi 1050719f5d7fSMiklos Szeredi out_free: 10518ffcb32eSDavid Howells mnt_free_id(mnt); 1052719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1053be34d1a3SDavid Howells return ERR_PTR(err); 10541da177e4SLinus Torvalds } 10551da177e4SLinus Torvalds 10569ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10579ea459e1SAl Viro { 10589ea459e1SAl Viro /* 10599ea459e1SAl Viro * This probably indicates that somebody messed 10609ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 10619ea459e1SAl Viro * happens, the filesystem was probably unable 10629ea459e1SAl Viro * to make r/w->r/o transitions. 10639ea459e1SAl Viro */ 10649ea459e1SAl Viro /* 10659ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 10669ea459e1SAl Viro * so mnt_get_writers() below is safe. 10679ea459e1SAl Viro */ 10689ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 10699ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 10709ea459e1SAl Viro mnt_pin_kill(mnt); 10719ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 10729ea459e1SAl Viro dput(mnt->mnt.mnt_root); 10739ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 10749ea459e1SAl Viro mnt_free_id(mnt); 10759ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 10769ea459e1SAl Viro } 10779ea459e1SAl Viro 10789ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 10799ea459e1SAl Viro { 10809ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 10819ea459e1SAl Viro } 10829ea459e1SAl Viro 10839ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 10849ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 10859ea459e1SAl Viro { 10869ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 10879ea459e1SAl Viro struct llist_node *next; 10889ea459e1SAl Viro 10899ea459e1SAl Viro for (; node; node = next) { 10909ea459e1SAl Viro next = llist_next(node); 10919ea459e1SAl Viro cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); 10929ea459e1SAl Viro } 10939ea459e1SAl Viro } 10949ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 10959ea459e1SAl Viro 1096900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 10977b7b1aceSAl Viro { 109848a066e7SAl Viro rcu_read_lock(); 1099aa9c0e07SAl Viro mnt_add_count(mnt, -1); 110048a066e7SAl Viro if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ 110148a066e7SAl Viro rcu_read_unlock(); 110299b7db7bSNick Piggin return; 1103b3e19d92SNick Piggin } 1104719ea2fbSAl Viro lock_mount_hash(); 1105b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 110648a066e7SAl Viro rcu_read_unlock(); 1107719ea2fbSAl Viro unlock_mount_hash(); 110899b7db7bSNick Piggin return; 110999b7db7bSNick Piggin } 111048a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 111148a066e7SAl Viro rcu_read_unlock(); 111248a066e7SAl Viro unlock_mount_hash(); 111348a066e7SAl Viro return; 111448a066e7SAl Viro } 111548a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 111648a066e7SAl Viro rcu_read_unlock(); 1117962830dfSAndi Kleen 111839f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1119ce07d891SEric W. Biederman 1120ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1121ce07d891SEric W. Biederman struct mount *p, *tmp; 1122ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1123ce07d891SEric W. Biederman umount_mnt(p); 1124ce07d891SEric W. Biederman } 1125ce07d891SEric W. Biederman } 1126719ea2fbSAl Viro unlock_mount_hash(); 1127649a795aSAl Viro 11289ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11299ea459e1SAl Viro struct task_struct *task = current; 11309ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11319ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11329ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11339ea459e1SAl Viro return; 11349ea459e1SAl Viro } 11359ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11369ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11379ea459e1SAl Viro return; 11389ea459e1SAl Viro } 11399ea459e1SAl Viro cleanup_mnt(mnt); 1140b3e19d92SNick Piggin } 1141b3e19d92SNick Piggin 1142b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1143b3e19d92SNick Piggin { 1144b3e19d92SNick Piggin if (mnt) { 1145863d684fSAl Viro struct mount *m = real_mount(mnt); 1146b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1147863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1148863d684fSAl Viro m->mnt_expiry_mark = 0; 1149863d684fSAl Viro mntput_no_expire(m); 1150b3e19d92SNick Piggin } 1151b3e19d92SNick Piggin } 1152b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1153b3e19d92SNick Piggin 1154b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1155b3e19d92SNick Piggin { 1156b3e19d92SNick Piggin if (mnt) 115783adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1158b3e19d92SNick Piggin return mnt; 1159b3e19d92SNick Piggin } 1160b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1161b3e19d92SNick Piggin 1162ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 11637b7b1aceSAl Viro { 11643064c356SAl Viro struct mount *p; 11653064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 11663064c356SAl Viro if (IS_ERR(p)) 11673064c356SAl Viro return ERR_CAST(p); 11683064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 11693064c356SAl Viro return &p->mnt; 11707b7b1aceSAl Viro } 11711da177e4SLinus Torvalds 1172b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 1173b3b304a2SMiklos Szeredi { 1174b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 1175b3b304a2SMiklos Szeredi } 1176b3b304a2SMiklos Szeredi 1177b3b304a2SMiklos Szeredi /* 1178b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 1179b3b304a2SMiklos Szeredi * implement more complex mount option showing. 1180b3b304a2SMiklos Szeredi * 1181b3b304a2SMiklos Szeredi * See also save_mount_options(). 1182b3b304a2SMiklos Szeredi */ 118334c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 1184b3b304a2SMiklos Szeredi { 11852a32cebdSAl Viro const char *options; 11862a32cebdSAl Viro 11872a32cebdSAl Viro rcu_read_lock(); 118834c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 1189b3b304a2SMiklos Szeredi 1190b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 1191b3b304a2SMiklos Szeredi seq_putc(m, ','); 1192b3b304a2SMiklos Szeredi mangle(m, options); 1193b3b304a2SMiklos Szeredi } 11942a32cebdSAl Viro rcu_read_unlock(); 1195b3b304a2SMiklos Szeredi 1196b3b304a2SMiklos Szeredi return 0; 1197b3b304a2SMiklos Szeredi } 1198b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 1199b3b304a2SMiklos Szeredi 1200b3b304a2SMiklos Szeredi /* 1201b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 1202b3b304a2SMiklos Szeredi * called from the fill_super() callback. 1203b3b304a2SMiklos Szeredi * 1204b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 1205b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1206b3b304a2SMiklos Szeredi * remount fails. 1207b3b304a2SMiklos Szeredi * 1208b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1209b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1210b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1211b3b304a2SMiklos Szeredi * any more. 1212b3b304a2SMiklos Szeredi */ 1213b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1214b3b304a2SMiklos Szeredi { 12152a32cebdSAl Viro BUG_ON(sb->s_options); 12162a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1217b3b304a2SMiklos Szeredi } 1218b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1219b3b304a2SMiklos Szeredi 12202a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 12212a32cebdSAl Viro { 12222a32cebdSAl Viro char *old = sb->s_options; 12232a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 12242a32cebdSAl Viro if (old) { 12252a32cebdSAl Viro synchronize_rcu(); 12262a32cebdSAl Viro kfree(old); 12272a32cebdSAl Viro } 12282a32cebdSAl Viro } 12292a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 12302a32cebdSAl Viro 1231a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12320226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12331da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12341da177e4SLinus Torvalds { 1235ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12361da177e4SLinus Torvalds 1237390c6843SRam Pai down_read(&namespace_sem); 1238c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1239c7999c36SAl Viro void *v = p->cached_mount; 1240c7999c36SAl Viro if (*pos == p->cached_index) 1241c7999c36SAl Viro return v; 1242c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1243c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1244c7999c36SAl Viro return p->cached_mount = v; 1245c7999c36SAl Viro } 1246c7999c36SAl Viro } 1247c7999c36SAl Viro 1248c7999c36SAl Viro p->cached_event = p->ns->event; 1249c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1250c7999c36SAl Viro p->cached_index = *pos; 1251c7999c36SAl Viro return p->cached_mount; 12521da177e4SLinus Torvalds } 12531da177e4SLinus Torvalds 12541da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12551da177e4SLinus Torvalds { 1256ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1257b0765fb8SPavel Emelianov 1258c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1259c7999c36SAl Viro p->cached_index = *pos; 1260c7999c36SAl Viro return p->cached_mount; 12611da177e4SLinus Torvalds } 12621da177e4SLinus Torvalds 12631da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12641da177e4SLinus Torvalds { 1265390c6843SRam Pai up_read(&namespace_sem); 12661da177e4SLinus Torvalds } 12671da177e4SLinus Torvalds 12680226f492SAl Viro static int m_show(struct seq_file *m, void *v) 12699f5596afSAl Viro { 1270ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12711a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 12720226f492SAl Viro return p->show(m, &r->mnt); 12731da177e4SLinus Torvalds } 12741da177e4SLinus Torvalds 1275a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 12761da177e4SLinus Torvalds .start = m_start, 12771da177e4SLinus Torvalds .next = m_next, 12781da177e4SLinus Torvalds .stop = m_stop, 12790226f492SAl Viro .show = m_show, 1280b4629fe2SChuck Lever }; 1281a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1282b4629fe2SChuck Lever 12831da177e4SLinus Torvalds /** 12841da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 12851da177e4SLinus Torvalds * @mnt: root of mount tree 12861da177e4SLinus Torvalds * 12871da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 12881da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 12891da177e4SLinus Torvalds * busy. 12901da177e4SLinus Torvalds */ 1291909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 12921da177e4SLinus Torvalds { 1293909b0a88SAl Viro struct mount *mnt = real_mount(m); 129436341f64SRam Pai int actual_refs = 0; 129536341f64SRam Pai int minimum_refs = 0; 1296315fc83eSAl Viro struct mount *p; 1297909b0a88SAl Viro BUG_ON(!m); 12981da177e4SLinus Torvalds 1299b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1300719ea2fbSAl Viro lock_mount_hash(); 1301909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 130283adc753SAl Viro actual_refs += mnt_get_count(p); 13031da177e4SLinus Torvalds minimum_refs += 2; 13041da177e4SLinus Torvalds } 1305719ea2fbSAl Viro unlock_mount_hash(); 13061da177e4SLinus Torvalds 13071da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13081da177e4SLinus Torvalds return 0; 1309e3474a8eSIan Kent 1310e3474a8eSIan Kent return 1; 13111da177e4SLinus Torvalds } 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13141da177e4SLinus Torvalds 13151da177e4SLinus Torvalds /** 13161da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13171da177e4SLinus Torvalds * @mnt: root of mount 13181da177e4SLinus Torvalds * 13191da177e4SLinus Torvalds * This is called to check if a mount point has any 13201da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13211da177e4SLinus Torvalds * mount has sub mounts this will return busy 13221da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13231da177e4SLinus Torvalds * 13241da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13251da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13261da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13271da177e4SLinus Torvalds */ 13281da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13291da177e4SLinus Torvalds { 1330e3474a8eSIan Kent int ret = 1; 13318ad08d8aSAl Viro down_read(&namespace_sem); 1332719ea2fbSAl Viro lock_mount_hash(); 13331ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1334e3474a8eSIan Kent ret = 0; 1335719ea2fbSAl Viro unlock_mount_hash(); 13368ad08d8aSAl Viro up_read(&namespace_sem); 1337a05964f3SRam Pai return ret; 13381da177e4SLinus Torvalds } 13391da177e4SLinus Torvalds 13401da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13411da177e4SLinus Torvalds 134238129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1343e3197d83SAl Viro 134497216be0SAl Viro static void namespace_unlock(void) 13451da177e4SLinus Torvalds { 1346a3b3c562SEric W. Biederman struct hlist_head head; 134797216be0SAl Viro 1348a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1349a3b3c562SEric W. Biederman 135097216be0SAl Viro up_write(&namespace_sem); 1351a3b3c562SEric W. Biederman 1352a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 135397216be0SAl Viro return; 135497216be0SAl Viro 135548a066e7SAl Viro synchronize_rcu(); 135648a066e7SAl Viro 135787b95ce0SAl Viro group_pin_kill(&head); 135870fbcdf4SRam Pai } 135970fbcdf4SRam Pai 136097216be0SAl Viro static inline void namespace_lock(void) 1361e3197d83SAl Viro { 136297216be0SAl Viro down_write(&namespace_sem); 1363e3197d83SAl Viro } 1364e3197d83SAl Viro 1365e819f152SEric W. Biederman enum umount_tree_flags { 1366e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1367e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1368e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1369e819f152SEric W. Biederman }; 1370f2d0a123SEric W. Biederman 1371f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1372f2d0a123SEric W. Biederman { 1373f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1374f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1375f2d0a123SEric W. Biederman return true; 1376f2d0a123SEric W. Biederman 1377f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1378f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1379f2d0a123SEric W. Biederman return true; 1380f2d0a123SEric W. Biederman 1381f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1382f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1383f2d0a123SEric W. Biederman * connected to mounted mounts. 1384f2d0a123SEric W. Biederman */ 1385f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1386f2d0a123SEric W. Biederman return true; 1387f2d0a123SEric W. Biederman 1388f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1389f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1390f2d0a123SEric W. Biederman return false; 1391f2d0a123SEric W. Biederman 1392f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1393f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1394f2d0a123SEric W. Biederman return false; 1395f2d0a123SEric W. Biederman 1396f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1397f2d0a123SEric W. Biederman return true; 1398f2d0a123SEric W. Biederman } 1399f2d0a123SEric W. Biederman 140099b7db7bSNick Piggin /* 140148a066e7SAl Viro * mount_lock must be held 140299b7db7bSNick Piggin * namespace_sem must be held for write 140399b7db7bSNick Piggin */ 1404e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 140570fbcdf4SRam Pai { 1406c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1407315fc83eSAl Viro struct mount *p; 140870fbcdf4SRam Pai 14095d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14105d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14115d88457eSEric W. Biederman 1412c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1413590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1414590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1415c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1416590ce4bcSEric W. Biederman } 1417c003b26fSEric W. Biederman 1418411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1419c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1420c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 142138129a13SAl Viro } 142270fbcdf4SRam Pai 1423c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1424e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14257b8a53fdSAl Viro propagate_umount(&tmp_list); 1426a05964f3SRam Pai 1427c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1428d2921684SEric W. Biederman struct mnt_namespace *ns; 1429ce07d891SEric W. Biederman bool disconnect; 1430c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 14316776db3dSAl Viro list_del_init(&p->mnt_expire); 14321a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1433d2921684SEric W. Biederman ns = p->mnt_ns; 1434d2921684SEric W. Biederman if (ns) { 1435d2921684SEric W. Biederman ns->mounts--; 1436d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1437d2921684SEric W. Biederman } 143863d37a84SAl Viro p->mnt_ns = NULL; 1439e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 144048a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 144187b95ce0SAl Viro 1442f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1443ce07d891SEric W. Biederman 1444ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1445ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1446676da58dSAl Viro if (mnt_has_parent(p)) { 144781b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1448ce07d891SEric W. Biederman if (!disconnect) { 1449ce07d891SEric W. Biederman /* Don't forget about p */ 1450ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1451ce07d891SEric W. Biederman } else { 14526a46c573SEric W. Biederman umount_mnt(p); 14537c4b93d8SAl Viro } 1454ce07d891SEric W. Biederman } 14550f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 145638129a13SAl Viro } 14571da177e4SLinus Torvalds } 14581da177e4SLinus Torvalds 1459b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1460c35038beSAl Viro 14611ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 14621da177e4SLinus Torvalds { 14631ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 14641da177e4SLinus Torvalds int retval; 14651da177e4SLinus Torvalds 14661ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 14671da177e4SLinus Torvalds if (retval) 14681da177e4SLinus Torvalds return retval; 14691da177e4SLinus Torvalds 14701da177e4SLinus Torvalds /* 14711da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 14721da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 14731da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 14741da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 14751da177e4SLinus Torvalds */ 14761da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 14771ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 14781da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 14791da177e4SLinus Torvalds return -EINVAL; 14801da177e4SLinus Torvalds 1481b3e19d92SNick Piggin /* 1482b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1483b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1484b3e19d92SNick Piggin */ 1485719ea2fbSAl Viro lock_mount_hash(); 148683adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1487719ea2fbSAl Viro unlock_mount_hash(); 14881da177e4SLinus Torvalds return -EBUSY; 1489b3e19d92SNick Piggin } 1490719ea2fbSAl Viro unlock_mount_hash(); 14911da177e4SLinus Torvalds 1492863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 14931da177e4SLinus Torvalds return -EAGAIN; 14941da177e4SLinus Torvalds } 14951da177e4SLinus Torvalds 14961da177e4SLinus Torvalds /* 14971da177e4SLinus Torvalds * If we may have to abort operations to get out of this 14981da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 14991da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 15001da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 15011da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 15021da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 15031da177e4SLinus Torvalds * about for the moment. 15041da177e4SLinus Torvalds */ 15051da177e4SLinus Torvalds 150642faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 150742faad99SAl Viro sb->s_op->umount_begin(sb); 150842faad99SAl Viro } 15091da177e4SLinus Torvalds 15101da177e4SLinus Torvalds /* 15111da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15121da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15131da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15141da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15151da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15161da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15171da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15181da177e4SLinus Torvalds */ 15191ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15201da177e4SLinus Torvalds /* 15211da177e4SLinus Torvalds * Special case for "unmounting" root ... 15221da177e4SLinus Torvalds * we just try to remount it readonly. 15231da177e4SLinus Torvalds */ 1524a1480dccSAndy Lutomirski if (!capable(CAP_SYS_ADMIN)) 1525a1480dccSAndy Lutomirski return -EPERM; 15261da177e4SLinus Torvalds down_write(&sb->s_umount); 15274aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 15281da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 15291da177e4SLinus Torvalds up_write(&sb->s_umount); 15301da177e4SLinus Torvalds return retval; 15311da177e4SLinus Torvalds } 15321da177e4SLinus Torvalds 153397216be0SAl Viro namespace_lock(); 1534719ea2fbSAl Viro lock_mount_hash(); 15355addc5ddSAl Viro event++; 15361da177e4SLinus Torvalds 153748a066e7SAl Viro if (flags & MNT_DETACH) { 153848a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1539e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 154048a066e7SAl Viro retval = 0; 154148a066e7SAl Viro } else { 1542b54b9be7SAl Viro shrink_submounts(mnt); 15431da177e4SLinus Torvalds retval = -EBUSY; 154448a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 15451a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1546e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 15471da177e4SLinus Torvalds retval = 0; 15481da177e4SLinus Torvalds } 154948a066e7SAl Viro } 1550719ea2fbSAl Viro unlock_mount_hash(); 1551e3197d83SAl Viro namespace_unlock(); 15521da177e4SLinus Torvalds return retval; 15531da177e4SLinus Torvalds } 15541da177e4SLinus Torvalds 15551da177e4SLinus Torvalds /* 155680b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 155780b5dce8SEric W. Biederman * 155880b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 155980b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 156080b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 156180b5dce8SEric W. Biederman * leaking them. 156280b5dce8SEric W. Biederman * 156380b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 156480b5dce8SEric W. Biederman */ 156580b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 156680b5dce8SEric W. Biederman { 156780b5dce8SEric W. Biederman struct mountpoint *mp; 156880b5dce8SEric W. Biederman struct mount *mnt; 156980b5dce8SEric W. Biederman 157080b5dce8SEric W. Biederman namespace_lock(); 157180b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1572f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 157380b5dce8SEric W. Biederman goto out_unlock; 157480b5dce8SEric W. Biederman 157580b5dce8SEric W. Biederman lock_mount_hash(); 1576e06b933eSAndrey Ulanov event++; 157780b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 157880b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1579ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1580fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1581fe78fcc8SEric W. Biederman umount_mnt(mnt); 1582ce07d891SEric W. Biederman } 1583e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 158480b5dce8SEric W. Biederman } 158580b5dce8SEric W. Biederman unlock_mount_hash(); 158680b5dce8SEric W. Biederman put_mountpoint(mp); 158780b5dce8SEric W. Biederman out_unlock: 158880b5dce8SEric W. Biederman namespace_unlock(); 158980b5dce8SEric W. Biederman } 159080b5dce8SEric W. Biederman 159180b5dce8SEric W. Biederman /* 15929b40bc90SAl Viro * Is the caller allowed to modify his namespace? 15939b40bc90SAl Viro */ 15949b40bc90SAl Viro static inline bool may_mount(void) 15959b40bc90SAl Viro { 15969b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 15979b40bc90SAl Viro } 15989b40bc90SAl Viro 15999e8925b6SJeff Layton static inline bool may_mandlock(void) 16009e8925b6SJeff Layton { 16019e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 16029e8925b6SJeff Layton return false; 16039e8925b6SJeff Layton #endif 160495ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 16059e8925b6SJeff Layton } 16069e8925b6SJeff Layton 16079b40bc90SAl Viro /* 16081da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16091da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16101da177e4SLinus Torvalds * 16111da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16121da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16131da177e4SLinus Torvalds */ 16141da177e4SLinus Torvalds 1615bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 16161da177e4SLinus Torvalds { 16172d8f3038SAl Viro struct path path; 1618900148dcSAl Viro struct mount *mnt; 16191da177e4SLinus Torvalds int retval; 1620db1f05bbSMiklos Szeredi int lookup_flags = 0; 16211da177e4SLinus Torvalds 1622db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1623db1f05bbSMiklos Szeredi return -EINVAL; 1624db1f05bbSMiklos Szeredi 16259b40bc90SAl Viro if (!may_mount()) 16269b40bc90SAl Viro return -EPERM; 16279b40bc90SAl Viro 1628db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1629db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1630db1f05bbSMiklos Szeredi 1631197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 16321da177e4SLinus Torvalds if (retval) 16331da177e4SLinus Torvalds goto out; 1634900148dcSAl Viro mnt = real_mount(path.mnt); 16351da177e4SLinus Torvalds retval = -EINVAL; 16362d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 16371da177e4SLinus Torvalds goto dput_and_out; 1638143c8c91SAl Viro if (!check_mnt(mnt)) 16391da177e4SLinus Torvalds goto dput_and_out; 16405ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 16415ff9d8a6SEric W. Biederman goto dput_and_out; 1642b2f5d4dcSEric W. Biederman retval = -EPERM; 1643b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1644b2f5d4dcSEric W. Biederman goto dput_and_out; 16451da177e4SLinus Torvalds 1646900148dcSAl Viro retval = do_umount(mnt, flags); 16471da177e4SLinus Torvalds dput_and_out: 1648429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 16492d8f3038SAl Viro dput(path.dentry); 1650900148dcSAl Viro mntput_no_expire(mnt); 16511da177e4SLinus Torvalds out: 16521da177e4SLinus Torvalds return retval; 16531da177e4SLinus Torvalds } 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds /* 16581da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 16591da177e4SLinus Torvalds */ 1660bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 16611da177e4SLinus Torvalds { 16621da177e4SLinus Torvalds return sys_umount(name, 0); 16631da177e4SLinus Torvalds } 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds #endif 16661da177e4SLinus Torvalds 16674ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 16688823c079SEric W. Biederman { 16694ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1670e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1671e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 16724ce5d2b1SEric W. Biederman } 16734ce5d2b1SEric W. Biederman 167458be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 167558be2825SAl Viro { 167658be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 167758be2825SAl Viro } 167858be2825SAl Viro 16794ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 16804ce5d2b1SEric W. Biederman { 16814ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 16824ce5d2b1SEric W. Biederman * mount namespace loop? 16834ce5d2b1SEric W. Biederman */ 16844ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 16854ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 16864ce5d2b1SEric W. Biederman return false; 16874ce5d2b1SEric W. Biederman 1688f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 16898823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 16908823c079SEric W. Biederman } 16918823c079SEric W. Biederman 169287129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 169336341f64SRam Pai int flag) 16941da177e4SLinus Torvalds { 169584d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 16961da177e4SLinus Torvalds 16974ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 16984ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 16994ce5d2b1SEric W. Biederman 17004ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1701be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 17029676f0c6SRam Pai 170336341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1704be34d1a3SDavid Howells if (IS_ERR(q)) 1705be34d1a3SDavid Howells return q; 1706be34d1a3SDavid Howells 1707a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17081da177e4SLinus Torvalds 17091da177e4SLinus Torvalds p = mnt; 17106b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1711315fc83eSAl Viro struct mount *s; 17127ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17131da177e4SLinus Torvalds continue; 17141da177e4SLinus Torvalds 1715909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 171612a5b529SAl Viro struct mount *t = NULL; 17174ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17184ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 17194ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17204ce5d2b1SEric W. Biederman continue; 17214ce5d2b1SEric W. Biederman } 17224ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17234ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17249676f0c6SRam Pai s = skip_mnt_tree(s); 17259676f0c6SRam Pai continue; 17269676f0c6SRam Pai } 17270714a533SAl Viro while (p != s->mnt_parent) { 17280714a533SAl Viro p = p->mnt_parent; 17290714a533SAl Viro q = q->mnt_parent; 17301da177e4SLinus Torvalds } 173187129cc0SAl Viro p = s; 173284d17192SAl Viro parent = q; 173387129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1734be34d1a3SDavid Howells if (IS_ERR(q)) 1735be34d1a3SDavid Howells goto out; 1736719ea2fbSAl Viro lock_mount_hash(); 17371a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 173812a5b529SAl Viro mnt_set_mountpoint(parent, p->mnt_mp, q); 173912a5b529SAl Viro if (!list_empty(&parent->mnt_mounts)) { 174012a5b529SAl Viro t = list_last_entry(&parent->mnt_mounts, 174112a5b529SAl Viro struct mount, mnt_child); 174212a5b529SAl Viro if (t->mnt_mp != p->mnt_mp) 174312a5b529SAl Viro t = NULL; 174412a5b529SAl Viro } 174512a5b529SAl Viro attach_shadowed(q, parent, t); 1746719ea2fbSAl Viro unlock_mount_hash(); 17471da177e4SLinus Torvalds } 17481da177e4SLinus Torvalds } 17491da177e4SLinus Torvalds return res; 1750be34d1a3SDavid Howells out: 17511da177e4SLinus Torvalds if (res) { 1752719ea2fbSAl Viro lock_mount_hash(); 1753e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1754719ea2fbSAl Viro unlock_mount_hash(); 17551da177e4SLinus Torvalds } 1756be34d1a3SDavid Howells return q; 17571da177e4SLinus Torvalds } 17581da177e4SLinus Torvalds 1759be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1760be34d1a3SDavid Howells 1761ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 17628aec0809SAl Viro { 1763cb338d06SAl Viro struct mount *tree; 176497216be0SAl Viro namespace_lock(); 1765cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1766cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1767cd4a4017SEric W. Biederman else 176887129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 176987129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1770328e6d90SAl Viro namespace_unlock(); 1771be34d1a3SDavid Howells if (IS_ERR(tree)) 177252e220d3SDan Carpenter return ERR_CAST(tree); 1773be34d1a3SDavid Howells return &tree->mnt; 17748aec0809SAl Viro } 17758aec0809SAl Viro 17768aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 17778aec0809SAl Viro { 177897216be0SAl Viro namespace_lock(); 1779719ea2fbSAl Viro lock_mount_hash(); 1780e819f152SEric W. Biederman umount_tree(real_mount(mnt), UMOUNT_SYNC); 1781719ea2fbSAl Viro unlock_mount_hash(); 17823ab6abeeSAl Viro namespace_unlock(); 17838aec0809SAl Viro } 17848aec0809SAl Viro 1785c771d683SMiklos Szeredi /** 1786c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1787c771d683SMiklos Szeredi * 1788c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1789c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1790c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1791c771d683SMiklos Szeredi * 1792c771d683SMiklos Szeredi * Release with mntput(). 1793c771d683SMiklos Szeredi */ 1794ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1795c771d683SMiklos Szeredi { 1796c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1797c771d683SMiklos Szeredi struct mount *new_mnt; 1798c771d683SMiklos Szeredi 1799c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1800c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1801c771d683SMiklos Szeredi 1802c771d683SMiklos Szeredi down_read(&namespace_sem); 1803c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1804c771d683SMiklos Szeredi up_read(&namespace_sem); 1805c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1806c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1807c771d683SMiklos Szeredi 1808c771d683SMiklos Szeredi return &new_mnt->mnt; 1809c771d683SMiklos Szeredi } 1810c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1811c771d683SMiklos Szeredi 18121f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18131f707137SAl Viro struct vfsmount *root) 18141f707137SAl Viro { 18151a4eeaf2SAl Viro struct mount *mnt; 18161f707137SAl Viro int res = f(root, arg); 18171f707137SAl Viro if (res) 18181f707137SAl Viro return res; 18191a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18201a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18211f707137SAl Viro if (res) 18221f707137SAl Viro return res; 18231f707137SAl Viro } 18241f707137SAl Viro return 0; 18251f707137SAl Viro } 18261f707137SAl Viro 18274b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1828719f5d7fSMiklos Szeredi { 1829315fc83eSAl Viro struct mount *p; 1830719f5d7fSMiklos Szeredi 1831909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1832fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 18334b8b21f4SAl Viro mnt_release_group_id(p); 1834719f5d7fSMiklos Szeredi } 1835719f5d7fSMiklos Szeredi } 1836719f5d7fSMiklos Szeredi 18374b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1838719f5d7fSMiklos Szeredi { 1839315fc83eSAl Viro struct mount *p; 1840719f5d7fSMiklos Szeredi 1841909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1842fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 18434b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1844719f5d7fSMiklos Szeredi if (err) { 18454b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1846719f5d7fSMiklos Szeredi return err; 1847719f5d7fSMiklos Szeredi } 1848719f5d7fSMiklos Szeredi } 1849719f5d7fSMiklos Szeredi } 1850719f5d7fSMiklos Szeredi 1851719f5d7fSMiklos Szeredi return 0; 1852719f5d7fSMiklos Szeredi } 1853719f5d7fSMiklos Szeredi 1854d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1855d2921684SEric W. Biederman { 1856d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1857d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1858d2921684SEric W. Biederman struct mount *p; 1859d2921684SEric W. Biederman 1860d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1861d2921684SEric W. Biederman mounts++; 1862d2921684SEric W. Biederman 1863d2921684SEric W. Biederman old = ns->mounts; 1864d2921684SEric W. Biederman pending = ns->pending_mounts; 1865d2921684SEric W. Biederman sum = old + pending; 1866d2921684SEric W. Biederman if ((old > sum) || 1867d2921684SEric W. Biederman (pending > sum) || 1868d2921684SEric W. Biederman (max < sum) || 1869d2921684SEric W. Biederman (mounts > (max - sum))) 1870d2921684SEric W. Biederman return -ENOSPC; 1871d2921684SEric W. Biederman 1872d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1873d2921684SEric W. Biederman return 0; 1874d2921684SEric W. Biederman } 1875d2921684SEric W. Biederman 1876b90fa9aeSRam Pai /* 1877b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1878b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 187921444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 188021444403SRam Pai * store the parent mount and mountpoint dentry. 188121444403SRam Pai * (done when source_mnt is moved) 1882b90fa9aeSRam Pai * 1883b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1884b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 18859676f0c6SRam Pai * --------------------------------------------------------------------------- 1886b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 18879676f0c6SRam Pai * |************************************************************************** 18889676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 18899676f0c6SRam Pai * | dest | | | | | 18909676f0c6SRam Pai * | | | | | | | 18919676f0c6SRam Pai * | v | | | | | 18929676f0c6SRam Pai * |************************************************************************** 18939676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 18945afe0022SRam Pai * | | | | | | 18959676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 18969676f0c6SRam Pai * *************************************************************************** 1897b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1898b90fa9aeSRam Pai * destination mount. 1899b90fa9aeSRam Pai * 1900b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1901b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1902b90fa9aeSRam Pai * the peer group of the source mount. 1903b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1904b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1905b90fa9aeSRam Pai * mount. 19065afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 19075afe0022SRam Pai * of the destination mount and the cloned mount is made slave 19085afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 19095afe0022SRam Pai * is marked as 'shared and slave'. 19105afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 19115afe0022SRam Pai * source mount. 19125afe0022SRam Pai * 19139676f0c6SRam Pai * --------------------------------------------------------------------------- 191421444403SRam Pai * | MOVE MOUNT OPERATION | 19159676f0c6SRam Pai * |************************************************************************** 19169676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19179676f0c6SRam Pai * | dest | | | | | 19189676f0c6SRam Pai * | | | | | | | 19199676f0c6SRam Pai * | v | | | | | 19209676f0c6SRam Pai * |************************************************************************** 19219676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 19225afe0022SRam Pai * | | | | | | 19239676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 19249676f0c6SRam Pai * *************************************************************************** 19255afe0022SRam Pai * 19265afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 19275afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 192821444403SRam Pai * (+*) the mount is moved to the destination. 19295afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 19305afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 19315afe0022SRam Pai * the mount is marked as 'shared and slave'. 19325afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1933b90fa9aeSRam Pai * 1934b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1935b90fa9aeSRam Pai * applied to each mount in the tree. 1936b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1937b90fa9aeSRam Pai * in allocations. 1938b90fa9aeSRam Pai */ 19390fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 194084d17192SAl Viro struct mount *dest_mnt, 194184d17192SAl Viro struct mountpoint *dest_mp, 194284d17192SAl Viro struct path *parent_path) 1943b90fa9aeSRam Pai { 194438129a13SAl Viro HLIST_HEAD(tree_list); 1945d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 1946315fc83eSAl Viro struct mount *child, *p; 194738129a13SAl Viro struct hlist_node *n; 1948719f5d7fSMiklos Szeredi int err; 1949b90fa9aeSRam Pai 1950d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 1951d2921684SEric W. Biederman if (!parent_path) { 1952d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 1953d2921684SEric W. Biederman if (err) 1954d2921684SEric W. Biederman goto out; 1955d2921684SEric W. Biederman } 1956d2921684SEric W. Biederman 1957fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 19580fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1959719f5d7fSMiklos Szeredi if (err) 1960719f5d7fSMiklos Szeredi goto out; 196184d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1962f2ebb3a9SAl Viro lock_mount_hash(); 1963719f5d7fSMiklos Szeredi if (err) 1964719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1965909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 19660f0afb1dSAl Viro set_mnt_shared(p); 19670b1b901bSAl Viro } else { 19680b1b901bSAl Viro lock_mount_hash(); 1969b90fa9aeSRam Pai } 19701a390689SAl Viro if (parent_path) { 19710fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 197284d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1973143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 197421444403SRam Pai } else { 197584d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 19761d6a32acSAl Viro commit_tree(source_mnt, NULL); 197721444403SRam Pai } 1978b90fa9aeSRam Pai 197938129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 19801d6a32acSAl Viro struct mount *q; 198138129a13SAl Viro hlist_del_init(&child->mnt_hash); 19821d6a32acSAl Viro q = __lookup_mnt_last(&child->mnt_parent->mnt, 19831d6a32acSAl Viro child->mnt_mountpoint); 19841d6a32acSAl Viro commit_tree(child, q); 1985b90fa9aeSRam Pai } 1986719ea2fbSAl Viro unlock_mount_hash(); 198799b7db7bSNick Piggin 1988b90fa9aeSRam Pai return 0; 1989719f5d7fSMiklos Szeredi 1990719f5d7fSMiklos Szeredi out_cleanup_ids: 1991f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 1992f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 1993d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 1994e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 1995f2ebb3a9SAl Viro } 1996f2ebb3a9SAl Viro unlock_mount_hash(); 19970fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1998719f5d7fSMiklos Szeredi out: 1999d2921684SEric W. Biederman ns->pending_mounts = 0; 2000719f5d7fSMiklos Szeredi return err; 2001b90fa9aeSRam Pai } 2002b90fa9aeSRam Pai 200384d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2004b12cea91SAl Viro { 2005b12cea91SAl Viro struct vfsmount *mnt; 200684d17192SAl Viro struct dentry *dentry = path->dentry; 2007b12cea91SAl Viro retry: 20085955102cSAl Viro inode_lock(dentry->d_inode); 200984d17192SAl Viro if (unlikely(cant_mount(dentry))) { 20105955102cSAl Viro inode_unlock(dentry->d_inode); 201184d17192SAl Viro return ERR_PTR(-ENOENT); 2012b12cea91SAl Viro } 201397216be0SAl Viro namespace_lock(); 2014b12cea91SAl Viro mnt = lookup_mnt(path); 201584d17192SAl Viro if (likely(!mnt)) { 2016e2dfa935SEric W. Biederman struct mountpoint *mp = lookup_mountpoint(dentry); 2017e2dfa935SEric W. Biederman if (!mp) 2018e2dfa935SEric W. Biederman mp = new_mountpoint(dentry); 201984d17192SAl Viro if (IS_ERR(mp)) { 202097216be0SAl Viro namespace_unlock(); 20215955102cSAl Viro inode_unlock(dentry->d_inode); 202284d17192SAl Viro return mp; 202384d17192SAl Viro } 202484d17192SAl Viro return mp; 202584d17192SAl Viro } 202697216be0SAl Viro namespace_unlock(); 20275955102cSAl Viro inode_unlock(path->dentry->d_inode); 2028b12cea91SAl Viro path_put(path); 2029b12cea91SAl Viro path->mnt = mnt; 203084d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2031b12cea91SAl Viro goto retry; 2032b12cea91SAl Viro } 2033b12cea91SAl Viro 203484d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2035b12cea91SAl Viro { 203684d17192SAl Viro struct dentry *dentry = where->m_dentry; 203784d17192SAl Viro put_mountpoint(where); 2038328e6d90SAl Viro namespace_unlock(); 20395955102cSAl Viro inode_unlock(dentry->d_inode); 2040b12cea91SAl Viro } 2041b12cea91SAl Viro 204284d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 20431da177e4SLinus Torvalds { 204495bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 20451da177e4SLinus Torvalds return -EINVAL; 20461da177e4SLinus Torvalds 2047e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2048e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 20491da177e4SLinus Torvalds return -ENOTDIR; 20501da177e4SLinus Torvalds 205184d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 20521da177e4SLinus Torvalds } 20531da177e4SLinus Torvalds 20541da177e4SLinus Torvalds /* 20557a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 20567a2e8a8fSValerie Aurora */ 20577a2e8a8fSValerie Aurora 20587a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 20597a2e8a8fSValerie Aurora { 20607c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 20617a2e8a8fSValerie Aurora 20627a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 20637a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 20647a2e8a8fSValerie Aurora return 0; 20657a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 20667a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 20677a2e8a8fSValerie Aurora return 0; 20687a2e8a8fSValerie Aurora return type; 20697a2e8a8fSValerie Aurora } 20707a2e8a8fSValerie Aurora 20717a2e8a8fSValerie Aurora /* 207207b20889SRam Pai * recursively change the type of the mountpoint. 207307b20889SRam Pai */ 20740a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 207507b20889SRam Pai { 2076315fc83eSAl Viro struct mount *m; 20774b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 207807b20889SRam Pai int recurse = flag & MS_REC; 20797a2e8a8fSValerie Aurora int type; 2080719f5d7fSMiklos Szeredi int err = 0; 208107b20889SRam Pai 20822d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 208307b20889SRam Pai return -EINVAL; 208407b20889SRam Pai 20857a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 20867a2e8a8fSValerie Aurora if (!type) 20877a2e8a8fSValerie Aurora return -EINVAL; 20887a2e8a8fSValerie Aurora 208997216be0SAl Viro namespace_lock(); 2090719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2091719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2092719f5d7fSMiklos Szeredi if (err) 2093719f5d7fSMiklos Szeredi goto out_unlock; 2094719f5d7fSMiklos Szeredi } 2095719f5d7fSMiklos Szeredi 2096719ea2fbSAl Viro lock_mount_hash(); 2097909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 20980f0afb1dSAl Viro change_mnt_propagation(m, type); 2099719ea2fbSAl Viro unlock_mount_hash(); 2100719f5d7fSMiklos Szeredi 2101719f5d7fSMiklos Szeredi out_unlock: 210297216be0SAl Viro namespace_unlock(); 2103719f5d7fSMiklos Szeredi return err; 210407b20889SRam Pai } 210507b20889SRam Pai 21065ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 21075ff9d8a6SEric W. Biederman { 21085ff9d8a6SEric W. Biederman struct mount *child; 21095ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 21105ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 21115ff9d8a6SEric W. Biederman continue; 21125ff9d8a6SEric W. Biederman 21135ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 21145ff9d8a6SEric W. Biederman return true; 21155ff9d8a6SEric W. Biederman } 21165ff9d8a6SEric W. Biederman return false; 21175ff9d8a6SEric W. Biederman } 21185ff9d8a6SEric W. Biederman 211907b20889SRam Pai /* 21201da177e4SLinus Torvalds * do loopback mount. 21211da177e4SLinus Torvalds */ 2122808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 21232dafe1c4SEric Sandeen int recurse) 21241da177e4SLinus Torvalds { 21252d92ab3cSAl Viro struct path old_path; 212684d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 212784d17192SAl Viro struct mountpoint *mp; 212857eccb83SAl Viro int err; 21291da177e4SLinus Torvalds if (!old_name || !*old_name) 21301da177e4SLinus Torvalds return -EINVAL; 2131815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 21321da177e4SLinus Torvalds if (err) 21331da177e4SLinus Torvalds return err; 21341da177e4SLinus Torvalds 21358823c079SEric W. Biederman err = -EINVAL; 21364ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 21378823c079SEric W. Biederman goto out; 21388823c079SEric W. Biederman 213984d17192SAl Viro mp = lock_mount(path); 214084d17192SAl Viro err = PTR_ERR(mp); 214184d17192SAl Viro if (IS_ERR(mp)) 21429676f0c6SRam Pai goto out; 21439676f0c6SRam Pai 214487129cc0SAl Viro old = real_mount(old_path.mnt); 214584d17192SAl Viro parent = real_mount(path->mnt); 214687129cc0SAl Viro 2147b12cea91SAl Viro err = -EINVAL; 2148fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2149b12cea91SAl Viro goto out2; 2150b12cea91SAl Viro 2151e149ed2bSAl Viro if (!check_mnt(parent)) 2152e149ed2bSAl Viro goto out2; 2153e149ed2bSAl Viro 2154e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2155b12cea91SAl Viro goto out2; 2156ccd48bc7SAl Viro 21575ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 21585ff9d8a6SEric W. Biederman goto out2; 21595ff9d8a6SEric W. Biederman 21601da177e4SLinus Torvalds if (recurse) 21614ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 21621da177e4SLinus Torvalds else 216387129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 21641da177e4SLinus Torvalds 2165be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2166be34d1a3SDavid Howells err = PTR_ERR(mnt); 2167e9c5d8a5SAndrey Vagin goto out2; 2168be34d1a3SDavid Howells } 2169ccd48bc7SAl Viro 21705ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 21715ff9d8a6SEric W. Biederman 217284d17192SAl Viro err = graft_tree(mnt, parent, mp); 21731da177e4SLinus Torvalds if (err) { 2174719ea2fbSAl Viro lock_mount_hash(); 2175e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2176719ea2fbSAl Viro unlock_mount_hash(); 21775b83d2c5SRam Pai } 2178b12cea91SAl Viro out2: 217984d17192SAl Viro unlock_mount(mp); 2180ccd48bc7SAl Viro out: 21812d92ab3cSAl Viro path_put(&old_path); 21821da177e4SLinus Torvalds return err; 21831da177e4SLinus Torvalds } 21841da177e4SLinus Torvalds 21852e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 21862e4b7fcdSDave Hansen { 21872e4b7fcdSDave Hansen int error = 0; 21882e4b7fcdSDave Hansen int readonly_request = 0; 21892e4b7fcdSDave Hansen 21902e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 21912e4b7fcdSDave Hansen readonly_request = 1; 21922e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 21932e4b7fcdSDave Hansen return 0; 21942e4b7fcdSDave Hansen 21952e4b7fcdSDave Hansen if (readonly_request) 219683adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 21972e4b7fcdSDave Hansen else 219883adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 21992e4b7fcdSDave Hansen return error; 22002e4b7fcdSDave Hansen } 22012e4b7fcdSDave Hansen 22021da177e4SLinus Torvalds /* 22031da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 22041da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 22051da177e4SLinus Torvalds * on it - tough luck. 22061da177e4SLinus Torvalds */ 22070a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 22081da177e4SLinus Torvalds void *data) 22091da177e4SLinus Torvalds { 22101da177e4SLinus Torvalds int err; 22112d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2212143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 22131da177e4SLinus Torvalds 2214143c8c91SAl Viro if (!check_mnt(mnt)) 22151da177e4SLinus Torvalds return -EINVAL; 22161da177e4SLinus Torvalds 22172d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 22181da177e4SLinus Torvalds return -EINVAL; 22191da177e4SLinus Torvalds 222007b64558SEric W. Biederman /* Don't allow changing of locked mnt flags. 222107b64558SEric W. Biederman * 222207b64558SEric W. Biederman * No locks need to be held here while testing the various 222307b64558SEric W. Biederman * MNT_LOCK flags because those flags can never be cleared 222407b64558SEric W. Biederman * once they are set. 222507b64558SEric W. Biederman */ 222607b64558SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 222707b64558SEric W. Biederman !(mnt_flags & MNT_READONLY)) { 222807b64558SEric W. Biederman return -EPERM; 222907b64558SEric W. Biederman } 22309566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 22319566d674SEric W. Biederman !(mnt_flags & MNT_NODEV)) { 22329566d674SEric W. Biederman return -EPERM; 22339566d674SEric W. Biederman } 22349566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && 22359566d674SEric W. Biederman !(mnt_flags & MNT_NOSUID)) { 22369566d674SEric W. Biederman return -EPERM; 22379566d674SEric W. Biederman } 22389566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && 22399566d674SEric W. Biederman !(mnt_flags & MNT_NOEXEC)) { 22409566d674SEric W. Biederman return -EPERM; 22419566d674SEric W. Biederman } 22429566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 22439566d674SEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { 22449566d674SEric W. Biederman return -EPERM; 22459566d674SEric W. Biederman } 22469566d674SEric W. Biederman 2247ff36fe2cSEric Paris err = security_sb_remount(sb, data); 2248ff36fe2cSEric Paris if (err) 2249ff36fe2cSEric Paris return err; 2250ff36fe2cSEric Paris 22511da177e4SLinus Torvalds down_write(&sb->s_umount); 22522e4b7fcdSDave Hansen if (flags & MS_BIND) 22532d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 225457eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 225557eccb83SAl Viro err = -EPERM; 22564aa98cf7SAl Viro else 22571da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 22587b43a79fSAl Viro if (!err) { 2259719ea2fbSAl Viro lock_mount_hash(); 2260a6138db8SEric W. Biederman mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 2261143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 2262143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2263719ea2fbSAl Viro unlock_mount_hash(); 22640e55a7ccSDan Williams } 22656339dab8SAl Viro up_write(&sb->s_umount); 22661da177e4SLinus Torvalds return err; 22671da177e4SLinus Torvalds } 22681da177e4SLinus Torvalds 2269cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 22709676f0c6SRam Pai { 2271315fc83eSAl Viro struct mount *p; 2272909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2273fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 22749676f0c6SRam Pai return 1; 22759676f0c6SRam Pai } 22769676f0c6SRam Pai return 0; 22779676f0c6SRam Pai } 22789676f0c6SRam Pai 2279808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 22801da177e4SLinus Torvalds { 22812d92ab3cSAl Viro struct path old_path, parent_path; 2282676da58dSAl Viro struct mount *p; 22830fb54e50SAl Viro struct mount *old; 228484d17192SAl Viro struct mountpoint *mp; 228557eccb83SAl Viro int err; 22861da177e4SLinus Torvalds if (!old_name || !*old_name) 22871da177e4SLinus Torvalds return -EINVAL; 22882d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 22891da177e4SLinus Torvalds if (err) 22901da177e4SLinus Torvalds return err; 22911da177e4SLinus Torvalds 229284d17192SAl Viro mp = lock_mount(path); 229384d17192SAl Viro err = PTR_ERR(mp); 229484d17192SAl Viro if (IS_ERR(mp)) 2295cc53ce53SDavid Howells goto out; 2296cc53ce53SDavid Howells 2297143c8c91SAl Viro old = real_mount(old_path.mnt); 2298fc7be130SAl Viro p = real_mount(path->mnt); 2299143c8c91SAl Viro 23001da177e4SLinus Torvalds err = -EINVAL; 2301fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 23021da177e4SLinus Torvalds goto out1; 23031da177e4SLinus Torvalds 23045ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 23055ff9d8a6SEric W. Biederman goto out1; 23065ff9d8a6SEric W. Biederman 23071da177e4SLinus Torvalds err = -EINVAL; 23082d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 230921444403SRam Pai goto out1; 23101da177e4SLinus Torvalds 2311676da58dSAl Viro if (!mnt_has_parent(old)) 231221444403SRam Pai goto out1; 23131da177e4SLinus Torvalds 2314e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2315e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 231621444403SRam Pai goto out1; 231721444403SRam Pai /* 231821444403SRam Pai * Don't move a mount residing in a shared parent. 231921444403SRam Pai */ 2320fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 232121444403SRam Pai goto out1; 23229676f0c6SRam Pai /* 23239676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 23249676f0c6SRam Pai * mount which is shared. 23259676f0c6SRam Pai */ 2326fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 23279676f0c6SRam Pai goto out1; 23281da177e4SLinus Torvalds err = -ELOOP; 2329fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2330676da58dSAl Viro if (p == old) 233121444403SRam Pai goto out1; 23321da177e4SLinus Torvalds 233384d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 23344ac91378SJan Blunck if (err) 233521444403SRam Pai goto out1; 23361da177e4SLinus Torvalds 23371da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 23381da177e4SLinus Torvalds * automatically */ 23396776db3dSAl Viro list_del_init(&old->mnt_expire); 23401da177e4SLinus Torvalds out1: 234184d17192SAl Viro unlock_mount(mp); 23421da177e4SLinus Torvalds out: 23431da177e4SLinus Torvalds if (!err) 23441a390689SAl Viro path_put(&parent_path); 23452d92ab3cSAl Viro path_put(&old_path); 23461da177e4SLinus Torvalds return err; 23471da177e4SLinus Torvalds } 23481da177e4SLinus Torvalds 23499d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 23509d412a43SAl Viro { 23519d412a43SAl Viro int err; 23529d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 23539d412a43SAl Viro if (subtype) { 23549d412a43SAl Viro subtype++; 23559d412a43SAl Viro err = -EINVAL; 23569d412a43SAl Viro if (!subtype[0]) 23579d412a43SAl Viro goto err; 23589d412a43SAl Viro } else 23599d412a43SAl Viro subtype = ""; 23609d412a43SAl Viro 23619d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 23629d412a43SAl Viro err = -ENOMEM; 23639d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 23649d412a43SAl Viro goto err; 23659d412a43SAl Viro return mnt; 23669d412a43SAl Viro 23679d412a43SAl Viro err: 23689d412a43SAl Viro mntput(mnt); 23699d412a43SAl Viro return ERR_PTR(err); 23709d412a43SAl Viro } 23719d412a43SAl Viro 23729d412a43SAl Viro /* 23739d412a43SAl Viro * add a mount into a namespace's mount tree 23749d412a43SAl Viro */ 237595bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 23769d412a43SAl Viro { 237784d17192SAl Viro struct mountpoint *mp; 237884d17192SAl Viro struct mount *parent; 23799d412a43SAl Viro int err; 23809d412a43SAl Viro 2381f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 23829d412a43SAl Viro 238384d17192SAl Viro mp = lock_mount(path); 238484d17192SAl Viro if (IS_ERR(mp)) 238584d17192SAl Viro return PTR_ERR(mp); 23869d412a43SAl Viro 238784d17192SAl Viro parent = real_mount(path->mnt); 23889d412a43SAl Viro err = -EINVAL; 238984d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2390156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2391156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 23929d412a43SAl Viro goto unlock; 2393156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 239484d17192SAl Viro if (!parent->mnt_ns) 2395156cacb1SAl Viro goto unlock; 2396156cacb1SAl Viro } 23979d412a43SAl Viro 23989d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 23999d412a43SAl Viro err = -EBUSY; 240095bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 24019d412a43SAl Viro path->mnt->mnt_root == path->dentry) 24029d412a43SAl Viro goto unlock; 24039d412a43SAl Viro 24049d412a43SAl Viro err = -EINVAL; 2405e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 24069d412a43SAl Viro goto unlock; 24079d412a43SAl Viro 240895bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 240984d17192SAl Viro err = graft_tree(newmnt, parent, mp); 24109d412a43SAl Viro 24119d412a43SAl Viro unlock: 241284d17192SAl Viro unlock_mount(mp); 24139d412a43SAl Viro return err; 24149d412a43SAl Viro } 2415b1e75df4SAl Viro 24168654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); 24171b852bceSEric W. Biederman 24181da177e4SLinus Torvalds /* 24191da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 24201da177e4SLinus Torvalds * namespace's tree 24211da177e4SLinus Torvalds */ 24220c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2423808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 24241da177e4SLinus Torvalds { 24250c55cfc4SEric W. Biederman struct file_system_type *type; 24261da177e4SLinus Torvalds struct vfsmount *mnt; 242715f9a3f3SAl Viro int err; 24281da177e4SLinus Torvalds 24290c55cfc4SEric W. Biederman if (!fstype) 24301da177e4SLinus Torvalds return -EINVAL; 24311da177e4SLinus Torvalds 24320c55cfc4SEric W. Biederman type = get_fs_type(fstype); 24330c55cfc4SEric W. Biederman if (!type) 24340c55cfc4SEric W. Biederman return -ENODEV; 24350c55cfc4SEric W. Biederman 24360c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 24370c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 24380c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 24390c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 24400c55cfc4SEric W. Biederman 24410c55cfc4SEric W. Biederman put_filesystem(type); 24421da177e4SLinus Torvalds if (IS_ERR(mnt)) 24431da177e4SLinus Torvalds return PTR_ERR(mnt); 24441da177e4SLinus Torvalds 24458654df4eSEric W. Biederman if (mount_too_revealing(mnt, &mnt_flags)) { 24468654df4eSEric W. Biederman mntput(mnt); 24478654df4eSEric W. Biederman return -EPERM; 24488654df4eSEric W. Biederman } 24498654df4eSEric W. Biederman 245095bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 245115f9a3f3SAl Viro if (err) 245215f9a3f3SAl Viro mntput(mnt); 245315f9a3f3SAl Viro return err; 24541da177e4SLinus Torvalds } 24551da177e4SLinus Torvalds 245619a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 245719a167afSAl Viro { 24586776db3dSAl Viro struct mount *mnt = real_mount(m); 245919a167afSAl Viro int err; 246019a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 246119a167afSAl Viro * expired before we get a chance to add it 246219a167afSAl Viro */ 24636776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 246419a167afSAl Viro 246519a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 246619a167afSAl Viro m->mnt_root == path->dentry) { 2467b1e75df4SAl Viro err = -ELOOP; 2468b1e75df4SAl Viro goto fail; 246919a167afSAl Viro } 247019a167afSAl Viro 247195bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2472b1e75df4SAl Viro if (!err) 2473b1e75df4SAl Viro return 0; 2474b1e75df4SAl Viro fail: 2475b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 24766776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 247797216be0SAl Viro namespace_lock(); 24786776db3dSAl Viro list_del_init(&mnt->mnt_expire); 247997216be0SAl Viro namespace_unlock(); 248019a167afSAl Viro } 2481b1e75df4SAl Viro mntput(m); 2482b1e75df4SAl Viro mntput(m); 248319a167afSAl Viro return err; 248419a167afSAl Viro } 248519a167afSAl Viro 2486ea5b778aSDavid Howells /** 2487ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2488ea5b778aSDavid Howells * @mnt: The mount to list. 2489ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2490ea5b778aSDavid Howells */ 2491ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2492ea5b778aSDavid Howells { 249397216be0SAl Viro namespace_lock(); 2494ea5b778aSDavid Howells 24956776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2496ea5b778aSDavid Howells 249797216be0SAl Viro namespace_unlock(); 2498ea5b778aSDavid Howells } 2499ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2500ea5b778aSDavid Howells 2501ea5b778aSDavid Howells /* 25021da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 25031da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 25041da177e4SLinus Torvalds * here 25051da177e4SLinus Torvalds */ 25061da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 25071da177e4SLinus Torvalds { 2508761d5c38SAl Viro struct mount *mnt, *next; 25091da177e4SLinus Torvalds LIST_HEAD(graveyard); 25101da177e4SLinus Torvalds 25111da177e4SLinus Torvalds if (list_empty(mounts)) 25121da177e4SLinus Torvalds return; 25131da177e4SLinus Torvalds 251497216be0SAl Viro namespace_lock(); 2515719ea2fbSAl Viro lock_mount_hash(); 25161da177e4SLinus Torvalds 25171da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 25181da177e4SLinus Torvalds * following criteria: 25191da177e4SLinus Torvalds * - only referenced by its parent vfsmount 25201da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 25211da177e4SLinus Torvalds * cleared by mntput()) 25221da177e4SLinus Torvalds */ 25236776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2524863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 25251ab59738SAl Viro propagate_mount_busy(mnt, 1)) 25261da177e4SLinus Torvalds continue; 25276776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 25281da177e4SLinus Torvalds } 2529bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 25306776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2531143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2532e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2533bcc5c7d2SAl Viro } 2534719ea2fbSAl Viro unlock_mount_hash(); 25353ab6abeeSAl Viro namespace_unlock(); 25361da177e4SLinus Torvalds } 25371da177e4SLinus Torvalds 25381da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 25391da177e4SLinus Torvalds 25401da177e4SLinus Torvalds /* 25415528f911STrond Myklebust * Ripoff of 'select_parent()' 25425528f911STrond Myklebust * 25435528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 25445528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 25455528f911STrond Myklebust */ 2546692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 25475528f911STrond Myklebust { 2548692afc31SAl Viro struct mount *this_parent = parent; 25495528f911STrond Myklebust struct list_head *next; 25505528f911STrond Myklebust int found = 0; 25515528f911STrond Myklebust 25525528f911STrond Myklebust repeat: 25536b41d536SAl Viro next = this_parent->mnt_mounts.next; 25545528f911STrond Myklebust resume: 25556b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 25565528f911STrond Myklebust struct list_head *tmp = next; 25576b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 25585528f911STrond Myklebust 25595528f911STrond Myklebust next = tmp->next; 2560692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 25615528f911STrond Myklebust continue; 25625528f911STrond Myklebust /* 25635528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 25645528f911STrond Myklebust */ 25656b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 25665528f911STrond Myklebust this_parent = mnt; 25675528f911STrond Myklebust goto repeat; 25685528f911STrond Myklebust } 25695528f911STrond Myklebust 25701ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 25716776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 25725528f911STrond Myklebust found++; 25735528f911STrond Myklebust } 25745528f911STrond Myklebust } 25755528f911STrond Myklebust /* 25765528f911STrond Myklebust * All done at this level ... ascend and resume the search 25775528f911STrond Myklebust */ 25785528f911STrond Myklebust if (this_parent != parent) { 25796b41d536SAl Viro next = this_parent->mnt_child.next; 25800714a533SAl Viro this_parent = this_parent->mnt_parent; 25815528f911STrond Myklebust goto resume; 25825528f911STrond Myklebust } 25835528f911STrond Myklebust return found; 25845528f911STrond Myklebust } 25855528f911STrond Myklebust 25865528f911STrond Myklebust /* 25875528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 25885528f911STrond Myklebust * submounts of a specific parent mountpoint 258999b7db7bSNick Piggin * 259048a066e7SAl Viro * mount_lock must be held for write 25915528f911STrond Myklebust */ 2592b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 25935528f911STrond Myklebust { 25945528f911STrond Myklebust LIST_HEAD(graveyard); 2595761d5c38SAl Viro struct mount *m; 25965528f911STrond Myklebust 25975528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2598c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2599bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2600761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 26016776db3dSAl Viro mnt_expire); 2602143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2603e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2604bcc5c7d2SAl Viro } 2605bcc5c7d2SAl Viro } 26065528f911STrond Myklebust } 26075528f911STrond Myklebust 26085528f911STrond Myklebust /* 26091da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 26101da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 26111da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 26121da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 26131da177e4SLinus Torvalds */ 2614b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2615b58fed8bSRam Pai unsigned long n) 26161da177e4SLinus Torvalds { 26171da177e4SLinus Torvalds char *t = to; 26181da177e4SLinus Torvalds const char __user *f = from; 26191da177e4SLinus Torvalds char c; 26201da177e4SLinus Torvalds 26211da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 26221da177e4SLinus Torvalds return n; 26231da177e4SLinus Torvalds 26241da177e4SLinus Torvalds while (n) { 26251da177e4SLinus Torvalds if (__get_user(c, f)) { 26261da177e4SLinus Torvalds memset(t, 0, n); 26271da177e4SLinus Torvalds break; 26281da177e4SLinus Torvalds } 26291da177e4SLinus Torvalds *t++ = c; 26301da177e4SLinus Torvalds f++; 26311da177e4SLinus Torvalds n--; 26321da177e4SLinus Torvalds } 26331da177e4SLinus Torvalds return n; 26341da177e4SLinus Torvalds } 26351da177e4SLinus Torvalds 2636b40ef869SAl Viro void *copy_mount_options(const void __user * data) 26371da177e4SLinus Torvalds { 26381da177e4SLinus Torvalds int i; 26391da177e4SLinus Torvalds unsigned long size; 2640b40ef869SAl Viro char *copy; 26411da177e4SLinus Torvalds 26421da177e4SLinus Torvalds if (!data) 2643b40ef869SAl Viro return NULL; 26441da177e4SLinus Torvalds 2645b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2646b40ef869SAl Viro if (!copy) 2647b40ef869SAl Viro return ERR_PTR(-ENOMEM); 26481da177e4SLinus Torvalds 26491da177e4SLinus Torvalds /* We only care that *some* data at the address the user 26501da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 26511da177e4SLinus Torvalds * the remainder of the page. 26521da177e4SLinus Torvalds */ 26531da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 26541da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 26551da177e4SLinus Torvalds if (size > PAGE_SIZE) 26561da177e4SLinus Torvalds size = PAGE_SIZE; 26571da177e4SLinus Torvalds 2658b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 26591da177e4SLinus Torvalds if (!i) { 2660b40ef869SAl Viro kfree(copy); 2661b40ef869SAl Viro return ERR_PTR(-EFAULT); 26621da177e4SLinus Torvalds } 26631da177e4SLinus Torvalds if (i != PAGE_SIZE) 2664b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2665b40ef869SAl Viro return copy; 26661da177e4SLinus Torvalds } 26671da177e4SLinus Torvalds 2668b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2669eca6f534SVegard Nossum { 2670b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2671eca6f534SVegard Nossum } 2672eca6f534SVegard Nossum 26731da177e4SLinus Torvalds /* 26741da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 26751da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 26761da177e4SLinus Torvalds * 26771da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 26781da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 26791da177e4SLinus Torvalds * information (or be NULL). 26801da177e4SLinus Torvalds * 26811da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 26821da177e4SLinus Torvalds * When the flags word was introduced its top half was required 26831da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 26841da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 26851da177e4SLinus Torvalds * and must be discarded. 26861da177e4SLinus Torvalds */ 26875e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2688808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 26891da177e4SLinus Torvalds { 26902d92ab3cSAl Viro struct path path; 26911da177e4SLinus Torvalds int retval = 0; 26921da177e4SLinus Torvalds int mnt_flags = 0; 26931da177e4SLinus Torvalds 26941da177e4SLinus Torvalds /* Discard magic */ 26951da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 26961da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 26971da177e4SLinus Torvalds 26981da177e4SLinus Torvalds /* Basic sanity checks */ 26991da177e4SLinus Torvalds if (data_page) 27001da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 27011da177e4SLinus Torvalds 2702a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 27035e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2704a27ab9f2STetsuo Handa if (retval) 2705a27ab9f2STetsuo Handa return retval; 2706a27ab9f2STetsuo Handa 2707a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2708a27ab9f2STetsuo Handa type_page, flags, data_page); 27090d5cadb8SAl Viro if (!retval && !may_mount()) 27100d5cadb8SAl Viro retval = -EPERM; 27119e8925b6SJeff Layton if (!retval && (flags & MS_MANDLOCK) && !may_mandlock()) 27129e8925b6SJeff Layton retval = -EPERM; 2713a27ab9f2STetsuo Handa if (retval) 2714a27ab9f2STetsuo Handa goto dput_out; 2715a27ab9f2STetsuo Handa 2716613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2717613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 27180a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 27190a1c01c9SMatthew Garrett 27201da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 27211da177e4SLinus Torvalds if (flags & MS_NOSUID) 27221da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 27231da177e4SLinus Torvalds if (flags & MS_NODEV) 27241da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 27251da177e4SLinus Torvalds if (flags & MS_NOEXEC) 27261da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2727fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2728fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2729fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2730fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2731d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2732d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 27332e4b7fcdSDave Hansen if (flags & MS_RDONLY) 27342e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2735fc33a7bbSChristoph Hellwig 2736ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2737ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2738ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2739ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2740ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2741ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2742ffbc6f0eSEric W. Biederman } 2743ffbc6f0eSEric W. Biederman 27447a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2745d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2746c568d683SMiklos Szeredi MS_STRICTATIME | MS_NOREMOTELOCK); 27471da177e4SLinus Torvalds 27481da177e4SLinus Torvalds if (flags & MS_REMOUNT) 27492d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 27501da177e4SLinus Torvalds data_page); 27511da177e4SLinus Torvalds else if (flags & MS_BIND) 27522d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 27539676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 27542d92ab3cSAl Viro retval = do_change_type(&path, flags); 27551da177e4SLinus Torvalds else if (flags & MS_MOVE) 27562d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 27571da177e4SLinus Torvalds else 27582d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 27591da177e4SLinus Torvalds dev_name, data_page); 27601da177e4SLinus Torvalds dput_out: 27612d92ab3cSAl Viro path_put(&path); 27621da177e4SLinus Torvalds return retval; 27631da177e4SLinus Torvalds } 27641da177e4SLinus Torvalds 2765537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 2766537f7ccbSEric W. Biederman { 2767537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 2768537f7ccbSEric W. Biederman } 2769537f7ccbSEric W. Biederman 2770537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 2771537f7ccbSEric W. Biederman { 2772537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 2773537f7ccbSEric W. Biederman } 2774537f7ccbSEric W. Biederman 2775771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2776771b1371SEric W. Biederman { 27776344c433SAl Viro ns_free_inum(&ns->ns); 2778537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 2779771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2780771b1371SEric W. Biederman kfree(ns); 2781771b1371SEric W. Biederman } 2782771b1371SEric W. Biederman 27838823c079SEric W. Biederman /* 27848823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 27858823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 27868823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 27878823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 27888823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 27898823c079SEric W. Biederman */ 27908823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 27918823c079SEric W. Biederman 2792771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2793cf8d2c11STrond Myklebust { 2794cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2795537f7ccbSEric W. Biederman struct ucounts *ucounts; 279698f842e6SEric W. Biederman int ret; 2797cf8d2c11STrond Myklebust 2798537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 2799537f7ccbSEric W. Biederman if (!ucounts) 2800df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 2801537f7ccbSEric W. Biederman 2802cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2803537f7ccbSEric W. Biederman if (!new_ns) { 2804537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 2805cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2806537f7ccbSEric W. Biederman } 28076344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 280898f842e6SEric W. Biederman if (ret) { 280998f842e6SEric W. Biederman kfree(new_ns); 2810537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 281198f842e6SEric W. Biederman return ERR_PTR(ret); 281298f842e6SEric W. Biederman } 281333c42940SAl Viro new_ns->ns.ops = &mntns_operations; 28148823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2815cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2816cf8d2c11STrond Myklebust new_ns->root = NULL; 2817cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2818cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2819cf8d2c11STrond Myklebust new_ns->event = 0; 2820771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2821537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 2822d2921684SEric W. Biederman new_ns->mounts = 0; 2823d2921684SEric W. Biederman new_ns->pending_mounts = 0; 2824cf8d2c11STrond Myklebust return new_ns; 2825cf8d2c11STrond Myklebust } 2826cf8d2c11STrond Myklebust 28270766f788SEmese Revfy __latent_entropy 28289559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 28299559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 28301da177e4SLinus Torvalds { 28316b3286edSKirill Korotaev struct mnt_namespace *new_ns; 28327f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2833315fc83eSAl Viro struct mount *p, *q; 28349559f689SAl Viro struct mount *old; 2835cb338d06SAl Viro struct mount *new; 28367a472ef4SEric W. Biederman int copy_flags; 28371da177e4SLinus Torvalds 28389559f689SAl Viro BUG_ON(!ns); 28399559f689SAl Viro 28409559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 28419559f689SAl Viro get_mnt_ns(ns); 28429559f689SAl Viro return ns; 28439559f689SAl Viro } 28449559f689SAl Viro 28459559f689SAl Viro old = ns->root; 28469559f689SAl Viro 2847771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2848cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2849cf8d2c11STrond Myklebust return new_ns; 28501da177e4SLinus Torvalds 285197216be0SAl Viro namespace_lock(); 28521da177e4SLinus Torvalds /* First pass: copy the tree topology */ 28534ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 28549559f689SAl Viro if (user_ns != ns->user_ns) 2855132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 28567a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2857be34d1a3SDavid Howells if (IS_ERR(new)) { 2858328e6d90SAl Viro namespace_unlock(); 2859771b1371SEric W. Biederman free_mnt_ns(new_ns); 2860be34d1a3SDavid Howells return ERR_CAST(new); 28611da177e4SLinus Torvalds } 2862be08d6d2SAl Viro new_ns->root = new; 28631a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 28641da177e4SLinus Torvalds 28651da177e4SLinus Torvalds /* 28661da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 28671da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 28681da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 28691da177e4SLinus Torvalds */ 2870909b0a88SAl Viro p = old; 2871cb338d06SAl Viro q = new; 28721da177e4SLinus Torvalds while (p) { 2873143c8c91SAl Viro q->mnt_ns = new_ns; 2874d2921684SEric W. Biederman new_ns->mounts++; 28759559f689SAl Viro if (new_fs) { 28769559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 28779559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2878315fc83eSAl Viro rootmnt = &p->mnt; 28791da177e4SLinus Torvalds } 28809559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 28819559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2882315fc83eSAl Viro pwdmnt = &p->mnt; 28831da177e4SLinus Torvalds } 28841da177e4SLinus Torvalds } 2885909b0a88SAl Viro p = next_mnt(p, old); 2886909b0a88SAl Viro q = next_mnt(q, new); 28874ce5d2b1SEric W. Biederman if (!q) 28884ce5d2b1SEric W. Biederman break; 28894ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 28904ce5d2b1SEric W. Biederman p = next_mnt(p, old); 28911da177e4SLinus Torvalds } 2892328e6d90SAl Viro namespace_unlock(); 28931da177e4SLinus Torvalds 28941da177e4SLinus Torvalds if (rootmnt) 2895f03c6599SAl Viro mntput(rootmnt); 28961da177e4SLinus Torvalds if (pwdmnt) 2897f03c6599SAl Viro mntput(pwdmnt); 28981da177e4SLinus Torvalds 2899741a2951SJANAK DESAI return new_ns; 2900741a2951SJANAK DESAI } 2901741a2951SJANAK DESAI 2902cf8d2c11STrond Myklebust /** 2903cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2904cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2905cf8d2c11STrond Myklebust */ 29061a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2907cf8d2c11STrond Myklebust { 2908771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2909cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 29101a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 29111a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2912be08d6d2SAl Viro new_ns->root = mnt; 2913d2921684SEric W. Biederman new_ns->mounts++; 2914b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2915c1334495SAl Viro } else { 29161a4eeaf2SAl Viro mntput(m); 2917cf8d2c11STrond Myklebust } 2918cf8d2c11STrond Myklebust return new_ns; 2919cf8d2c11STrond Myklebust } 2920cf8d2c11STrond Myklebust 2921ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2922ea441d11SAl Viro { 2923ea441d11SAl Viro struct mnt_namespace *ns; 2924d31da0f0SAl Viro struct super_block *s; 2925ea441d11SAl Viro struct path path; 2926ea441d11SAl Viro int err; 2927ea441d11SAl Viro 2928ea441d11SAl Viro ns = create_mnt_ns(mnt); 2929ea441d11SAl Viro if (IS_ERR(ns)) 2930ea441d11SAl Viro return ERR_CAST(ns); 2931ea441d11SAl Viro 2932ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2933ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2934ea441d11SAl Viro 2935ea441d11SAl Viro put_mnt_ns(ns); 2936ea441d11SAl Viro 2937ea441d11SAl Viro if (err) 2938ea441d11SAl Viro return ERR_PTR(err); 2939ea441d11SAl Viro 2940ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2941d31da0f0SAl Viro s = path.mnt->mnt_sb; 2942d31da0f0SAl Viro atomic_inc(&s->s_active); 2943ea441d11SAl Viro mntput(path.mnt); 2944ea441d11SAl Viro /* lock the sucker */ 2945d31da0f0SAl Viro down_write(&s->s_umount); 2946ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2947ea441d11SAl Viro return path.dentry; 2948ea441d11SAl Viro } 2949ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2950ea441d11SAl Viro 2951bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2952bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 29531da177e4SLinus Torvalds { 2954eca6f534SVegard Nossum int ret; 2955eca6f534SVegard Nossum char *kernel_type; 2956eca6f534SVegard Nossum char *kernel_dev; 2957b40ef869SAl Viro void *options; 29581da177e4SLinus Torvalds 2959b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 2960b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 2961b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 2962eca6f534SVegard Nossum goto out_type; 29631da177e4SLinus Torvalds 2964b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 2965b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 2966b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 2967eca6f534SVegard Nossum goto out_dev; 29681da177e4SLinus Torvalds 2969b40ef869SAl Viro options = copy_mount_options(data); 2970b40ef869SAl Viro ret = PTR_ERR(options); 2971b40ef869SAl Viro if (IS_ERR(options)) 2972eca6f534SVegard Nossum goto out_data; 29731da177e4SLinus Torvalds 2974b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 2975eca6f534SVegard Nossum 2976b40ef869SAl Viro kfree(options); 2977eca6f534SVegard Nossum out_data: 2978eca6f534SVegard Nossum kfree(kernel_dev); 2979eca6f534SVegard Nossum out_dev: 2980eca6f534SVegard Nossum kfree(kernel_type); 2981eca6f534SVegard Nossum out_type: 2982eca6f534SVegard Nossum return ret; 29831da177e4SLinus Torvalds } 29841da177e4SLinus Torvalds 29851da177e4SLinus Torvalds /* 2986afac7cbaSAl Viro * Return true if path is reachable from root 2987afac7cbaSAl Viro * 298848a066e7SAl Viro * namespace_sem or mount_lock is held 2989afac7cbaSAl Viro */ 2990643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2991afac7cbaSAl Viro const struct path *root) 2992afac7cbaSAl Viro { 2993643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2994a73324daSAl Viro dentry = mnt->mnt_mountpoint; 29950714a533SAl Viro mnt = mnt->mnt_parent; 2996afac7cbaSAl Viro } 2997643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2998afac7cbaSAl Viro } 2999afac7cbaSAl Viro 3000640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3001afac7cbaSAl Viro { 300225ab4c9bSYaowei Bai bool res; 300348a066e7SAl Viro read_seqlock_excl(&mount_lock); 3004643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 300548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3006afac7cbaSAl Viro return res; 3007afac7cbaSAl Viro } 3008afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3009afac7cbaSAl Viro 3010afac7cbaSAl Viro /* 30111da177e4SLinus Torvalds * pivot_root Semantics: 30121da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 30131da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 30141da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 30151da177e4SLinus Torvalds * 30161da177e4SLinus Torvalds * Restrictions: 30171da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 30181da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 30191da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 30201da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 30211da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 30221da177e4SLinus Torvalds * 30234a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 30244a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 30254a0d11faSNeil Brown * in this situation. 30264a0d11faSNeil Brown * 30271da177e4SLinus Torvalds * Notes: 30281da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 30291da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 30301da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 30311da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 30321da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 30331da177e4SLinus Torvalds * first. 30341da177e4SLinus Torvalds */ 30353480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 30363480b257SHeiko Carstens const char __user *, put_old) 30371da177e4SLinus Torvalds { 30382d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 303984d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 304084d17192SAl Viro struct mountpoint *old_mp, *root_mp; 30411da177e4SLinus Torvalds int error; 30421da177e4SLinus Torvalds 30439b40bc90SAl Viro if (!may_mount()) 30441da177e4SLinus Torvalds return -EPERM; 30451da177e4SLinus Torvalds 30462d8f3038SAl Viro error = user_path_dir(new_root, &new); 30471da177e4SLinus Torvalds if (error) 30481da177e4SLinus Torvalds goto out0; 30491da177e4SLinus Torvalds 30502d8f3038SAl Viro error = user_path_dir(put_old, &old); 30511da177e4SLinus Torvalds if (error) 30521da177e4SLinus Torvalds goto out1; 30531da177e4SLinus Torvalds 30542d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3055b12cea91SAl Viro if (error) 3056b12cea91SAl Viro goto out2; 30571da177e4SLinus Torvalds 3058f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 305984d17192SAl Viro old_mp = lock_mount(&old); 306084d17192SAl Viro error = PTR_ERR(old_mp); 306184d17192SAl Viro if (IS_ERR(old_mp)) 3062b12cea91SAl Viro goto out3; 3063b12cea91SAl Viro 30641da177e4SLinus Torvalds error = -EINVAL; 3065419148daSAl Viro new_mnt = real_mount(new.mnt); 3066419148daSAl Viro root_mnt = real_mount(root.mnt); 306784d17192SAl Viro old_mnt = real_mount(old.mnt); 306884d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3069fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3070fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3071b12cea91SAl Viro goto out4; 3072143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3073b12cea91SAl Viro goto out4; 30745ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 30755ff9d8a6SEric W. Biederman goto out4; 30761da177e4SLinus Torvalds error = -ENOENT; 3077f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3078b12cea91SAl Viro goto out4; 30791da177e4SLinus Torvalds error = -EBUSY; 308084d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3081b12cea91SAl Viro goto out4; /* loop, on the same file system */ 30821da177e4SLinus Torvalds error = -EINVAL; 30838c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3084b12cea91SAl Viro goto out4; /* not a mountpoint */ 3085676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3086b12cea91SAl Viro goto out4; /* not attached */ 308784d17192SAl Viro root_mp = root_mnt->mnt_mp; 30882d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3089b12cea91SAl Viro goto out4; /* not a mountpoint */ 3090676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3091b12cea91SAl Viro goto out4; /* not attached */ 30924ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 309384d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3094b12cea91SAl Viro goto out4; 30950d082601SEric W. Biederman /* make certain new is below the root */ 30960d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 30970d082601SEric W. Biederman goto out4; 309884d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3099719ea2fbSAl Viro lock_mount_hash(); 3100419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3101419148daSAl Viro detach_mnt(root_mnt, &root_parent); 31025ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 31035ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 31045ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 31055ff9d8a6SEric W. Biederman } 31064ac91378SJan Blunck /* mount old root on put_old */ 310784d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 31084ac91378SJan Blunck /* mount new_root on / */ 310984d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 31106b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 31114fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 31124fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 3113719ea2fbSAl Viro unlock_mount_hash(); 31142d8f3038SAl Viro chroot_fs_refs(&root, &new); 311584d17192SAl Viro put_mountpoint(root_mp); 31161da177e4SLinus Torvalds error = 0; 3117b12cea91SAl Viro out4: 311884d17192SAl Viro unlock_mount(old_mp); 3119b12cea91SAl Viro if (!error) { 31201a390689SAl Viro path_put(&root_parent); 31211a390689SAl Viro path_put(&parent_path); 3122b12cea91SAl Viro } 3123b12cea91SAl Viro out3: 31248c3ee42eSAl Viro path_put(&root); 3125b12cea91SAl Viro out2: 31262d8f3038SAl Viro path_put(&old); 31271da177e4SLinus Torvalds out1: 31282d8f3038SAl Viro path_put(&new); 31291da177e4SLinus Torvalds out0: 31301da177e4SLinus Torvalds return error; 31311da177e4SLinus Torvalds } 31321da177e4SLinus Torvalds 31331da177e4SLinus Torvalds static void __init init_mount_tree(void) 31341da177e4SLinus Torvalds { 31351da177e4SLinus Torvalds struct vfsmount *mnt; 31366b3286edSKirill Korotaev struct mnt_namespace *ns; 3137ac748a09SJan Blunck struct path root; 31380c55cfc4SEric W. Biederman struct file_system_type *type; 31391da177e4SLinus Torvalds 31400c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 31410c55cfc4SEric W. Biederman if (!type) 31420c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 31430c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 31440c55cfc4SEric W. Biederman put_filesystem(type); 31451da177e4SLinus Torvalds if (IS_ERR(mnt)) 31461da177e4SLinus Torvalds panic("Can't create rootfs"); 3147b3e19d92SNick Piggin 31483b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 31493b22edc5STrond Myklebust if (IS_ERR(ns)) 31501da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 31511da177e4SLinus Torvalds 31526b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 31536b3286edSKirill Korotaev get_mnt_ns(ns); 31541da177e4SLinus Torvalds 3155be08d6d2SAl Viro root.mnt = mnt; 3156be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3157da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3158ac748a09SJan Blunck 3159ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3160ac748a09SJan Blunck set_fs_root(current->fs, &root); 31611da177e4SLinus Torvalds } 31621da177e4SLinus Torvalds 316374bf17cfSDenis Cheng void __init mnt_init(void) 31641da177e4SLinus Torvalds { 316513f14b4dSEric Dumazet unsigned u; 316615a67dd8SRandy Dunlap int err; 31671da177e4SLinus Torvalds 31687d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 316920c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 31701da177e4SLinus Torvalds 31710818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 317238129a13SAl Viro sizeof(struct hlist_head), 31730818bf27SAl Viro mhash_entries, 19, 31740818bf27SAl Viro 0, 31750818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 31760818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 31770818bf27SAl Viro sizeof(struct hlist_head), 31780818bf27SAl Viro mphash_entries, 19, 31790818bf27SAl Viro 0, 31800818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 31811da177e4SLinus Torvalds 318284d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 31831da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 31841da177e4SLinus Torvalds 31850818bf27SAl Viro for (u = 0; u <= m_hash_mask; u++) 318638129a13SAl Viro INIT_HLIST_HEAD(&mount_hashtable[u]); 31870818bf27SAl Viro for (u = 0; u <= mp_hash_mask; u++) 31880818bf27SAl Viro INIT_HLIST_HEAD(&mountpoint_hashtable[u]); 31891da177e4SLinus Torvalds 31904b93dc9bSTejun Heo kernfs_init(); 31914b93dc9bSTejun Heo 319215a67dd8SRandy Dunlap err = sysfs_init(); 319315a67dd8SRandy Dunlap if (err) 319415a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 31958e24eea7SHarvey Harrison __func__, err); 319600d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 319700d26666SGreg Kroah-Hartman if (!fs_kobj) 31988e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 31991da177e4SLinus Torvalds init_rootfs(); 32001da177e4SLinus Torvalds init_mount_tree(); 32011da177e4SLinus Torvalds } 32021da177e4SLinus Torvalds 3203616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 32041da177e4SLinus Torvalds { 3205d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3206616511d0STrond Myklebust return; 32077b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3208771b1371SEric W. Biederman free_mnt_ns(ns); 32091da177e4SLinus Torvalds } 32109d412a43SAl Viro 32119d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 32129d412a43SAl Viro { 3213423e0ab0STim Chen struct vfsmount *mnt; 3214423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 3215423e0ab0STim Chen if (!IS_ERR(mnt)) { 3216423e0ab0STim Chen /* 3217423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3218423e0ab0STim Chen * we unmount before file sys is unregistered 3219423e0ab0STim Chen */ 3220f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3221423e0ab0STim Chen } 3222423e0ab0STim Chen return mnt; 32239d412a43SAl Viro } 32249d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3225423e0ab0STim Chen 3226423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3227423e0ab0STim Chen { 3228423e0ab0STim Chen /* release long term mount so mount point can be released */ 3229423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3230f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 323148a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3232423e0ab0STim Chen mntput(mnt); 3233423e0ab0STim Chen } 3234423e0ab0STim Chen } 3235423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 323602125a82SAl Viro 323702125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 323802125a82SAl Viro { 3239143c8c91SAl Viro return check_mnt(real_mount(mnt)); 324002125a82SAl Viro } 32418823c079SEric W. Biederman 32423151527eSEric W. Biederman bool current_chrooted(void) 32433151527eSEric W. Biederman { 32443151527eSEric W. Biederman /* Does the current process have a non-standard root */ 32453151527eSEric W. Biederman struct path ns_root; 32463151527eSEric W. Biederman struct path fs_root; 32473151527eSEric W. Biederman bool chrooted; 32483151527eSEric W. Biederman 32493151527eSEric W. Biederman /* Find the namespace root */ 32503151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 32513151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 32523151527eSEric W. Biederman path_get(&ns_root); 32533151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 32543151527eSEric W. Biederman ; 32553151527eSEric W. Biederman 32563151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 32573151527eSEric W. Biederman 32583151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 32593151527eSEric W. Biederman 32603151527eSEric W. Biederman path_put(&fs_root); 32613151527eSEric W. Biederman path_put(&ns_root); 32623151527eSEric W. Biederman 32633151527eSEric W. Biederman return chrooted; 32643151527eSEric W. Biederman } 32653151527eSEric W. Biederman 32668654df4eSEric W. Biederman static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, 32678654df4eSEric W. Biederman int *new_mnt_flags) 326887a8ebd6SEric W. Biederman { 32698c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 327087a8ebd6SEric W. Biederman struct mount *mnt; 3271e51db735SEric W. Biederman bool visible = false; 327287a8ebd6SEric W. Biederman 327344bb4385SAl Viro down_read(&namespace_sem); 327487a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3275e51db735SEric W. Biederman struct mount *child; 327677b1a97dSEric W. Biederman int mnt_flags; 327777b1a97dSEric W. Biederman 32788654df4eSEric W. Biederman if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) 3279e51db735SEric W. Biederman continue; 3280e51db735SEric W. Biederman 32817e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 32827e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 32837e96c1b0SEric W. Biederman */ 32847e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 32857e96c1b0SEric W. Biederman continue; 32867e96c1b0SEric W. Biederman 3287a1935c17SEric W. Biederman /* A local view of the mount flags */ 328877b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 328977b1a97dSEric W. Biederman 3290695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3291695e9df0SEric W. Biederman if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY) 3292695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3293695e9df0SEric W. Biederman 32948c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 32958c6cf9ccSEric W. Biederman * than the proposed new mount. 32968c6cf9ccSEric W. Biederman */ 329777b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 32988c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 32998c6cf9ccSEric W. Biederman continue; 330077b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 330177b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 33028c6cf9ccSEric W. Biederman continue; 33038c6cf9ccSEric W. Biederman 3304ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3305ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3306ceeb0e5dSEric W. Biederman * empty directories. 3307e51db735SEric W. Biederman */ 3308e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3309e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3310ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3311d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3312ceeb0e5dSEric W. Biederman continue; 33137236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 33147236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3315e51db735SEric W. Biederman goto next; 331687a8ebd6SEric W. Biederman } 33178c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 331877b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 33198c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3320e51db735SEric W. Biederman visible = true; 3321e51db735SEric W. Biederman goto found; 3322e51db735SEric W. Biederman next: ; 332387a8ebd6SEric W. Biederman } 3324e51db735SEric W. Biederman found: 332544bb4385SAl Viro up_read(&namespace_sem); 3326e51db735SEric W. Biederman return visible; 332787a8ebd6SEric W. Biederman } 332887a8ebd6SEric W. Biederman 33298654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) 33308654df4eSEric W. Biederman { 3331a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 33328654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 33338654df4eSEric W. Biederman unsigned long s_iflags; 33348654df4eSEric W. Biederman 33358654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 33368654df4eSEric W. Biederman return false; 33378654df4eSEric W. Biederman 33388654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 33398654df4eSEric W. Biederman s_iflags = mnt->mnt_sb->s_iflags; 33408654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 33418654df4eSEric W. Biederman return false; 33428654df4eSEric W. Biederman 3343a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3344a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3345a1935c17SEric W. Biederman required_iflags); 3346a1935c17SEric W. Biederman return true; 3347a1935c17SEric W. Biederman } 3348a1935c17SEric W. Biederman 33498654df4eSEric W. Biederman return !mnt_already_visible(ns, mnt, new_mnt_flags); 33508654df4eSEric W. Biederman } 33518654df4eSEric W. Biederman 3352380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3353380cf5baSAndy Lutomirski { 3354380cf5baSAndy Lutomirski /* 3355380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3356380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3357380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3358380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3359380cf5baSAndy Lutomirski * in other namespaces. 3360380cf5baSAndy Lutomirski */ 3361380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3362380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3363380cf5baSAndy Lutomirski } 3364380cf5baSAndy Lutomirski 336564964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 33668823c079SEric W. Biederman { 336758be2825SAl Viro struct ns_common *ns = NULL; 33688823c079SEric W. Biederman struct nsproxy *nsproxy; 33698823c079SEric W. Biederman 3370728dba3aSEric W. Biederman task_lock(task); 3371728dba3aSEric W. Biederman nsproxy = task->nsproxy; 33728823c079SEric W. Biederman if (nsproxy) { 337358be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 337458be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 33758823c079SEric W. Biederman } 3376728dba3aSEric W. Biederman task_unlock(task); 33778823c079SEric W. Biederman 33788823c079SEric W. Biederman return ns; 33798823c079SEric W. Biederman } 33808823c079SEric W. Biederman 338164964528SAl Viro static void mntns_put(struct ns_common *ns) 33828823c079SEric W. Biederman { 338358be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 33848823c079SEric W. Biederman } 33858823c079SEric W. Biederman 338664964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 33878823c079SEric W. Biederman { 33888823c079SEric W. Biederman struct fs_struct *fs = current->fs; 338958be2825SAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns); 33908823c079SEric W. Biederman struct path root; 33918823c079SEric W. Biederman 33920c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3393c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3394c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3395ae11e0f1SZhao Hongjiang return -EPERM; 33968823c079SEric W. Biederman 33978823c079SEric W. Biederman if (fs->users != 1) 33988823c079SEric W. Biederman return -EINVAL; 33998823c079SEric W. Biederman 34008823c079SEric W. Biederman get_mnt_ns(mnt_ns); 34018823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 34028823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 34038823c079SEric W. Biederman 34048823c079SEric W. Biederman /* Find the root */ 34058823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 34068823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 34078823c079SEric W. Biederman path_get(&root); 34088823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 34098823c079SEric W. Biederman ; 34108823c079SEric W. Biederman 34118823c079SEric W. Biederman /* Update the pwd and root */ 34128823c079SEric W. Biederman set_fs_pwd(fs, &root); 34138823c079SEric W. Biederman set_fs_root(fs, &root); 34148823c079SEric W. Biederman 34158823c079SEric W. Biederman path_put(&root); 34168823c079SEric W. Biederman return 0; 34178823c079SEric W. Biederman } 34188823c079SEric W. Biederman 3419bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3420bcac25a5SAndrey Vagin { 3421bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3422bcac25a5SAndrey Vagin } 3423bcac25a5SAndrey Vagin 34248823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 34258823c079SEric W. Biederman .name = "mnt", 34268823c079SEric W. Biederman .type = CLONE_NEWNS, 34278823c079SEric W. Biederman .get = mntns_get, 34288823c079SEric W. Biederman .put = mntns_put, 34298823c079SEric W. Biederman .install = mntns_install, 3430bcac25a5SAndrey Vagin .owner = mntns_owner, 34318823c079SEric W. Biederman }; 3432