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 300818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 310818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 320818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 330818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 340818bf27SAl Viro 350818bf27SAl Viro static __initdata unsigned long mhash_entries; 360818bf27SAl Viro static int __init set_mhash_entries(char *str) 370818bf27SAl Viro { 380818bf27SAl Viro if (!str) 390818bf27SAl Viro return 0; 400818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 410818bf27SAl Viro return 1; 420818bf27SAl Viro } 430818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 440818bf27SAl Viro 450818bf27SAl Viro static __initdata unsigned long mphash_entries; 460818bf27SAl Viro static int __init set_mphash_entries(char *str) 470818bf27SAl Viro { 480818bf27SAl Viro if (!str) 490818bf27SAl Viro return 0; 500818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 510818bf27SAl Viro return 1; 520818bf27SAl Viro } 530818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 5413f14b4dSEric Dumazet 55c7999c36SAl Viro static u64 event; 5673cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 57719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 5899b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 59f21f6220SAl Viro static int mnt_id_start = 0; 60f21f6220SAl Viro static int mnt_group_start = 1; 615addc5ddSAl Viro 6238129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 630818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 64e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 6559aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 661da177e4SLinus Torvalds 67f87fd4c2SMiklos Szeredi /* /sys/fs */ 6800d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 6900d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 70f87fd4c2SMiklos Szeredi 7199b7db7bSNick Piggin /* 7299b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7399b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 7499b7db7bSNick Piggin * up the tree. 7599b7db7bSNick Piggin * 7699b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 7799b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 7899b7db7bSNick Piggin */ 7948a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8099b7db7bSNick Piggin 8138129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 821da177e4SLinus Torvalds { 831da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 841da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 850818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 860818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 870818bf27SAl Viro } 880818bf27SAl Viro 890818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 900818bf27SAl Viro { 910818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 920818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 930818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 941da177e4SLinus Torvalds } 951da177e4SLinus Torvalds 9699b7db7bSNick Piggin /* 9799b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 9899b7db7bSNick Piggin * serialize with freeing. 9999b7db7bSNick Piggin */ 100b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10173cd49ecSMiklos Szeredi { 10273cd49ecSMiklos Szeredi int res; 10373cd49ecSMiklos Szeredi 10473cd49ecSMiklos Szeredi retry: 10573cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 10699b7db7bSNick Piggin spin_lock(&mnt_id_lock); 10715169fe7SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 108f21f6220SAl Viro if (!res) 10915169fe7SAl Viro mnt_id_start = mnt->mnt_id + 1; 11099b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 11173cd49ecSMiklos Szeredi if (res == -EAGAIN) 11273cd49ecSMiklos Szeredi goto retry; 11373cd49ecSMiklos Szeredi 11473cd49ecSMiklos Szeredi return res; 11573cd49ecSMiklos Szeredi } 11673cd49ecSMiklos Szeredi 117b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11873cd49ecSMiklos Szeredi { 11915169fe7SAl Viro int id = mnt->mnt_id; 12099b7db7bSNick Piggin spin_lock(&mnt_id_lock); 121f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 122f21f6220SAl Viro if (mnt_id_start > id) 123f21f6220SAl Viro mnt_id_start = id; 12499b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 12573cd49ecSMiklos Szeredi } 12673cd49ecSMiklos Szeredi 127719f5d7fSMiklos Szeredi /* 128719f5d7fSMiklos Szeredi * Allocate a new peer group ID 129719f5d7fSMiklos Szeredi * 130719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 131719f5d7fSMiklos Szeredi */ 1324b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 133719f5d7fSMiklos Szeredi { 134f21f6220SAl Viro int res; 135f21f6220SAl Viro 136719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 137719f5d7fSMiklos Szeredi return -ENOMEM; 138719f5d7fSMiklos Szeredi 139f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 140f21f6220SAl Viro mnt_group_start, 14115169fe7SAl Viro &mnt->mnt_group_id); 142f21f6220SAl Viro if (!res) 14315169fe7SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 144f21f6220SAl Viro 145f21f6220SAl Viro return res; 146719f5d7fSMiklos Szeredi } 147719f5d7fSMiklos Szeredi 148719f5d7fSMiklos Szeredi /* 149719f5d7fSMiklos Szeredi * Release a peer group ID 150719f5d7fSMiklos Szeredi */ 1514b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 152719f5d7fSMiklos Szeredi { 15315169fe7SAl Viro int id = mnt->mnt_group_id; 154f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 155f21f6220SAl Viro if (mnt_group_start > id) 156f21f6220SAl Viro mnt_group_start = id; 15715169fe7SAl Viro mnt->mnt_group_id = 0; 158719f5d7fSMiklos Szeredi } 159719f5d7fSMiklos Szeredi 160b3e19d92SNick Piggin /* 161b3e19d92SNick Piggin * vfsmount lock must be held for read 162b3e19d92SNick Piggin */ 16383adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 164b3e19d92SNick Piggin { 165b3e19d92SNick Piggin #ifdef CONFIG_SMP 16668e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 167b3e19d92SNick Piggin #else 168b3e19d92SNick Piggin preempt_disable(); 16968e8a9feSAl Viro mnt->mnt_count += n; 170b3e19d92SNick Piggin preempt_enable(); 171b3e19d92SNick Piggin #endif 172b3e19d92SNick Piggin } 173b3e19d92SNick Piggin 174b3e19d92SNick Piggin /* 175b3e19d92SNick Piggin * vfsmount lock must be held for write 176b3e19d92SNick Piggin */ 17783adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 178b3e19d92SNick Piggin { 179b3e19d92SNick Piggin #ifdef CONFIG_SMP 180f03c6599SAl Viro unsigned int count = 0; 181b3e19d92SNick Piggin int cpu; 182b3e19d92SNick Piggin 183b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 18468e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 185b3e19d92SNick Piggin } 186b3e19d92SNick Piggin 187b3e19d92SNick Piggin return count; 188b3e19d92SNick Piggin #else 18968e8a9feSAl Viro return mnt->mnt_count; 190b3e19d92SNick Piggin #endif 191b3e19d92SNick Piggin } 192b3e19d92SNick Piggin 19387b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 19487b95ce0SAl Viro { 19587b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 19687b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 19787b95ce0SAl Viro pin_remove(p); 19887b95ce0SAl Viro mntput(&m->mnt); 19987b95ce0SAl Viro } 20087b95ce0SAl Viro 201b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 2021da177e4SLinus Torvalds { 203c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 204c63181e6SAl Viro if (mnt) { 20573cd49ecSMiklos Szeredi int err; 20673cd49ecSMiklos Szeredi 207c63181e6SAl Viro err = mnt_alloc_id(mnt); 20888b38782SLi Zefan if (err) 20988b38782SLi Zefan goto out_free_cache; 21088b38782SLi Zefan 21188b38782SLi Zefan if (name) { 212fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 213c63181e6SAl Viro if (!mnt->mnt_devname) 21488b38782SLi Zefan goto out_free_id; 21573cd49ecSMiklos Szeredi } 21673cd49ecSMiklos Szeredi 217b3e19d92SNick Piggin #ifdef CONFIG_SMP 218c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 219c63181e6SAl Viro if (!mnt->mnt_pcp) 220b3e19d92SNick Piggin goto out_free_devname; 221b3e19d92SNick Piggin 222c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 223b3e19d92SNick Piggin #else 224c63181e6SAl Viro mnt->mnt_count = 1; 225c63181e6SAl Viro mnt->mnt_writers = 0; 226b3e19d92SNick Piggin #endif 227b3e19d92SNick Piggin 22838129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 229c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 230c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 231c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 232c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 233c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 234c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 235c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2360a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 2372504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 2382504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 2392504c5d6SAndreas Gruenbacher #endif 24087b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2411da177e4SLinus Torvalds } 242c63181e6SAl Viro return mnt; 24388b38782SLi Zefan 244d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 245d3ef3d73Snpiggin@suse.de out_free_devname: 246fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 247d3ef3d73Snpiggin@suse.de #endif 24888b38782SLi Zefan out_free_id: 249c63181e6SAl Viro mnt_free_id(mnt); 25088b38782SLi Zefan out_free_cache: 251c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 25288b38782SLi Zefan return NULL; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds 2558366025eSDave Hansen /* 2568366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2578366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2588366025eSDave Hansen * We must keep track of when those operations start 2598366025eSDave Hansen * (for permission checks) and when they end, so that 2608366025eSDave Hansen * we can determine when writes are able to occur to 2618366025eSDave Hansen * a filesystem. 2628366025eSDave Hansen */ 2633d733633SDave Hansen /* 2643d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2653d733633SDave Hansen * @mnt: the mount to check for its write status 2663d733633SDave Hansen * 2673d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2683d733633SDave Hansen * It does not guarantee that the filesystem will stay 2693d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2703d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2713d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2723d733633SDave Hansen * r/w. 2733d733633SDave Hansen */ 2743d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2753d733633SDave Hansen { 2762e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2772e4b7fcdSDave Hansen return 1; 2782e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2792e4b7fcdSDave Hansen return 1; 2802e4b7fcdSDave Hansen return 0; 2813d733633SDave Hansen } 2823d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2833d733633SDave Hansen 28483adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2853d733633SDave Hansen { 286d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 28768e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 288d3ef3d73Snpiggin@suse.de #else 28968e8a9feSAl Viro mnt->mnt_writers++; 290d3ef3d73Snpiggin@suse.de #endif 2913d733633SDave Hansen } 2923d733633SDave Hansen 29383adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2943d733633SDave Hansen { 295d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 29668e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 297d3ef3d73Snpiggin@suse.de #else 29868e8a9feSAl Viro mnt->mnt_writers--; 299d3ef3d73Snpiggin@suse.de #endif 300d3ef3d73Snpiggin@suse.de } 301d3ef3d73Snpiggin@suse.de 30283adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 303d3ef3d73Snpiggin@suse.de { 304d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 305d3ef3d73Snpiggin@suse.de unsigned int count = 0; 3063d733633SDave Hansen int cpu; 3073d733633SDave Hansen 3083d733633SDave Hansen for_each_possible_cpu(cpu) { 30968e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 3103d733633SDave Hansen } 3113d733633SDave Hansen 312d3ef3d73Snpiggin@suse.de return count; 313d3ef3d73Snpiggin@suse.de #else 314d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 315d3ef3d73Snpiggin@suse.de #endif 3163d733633SDave Hansen } 3173d733633SDave Hansen 3184ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3194ed5e82fSMiklos Szeredi { 3204ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3214ed5e82fSMiklos Szeredi return 1; 3224ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3234ed5e82fSMiklos Szeredi smp_rmb(); 3244ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3254ed5e82fSMiklos Szeredi } 3264ed5e82fSMiklos Szeredi 3273d733633SDave Hansen /* 328eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 329eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 330eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 331eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3323d733633SDave Hansen */ 3338366025eSDave Hansen /** 334eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 33583adc753SAl Viro * @m: the mount on which to take a write 3368366025eSDave Hansen * 337eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 338eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 339eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 340eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 341eb04c282SJan Kara * called. This is effectively a refcount. 3428366025eSDave Hansen */ 343eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3448366025eSDave Hansen { 34583adc753SAl Viro struct mount *mnt = real_mount(m); 3463d733633SDave Hansen int ret = 0; 3473d733633SDave Hansen 348d3ef3d73Snpiggin@suse.de preempt_disable(); 349c6653a83SNick Piggin mnt_inc_writers(mnt); 350d3ef3d73Snpiggin@suse.de /* 351c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 352d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 353d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 354d3ef3d73Snpiggin@suse.de */ 355d3ef3d73Snpiggin@suse.de smp_mb(); 3561e75529eSMiao Xie while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 357d3ef3d73Snpiggin@suse.de cpu_relax(); 358d3ef3d73Snpiggin@suse.de /* 359d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 360d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 361d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 362d3ef3d73Snpiggin@suse.de */ 363d3ef3d73Snpiggin@suse.de smp_rmb(); 3644ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 365c6653a83SNick Piggin mnt_dec_writers(mnt); 3663d733633SDave Hansen ret = -EROFS; 3673d733633SDave Hansen } 368d3ef3d73Snpiggin@suse.de preempt_enable(); 369eb04c282SJan Kara 370eb04c282SJan Kara return ret; 371eb04c282SJan Kara } 372eb04c282SJan Kara 373eb04c282SJan Kara /** 374eb04c282SJan Kara * mnt_want_write - get write access to a mount 375eb04c282SJan Kara * @m: the mount on which to take a write 376eb04c282SJan Kara * 377eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 378eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 379eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 380eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 381eb04c282SJan Kara */ 382eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 383eb04c282SJan Kara { 384eb04c282SJan Kara int ret; 385eb04c282SJan Kara 386eb04c282SJan Kara sb_start_write(m->mnt_sb); 387eb04c282SJan Kara ret = __mnt_want_write(m); 388eb04c282SJan Kara if (ret) 389eb04c282SJan Kara sb_end_write(m->mnt_sb); 3903d733633SDave Hansen return ret; 3918366025eSDave Hansen } 3928366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3938366025eSDave Hansen 3948366025eSDave Hansen /** 39596029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 39696029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 39796029c4eSnpiggin@suse.de * 39896029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 39996029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 40096029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 40196029c4eSnpiggin@suse.de * on it. This allows some optimisation. 40296029c4eSnpiggin@suse.de * 40396029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 40496029c4eSnpiggin@suse.de * drop the reference. 40596029c4eSnpiggin@suse.de */ 40696029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 40796029c4eSnpiggin@suse.de { 40896029c4eSnpiggin@suse.de /* superblock may be r/o */ 40996029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 41096029c4eSnpiggin@suse.de return -EROFS; 41196029c4eSnpiggin@suse.de preempt_disable(); 41283adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 41396029c4eSnpiggin@suse.de preempt_enable(); 41496029c4eSnpiggin@suse.de return 0; 41596029c4eSnpiggin@suse.de } 41696029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 41796029c4eSnpiggin@suse.de 41896029c4eSnpiggin@suse.de /** 419eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 420eb04c282SJan Kara * @file: the file who's mount on which to take a write 421eb04c282SJan Kara * 422eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 423eb04c282SJan Kara * do some optimisations if the file is open for write already 424eb04c282SJan Kara */ 425eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 426eb04c282SJan Kara { 42783f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 428eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 429eb04c282SJan Kara else 430eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 431eb04c282SJan Kara } 432eb04c282SJan Kara 433eb04c282SJan Kara /** 43496029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 43596029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 43696029c4eSnpiggin@suse.de * 43796029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 43896029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 43996029c4eSnpiggin@suse.de */ 44096029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 44196029c4eSnpiggin@suse.de { 442eb04c282SJan Kara int ret; 443eb04c282SJan Kara 444eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 445eb04c282SJan Kara ret = __mnt_want_write_file(file); 446eb04c282SJan Kara if (ret) 447eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 448eb04c282SJan Kara return ret; 44996029c4eSnpiggin@suse.de } 45096029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 45196029c4eSnpiggin@suse.de 45296029c4eSnpiggin@suse.de /** 453eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4548366025eSDave Hansen * @mnt: the mount on which to give up write access 4558366025eSDave Hansen * 4568366025eSDave Hansen * Tells the low-level filesystem that we are done 4578366025eSDave Hansen * performing writes to it. Must be matched with 458eb04c282SJan Kara * __mnt_want_write() call above. 4598366025eSDave Hansen */ 460eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4618366025eSDave Hansen { 462d3ef3d73Snpiggin@suse.de preempt_disable(); 46383adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 464d3ef3d73Snpiggin@suse.de preempt_enable(); 4658366025eSDave Hansen } 466eb04c282SJan Kara 467eb04c282SJan Kara /** 468eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 469eb04c282SJan Kara * @mnt: the mount on which to give up write access 470eb04c282SJan Kara * 471eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 472eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 473eb04c282SJan Kara * mnt_want_write() call above. 474eb04c282SJan Kara */ 475eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 476eb04c282SJan Kara { 477eb04c282SJan Kara __mnt_drop_write(mnt); 478eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 479eb04c282SJan Kara } 4808366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4818366025eSDave Hansen 482eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 483eb04c282SJan Kara { 484eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 485eb04c282SJan Kara } 486eb04c282SJan Kara 4872a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4882a79f17eSAl Viro { 4892a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4902a79f17eSAl Viro } 4912a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4922a79f17eSAl Viro 49383adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4948366025eSDave Hansen { 4953d733633SDave Hansen int ret = 0; 4963d733633SDave Hansen 497719ea2fbSAl Viro lock_mount_hash(); 49883adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 499d3ef3d73Snpiggin@suse.de /* 500d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 501d3ef3d73Snpiggin@suse.de * should be visible before we do. 502d3ef3d73Snpiggin@suse.de */ 503d3ef3d73Snpiggin@suse.de smp_mb(); 504d3ef3d73Snpiggin@suse.de 505d3ef3d73Snpiggin@suse.de /* 506d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 507d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 508d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 509d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 510d3ef3d73Snpiggin@suse.de * 511d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 512d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 513d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 514d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 515d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 516d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 517d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 518d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 519d3ef3d73Snpiggin@suse.de * we're counting up here. 520d3ef3d73Snpiggin@suse.de */ 521c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 522d3ef3d73Snpiggin@suse.de ret = -EBUSY; 523d3ef3d73Snpiggin@suse.de else 52483adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 525d3ef3d73Snpiggin@suse.de /* 526d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 527d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 528d3ef3d73Snpiggin@suse.de */ 529d3ef3d73Snpiggin@suse.de smp_wmb(); 53083adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 531719ea2fbSAl Viro unlock_mount_hash(); 5323d733633SDave Hansen return ret; 5333d733633SDave Hansen } 5348366025eSDave Hansen 53583adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5362e4b7fcdSDave Hansen { 537719ea2fbSAl Viro lock_mount_hash(); 53883adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 539719ea2fbSAl Viro unlock_mount_hash(); 5402e4b7fcdSDave Hansen } 5412e4b7fcdSDave Hansen 5424ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5434ed5e82fSMiklos Szeredi { 5444ed5e82fSMiklos Szeredi struct mount *mnt; 5454ed5e82fSMiklos Szeredi int err = 0; 5464ed5e82fSMiklos Szeredi 5478e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5488e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5498e8b8796SMiklos Szeredi return -EBUSY; 5508e8b8796SMiklos Szeredi 551719ea2fbSAl Viro lock_mount_hash(); 5524ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5534ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5544ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5554ed5e82fSMiklos Szeredi smp_mb(); 5564ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5574ed5e82fSMiklos Szeredi err = -EBUSY; 5584ed5e82fSMiklos Szeredi break; 5594ed5e82fSMiklos Szeredi } 5604ed5e82fSMiklos Szeredi } 5614ed5e82fSMiklos Szeredi } 5628e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5638e8b8796SMiklos Szeredi err = -EBUSY; 5648e8b8796SMiklos Szeredi 5654ed5e82fSMiklos Szeredi if (!err) { 5664ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5674ed5e82fSMiklos Szeredi smp_wmb(); 5684ed5e82fSMiklos Szeredi } 5694ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5704ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5714ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5724ed5e82fSMiklos Szeredi } 573719ea2fbSAl Viro unlock_mount_hash(); 5744ed5e82fSMiklos Szeredi 5754ed5e82fSMiklos Szeredi return err; 5764ed5e82fSMiklos Szeredi } 5774ed5e82fSMiklos Szeredi 578b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5791da177e4SLinus Torvalds { 580fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 581d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 58268e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 583d3ef3d73Snpiggin@suse.de #endif 584b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5851da177e4SLinus Torvalds } 5861da177e4SLinus Torvalds 5878ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5888ffcb32eSDavid Howells { 5898ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5908ffcb32eSDavid Howells } 5918ffcb32eSDavid Howells 59248a066e7SAl Viro /* call under rcu_read_lock */ 593294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 59448a066e7SAl Viro { 59548a066e7SAl Viro struct mount *mnt; 59648a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 597294d71ffSAl Viro return 1; 59848a066e7SAl Viro if (bastard == NULL) 599294d71ffSAl Viro return 0; 60048a066e7SAl Viro mnt = real_mount(bastard); 60148a066e7SAl Viro mnt_add_count(mnt, 1); 60248a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 603294d71ffSAl Viro return 0; 60448a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 60548a066e7SAl Viro mnt_add_count(mnt, -1); 606294d71ffSAl Viro return 1; 60748a066e7SAl Viro } 608294d71ffSAl Viro return -1; 609294d71ffSAl Viro } 610294d71ffSAl Viro 611294d71ffSAl Viro /* call under rcu_read_lock */ 612294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 613294d71ffSAl Viro { 614294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 615294d71ffSAl Viro if (likely(!res)) 616294d71ffSAl Viro return true; 617294d71ffSAl Viro if (unlikely(res < 0)) { 61848a066e7SAl Viro rcu_read_unlock(); 61948a066e7SAl Viro mntput(bastard); 62048a066e7SAl Viro rcu_read_lock(); 621294d71ffSAl Viro } 62248a066e7SAl Viro return false; 62348a066e7SAl Viro } 62448a066e7SAl Viro 6251da177e4SLinus Torvalds /* 626474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 62748a066e7SAl Viro * call under rcu_read_lock() 6281da177e4SLinus Torvalds */ 629474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6301da177e4SLinus Torvalds { 63138129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 632474279dcSAl Viro struct mount *p; 6331da177e4SLinus Torvalds 63438129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 635474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 636474279dcSAl Viro return p; 637474279dcSAl Viro return NULL; 6381da177e4SLinus Torvalds } 639474279dcSAl Viro 640474279dcSAl Viro /* 641474279dcSAl Viro * find the last mount at @dentry on vfsmount @mnt. 64248a066e7SAl Viro * mount_lock must be held. 643474279dcSAl Viro */ 644474279dcSAl Viro struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) 645474279dcSAl Viro { 646411a938bSEric W. Biederman struct mount *p, *res = NULL; 647411a938bSEric W. Biederman p = __lookup_mnt(mnt, dentry); 64838129a13SAl Viro if (!p) 64938129a13SAl Viro goto out; 650411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 651411a938bSEric W. Biederman res = p; 65238129a13SAl Viro hlist_for_each_entry_continue(p, mnt_hash) { 6531d6a32acSAl Viro if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) 6541d6a32acSAl Viro break; 655411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 6561d6a32acSAl Viro res = p; 6571d6a32acSAl Viro } 65838129a13SAl Viro out: 6591d6a32acSAl Viro return res; 6601da177e4SLinus Torvalds } 6611da177e4SLinus Torvalds 662a05964f3SRam Pai /* 663f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 664f015f126SDavid Howells * 665f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 666f015f126SDavid Howells * following mounts: 667f015f126SDavid Howells * 668f015f126SDavid Howells * mount /dev/sda1 /mnt 669f015f126SDavid Howells * mount /dev/sda2 /mnt 670f015f126SDavid Howells * mount /dev/sda3 /mnt 671f015f126SDavid Howells * 672f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 673f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 674f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 675f015f126SDavid Howells * 676f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 677a05964f3SRam Pai */ 6781c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 679a05964f3SRam Pai { 680c7105365SAl Viro struct mount *child_mnt; 68148a066e7SAl Viro struct vfsmount *m; 68248a066e7SAl Viro unsigned seq; 68399b7db7bSNick Piggin 68448a066e7SAl Viro rcu_read_lock(); 68548a066e7SAl Viro do { 68648a066e7SAl Viro seq = read_seqbegin(&mount_lock); 687474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 68848a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 68948a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 69048a066e7SAl Viro rcu_read_unlock(); 69148a066e7SAl Viro return m; 692a05964f3SRam Pai } 693a05964f3SRam Pai 6947af1364fSEric W. Biederman /* 6957af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6967af1364fSEric W. Biederman * current mount namespace. 6977af1364fSEric W. Biederman * 6987af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6997af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 7007af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 7017af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 7027af1364fSEric W. Biederman * is a mountpoint. 7037af1364fSEric W. Biederman * 7047af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 7057af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 7067af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 7077af1364fSEric W. Biederman * parent mount. 7087af1364fSEric W. Biederman */ 7097af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 7107af1364fSEric W. Biederman { 7117af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7127af1364fSEric W. Biederman struct mount *mnt; 7137af1364fSEric W. Biederman bool is_covered = false; 7147af1364fSEric W. Biederman 7157af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 7167af1364fSEric W. Biederman goto out; 7177af1364fSEric W. Biederman 7187af1364fSEric W. Biederman down_read(&namespace_sem); 7197af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7207af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7217af1364fSEric W. Biederman if (is_covered) 7227af1364fSEric W. Biederman break; 7237af1364fSEric W. Biederman } 7247af1364fSEric W. Biederman up_read(&namespace_sem); 7257af1364fSEric W. Biederman out: 7267af1364fSEric W. Biederman return is_covered; 7277af1364fSEric W. Biederman } 7287af1364fSEric W. Biederman 729e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 73084d17192SAl Viro { 7310818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 73284d17192SAl Viro struct mountpoint *mp; 73384d17192SAl Viro 7340818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 73584d17192SAl Viro if (mp->m_dentry == dentry) { 73684d17192SAl Viro /* might be worth a WARN_ON() */ 73784d17192SAl Viro if (d_unlinked(dentry)) 73884d17192SAl Viro return ERR_PTR(-ENOENT); 73984d17192SAl Viro mp->m_count++; 74084d17192SAl Viro return mp; 74184d17192SAl Viro } 74284d17192SAl Viro } 743e2dfa935SEric W. Biederman return NULL; 744e2dfa935SEric W. Biederman } 745e2dfa935SEric W. Biederman 746e2dfa935SEric W. Biederman static struct mountpoint *new_mountpoint(struct dentry *dentry) 747e2dfa935SEric W. Biederman { 748e2dfa935SEric W. Biederman struct hlist_head *chain = mp_hash(dentry); 749e2dfa935SEric W. Biederman struct mountpoint *mp; 750e2dfa935SEric W. Biederman int ret; 75184d17192SAl Viro 75284d17192SAl Viro mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 75384d17192SAl Viro if (!mp) 75484d17192SAl Viro return ERR_PTR(-ENOMEM); 75584d17192SAl Viro 756eed81007SMiklos Szeredi ret = d_set_mounted(dentry); 757eed81007SMiklos Szeredi if (ret) { 75884d17192SAl Viro kfree(mp); 759eed81007SMiklos Szeredi return ERR_PTR(ret); 76084d17192SAl Viro } 761eed81007SMiklos Szeredi 76284d17192SAl Viro mp->m_dentry = dentry; 76384d17192SAl Viro mp->m_count = 1; 7640818bf27SAl Viro hlist_add_head(&mp->m_hash, chain); 7650a5eb7c8SEric W. Biederman INIT_HLIST_HEAD(&mp->m_list); 76684d17192SAl Viro return mp; 76784d17192SAl Viro } 76884d17192SAl Viro 76984d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 77084d17192SAl Viro { 77184d17192SAl Viro if (!--mp->m_count) { 77284d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7730a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 77484d17192SAl Viro spin_lock(&dentry->d_lock); 77584d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 77684d17192SAl Viro spin_unlock(&dentry->d_lock); 7770818bf27SAl Viro hlist_del(&mp->m_hash); 77884d17192SAl Viro kfree(mp); 77984d17192SAl Viro } 78084d17192SAl Viro } 78184d17192SAl Viro 782143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7831da177e4SLinus Torvalds { 7846b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7851da177e4SLinus Torvalds } 7861da177e4SLinus Torvalds 78799b7db7bSNick Piggin /* 78899b7db7bSNick Piggin * vfsmount lock must be held for write 78999b7db7bSNick Piggin */ 7906b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7915addc5ddSAl Viro { 7925addc5ddSAl Viro if (ns) { 7935addc5ddSAl Viro ns->event = ++event; 7945addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7955addc5ddSAl Viro } 7965addc5ddSAl Viro } 7975addc5ddSAl Viro 79899b7db7bSNick Piggin /* 79999b7db7bSNick Piggin * vfsmount lock must be held for write 80099b7db7bSNick Piggin */ 8016b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8025addc5ddSAl Viro { 8035addc5ddSAl Viro if (ns && ns->event != event) { 8045addc5ddSAl Viro ns->event = event; 8055addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8065addc5ddSAl Viro } 8075addc5ddSAl Viro } 8085addc5ddSAl Viro 80999b7db7bSNick Piggin /* 81099b7db7bSNick Piggin * vfsmount lock must be held for write 81199b7db7bSNick Piggin */ 8127bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 8131da177e4SLinus Torvalds { 8140714a533SAl Viro mnt->mnt_parent = mnt; 815a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8166b41d536SAl Viro list_del_init(&mnt->mnt_child); 81738129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8180a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 81984d17192SAl Viro put_mountpoint(mnt->mnt_mp); 82084d17192SAl Viro mnt->mnt_mp = NULL; 8211da177e4SLinus Torvalds } 8221da177e4SLinus Torvalds 82399b7db7bSNick Piggin /* 82499b7db7bSNick Piggin * vfsmount lock must be held for write 82599b7db7bSNick Piggin */ 8267bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8277bdb11deSEric W. Biederman { 8287bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8297bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8307bdb11deSEric W. Biederman unhash_mnt(mnt); 8317bdb11deSEric W. Biederman } 8327bdb11deSEric W. Biederman 8337bdb11deSEric W. Biederman /* 8347bdb11deSEric W. Biederman * vfsmount lock must be held for write 8357bdb11deSEric W. Biederman */ 8366a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8376a46c573SEric W. Biederman { 8386a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8396a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8406a46c573SEric W. Biederman unhash_mnt(mnt); 8416a46c573SEric W. Biederman } 8426a46c573SEric W. Biederman 8436a46c573SEric W. Biederman /* 8446a46c573SEric W. Biederman * vfsmount lock must be held for write 8456a46c573SEric W. Biederman */ 84684d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 84784d17192SAl Viro struct mountpoint *mp, 84844d964d6SAl Viro struct mount *child_mnt) 849b90fa9aeSRam Pai { 85084d17192SAl Viro mp->m_count++; 8513a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 85284d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8533a2393d7SAl Viro child_mnt->mnt_parent = mnt; 85484d17192SAl Viro child_mnt->mnt_mp = mp; 8550a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 856b90fa9aeSRam Pai } 857b90fa9aeSRam Pai 85899b7db7bSNick Piggin /* 85999b7db7bSNick Piggin * vfsmount lock must be held for write 86099b7db7bSNick Piggin */ 86184d17192SAl Viro static void attach_mnt(struct mount *mnt, 86284d17192SAl Viro struct mount *parent, 86384d17192SAl Viro struct mountpoint *mp) 8641da177e4SLinus Torvalds { 86584d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 86638129a13SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); 86784d17192SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 868b90fa9aeSRam Pai } 869b90fa9aeSRam Pai 87012a5b529SAl Viro static void attach_shadowed(struct mount *mnt, 87112a5b529SAl Viro struct mount *parent, 87212a5b529SAl Viro struct mount *shadows) 87312a5b529SAl Viro { 87412a5b529SAl Viro if (shadows) { 875f6f99332SLinus Torvalds hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); 87612a5b529SAl Viro list_add(&mnt->mnt_child, &shadows->mnt_child); 87712a5b529SAl Viro } else { 87812a5b529SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, 87912a5b529SAl Viro m_hash(&parent->mnt, mnt->mnt_mountpoint)); 88012a5b529SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 88112a5b529SAl Viro } 88212a5b529SAl Viro } 88312a5b529SAl Viro 884b90fa9aeSRam Pai /* 88599b7db7bSNick Piggin * vfsmount lock must be held for write 886b90fa9aeSRam Pai */ 8871d6a32acSAl Viro static void commit_tree(struct mount *mnt, struct mount *shadows) 888b90fa9aeSRam Pai { 8890714a533SAl Viro struct mount *parent = mnt->mnt_parent; 89083adc753SAl Viro struct mount *m; 891b90fa9aeSRam Pai LIST_HEAD(head); 892143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 893b90fa9aeSRam Pai 8940714a533SAl Viro BUG_ON(parent == mnt); 895b90fa9aeSRam Pai 8961a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 897f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 898143c8c91SAl Viro m->mnt_ns = n; 899f03c6599SAl Viro 900b90fa9aeSRam Pai list_splice(&head, n->list.prev); 901b90fa9aeSRam Pai 90212a5b529SAl Viro attach_shadowed(mnt, parent, shadows); 9036b3286edSKirill Korotaev touch_mnt_namespace(n); 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds 906909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9071da177e4SLinus Torvalds { 9086b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9096b41d536SAl Viro if (next == &p->mnt_mounts) { 9101da177e4SLinus Torvalds while (1) { 911909b0a88SAl Viro if (p == root) 9121da177e4SLinus Torvalds return NULL; 9136b41d536SAl Viro next = p->mnt_child.next; 9146b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9151da177e4SLinus Torvalds break; 9160714a533SAl Viro p = p->mnt_parent; 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds } 9196b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9201da177e4SLinus Torvalds } 9211da177e4SLinus Torvalds 922315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9239676f0c6SRam Pai { 9246b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9256b41d536SAl Viro while (prev != &p->mnt_mounts) { 9266b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9276b41d536SAl Viro prev = p->mnt_mounts.prev; 9289676f0c6SRam Pai } 9299676f0c6SRam Pai return p; 9309676f0c6SRam Pai } 9319676f0c6SRam Pai 9329d412a43SAl Viro struct vfsmount * 9339d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 9349d412a43SAl Viro { 935b105e270SAl Viro struct mount *mnt; 9369d412a43SAl Viro struct dentry *root; 9379d412a43SAl Viro 9389d412a43SAl Viro if (!type) 9399d412a43SAl Viro return ERR_PTR(-ENODEV); 9409d412a43SAl Viro 9419d412a43SAl Viro mnt = alloc_vfsmnt(name); 9429d412a43SAl Viro if (!mnt) 9439d412a43SAl Viro return ERR_PTR(-ENOMEM); 9449d412a43SAl Viro 9459d412a43SAl Viro if (flags & MS_KERNMOUNT) 946b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9479d412a43SAl Viro 9489d412a43SAl Viro root = mount_fs(type, flags, name, data); 9499d412a43SAl Viro if (IS_ERR(root)) { 9508ffcb32eSDavid Howells mnt_free_id(mnt); 9519d412a43SAl Viro free_vfsmnt(mnt); 9529d412a43SAl Viro return ERR_CAST(root); 9539d412a43SAl Viro } 9549d412a43SAl Viro 955b105e270SAl Viro mnt->mnt.mnt_root = root; 956b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 957a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9580714a533SAl Viro mnt->mnt_parent = mnt; 959719ea2fbSAl Viro lock_mount_hash(); 96039f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 961719ea2fbSAl Viro unlock_mount_hash(); 962b105e270SAl Viro return &mnt->mnt; 9639d412a43SAl Viro } 9649d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 9659d412a43SAl Viro 96687129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 96736341f64SRam Pai int flag) 9681da177e4SLinus Torvalds { 96987129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 970be34d1a3SDavid Howells struct mount *mnt; 971be34d1a3SDavid Howells int err; 9721da177e4SLinus Torvalds 973be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 974be34d1a3SDavid Howells if (!mnt) 975be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 976be34d1a3SDavid Howells 9777a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 97815169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 979719f5d7fSMiklos Szeredi else 98015169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 981719f5d7fSMiklos Szeredi 98215169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 983be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 984719f5d7fSMiklos Szeredi if (err) 985719f5d7fSMiklos Szeredi goto out_free; 986719f5d7fSMiklos Szeredi } 987719f5d7fSMiklos Szeredi 988f2ebb3a9SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); 989132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 9909566d674SEric W. Biederman if (flag & CL_UNPRIVILEGED) { 9919566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; 9929566d674SEric W. Biederman 9939566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_READONLY) 994132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 995132c94e3SEric W. Biederman 9969566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NODEV) 9979566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; 9989566d674SEric W. Biederman 9999566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOSUID) 10009566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; 10019566d674SEric W. Biederman 10029566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOEXEC) 10039566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; 10049566d674SEric W. Biederman } 10059566d674SEric W. Biederman 10065ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 1007381cacb1SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && 1008381cacb1SEric W. Biederman (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) 10095ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 10105ff9d8a6SEric W. Biederman 10111da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1012b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1013b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1014a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10150714a533SAl Viro mnt->mnt_parent = mnt; 1016719ea2fbSAl Viro lock_mount_hash(); 101739f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1018719ea2fbSAl Viro unlock_mount_hash(); 1019b90fa9aeSRam Pai 10207a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10217a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10226776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 102332301920SAl Viro mnt->mnt_master = old; 1024fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10258aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1026fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10276776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1028d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10296776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1030d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10315afe0022SRam Pai } 1032b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10330f0afb1dSAl Viro set_mnt_shared(mnt); 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10361da177e4SLinus Torvalds * as the original if that was on one */ 103736341f64SRam Pai if (flag & CL_EXPIRE) { 10386776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10396776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10401da177e4SLinus Torvalds } 1041be34d1a3SDavid Howells 1042cb338d06SAl Viro return mnt; 1043719f5d7fSMiklos Szeredi 1044719f5d7fSMiklos Szeredi out_free: 10458ffcb32eSDavid Howells mnt_free_id(mnt); 1046719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1047be34d1a3SDavid Howells return ERR_PTR(err); 10481da177e4SLinus Torvalds } 10491da177e4SLinus Torvalds 10509ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10519ea459e1SAl Viro { 10529ea459e1SAl Viro /* 10539ea459e1SAl Viro * This probably indicates that somebody messed 10549ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 10559ea459e1SAl Viro * happens, the filesystem was probably unable 10569ea459e1SAl Viro * to make r/w->r/o transitions. 10579ea459e1SAl Viro */ 10589ea459e1SAl Viro /* 10599ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 10609ea459e1SAl Viro * so mnt_get_writers() below is safe. 10619ea459e1SAl Viro */ 10629ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 10639ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 10649ea459e1SAl Viro mnt_pin_kill(mnt); 10659ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 10669ea459e1SAl Viro dput(mnt->mnt.mnt_root); 10679ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 10689ea459e1SAl Viro mnt_free_id(mnt); 10699ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 10709ea459e1SAl Viro } 10719ea459e1SAl Viro 10729ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 10739ea459e1SAl Viro { 10749ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 10759ea459e1SAl Viro } 10769ea459e1SAl Viro 10779ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 10789ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 10799ea459e1SAl Viro { 10809ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 10819ea459e1SAl Viro struct llist_node *next; 10829ea459e1SAl Viro 10839ea459e1SAl Viro for (; node; node = next) { 10849ea459e1SAl Viro next = llist_next(node); 10859ea459e1SAl Viro cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); 10869ea459e1SAl Viro } 10879ea459e1SAl Viro } 10889ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 10899ea459e1SAl Viro 1090900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 10917b7b1aceSAl Viro { 109248a066e7SAl Viro rcu_read_lock(); 1093aa9c0e07SAl Viro mnt_add_count(mnt, -1); 109448a066e7SAl Viro if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ 109548a066e7SAl Viro rcu_read_unlock(); 109699b7db7bSNick Piggin return; 1097b3e19d92SNick Piggin } 1098719ea2fbSAl Viro lock_mount_hash(); 1099b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 110048a066e7SAl Viro rcu_read_unlock(); 1101719ea2fbSAl Viro unlock_mount_hash(); 110299b7db7bSNick Piggin return; 110399b7db7bSNick Piggin } 110448a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 110548a066e7SAl Viro rcu_read_unlock(); 110648a066e7SAl Viro unlock_mount_hash(); 110748a066e7SAl Viro return; 110848a066e7SAl Viro } 110948a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 111048a066e7SAl Viro rcu_read_unlock(); 1111962830dfSAndi Kleen 111239f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1113ce07d891SEric W. Biederman 1114ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1115ce07d891SEric W. Biederman struct mount *p, *tmp; 1116ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1117ce07d891SEric W. Biederman umount_mnt(p); 1118ce07d891SEric W. Biederman } 1119ce07d891SEric W. Biederman } 1120719ea2fbSAl Viro unlock_mount_hash(); 1121649a795aSAl Viro 11229ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11239ea459e1SAl Viro struct task_struct *task = current; 11249ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11259ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11269ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11279ea459e1SAl Viro return; 11289ea459e1SAl Viro } 11299ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11309ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11319ea459e1SAl Viro return; 11329ea459e1SAl Viro } 11339ea459e1SAl Viro cleanup_mnt(mnt); 1134b3e19d92SNick Piggin } 1135b3e19d92SNick Piggin 1136b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1137b3e19d92SNick Piggin { 1138b3e19d92SNick Piggin if (mnt) { 1139863d684fSAl Viro struct mount *m = real_mount(mnt); 1140b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1141863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1142863d684fSAl Viro m->mnt_expiry_mark = 0; 1143863d684fSAl Viro mntput_no_expire(m); 1144b3e19d92SNick Piggin } 1145b3e19d92SNick Piggin } 1146b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1147b3e19d92SNick Piggin 1148b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1149b3e19d92SNick Piggin { 1150b3e19d92SNick Piggin if (mnt) 115183adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1152b3e19d92SNick Piggin return mnt; 1153b3e19d92SNick Piggin } 1154b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1155b3e19d92SNick Piggin 11563064c356SAl Viro struct vfsmount *mnt_clone_internal(struct path *path) 11577b7b1aceSAl Viro { 11583064c356SAl Viro struct mount *p; 11593064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 11603064c356SAl Viro if (IS_ERR(p)) 11613064c356SAl Viro return ERR_CAST(p); 11623064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 11633064c356SAl Viro return &p->mnt; 11647b7b1aceSAl Viro } 11651da177e4SLinus Torvalds 1166b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 1167b3b304a2SMiklos Szeredi { 1168b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 1169b3b304a2SMiklos Szeredi } 1170b3b304a2SMiklos Szeredi 1171b3b304a2SMiklos Szeredi /* 1172b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 1173b3b304a2SMiklos Szeredi * implement more complex mount option showing. 1174b3b304a2SMiklos Szeredi * 1175b3b304a2SMiklos Szeredi * See also save_mount_options(). 1176b3b304a2SMiklos Szeredi */ 117734c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 1178b3b304a2SMiklos Szeredi { 11792a32cebdSAl Viro const char *options; 11802a32cebdSAl Viro 11812a32cebdSAl Viro rcu_read_lock(); 118234c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 1183b3b304a2SMiklos Szeredi 1184b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 1185b3b304a2SMiklos Szeredi seq_putc(m, ','); 1186b3b304a2SMiklos Szeredi mangle(m, options); 1187b3b304a2SMiklos Szeredi } 11882a32cebdSAl Viro rcu_read_unlock(); 1189b3b304a2SMiklos Szeredi 1190b3b304a2SMiklos Szeredi return 0; 1191b3b304a2SMiklos Szeredi } 1192b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 1193b3b304a2SMiklos Szeredi 1194b3b304a2SMiklos Szeredi /* 1195b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 1196b3b304a2SMiklos Szeredi * called from the fill_super() callback. 1197b3b304a2SMiklos Szeredi * 1198b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 1199b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1200b3b304a2SMiklos Szeredi * remount fails. 1201b3b304a2SMiklos Szeredi * 1202b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1203b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1204b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1205b3b304a2SMiklos Szeredi * any more. 1206b3b304a2SMiklos Szeredi */ 1207b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1208b3b304a2SMiklos Szeredi { 12092a32cebdSAl Viro BUG_ON(sb->s_options); 12102a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1211b3b304a2SMiklos Szeredi } 1212b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1213b3b304a2SMiklos Szeredi 12142a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 12152a32cebdSAl Viro { 12162a32cebdSAl Viro char *old = sb->s_options; 12172a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 12182a32cebdSAl Viro if (old) { 12192a32cebdSAl Viro synchronize_rcu(); 12202a32cebdSAl Viro kfree(old); 12212a32cebdSAl Viro } 12222a32cebdSAl Viro } 12232a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 12242a32cebdSAl Viro 1225a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12260226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12271da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12281da177e4SLinus Torvalds { 1229ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12301da177e4SLinus Torvalds 1231390c6843SRam Pai down_read(&namespace_sem); 1232c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1233c7999c36SAl Viro void *v = p->cached_mount; 1234c7999c36SAl Viro if (*pos == p->cached_index) 1235c7999c36SAl Viro return v; 1236c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1237c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1238c7999c36SAl Viro return p->cached_mount = v; 1239c7999c36SAl Viro } 1240c7999c36SAl Viro } 1241c7999c36SAl Viro 1242c7999c36SAl Viro p->cached_event = p->ns->event; 1243c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1244c7999c36SAl Viro p->cached_index = *pos; 1245c7999c36SAl Viro return p->cached_mount; 12461da177e4SLinus Torvalds } 12471da177e4SLinus Torvalds 12481da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12491da177e4SLinus Torvalds { 1250ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1251b0765fb8SPavel Emelianov 1252c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1253c7999c36SAl Viro p->cached_index = *pos; 1254c7999c36SAl Viro return p->cached_mount; 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12581da177e4SLinus Torvalds { 1259390c6843SRam Pai up_read(&namespace_sem); 12601da177e4SLinus Torvalds } 12611da177e4SLinus Torvalds 12620226f492SAl Viro static int m_show(struct seq_file *m, void *v) 12639f5596afSAl Viro { 1264ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12651a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 12660226f492SAl Viro return p->show(m, &r->mnt); 12671da177e4SLinus Torvalds } 12681da177e4SLinus Torvalds 1269a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 12701da177e4SLinus Torvalds .start = m_start, 12711da177e4SLinus Torvalds .next = m_next, 12721da177e4SLinus Torvalds .stop = m_stop, 12730226f492SAl Viro .show = m_show, 1274b4629fe2SChuck Lever }; 1275a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1276b4629fe2SChuck Lever 12771da177e4SLinus Torvalds /** 12781da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 12791da177e4SLinus Torvalds * @mnt: root of mount tree 12801da177e4SLinus Torvalds * 12811da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 12821da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 12831da177e4SLinus Torvalds * busy. 12841da177e4SLinus Torvalds */ 1285909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 12861da177e4SLinus Torvalds { 1287909b0a88SAl Viro struct mount *mnt = real_mount(m); 128836341f64SRam Pai int actual_refs = 0; 128936341f64SRam Pai int minimum_refs = 0; 1290315fc83eSAl Viro struct mount *p; 1291909b0a88SAl Viro BUG_ON(!m); 12921da177e4SLinus Torvalds 1293b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1294719ea2fbSAl Viro lock_mount_hash(); 1295909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 129683adc753SAl Viro actual_refs += mnt_get_count(p); 12971da177e4SLinus Torvalds minimum_refs += 2; 12981da177e4SLinus Torvalds } 1299719ea2fbSAl Viro unlock_mount_hash(); 13001da177e4SLinus Torvalds 13011da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13021da177e4SLinus Torvalds return 0; 1303e3474a8eSIan Kent 1304e3474a8eSIan Kent return 1; 13051da177e4SLinus Torvalds } 13061da177e4SLinus Torvalds 13071da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13081da177e4SLinus Torvalds 13091da177e4SLinus Torvalds /** 13101da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13111da177e4SLinus Torvalds * @mnt: root of mount 13121da177e4SLinus Torvalds * 13131da177e4SLinus Torvalds * This is called to check if a mount point has any 13141da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13151da177e4SLinus Torvalds * mount has sub mounts this will return busy 13161da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13171da177e4SLinus Torvalds * 13181da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13191da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13201da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13211da177e4SLinus Torvalds */ 13221da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13231da177e4SLinus Torvalds { 1324e3474a8eSIan Kent int ret = 1; 13258ad08d8aSAl Viro down_read(&namespace_sem); 1326719ea2fbSAl Viro lock_mount_hash(); 13271ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1328e3474a8eSIan Kent ret = 0; 1329719ea2fbSAl Viro unlock_mount_hash(); 13308ad08d8aSAl Viro up_read(&namespace_sem); 1331a05964f3SRam Pai return ret; 13321da177e4SLinus Torvalds } 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13351da177e4SLinus Torvalds 133638129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1337e3197d83SAl Viro 133897216be0SAl Viro static void namespace_unlock(void) 13391da177e4SLinus Torvalds { 1340a3b3c562SEric W. Biederman struct hlist_head head; 134197216be0SAl Viro 1342a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1343a3b3c562SEric W. Biederman 134497216be0SAl Viro up_write(&namespace_sem); 1345a3b3c562SEric W. Biederman 1346a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 134797216be0SAl Viro return; 134897216be0SAl Viro 134948a066e7SAl Viro synchronize_rcu(); 135048a066e7SAl Viro 135187b95ce0SAl Viro group_pin_kill(&head); 135270fbcdf4SRam Pai } 135370fbcdf4SRam Pai 135497216be0SAl Viro static inline void namespace_lock(void) 1355e3197d83SAl Viro { 135697216be0SAl Viro down_write(&namespace_sem); 1357e3197d83SAl Viro } 1358e3197d83SAl Viro 1359e819f152SEric W. Biederman enum umount_tree_flags { 1360e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1361e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1362e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1363e819f152SEric W. Biederman }; 1364f2d0a123SEric W. Biederman 1365f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1366f2d0a123SEric W. Biederman { 1367f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1368f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1369f2d0a123SEric W. Biederman return true; 1370f2d0a123SEric W. Biederman 1371f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1372f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1373f2d0a123SEric W. Biederman return true; 1374f2d0a123SEric W. Biederman 1375f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1376f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1377f2d0a123SEric W. Biederman * connected to mounted mounts. 1378f2d0a123SEric W. Biederman */ 1379f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1380f2d0a123SEric W. Biederman return true; 1381f2d0a123SEric W. Biederman 1382f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1383f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1384f2d0a123SEric W. Biederman return false; 1385f2d0a123SEric W. Biederman 1386f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1387f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1388f2d0a123SEric W. Biederman return false; 1389f2d0a123SEric W. Biederman 1390f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1391f2d0a123SEric W. Biederman return true; 1392f2d0a123SEric W. Biederman } 1393f2d0a123SEric W. Biederman 139499b7db7bSNick Piggin /* 139548a066e7SAl Viro * mount_lock must be held 139699b7db7bSNick Piggin * namespace_sem must be held for write 139799b7db7bSNick Piggin */ 1398e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 139970fbcdf4SRam Pai { 1400c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1401315fc83eSAl Viro struct mount *p; 140270fbcdf4SRam Pai 14035d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14045d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14055d88457eSEric W. Biederman 1406c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1407590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1408590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1409c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1410590ce4bcSEric W. Biederman } 1411c003b26fSEric W. Biederman 1412411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1413c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1414c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 141538129a13SAl Viro } 141670fbcdf4SRam Pai 1417c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1418e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14197b8a53fdSAl Viro propagate_umount(&tmp_list); 1420a05964f3SRam Pai 1421c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1422ce07d891SEric W. Biederman bool disconnect; 1423c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 14246776db3dSAl Viro list_del_init(&p->mnt_expire); 14251a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1426143c8c91SAl Viro __touch_mnt_namespace(p->mnt_ns); 142763d37a84SAl Viro p->mnt_ns = NULL; 1428e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 142948a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 143087b95ce0SAl Viro 1431f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1432ce07d891SEric W. Biederman 1433ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1434ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1435676da58dSAl Viro if (mnt_has_parent(p)) { 143681b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1437ce07d891SEric W. Biederman if (!disconnect) { 1438ce07d891SEric W. Biederman /* Don't forget about p */ 1439ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1440ce07d891SEric W. Biederman } else { 14416a46c573SEric W. Biederman umount_mnt(p); 14427c4b93d8SAl Viro } 1443ce07d891SEric W. Biederman } 14440f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 144538129a13SAl Viro } 14461da177e4SLinus Torvalds } 14471da177e4SLinus Torvalds 1448b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1449c35038beSAl Viro 14501ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 14511da177e4SLinus Torvalds { 14521ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 14531da177e4SLinus Torvalds int retval; 14541da177e4SLinus Torvalds 14551ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 14561da177e4SLinus Torvalds if (retval) 14571da177e4SLinus Torvalds return retval; 14581da177e4SLinus Torvalds 14591da177e4SLinus Torvalds /* 14601da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 14611da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 14621da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 14631da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 14641da177e4SLinus Torvalds */ 14651da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 14661ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 14671da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 14681da177e4SLinus Torvalds return -EINVAL; 14691da177e4SLinus Torvalds 1470b3e19d92SNick Piggin /* 1471b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1472b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1473b3e19d92SNick Piggin */ 1474719ea2fbSAl Viro lock_mount_hash(); 147583adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1476719ea2fbSAl Viro unlock_mount_hash(); 14771da177e4SLinus Torvalds return -EBUSY; 1478b3e19d92SNick Piggin } 1479719ea2fbSAl Viro unlock_mount_hash(); 14801da177e4SLinus Torvalds 1481863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 14821da177e4SLinus Torvalds return -EAGAIN; 14831da177e4SLinus Torvalds } 14841da177e4SLinus Torvalds 14851da177e4SLinus Torvalds /* 14861da177e4SLinus Torvalds * If we may have to abort operations to get out of this 14871da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 14881da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 14891da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 14901da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 14911da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 14921da177e4SLinus Torvalds * about for the moment. 14931da177e4SLinus Torvalds */ 14941da177e4SLinus Torvalds 149542faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 149642faad99SAl Viro sb->s_op->umount_begin(sb); 149742faad99SAl Viro } 14981da177e4SLinus Torvalds 14991da177e4SLinus Torvalds /* 15001da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15011da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15021da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15031da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15041da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15051da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15061da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15071da177e4SLinus Torvalds */ 15081ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15091da177e4SLinus Torvalds /* 15101da177e4SLinus Torvalds * Special case for "unmounting" root ... 15111da177e4SLinus Torvalds * we just try to remount it readonly. 15121da177e4SLinus Torvalds */ 1513a1480dccSAndy Lutomirski if (!capable(CAP_SYS_ADMIN)) 1514a1480dccSAndy Lutomirski return -EPERM; 15151da177e4SLinus Torvalds down_write(&sb->s_umount); 15164aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 15171da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 15181da177e4SLinus Torvalds up_write(&sb->s_umount); 15191da177e4SLinus Torvalds return retval; 15201da177e4SLinus Torvalds } 15211da177e4SLinus Torvalds 152297216be0SAl Viro namespace_lock(); 1523719ea2fbSAl Viro lock_mount_hash(); 15245addc5ddSAl Viro event++; 15251da177e4SLinus Torvalds 152648a066e7SAl Viro if (flags & MNT_DETACH) { 152748a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1528e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 152948a066e7SAl Viro retval = 0; 153048a066e7SAl Viro } else { 1531b54b9be7SAl Viro shrink_submounts(mnt); 15321da177e4SLinus Torvalds retval = -EBUSY; 153348a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 15341a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1535e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 15361da177e4SLinus Torvalds retval = 0; 15371da177e4SLinus Torvalds } 153848a066e7SAl Viro } 1539719ea2fbSAl Viro unlock_mount_hash(); 1540e3197d83SAl Viro namespace_unlock(); 15411da177e4SLinus Torvalds return retval; 15421da177e4SLinus Torvalds } 15431da177e4SLinus Torvalds 15441da177e4SLinus Torvalds /* 154580b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 154680b5dce8SEric W. Biederman * 154780b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 154880b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 154980b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 155080b5dce8SEric W. Biederman * leaking them. 155180b5dce8SEric W. Biederman * 155280b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 155380b5dce8SEric W. Biederman */ 155480b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 155580b5dce8SEric W. Biederman { 155680b5dce8SEric W. Biederman struct mountpoint *mp; 155780b5dce8SEric W. Biederman struct mount *mnt; 155880b5dce8SEric W. Biederman 155980b5dce8SEric W. Biederman namespace_lock(); 156080b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1561f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 156280b5dce8SEric W. Biederman goto out_unlock; 156380b5dce8SEric W. Biederman 156480b5dce8SEric W. Biederman lock_mount_hash(); 156580b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 156680b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1567ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1568fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1569fe78fcc8SEric W. Biederman umount_mnt(mnt); 1570ce07d891SEric W. Biederman } 1571e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 157280b5dce8SEric W. Biederman } 157380b5dce8SEric W. Biederman unlock_mount_hash(); 157480b5dce8SEric W. Biederman put_mountpoint(mp); 157580b5dce8SEric W. Biederman out_unlock: 157680b5dce8SEric W. Biederman namespace_unlock(); 157780b5dce8SEric W. Biederman } 157880b5dce8SEric W. Biederman 157980b5dce8SEric W. Biederman /* 15809b40bc90SAl Viro * Is the caller allowed to modify his namespace? 15819b40bc90SAl Viro */ 15829b40bc90SAl Viro static inline bool may_mount(void) 15839b40bc90SAl Viro { 15849b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 15859b40bc90SAl Viro } 15869b40bc90SAl Viro 15879e8925b6SJeff Layton static inline bool may_mandlock(void) 15889e8925b6SJeff Layton { 15899e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 15909e8925b6SJeff Layton return false; 15919e8925b6SJeff Layton #endif 159295ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 15939e8925b6SJeff Layton } 15949e8925b6SJeff Layton 15959b40bc90SAl Viro /* 15961da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 15971da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 15981da177e4SLinus Torvalds * 15991da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16001da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16011da177e4SLinus Torvalds */ 16021da177e4SLinus Torvalds 1603bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 16041da177e4SLinus Torvalds { 16052d8f3038SAl Viro struct path path; 1606900148dcSAl Viro struct mount *mnt; 16071da177e4SLinus Torvalds int retval; 1608db1f05bbSMiklos Szeredi int lookup_flags = 0; 16091da177e4SLinus Torvalds 1610db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1611db1f05bbSMiklos Szeredi return -EINVAL; 1612db1f05bbSMiklos Szeredi 16139b40bc90SAl Viro if (!may_mount()) 16149b40bc90SAl Viro return -EPERM; 16159b40bc90SAl Viro 1616db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1617db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1618db1f05bbSMiklos Szeredi 1619197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 16201da177e4SLinus Torvalds if (retval) 16211da177e4SLinus Torvalds goto out; 1622900148dcSAl Viro mnt = real_mount(path.mnt); 16231da177e4SLinus Torvalds retval = -EINVAL; 16242d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 16251da177e4SLinus Torvalds goto dput_and_out; 1626143c8c91SAl Viro if (!check_mnt(mnt)) 16271da177e4SLinus Torvalds goto dput_and_out; 16285ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 16295ff9d8a6SEric W. Biederman goto dput_and_out; 1630b2f5d4dcSEric W. Biederman retval = -EPERM; 1631b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1632b2f5d4dcSEric W. Biederman goto dput_and_out; 16331da177e4SLinus Torvalds 1634900148dcSAl Viro retval = do_umount(mnt, flags); 16351da177e4SLinus Torvalds dput_and_out: 1636429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 16372d8f3038SAl Viro dput(path.dentry); 1638900148dcSAl Viro mntput_no_expire(mnt); 16391da177e4SLinus Torvalds out: 16401da177e4SLinus Torvalds return retval; 16411da177e4SLinus Torvalds } 16421da177e4SLinus Torvalds 16431da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 16441da177e4SLinus Torvalds 16451da177e4SLinus Torvalds /* 16461da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 16471da177e4SLinus Torvalds */ 1648bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 16491da177e4SLinus Torvalds { 16501da177e4SLinus Torvalds return sys_umount(name, 0); 16511da177e4SLinus Torvalds } 16521da177e4SLinus Torvalds 16531da177e4SLinus Torvalds #endif 16541da177e4SLinus Torvalds 16554ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 16568823c079SEric W. Biederman { 16574ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1658e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1659e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 16604ce5d2b1SEric W. Biederman } 16614ce5d2b1SEric W. Biederman 166258be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 166358be2825SAl Viro { 166458be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 166558be2825SAl Viro } 166658be2825SAl Viro 16674ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 16684ce5d2b1SEric W. Biederman { 16694ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 16704ce5d2b1SEric W. Biederman * mount namespace loop? 16714ce5d2b1SEric W. Biederman */ 16724ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 16734ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 16744ce5d2b1SEric W. Biederman return false; 16754ce5d2b1SEric W. Biederman 1676f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 16778823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 16788823c079SEric W. Biederman } 16798823c079SEric W. Biederman 168087129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 168136341f64SRam Pai int flag) 16821da177e4SLinus Torvalds { 168384d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 16841da177e4SLinus Torvalds 16854ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 16864ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 16874ce5d2b1SEric W. Biederman 16884ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1689be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 16909676f0c6SRam Pai 169136341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1692be34d1a3SDavid Howells if (IS_ERR(q)) 1693be34d1a3SDavid Howells return q; 1694be34d1a3SDavid Howells 1695a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 16961da177e4SLinus Torvalds 16971da177e4SLinus Torvalds p = mnt; 16986b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1699315fc83eSAl Viro struct mount *s; 17007ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17011da177e4SLinus Torvalds continue; 17021da177e4SLinus Torvalds 1703909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 170412a5b529SAl Viro struct mount *t = NULL; 17054ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17064ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 17074ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17084ce5d2b1SEric W. Biederman continue; 17094ce5d2b1SEric W. Biederman } 17104ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17114ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17129676f0c6SRam Pai s = skip_mnt_tree(s); 17139676f0c6SRam Pai continue; 17149676f0c6SRam Pai } 17150714a533SAl Viro while (p != s->mnt_parent) { 17160714a533SAl Viro p = p->mnt_parent; 17170714a533SAl Viro q = q->mnt_parent; 17181da177e4SLinus Torvalds } 171987129cc0SAl Viro p = s; 172084d17192SAl Viro parent = q; 172187129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1722be34d1a3SDavid Howells if (IS_ERR(q)) 1723be34d1a3SDavid Howells goto out; 1724719ea2fbSAl Viro lock_mount_hash(); 17251a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 172612a5b529SAl Viro mnt_set_mountpoint(parent, p->mnt_mp, q); 172712a5b529SAl Viro if (!list_empty(&parent->mnt_mounts)) { 172812a5b529SAl Viro t = list_last_entry(&parent->mnt_mounts, 172912a5b529SAl Viro struct mount, mnt_child); 173012a5b529SAl Viro if (t->mnt_mp != p->mnt_mp) 173112a5b529SAl Viro t = NULL; 173212a5b529SAl Viro } 173312a5b529SAl Viro attach_shadowed(q, parent, t); 1734719ea2fbSAl Viro unlock_mount_hash(); 17351da177e4SLinus Torvalds } 17361da177e4SLinus Torvalds } 17371da177e4SLinus Torvalds return res; 1738be34d1a3SDavid Howells out: 17391da177e4SLinus Torvalds if (res) { 1740719ea2fbSAl Viro lock_mount_hash(); 1741e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1742719ea2fbSAl Viro unlock_mount_hash(); 17431da177e4SLinus Torvalds } 1744be34d1a3SDavid Howells return q; 17451da177e4SLinus Torvalds } 17461da177e4SLinus Torvalds 1747be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1748be34d1a3SDavid Howells 1749589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 17508aec0809SAl Viro { 1751cb338d06SAl Viro struct mount *tree; 175297216be0SAl Viro namespace_lock(); 1753cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1754cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1755cd4a4017SEric W. Biederman else 175687129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 175787129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1758328e6d90SAl Viro namespace_unlock(); 1759be34d1a3SDavid Howells if (IS_ERR(tree)) 176052e220d3SDan Carpenter return ERR_CAST(tree); 1761be34d1a3SDavid Howells return &tree->mnt; 17628aec0809SAl Viro } 17638aec0809SAl Viro 17648aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 17658aec0809SAl Viro { 176697216be0SAl Viro namespace_lock(); 1767719ea2fbSAl Viro lock_mount_hash(); 1768e819f152SEric W. Biederman umount_tree(real_mount(mnt), UMOUNT_SYNC); 1769719ea2fbSAl Viro unlock_mount_hash(); 17703ab6abeeSAl Viro namespace_unlock(); 17718aec0809SAl Viro } 17728aec0809SAl Viro 1773c771d683SMiklos Szeredi /** 1774c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1775c771d683SMiklos Szeredi * 1776c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1777c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1778c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1779c771d683SMiklos Szeredi * 1780c771d683SMiklos Szeredi * Release with mntput(). 1781c771d683SMiklos Szeredi */ 1782c771d683SMiklos Szeredi struct vfsmount *clone_private_mount(struct path *path) 1783c771d683SMiklos Szeredi { 1784c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1785c771d683SMiklos Szeredi struct mount *new_mnt; 1786c771d683SMiklos Szeredi 1787c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1788c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1789c771d683SMiklos Szeredi 1790c771d683SMiklos Szeredi down_read(&namespace_sem); 1791c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1792c771d683SMiklos Szeredi up_read(&namespace_sem); 1793c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1794c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1795c771d683SMiklos Szeredi 1796c771d683SMiklos Szeredi return &new_mnt->mnt; 1797c771d683SMiklos Szeredi } 1798c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1799c771d683SMiklos Szeredi 18001f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18011f707137SAl Viro struct vfsmount *root) 18021f707137SAl Viro { 18031a4eeaf2SAl Viro struct mount *mnt; 18041f707137SAl Viro int res = f(root, arg); 18051f707137SAl Viro if (res) 18061f707137SAl Viro return res; 18071a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18081a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18091f707137SAl Viro if (res) 18101f707137SAl Viro return res; 18111f707137SAl Viro } 18121f707137SAl Viro return 0; 18131f707137SAl Viro } 18141f707137SAl Viro 18154b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1816719f5d7fSMiklos Szeredi { 1817315fc83eSAl Viro struct mount *p; 1818719f5d7fSMiklos Szeredi 1819909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1820fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 18214b8b21f4SAl Viro mnt_release_group_id(p); 1822719f5d7fSMiklos Szeredi } 1823719f5d7fSMiklos Szeredi } 1824719f5d7fSMiklos Szeredi 18254b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1826719f5d7fSMiklos Szeredi { 1827315fc83eSAl Viro struct mount *p; 1828719f5d7fSMiklos Szeredi 1829909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1830fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 18314b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1832719f5d7fSMiklos Szeredi if (err) { 18334b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1834719f5d7fSMiklos Szeredi return err; 1835719f5d7fSMiklos Szeredi } 1836719f5d7fSMiklos Szeredi } 1837719f5d7fSMiklos Szeredi } 1838719f5d7fSMiklos Szeredi 1839719f5d7fSMiklos Szeredi return 0; 1840719f5d7fSMiklos Szeredi } 1841719f5d7fSMiklos Szeredi 1842b90fa9aeSRam Pai /* 1843b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1844b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 184521444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 184621444403SRam Pai * store the parent mount and mountpoint dentry. 184721444403SRam Pai * (done when source_mnt is moved) 1848b90fa9aeSRam Pai * 1849b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1850b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 18519676f0c6SRam Pai * --------------------------------------------------------------------------- 1852b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 18539676f0c6SRam Pai * |************************************************************************** 18549676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 18559676f0c6SRam Pai * | dest | | | | | 18569676f0c6SRam Pai * | | | | | | | 18579676f0c6SRam Pai * | v | | | | | 18589676f0c6SRam Pai * |************************************************************************** 18599676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 18605afe0022SRam Pai * | | | | | | 18619676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 18629676f0c6SRam Pai * *************************************************************************** 1863b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1864b90fa9aeSRam Pai * destination mount. 1865b90fa9aeSRam Pai * 1866b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1867b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1868b90fa9aeSRam Pai * the peer group of the source mount. 1869b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1870b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1871b90fa9aeSRam Pai * mount. 18725afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 18735afe0022SRam Pai * of the destination mount and the cloned mount is made slave 18745afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 18755afe0022SRam Pai * is marked as 'shared and slave'. 18765afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 18775afe0022SRam Pai * source mount. 18785afe0022SRam Pai * 18799676f0c6SRam Pai * --------------------------------------------------------------------------- 188021444403SRam Pai * | MOVE MOUNT OPERATION | 18819676f0c6SRam Pai * |************************************************************************** 18829676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 18839676f0c6SRam Pai * | dest | | | | | 18849676f0c6SRam Pai * | | | | | | | 18859676f0c6SRam Pai * | v | | | | | 18869676f0c6SRam Pai * |************************************************************************** 18879676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 18885afe0022SRam Pai * | | | | | | 18899676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 18909676f0c6SRam Pai * *************************************************************************** 18915afe0022SRam Pai * 18925afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 18935afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 189421444403SRam Pai * (+*) the mount is moved to the destination. 18955afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 18965afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 18975afe0022SRam Pai * the mount is marked as 'shared and slave'. 18985afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1899b90fa9aeSRam Pai * 1900b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1901b90fa9aeSRam Pai * applied to each mount in the tree. 1902b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1903b90fa9aeSRam Pai * in allocations. 1904b90fa9aeSRam Pai */ 19050fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 190684d17192SAl Viro struct mount *dest_mnt, 190784d17192SAl Viro struct mountpoint *dest_mp, 190884d17192SAl Viro struct path *parent_path) 1909b90fa9aeSRam Pai { 191038129a13SAl Viro HLIST_HEAD(tree_list); 1911315fc83eSAl Viro struct mount *child, *p; 191238129a13SAl Viro struct hlist_node *n; 1913719f5d7fSMiklos Szeredi int err; 1914b90fa9aeSRam Pai 1915fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 19160fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1917719f5d7fSMiklos Szeredi if (err) 1918719f5d7fSMiklos Szeredi goto out; 191984d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1920f2ebb3a9SAl Viro lock_mount_hash(); 1921719f5d7fSMiklos Szeredi if (err) 1922719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1923909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 19240f0afb1dSAl Viro set_mnt_shared(p); 19250b1b901bSAl Viro } else { 19260b1b901bSAl Viro lock_mount_hash(); 1927b90fa9aeSRam Pai } 19281a390689SAl Viro if (parent_path) { 19290fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 193084d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1931143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 193221444403SRam Pai } else { 193384d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 19341d6a32acSAl Viro commit_tree(source_mnt, NULL); 193521444403SRam Pai } 1936b90fa9aeSRam Pai 193738129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 19381d6a32acSAl Viro struct mount *q; 193938129a13SAl Viro hlist_del_init(&child->mnt_hash); 19401d6a32acSAl Viro q = __lookup_mnt_last(&child->mnt_parent->mnt, 19411d6a32acSAl Viro child->mnt_mountpoint); 19421d6a32acSAl Viro commit_tree(child, q); 1943b90fa9aeSRam Pai } 1944719ea2fbSAl Viro unlock_mount_hash(); 194599b7db7bSNick Piggin 1946b90fa9aeSRam Pai return 0; 1947719f5d7fSMiklos Szeredi 1948719f5d7fSMiklos Szeredi out_cleanup_ids: 1949f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 1950f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 1951e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 1952f2ebb3a9SAl Viro } 1953f2ebb3a9SAl Viro unlock_mount_hash(); 19540fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1955719f5d7fSMiklos Szeredi out: 1956719f5d7fSMiklos Szeredi return err; 1957b90fa9aeSRam Pai } 1958b90fa9aeSRam Pai 195984d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 1960b12cea91SAl Viro { 1961b12cea91SAl Viro struct vfsmount *mnt; 196284d17192SAl Viro struct dentry *dentry = path->dentry; 1963b12cea91SAl Viro retry: 19645955102cSAl Viro inode_lock(dentry->d_inode); 196584d17192SAl Viro if (unlikely(cant_mount(dentry))) { 19665955102cSAl Viro inode_unlock(dentry->d_inode); 196784d17192SAl Viro return ERR_PTR(-ENOENT); 1968b12cea91SAl Viro } 196997216be0SAl Viro namespace_lock(); 1970b12cea91SAl Viro mnt = lookup_mnt(path); 197184d17192SAl Viro if (likely(!mnt)) { 1972e2dfa935SEric W. Biederman struct mountpoint *mp = lookup_mountpoint(dentry); 1973e2dfa935SEric W. Biederman if (!mp) 1974e2dfa935SEric W. Biederman mp = new_mountpoint(dentry); 197584d17192SAl Viro if (IS_ERR(mp)) { 197697216be0SAl Viro namespace_unlock(); 19775955102cSAl Viro inode_unlock(dentry->d_inode); 197884d17192SAl Viro return mp; 197984d17192SAl Viro } 198084d17192SAl Viro return mp; 198184d17192SAl Viro } 198297216be0SAl Viro namespace_unlock(); 19835955102cSAl Viro inode_unlock(path->dentry->d_inode); 1984b12cea91SAl Viro path_put(path); 1985b12cea91SAl Viro path->mnt = mnt; 198684d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 1987b12cea91SAl Viro goto retry; 1988b12cea91SAl Viro } 1989b12cea91SAl Viro 199084d17192SAl Viro static void unlock_mount(struct mountpoint *where) 1991b12cea91SAl Viro { 199284d17192SAl Viro struct dentry *dentry = where->m_dentry; 199384d17192SAl Viro put_mountpoint(where); 1994328e6d90SAl Viro namespace_unlock(); 19955955102cSAl Viro inode_unlock(dentry->d_inode); 1996b12cea91SAl Viro } 1997b12cea91SAl Viro 199884d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 19991da177e4SLinus Torvalds { 200095bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 20011da177e4SLinus Torvalds return -EINVAL; 20021da177e4SLinus Torvalds 2003e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2004e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 20051da177e4SLinus Torvalds return -ENOTDIR; 20061da177e4SLinus Torvalds 200784d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 20081da177e4SLinus Torvalds } 20091da177e4SLinus Torvalds 20101da177e4SLinus Torvalds /* 20117a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 20127a2e8a8fSValerie Aurora */ 20137a2e8a8fSValerie Aurora 20147a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 20157a2e8a8fSValerie Aurora { 20167c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 20177a2e8a8fSValerie Aurora 20187a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 20197a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 20207a2e8a8fSValerie Aurora return 0; 20217a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 20227a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 20237a2e8a8fSValerie Aurora return 0; 20247a2e8a8fSValerie Aurora return type; 20257a2e8a8fSValerie Aurora } 20267a2e8a8fSValerie Aurora 20277a2e8a8fSValerie Aurora /* 202807b20889SRam Pai * recursively change the type of the mountpoint. 202907b20889SRam Pai */ 20300a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 203107b20889SRam Pai { 2032315fc83eSAl Viro struct mount *m; 20334b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 203407b20889SRam Pai int recurse = flag & MS_REC; 20357a2e8a8fSValerie Aurora int type; 2036719f5d7fSMiklos Szeredi int err = 0; 203707b20889SRam Pai 20382d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 203907b20889SRam Pai return -EINVAL; 204007b20889SRam Pai 20417a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 20427a2e8a8fSValerie Aurora if (!type) 20437a2e8a8fSValerie Aurora return -EINVAL; 20447a2e8a8fSValerie Aurora 204597216be0SAl Viro namespace_lock(); 2046719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2047719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2048719f5d7fSMiklos Szeredi if (err) 2049719f5d7fSMiklos Szeredi goto out_unlock; 2050719f5d7fSMiklos Szeredi } 2051719f5d7fSMiklos Szeredi 2052719ea2fbSAl Viro lock_mount_hash(); 2053909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 20540f0afb1dSAl Viro change_mnt_propagation(m, type); 2055719ea2fbSAl Viro unlock_mount_hash(); 2056719f5d7fSMiklos Szeredi 2057719f5d7fSMiklos Szeredi out_unlock: 205897216be0SAl Viro namespace_unlock(); 2059719f5d7fSMiklos Szeredi return err; 206007b20889SRam Pai } 206107b20889SRam Pai 20625ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 20635ff9d8a6SEric W. Biederman { 20645ff9d8a6SEric W. Biederman struct mount *child; 20655ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 20665ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 20675ff9d8a6SEric W. Biederman continue; 20685ff9d8a6SEric W. Biederman 20695ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 20705ff9d8a6SEric W. Biederman return true; 20715ff9d8a6SEric W. Biederman } 20725ff9d8a6SEric W. Biederman return false; 20735ff9d8a6SEric W. Biederman } 20745ff9d8a6SEric W. Biederman 207507b20889SRam Pai /* 20761da177e4SLinus Torvalds * do loopback mount. 20771da177e4SLinus Torvalds */ 2078808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 20792dafe1c4SEric Sandeen int recurse) 20801da177e4SLinus Torvalds { 20812d92ab3cSAl Viro struct path old_path; 208284d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 208384d17192SAl Viro struct mountpoint *mp; 208457eccb83SAl Viro int err; 20851da177e4SLinus Torvalds if (!old_name || !*old_name) 20861da177e4SLinus Torvalds return -EINVAL; 2087815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 20881da177e4SLinus Torvalds if (err) 20891da177e4SLinus Torvalds return err; 20901da177e4SLinus Torvalds 20918823c079SEric W. Biederman err = -EINVAL; 20924ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 20938823c079SEric W. Biederman goto out; 20948823c079SEric W. Biederman 209584d17192SAl Viro mp = lock_mount(path); 209684d17192SAl Viro err = PTR_ERR(mp); 209784d17192SAl Viro if (IS_ERR(mp)) 20989676f0c6SRam Pai goto out; 20999676f0c6SRam Pai 210087129cc0SAl Viro old = real_mount(old_path.mnt); 210184d17192SAl Viro parent = real_mount(path->mnt); 210287129cc0SAl Viro 2103b12cea91SAl Viro err = -EINVAL; 2104fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2105b12cea91SAl Viro goto out2; 2106b12cea91SAl Viro 2107e149ed2bSAl Viro if (!check_mnt(parent)) 2108e149ed2bSAl Viro goto out2; 2109e149ed2bSAl Viro 2110e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2111b12cea91SAl Viro goto out2; 2112ccd48bc7SAl Viro 21135ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 21145ff9d8a6SEric W. Biederman goto out2; 21155ff9d8a6SEric W. Biederman 21161da177e4SLinus Torvalds if (recurse) 21174ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 21181da177e4SLinus Torvalds else 211987129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 21201da177e4SLinus Torvalds 2121be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2122be34d1a3SDavid Howells err = PTR_ERR(mnt); 2123e9c5d8a5SAndrey Vagin goto out2; 2124be34d1a3SDavid Howells } 2125ccd48bc7SAl Viro 21265ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 21275ff9d8a6SEric W. Biederman 212884d17192SAl Viro err = graft_tree(mnt, parent, mp); 21291da177e4SLinus Torvalds if (err) { 2130719ea2fbSAl Viro lock_mount_hash(); 2131e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2132719ea2fbSAl Viro unlock_mount_hash(); 21335b83d2c5SRam Pai } 2134b12cea91SAl Viro out2: 213584d17192SAl Viro unlock_mount(mp); 2136ccd48bc7SAl Viro out: 21372d92ab3cSAl Viro path_put(&old_path); 21381da177e4SLinus Torvalds return err; 21391da177e4SLinus Torvalds } 21401da177e4SLinus Torvalds 21412e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 21422e4b7fcdSDave Hansen { 21432e4b7fcdSDave Hansen int error = 0; 21442e4b7fcdSDave Hansen int readonly_request = 0; 21452e4b7fcdSDave Hansen 21462e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 21472e4b7fcdSDave Hansen readonly_request = 1; 21482e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 21492e4b7fcdSDave Hansen return 0; 21502e4b7fcdSDave Hansen 21512e4b7fcdSDave Hansen if (readonly_request) 215283adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 21532e4b7fcdSDave Hansen else 215483adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 21552e4b7fcdSDave Hansen return error; 21562e4b7fcdSDave Hansen } 21572e4b7fcdSDave Hansen 21581da177e4SLinus Torvalds /* 21591da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 21601da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 21611da177e4SLinus Torvalds * on it - tough luck. 21621da177e4SLinus Torvalds */ 21630a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 21641da177e4SLinus Torvalds void *data) 21651da177e4SLinus Torvalds { 21661da177e4SLinus Torvalds int err; 21672d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2168143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 21691da177e4SLinus Torvalds 2170143c8c91SAl Viro if (!check_mnt(mnt)) 21711da177e4SLinus Torvalds return -EINVAL; 21721da177e4SLinus Torvalds 21732d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 21741da177e4SLinus Torvalds return -EINVAL; 21751da177e4SLinus Torvalds 217607b64558SEric W. Biederman /* Don't allow changing of locked mnt flags. 217707b64558SEric W. Biederman * 217807b64558SEric W. Biederman * No locks need to be held here while testing the various 217907b64558SEric W. Biederman * MNT_LOCK flags because those flags can never be cleared 218007b64558SEric W. Biederman * once they are set. 218107b64558SEric W. Biederman */ 218207b64558SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 218307b64558SEric W. Biederman !(mnt_flags & MNT_READONLY)) { 218407b64558SEric W. Biederman return -EPERM; 218507b64558SEric W. Biederman } 21869566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 21879566d674SEric W. Biederman !(mnt_flags & MNT_NODEV)) { 21883e186641SEric W. Biederman /* Was the nodev implicitly added in mount? */ 21893e186641SEric W. Biederman if ((mnt->mnt_ns->user_ns != &init_user_ns) && 21903e186641SEric W. Biederman !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) { 21913e186641SEric W. Biederman mnt_flags |= MNT_NODEV; 21923e186641SEric W. Biederman } else { 21939566d674SEric W. Biederman return -EPERM; 21949566d674SEric W. Biederman } 21953e186641SEric W. Biederman } 21969566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && 21979566d674SEric W. Biederman !(mnt_flags & MNT_NOSUID)) { 21989566d674SEric W. Biederman return -EPERM; 21999566d674SEric W. Biederman } 22009566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && 22019566d674SEric W. Biederman !(mnt_flags & MNT_NOEXEC)) { 22029566d674SEric W. Biederman return -EPERM; 22039566d674SEric W. Biederman } 22049566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 22059566d674SEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { 22069566d674SEric W. Biederman return -EPERM; 22079566d674SEric W. Biederman } 22089566d674SEric W. Biederman 2209ff36fe2cSEric Paris err = security_sb_remount(sb, data); 2210ff36fe2cSEric Paris if (err) 2211ff36fe2cSEric Paris return err; 2212ff36fe2cSEric Paris 22131da177e4SLinus Torvalds down_write(&sb->s_umount); 22142e4b7fcdSDave Hansen if (flags & MS_BIND) 22152d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 221657eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 221757eccb83SAl Viro err = -EPERM; 22184aa98cf7SAl Viro else 22191da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 22207b43a79fSAl Viro if (!err) { 2221719ea2fbSAl Viro lock_mount_hash(); 2222a6138db8SEric W. Biederman mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 2223143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 2224143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2225719ea2fbSAl Viro unlock_mount_hash(); 22260e55a7ccSDan Williams } 22276339dab8SAl Viro up_write(&sb->s_umount); 22281da177e4SLinus Torvalds return err; 22291da177e4SLinus Torvalds } 22301da177e4SLinus Torvalds 2231cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 22329676f0c6SRam Pai { 2233315fc83eSAl Viro struct mount *p; 2234909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2235fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 22369676f0c6SRam Pai return 1; 22379676f0c6SRam Pai } 22389676f0c6SRam Pai return 0; 22399676f0c6SRam Pai } 22409676f0c6SRam Pai 2241808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 22421da177e4SLinus Torvalds { 22432d92ab3cSAl Viro struct path old_path, parent_path; 2244676da58dSAl Viro struct mount *p; 22450fb54e50SAl Viro struct mount *old; 224684d17192SAl Viro struct mountpoint *mp; 224757eccb83SAl Viro int err; 22481da177e4SLinus Torvalds if (!old_name || !*old_name) 22491da177e4SLinus Torvalds return -EINVAL; 22502d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 22511da177e4SLinus Torvalds if (err) 22521da177e4SLinus Torvalds return err; 22531da177e4SLinus Torvalds 225484d17192SAl Viro mp = lock_mount(path); 225584d17192SAl Viro err = PTR_ERR(mp); 225684d17192SAl Viro if (IS_ERR(mp)) 2257cc53ce53SDavid Howells goto out; 2258cc53ce53SDavid Howells 2259143c8c91SAl Viro old = real_mount(old_path.mnt); 2260fc7be130SAl Viro p = real_mount(path->mnt); 2261143c8c91SAl Viro 22621da177e4SLinus Torvalds err = -EINVAL; 2263fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 22641da177e4SLinus Torvalds goto out1; 22651da177e4SLinus Torvalds 22665ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 22675ff9d8a6SEric W. Biederman goto out1; 22685ff9d8a6SEric W. Biederman 22691da177e4SLinus Torvalds err = -EINVAL; 22702d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 227121444403SRam Pai goto out1; 22721da177e4SLinus Torvalds 2273676da58dSAl Viro if (!mnt_has_parent(old)) 227421444403SRam Pai goto out1; 22751da177e4SLinus Torvalds 2276e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2277e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 227821444403SRam Pai goto out1; 227921444403SRam Pai /* 228021444403SRam Pai * Don't move a mount residing in a shared parent. 228121444403SRam Pai */ 2282fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 228321444403SRam Pai goto out1; 22849676f0c6SRam Pai /* 22859676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 22869676f0c6SRam Pai * mount which is shared. 22879676f0c6SRam Pai */ 2288fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 22899676f0c6SRam Pai goto out1; 22901da177e4SLinus Torvalds err = -ELOOP; 2291fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2292676da58dSAl Viro if (p == old) 229321444403SRam Pai goto out1; 22941da177e4SLinus Torvalds 229584d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 22964ac91378SJan Blunck if (err) 229721444403SRam Pai goto out1; 22981da177e4SLinus Torvalds 22991da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 23001da177e4SLinus Torvalds * automatically */ 23016776db3dSAl Viro list_del_init(&old->mnt_expire); 23021da177e4SLinus Torvalds out1: 230384d17192SAl Viro unlock_mount(mp); 23041da177e4SLinus Torvalds out: 23051da177e4SLinus Torvalds if (!err) 23061a390689SAl Viro path_put(&parent_path); 23072d92ab3cSAl Viro path_put(&old_path); 23081da177e4SLinus Torvalds return err; 23091da177e4SLinus Torvalds } 23101da177e4SLinus Torvalds 23119d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 23129d412a43SAl Viro { 23139d412a43SAl Viro int err; 23149d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 23159d412a43SAl Viro if (subtype) { 23169d412a43SAl Viro subtype++; 23179d412a43SAl Viro err = -EINVAL; 23189d412a43SAl Viro if (!subtype[0]) 23199d412a43SAl Viro goto err; 23209d412a43SAl Viro } else 23219d412a43SAl Viro subtype = ""; 23229d412a43SAl Viro 23239d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 23249d412a43SAl Viro err = -ENOMEM; 23259d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 23269d412a43SAl Viro goto err; 23279d412a43SAl Viro return mnt; 23289d412a43SAl Viro 23299d412a43SAl Viro err: 23309d412a43SAl Viro mntput(mnt); 23319d412a43SAl Viro return ERR_PTR(err); 23329d412a43SAl Viro } 23339d412a43SAl Viro 23349d412a43SAl Viro /* 23359d412a43SAl Viro * add a mount into a namespace's mount tree 23369d412a43SAl Viro */ 233795bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 23389d412a43SAl Viro { 233984d17192SAl Viro struct mountpoint *mp; 234084d17192SAl Viro struct mount *parent; 23419d412a43SAl Viro int err; 23429d412a43SAl Viro 2343f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 23449d412a43SAl Viro 234584d17192SAl Viro mp = lock_mount(path); 234684d17192SAl Viro if (IS_ERR(mp)) 234784d17192SAl Viro return PTR_ERR(mp); 23489d412a43SAl Viro 234984d17192SAl Viro parent = real_mount(path->mnt); 23509d412a43SAl Viro err = -EINVAL; 235184d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2352156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2353156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 23549d412a43SAl Viro goto unlock; 2355156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 235684d17192SAl Viro if (!parent->mnt_ns) 2357156cacb1SAl Viro goto unlock; 2358156cacb1SAl Viro } 23599d412a43SAl Viro 23609d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 23619d412a43SAl Viro err = -EBUSY; 236295bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 23639d412a43SAl Viro path->mnt->mnt_root == path->dentry) 23649d412a43SAl Viro goto unlock; 23659d412a43SAl Viro 23669d412a43SAl Viro err = -EINVAL; 2367e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 23689d412a43SAl Viro goto unlock; 23699d412a43SAl Viro 237095bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 237184d17192SAl Viro err = graft_tree(newmnt, parent, mp); 23729d412a43SAl Viro 23739d412a43SAl Viro unlock: 237484d17192SAl Viro unlock_mount(mp); 23759d412a43SAl Viro return err; 23769d412a43SAl Viro } 2377b1e75df4SAl Viro 23788654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); 23791b852bceSEric W. Biederman 23801da177e4SLinus Torvalds /* 23811da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 23821da177e4SLinus Torvalds * namespace's tree 23831da177e4SLinus Torvalds */ 23840c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2385808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 23861da177e4SLinus Torvalds { 23870c55cfc4SEric W. Biederman struct file_system_type *type; 23889b40bc90SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 23891da177e4SLinus Torvalds struct vfsmount *mnt; 239015f9a3f3SAl Viro int err; 23911da177e4SLinus Torvalds 23920c55cfc4SEric W. Biederman if (!fstype) 23931da177e4SLinus Torvalds return -EINVAL; 23941da177e4SLinus Torvalds 23950c55cfc4SEric W. Biederman type = get_fs_type(fstype); 23960c55cfc4SEric W. Biederman if (!type) 23970c55cfc4SEric W. Biederman return -ENODEV; 23980c55cfc4SEric W. Biederman 23990c55cfc4SEric W. Biederman if (user_ns != &init_user_ns) { 24000c55cfc4SEric W. Biederman /* Only in special cases allow devices from mounts 24010c55cfc4SEric W. Biederman * created outside the initial user namespace. 24020c55cfc4SEric W. Biederman */ 24030c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { 24040c55cfc4SEric W. Biederman flags |= MS_NODEV; 24059566d674SEric W. Biederman mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; 24060c55cfc4SEric W. Biederman } 240797c1df3eSEric W. Biederman } 24080c55cfc4SEric W. Biederman 24090c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 24100c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 24110c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 24120c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 24130c55cfc4SEric W. Biederman 24140c55cfc4SEric W. Biederman put_filesystem(type); 24151da177e4SLinus Torvalds if (IS_ERR(mnt)) 24161da177e4SLinus Torvalds return PTR_ERR(mnt); 24171da177e4SLinus Torvalds 24188654df4eSEric W. Biederman if (mount_too_revealing(mnt, &mnt_flags)) { 24198654df4eSEric W. Biederman mntput(mnt); 24208654df4eSEric W. Biederman return -EPERM; 24218654df4eSEric W. Biederman } 24228654df4eSEric W. Biederman 242395bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 242415f9a3f3SAl Viro if (err) 242515f9a3f3SAl Viro mntput(mnt); 242615f9a3f3SAl Viro return err; 24271da177e4SLinus Torvalds } 24281da177e4SLinus Torvalds 242919a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 243019a167afSAl Viro { 24316776db3dSAl Viro struct mount *mnt = real_mount(m); 243219a167afSAl Viro int err; 243319a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 243419a167afSAl Viro * expired before we get a chance to add it 243519a167afSAl Viro */ 24366776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 243719a167afSAl Viro 243819a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 243919a167afSAl Viro m->mnt_root == path->dentry) { 2440b1e75df4SAl Viro err = -ELOOP; 2441b1e75df4SAl Viro goto fail; 244219a167afSAl Viro } 244319a167afSAl Viro 244495bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2445b1e75df4SAl Viro if (!err) 2446b1e75df4SAl Viro return 0; 2447b1e75df4SAl Viro fail: 2448b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 24496776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 245097216be0SAl Viro namespace_lock(); 24516776db3dSAl Viro list_del_init(&mnt->mnt_expire); 245297216be0SAl Viro namespace_unlock(); 245319a167afSAl Viro } 2454b1e75df4SAl Viro mntput(m); 2455b1e75df4SAl Viro mntput(m); 245619a167afSAl Viro return err; 245719a167afSAl Viro } 245819a167afSAl Viro 2459ea5b778aSDavid Howells /** 2460ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2461ea5b778aSDavid Howells * @mnt: The mount to list. 2462ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2463ea5b778aSDavid Howells */ 2464ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2465ea5b778aSDavid Howells { 246697216be0SAl Viro namespace_lock(); 2467ea5b778aSDavid Howells 24686776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2469ea5b778aSDavid Howells 247097216be0SAl Viro namespace_unlock(); 2471ea5b778aSDavid Howells } 2472ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2473ea5b778aSDavid Howells 2474ea5b778aSDavid Howells /* 24751da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 24761da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 24771da177e4SLinus Torvalds * here 24781da177e4SLinus Torvalds */ 24791da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 24801da177e4SLinus Torvalds { 2481761d5c38SAl Viro struct mount *mnt, *next; 24821da177e4SLinus Torvalds LIST_HEAD(graveyard); 24831da177e4SLinus Torvalds 24841da177e4SLinus Torvalds if (list_empty(mounts)) 24851da177e4SLinus Torvalds return; 24861da177e4SLinus Torvalds 248797216be0SAl Viro namespace_lock(); 2488719ea2fbSAl Viro lock_mount_hash(); 24891da177e4SLinus Torvalds 24901da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 24911da177e4SLinus Torvalds * following criteria: 24921da177e4SLinus Torvalds * - only referenced by its parent vfsmount 24931da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 24941da177e4SLinus Torvalds * cleared by mntput()) 24951da177e4SLinus Torvalds */ 24966776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2497863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 24981ab59738SAl Viro propagate_mount_busy(mnt, 1)) 24991da177e4SLinus Torvalds continue; 25006776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 25011da177e4SLinus Torvalds } 2502bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 25036776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2504143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2505e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2506bcc5c7d2SAl Viro } 2507719ea2fbSAl Viro unlock_mount_hash(); 25083ab6abeeSAl Viro namespace_unlock(); 25091da177e4SLinus Torvalds } 25101da177e4SLinus Torvalds 25111da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 25121da177e4SLinus Torvalds 25131da177e4SLinus Torvalds /* 25145528f911STrond Myklebust * Ripoff of 'select_parent()' 25155528f911STrond Myklebust * 25165528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 25175528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 25185528f911STrond Myklebust */ 2519692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 25205528f911STrond Myklebust { 2521692afc31SAl Viro struct mount *this_parent = parent; 25225528f911STrond Myklebust struct list_head *next; 25235528f911STrond Myklebust int found = 0; 25245528f911STrond Myklebust 25255528f911STrond Myklebust repeat: 25266b41d536SAl Viro next = this_parent->mnt_mounts.next; 25275528f911STrond Myklebust resume: 25286b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 25295528f911STrond Myklebust struct list_head *tmp = next; 25306b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 25315528f911STrond Myklebust 25325528f911STrond Myklebust next = tmp->next; 2533692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 25345528f911STrond Myklebust continue; 25355528f911STrond Myklebust /* 25365528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 25375528f911STrond Myklebust */ 25386b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 25395528f911STrond Myklebust this_parent = mnt; 25405528f911STrond Myklebust goto repeat; 25415528f911STrond Myklebust } 25425528f911STrond Myklebust 25431ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 25446776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 25455528f911STrond Myklebust found++; 25465528f911STrond Myklebust } 25475528f911STrond Myklebust } 25485528f911STrond Myklebust /* 25495528f911STrond Myklebust * All done at this level ... ascend and resume the search 25505528f911STrond Myklebust */ 25515528f911STrond Myklebust if (this_parent != parent) { 25526b41d536SAl Viro next = this_parent->mnt_child.next; 25530714a533SAl Viro this_parent = this_parent->mnt_parent; 25545528f911STrond Myklebust goto resume; 25555528f911STrond Myklebust } 25565528f911STrond Myklebust return found; 25575528f911STrond Myklebust } 25585528f911STrond Myklebust 25595528f911STrond Myklebust /* 25605528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 25615528f911STrond Myklebust * submounts of a specific parent mountpoint 256299b7db7bSNick Piggin * 256348a066e7SAl Viro * mount_lock must be held for write 25645528f911STrond Myklebust */ 2565b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 25665528f911STrond Myklebust { 25675528f911STrond Myklebust LIST_HEAD(graveyard); 2568761d5c38SAl Viro struct mount *m; 25695528f911STrond Myklebust 25705528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2571c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2572bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2573761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 25746776db3dSAl Viro mnt_expire); 2575143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2576e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2577bcc5c7d2SAl Viro } 2578bcc5c7d2SAl Viro } 25795528f911STrond Myklebust } 25805528f911STrond Myklebust 25815528f911STrond Myklebust /* 25821da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 25831da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 25841da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 25851da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 25861da177e4SLinus Torvalds */ 2587b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2588b58fed8bSRam Pai unsigned long n) 25891da177e4SLinus Torvalds { 25901da177e4SLinus Torvalds char *t = to; 25911da177e4SLinus Torvalds const char __user *f = from; 25921da177e4SLinus Torvalds char c; 25931da177e4SLinus Torvalds 25941da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 25951da177e4SLinus Torvalds return n; 25961da177e4SLinus Torvalds 25971da177e4SLinus Torvalds while (n) { 25981da177e4SLinus Torvalds if (__get_user(c, f)) { 25991da177e4SLinus Torvalds memset(t, 0, n); 26001da177e4SLinus Torvalds break; 26011da177e4SLinus Torvalds } 26021da177e4SLinus Torvalds *t++ = c; 26031da177e4SLinus Torvalds f++; 26041da177e4SLinus Torvalds n--; 26051da177e4SLinus Torvalds } 26061da177e4SLinus Torvalds return n; 26071da177e4SLinus Torvalds } 26081da177e4SLinus Torvalds 2609b40ef869SAl Viro void *copy_mount_options(const void __user * data) 26101da177e4SLinus Torvalds { 26111da177e4SLinus Torvalds int i; 26121da177e4SLinus Torvalds unsigned long size; 2613b40ef869SAl Viro char *copy; 26141da177e4SLinus Torvalds 26151da177e4SLinus Torvalds if (!data) 2616b40ef869SAl Viro return NULL; 26171da177e4SLinus Torvalds 2618b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2619b40ef869SAl Viro if (!copy) 2620b40ef869SAl Viro return ERR_PTR(-ENOMEM); 26211da177e4SLinus Torvalds 26221da177e4SLinus Torvalds /* We only care that *some* data at the address the user 26231da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 26241da177e4SLinus Torvalds * the remainder of the page. 26251da177e4SLinus Torvalds */ 26261da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 26271da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 26281da177e4SLinus Torvalds if (size > PAGE_SIZE) 26291da177e4SLinus Torvalds size = PAGE_SIZE; 26301da177e4SLinus Torvalds 2631b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 26321da177e4SLinus Torvalds if (!i) { 2633b40ef869SAl Viro kfree(copy); 2634b40ef869SAl Viro return ERR_PTR(-EFAULT); 26351da177e4SLinus Torvalds } 26361da177e4SLinus Torvalds if (i != PAGE_SIZE) 2637b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2638b40ef869SAl Viro return copy; 26391da177e4SLinus Torvalds } 26401da177e4SLinus Torvalds 2641b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2642eca6f534SVegard Nossum { 2643b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2644eca6f534SVegard Nossum } 2645eca6f534SVegard Nossum 26461da177e4SLinus Torvalds /* 26471da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 26481da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 26491da177e4SLinus Torvalds * 26501da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 26511da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 26521da177e4SLinus Torvalds * information (or be NULL). 26531da177e4SLinus Torvalds * 26541da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 26551da177e4SLinus Torvalds * When the flags word was introduced its top half was required 26561da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 26571da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 26581da177e4SLinus Torvalds * and must be discarded. 26591da177e4SLinus Torvalds */ 26605e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2661808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 26621da177e4SLinus Torvalds { 26632d92ab3cSAl Viro struct path path; 26641da177e4SLinus Torvalds int retval = 0; 26651da177e4SLinus Torvalds int mnt_flags = 0; 26661da177e4SLinus Torvalds 26671da177e4SLinus Torvalds /* Discard magic */ 26681da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 26691da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 26701da177e4SLinus Torvalds 26711da177e4SLinus Torvalds /* Basic sanity checks */ 26721da177e4SLinus Torvalds if (data_page) 26731da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 26741da177e4SLinus Torvalds 2675a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 26765e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2677a27ab9f2STetsuo Handa if (retval) 2678a27ab9f2STetsuo Handa return retval; 2679a27ab9f2STetsuo Handa 2680a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2681a27ab9f2STetsuo Handa type_page, flags, data_page); 26820d5cadb8SAl Viro if (!retval && !may_mount()) 26830d5cadb8SAl Viro retval = -EPERM; 26849e8925b6SJeff Layton if (!retval && (flags & MS_MANDLOCK) && !may_mandlock()) 26859e8925b6SJeff Layton retval = -EPERM; 2686a27ab9f2STetsuo Handa if (retval) 2687a27ab9f2STetsuo Handa goto dput_out; 2688a27ab9f2STetsuo Handa 2689613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2690613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 26910a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 26920a1c01c9SMatthew Garrett 26931da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 26941da177e4SLinus Torvalds if (flags & MS_NOSUID) 26951da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 26961da177e4SLinus Torvalds if (flags & MS_NODEV) 26971da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 26981da177e4SLinus Torvalds if (flags & MS_NOEXEC) 26991da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2700fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2701fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2702fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2703fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2704d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2705d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 27062e4b7fcdSDave Hansen if (flags & MS_RDONLY) 27072e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2708fc33a7bbSChristoph Hellwig 2709ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2710ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2711ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2712ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2713ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2714ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2715ffbc6f0eSEric W. Biederman } 2716ffbc6f0eSEric W. Biederman 27177a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2718d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2719d0adde57SMatthew Garrett MS_STRICTATIME); 27201da177e4SLinus Torvalds 27211da177e4SLinus Torvalds if (flags & MS_REMOUNT) 27222d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 27231da177e4SLinus Torvalds data_page); 27241da177e4SLinus Torvalds else if (flags & MS_BIND) 27252d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 27269676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 27272d92ab3cSAl Viro retval = do_change_type(&path, flags); 27281da177e4SLinus Torvalds else if (flags & MS_MOVE) 27292d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 27301da177e4SLinus Torvalds else 27312d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 27321da177e4SLinus Torvalds dev_name, data_page); 27331da177e4SLinus Torvalds dput_out: 27342d92ab3cSAl Viro path_put(&path); 27351da177e4SLinus Torvalds return retval; 27361da177e4SLinus Torvalds } 27371da177e4SLinus Torvalds 2738771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2739771b1371SEric W. Biederman { 27406344c433SAl Viro ns_free_inum(&ns->ns); 2741771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2742771b1371SEric W. Biederman kfree(ns); 2743771b1371SEric W. Biederman } 2744771b1371SEric W. Biederman 27458823c079SEric W. Biederman /* 27468823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 27478823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 27488823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 27498823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 27508823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 27518823c079SEric W. Biederman */ 27528823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 27538823c079SEric W. Biederman 2754771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2755cf8d2c11STrond Myklebust { 2756cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 275798f842e6SEric W. Biederman int ret; 2758cf8d2c11STrond Myklebust 2759cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2760cf8d2c11STrond Myklebust if (!new_ns) 2761cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 27626344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 276398f842e6SEric W. Biederman if (ret) { 276498f842e6SEric W. Biederman kfree(new_ns); 276598f842e6SEric W. Biederman return ERR_PTR(ret); 276698f842e6SEric W. Biederman } 276733c42940SAl Viro new_ns->ns.ops = &mntns_operations; 27688823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2769cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2770cf8d2c11STrond Myklebust new_ns->root = NULL; 2771cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2772cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2773cf8d2c11STrond Myklebust new_ns->event = 0; 2774771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2775cf8d2c11STrond Myklebust return new_ns; 2776cf8d2c11STrond Myklebust } 2777cf8d2c11STrond Myklebust 27789559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 27799559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 27801da177e4SLinus Torvalds { 27816b3286edSKirill Korotaev struct mnt_namespace *new_ns; 27827f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2783315fc83eSAl Viro struct mount *p, *q; 27849559f689SAl Viro struct mount *old; 2785cb338d06SAl Viro struct mount *new; 27867a472ef4SEric W. Biederman int copy_flags; 27871da177e4SLinus Torvalds 27889559f689SAl Viro BUG_ON(!ns); 27899559f689SAl Viro 27909559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 27919559f689SAl Viro get_mnt_ns(ns); 27929559f689SAl Viro return ns; 27939559f689SAl Viro } 27949559f689SAl Viro 27959559f689SAl Viro old = ns->root; 27969559f689SAl Viro 2797771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2798cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2799cf8d2c11STrond Myklebust return new_ns; 28001da177e4SLinus Torvalds 280197216be0SAl Viro namespace_lock(); 28021da177e4SLinus Torvalds /* First pass: copy the tree topology */ 28034ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 28049559f689SAl Viro if (user_ns != ns->user_ns) 2805132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 28067a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2807be34d1a3SDavid Howells if (IS_ERR(new)) { 2808328e6d90SAl Viro namespace_unlock(); 2809771b1371SEric W. Biederman free_mnt_ns(new_ns); 2810be34d1a3SDavid Howells return ERR_CAST(new); 28111da177e4SLinus Torvalds } 2812be08d6d2SAl Viro new_ns->root = new; 28131a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 28141da177e4SLinus Torvalds 28151da177e4SLinus Torvalds /* 28161da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 28171da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 28181da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 28191da177e4SLinus Torvalds */ 2820909b0a88SAl Viro p = old; 2821cb338d06SAl Viro q = new; 28221da177e4SLinus Torvalds while (p) { 2823143c8c91SAl Viro q->mnt_ns = new_ns; 28249559f689SAl Viro if (new_fs) { 28259559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 28269559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2827315fc83eSAl Viro rootmnt = &p->mnt; 28281da177e4SLinus Torvalds } 28299559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 28309559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2831315fc83eSAl Viro pwdmnt = &p->mnt; 28321da177e4SLinus Torvalds } 28331da177e4SLinus Torvalds } 2834909b0a88SAl Viro p = next_mnt(p, old); 2835909b0a88SAl Viro q = next_mnt(q, new); 28364ce5d2b1SEric W. Biederman if (!q) 28374ce5d2b1SEric W. Biederman break; 28384ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 28394ce5d2b1SEric W. Biederman p = next_mnt(p, old); 28401da177e4SLinus Torvalds } 2841328e6d90SAl Viro namespace_unlock(); 28421da177e4SLinus Torvalds 28431da177e4SLinus Torvalds if (rootmnt) 2844f03c6599SAl Viro mntput(rootmnt); 28451da177e4SLinus Torvalds if (pwdmnt) 2846f03c6599SAl Viro mntput(pwdmnt); 28471da177e4SLinus Torvalds 2848741a2951SJANAK DESAI return new_ns; 2849741a2951SJANAK DESAI } 2850741a2951SJANAK DESAI 2851cf8d2c11STrond Myklebust /** 2852cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2853cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2854cf8d2c11STrond Myklebust */ 28551a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2856cf8d2c11STrond Myklebust { 2857771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2858cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 28591a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 28601a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2861be08d6d2SAl Viro new_ns->root = mnt; 2862b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2863c1334495SAl Viro } else { 28641a4eeaf2SAl Viro mntput(m); 2865cf8d2c11STrond Myklebust } 2866cf8d2c11STrond Myklebust return new_ns; 2867cf8d2c11STrond Myklebust } 2868cf8d2c11STrond Myklebust 2869ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2870ea441d11SAl Viro { 2871ea441d11SAl Viro struct mnt_namespace *ns; 2872d31da0f0SAl Viro struct super_block *s; 2873ea441d11SAl Viro struct path path; 2874ea441d11SAl Viro int err; 2875ea441d11SAl Viro 2876ea441d11SAl Viro ns = create_mnt_ns(mnt); 2877ea441d11SAl Viro if (IS_ERR(ns)) 2878ea441d11SAl Viro return ERR_CAST(ns); 2879ea441d11SAl Viro 2880ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2881ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2882ea441d11SAl Viro 2883ea441d11SAl Viro put_mnt_ns(ns); 2884ea441d11SAl Viro 2885ea441d11SAl Viro if (err) 2886ea441d11SAl Viro return ERR_PTR(err); 2887ea441d11SAl Viro 2888ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2889d31da0f0SAl Viro s = path.mnt->mnt_sb; 2890d31da0f0SAl Viro atomic_inc(&s->s_active); 2891ea441d11SAl Viro mntput(path.mnt); 2892ea441d11SAl Viro /* lock the sucker */ 2893d31da0f0SAl Viro down_write(&s->s_umount); 2894ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2895ea441d11SAl Viro return path.dentry; 2896ea441d11SAl Viro } 2897ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2898ea441d11SAl Viro 2899bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2900bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 29011da177e4SLinus Torvalds { 2902eca6f534SVegard Nossum int ret; 2903eca6f534SVegard Nossum char *kernel_type; 2904eca6f534SVegard Nossum char *kernel_dev; 2905b40ef869SAl Viro void *options; 29061da177e4SLinus Torvalds 2907b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 2908b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 2909b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 2910eca6f534SVegard Nossum goto out_type; 29111da177e4SLinus Torvalds 2912b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 2913b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 2914b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 2915eca6f534SVegard Nossum goto out_dev; 29161da177e4SLinus Torvalds 2917b40ef869SAl Viro options = copy_mount_options(data); 2918b40ef869SAl Viro ret = PTR_ERR(options); 2919b40ef869SAl Viro if (IS_ERR(options)) 2920eca6f534SVegard Nossum goto out_data; 29211da177e4SLinus Torvalds 2922b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 2923eca6f534SVegard Nossum 2924b40ef869SAl Viro kfree(options); 2925eca6f534SVegard Nossum out_data: 2926eca6f534SVegard Nossum kfree(kernel_dev); 2927eca6f534SVegard Nossum out_dev: 2928eca6f534SVegard Nossum kfree(kernel_type); 2929eca6f534SVegard Nossum out_type: 2930eca6f534SVegard Nossum return ret; 29311da177e4SLinus Torvalds } 29321da177e4SLinus Torvalds 29331da177e4SLinus Torvalds /* 2934afac7cbaSAl Viro * Return true if path is reachable from root 2935afac7cbaSAl Viro * 293648a066e7SAl Viro * namespace_sem or mount_lock is held 2937afac7cbaSAl Viro */ 2938643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2939afac7cbaSAl Viro const struct path *root) 2940afac7cbaSAl Viro { 2941643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2942a73324daSAl Viro dentry = mnt->mnt_mountpoint; 29430714a533SAl Viro mnt = mnt->mnt_parent; 2944afac7cbaSAl Viro } 2945643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2946afac7cbaSAl Viro } 2947afac7cbaSAl Viro 294825ab4c9bSYaowei Bai bool path_is_under(struct path *path1, struct path *path2) 2949afac7cbaSAl Viro { 295025ab4c9bSYaowei Bai bool res; 295148a066e7SAl Viro read_seqlock_excl(&mount_lock); 2952643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 295348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 2954afac7cbaSAl Viro return res; 2955afac7cbaSAl Viro } 2956afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 2957afac7cbaSAl Viro 2958afac7cbaSAl Viro /* 29591da177e4SLinus Torvalds * pivot_root Semantics: 29601da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 29611da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 29621da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 29631da177e4SLinus Torvalds * 29641da177e4SLinus Torvalds * Restrictions: 29651da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 29661da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 29671da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 29681da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 29691da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 29701da177e4SLinus Torvalds * 29714a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 29724a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 29734a0d11faSNeil Brown * in this situation. 29744a0d11faSNeil Brown * 29751da177e4SLinus Torvalds * Notes: 29761da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 29771da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 29781da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 29791da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 29801da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 29811da177e4SLinus Torvalds * first. 29821da177e4SLinus Torvalds */ 29833480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 29843480b257SHeiko Carstens const char __user *, put_old) 29851da177e4SLinus Torvalds { 29862d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 298784d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 298884d17192SAl Viro struct mountpoint *old_mp, *root_mp; 29891da177e4SLinus Torvalds int error; 29901da177e4SLinus Torvalds 29919b40bc90SAl Viro if (!may_mount()) 29921da177e4SLinus Torvalds return -EPERM; 29931da177e4SLinus Torvalds 29942d8f3038SAl Viro error = user_path_dir(new_root, &new); 29951da177e4SLinus Torvalds if (error) 29961da177e4SLinus Torvalds goto out0; 29971da177e4SLinus Torvalds 29982d8f3038SAl Viro error = user_path_dir(put_old, &old); 29991da177e4SLinus Torvalds if (error) 30001da177e4SLinus Torvalds goto out1; 30011da177e4SLinus Torvalds 30022d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3003b12cea91SAl Viro if (error) 3004b12cea91SAl Viro goto out2; 30051da177e4SLinus Torvalds 3006f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 300784d17192SAl Viro old_mp = lock_mount(&old); 300884d17192SAl Viro error = PTR_ERR(old_mp); 300984d17192SAl Viro if (IS_ERR(old_mp)) 3010b12cea91SAl Viro goto out3; 3011b12cea91SAl Viro 30121da177e4SLinus Torvalds error = -EINVAL; 3013419148daSAl Viro new_mnt = real_mount(new.mnt); 3014419148daSAl Viro root_mnt = real_mount(root.mnt); 301584d17192SAl Viro old_mnt = real_mount(old.mnt); 301684d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3017fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3018fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3019b12cea91SAl Viro goto out4; 3020143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3021b12cea91SAl Viro goto out4; 30225ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 30235ff9d8a6SEric W. Biederman goto out4; 30241da177e4SLinus Torvalds error = -ENOENT; 3025f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3026b12cea91SAl Viro goto out4; 30271da177e4SLinus Torvalds error = -EBUSY; 302884d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3029b12cea91SAl Viro goto out4; /* loop, on the same file system */ 30301da177e4SLinus Torvalds error = -EINVAL; 30318c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3032b12cea91SAl Viro goto out4; /* not a mountpoint */ 3033676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3034b12cea91SAl Viro goto out4; /* not attached */ 303584d17192SAl Viro root_mp = root_mnt->mnt_mp; 30362d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3037b12cea91SAl Viro goto out4; /* not a mountpoint */ 3038676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3039b12cea91SAl Viro goto out4; /* not attached */ 30404ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 304184d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3042b12cea91SAl Viro goto out4; 30430d082601SEric W. Biederman /* make certain new is below the root */ 30440d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 30450d082601SEric W. Biederman goto out4; 304684d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3047719ea2fbSAl Viro lock_mount_hash(); 3048419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3049419148daSAl Viro detach_mnt(root_mnt, &root_parent); 30505ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 30515ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 30525ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 30535ff9d8a6SEric W. Biederman } 30544ac91378SJan Blunck /* mount old root on put_old */ 305584d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 30564ac91378SJan Blunck /* mount new_root on / */ 305784d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 30586b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 30594fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 30604fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 3061719ea2fbSAl Viro unlock_mount_hash(); 30622d8f3038SAl Viro chroot_fs_refs(&root, &new); 306384d17192SAl Viro put_mountpoint(root_mp); 30641da177e4SLinus Torvalds error = 0; 3065b12cea91SAl Viro out4: 306684d17192SAl Viro unlock_mount(old_mp); 3067b12cea91SAl Viro if (!error) { 30681a390689SAl Viro path_put(&root_parent); 30691a390689SAl Viro path_put(&parent_path); 3070b12cea91SAl Viro } 3071b12cea91SAl Viro out3: 30728c3ee42eSAl Viro path_put(&root); 3073b12cea91SAl Viro out2: 30742d8f3038SAl Viro path_put(&old); 30751da177e4SLinus Torvalds out1: 30762d8f3038SAl Viro path_put(&new); 30771da177e4SLinus Torvalds out0: 30781da177e4SLinus Torvalds return error; 30791da177e4SLinus Torvalds } 30801da177e4SLinus Torvalds 30811da177e4SLinus Torvalds static void __init init_mount_tree(void) 30821da177e4SLinus Torvalds { 30831da177e4SLinus Torvalds struct vfsmount *mnt; 30846b3286edSKirill Korotaev struct mnt_namespace *ns; 3085ac748a09SJan Blunck struct path root; 30860c55cfc4SEric W. Biederman struct file_system_type *type; 30871da177e4SLinus Torvalds 30880c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 30890c55cfc4SEric W. Biederman if (!type) 30900c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 30910c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 30920c55cfc4SEric W. Biederman put_filesystem(type); 30931da177e4SLinus Torvalds if (IS_ERR(mnt)) 30941da177e4SLinus Torvalds panic("Can't create rootfs"); 3095b3e19d92SNick Piggin 30963b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 30973b22edc5STrond Myklebust if (IS_ERR(ns)) 30981da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 30991da177e4SLinus Torvalds 31006b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 31016b3286edSKirill Korotaev get_mnt_ns(ns); 31021da177e4SLinus Torvalds 3103be08d6d2SAl Viro root.mnt = mnt; 3104be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3105da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3106ac748a09SJan Blunck 3107ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3108ac748a09SJan Blunck set_fs_root(current->fs, &root); 31091da177e4SLinus Torvalds } 31101da177e4SLinus Torvalds 311174bf17cfSDenis Cheng void __init mnt_init(void) 31121da177e4SLinus Torvalds { 311313f14b4dSEric Dumazet unsigned u; 311415a67dd8SRandy Dunlap int err; 31151da177e4SLinus Torvalds 31167d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 311720c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 31181da177e4SLinus Torvalds 31190818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 312038129a13SAl Viro sizeof(struct hlist_head), 31210818bf27SAl Viro mhash_entries, 19, 31220818bf27SAl Viro 0, 31230818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 31240818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 31250818bf27SAl Viro sizeof(struct hlist_head), 31260818bf27SAl Viro mphash_entries, 19, 31270818bf27SAl Viro 0, 31280818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 31291da177e4SLinus Torvalds 313084d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 31311da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 31321da177e4SLinus Torvalds 31330818bf27SAl Viro for (u = 0; u <= m_hash_mask; u++) 313438129a13SAl Viro INIT_HLIST_HEAD(&mount_hashtable[u]); 31350818bf27SAl Viro for (u = 0; u <= mp_hash_mask; u++) 31360818bf27SAl Viro INIT_HLIST_HEAD(&mountpoint_hashtable[u]); 31371da177e4SLinus Torvalds 31384b93dc9bSTejun Heo kernfs_init(); 31394b93dc9bSTejun Heo 314015a67dd8SRandy Dunlap err = sysfs_init(); 314115a67dd8SRandy Dunlap if (err) 314215a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 31438e24eea7SHarvey Harrison __func__, err); 314400d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 314500d26666SGreg Kroah-Hartman if (!fs_kobj) 31468e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 31471da177e4SLinus Torvalds init_rootfs(); 31481da177e4SLinus Torvalds init_mount_tree(); 31491da177e4SLinus Torvalds } 31501da177e4SLinus Torvalds 3151616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 31521da177e4SLinus Torvalds { 3153d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3154616511d0STrond Myklebust return; 31557b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3156771b1371SEric W. Biederman free_mnt_ns(ns); 31571da177e4SLinus Torvalds } 31589d412a43SAl Viro 31599d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 31609d412a43SAl Viro { 3161423e0ab0STim Chen struct vfsmount *mnt; 3162423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 3163423e0ab0STim Chen if (!IS_ERR(mnt)) { 3164423e0ab0STim Chen /* 3165423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3166423e0ab0STim Chen * we unmount before file sys is unregistered 3167423e0ab0STim Chen */ 3168f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3169423e0ab0STim Chen } 3170423e0ab0STim Chen return mnt; 31719d412a43SAl Viro } 31729d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3173423e0ab0STim Chen 3174423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3175423e0ab0STim Chen { 3176423e0ab0STim Chen /* release long term mount so mount point can be released */ 3177423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3178f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 317948a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3180423e0ab0STim Chen mntput(mnt); 3181423e0ab0STim Chen } 3182423e0ab0STim Chen } 3183423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 318402125a82SAl Viro 318502125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 318602125a82SAl Viro { 3187143c8c91SAl Viro return check_mnt(real_mount(mnt)); 318802125a82SAl Viro } 31898823c079SEric W. Biederman 31903151527eSEric W. Biederman bool current_chrooted(void) 31913151527eSEric W. Biederman { 31923151527eSEric W. Biederman /* Does the current process have a non-standard root */ 31933151527eSEric W. Biederman struct path ns_root; 31943151527eSEric W. Biederman struct path fs_root; 31953151527eSEric W. Biederman bool chrooted; 31963151527eSEric W. Biederman 31973151527eSEric W. Biederman /* Find the namespace root */ 31983151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 31993151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 32003151527eSEric W. Biederman path_get(&ns_root); 32013151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 32023151527eSEric W. Biederman ; 32033151527eSEric W. Biederman 32043151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 32053151527eSEric W. Biederman 32063151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 32073151527eSEric W. Biederman 32083151527eSEric W. Biederman path_put(&fs_root); 32093151527eSEric W. Biederman path_put(&ns_root); 32103151527eSEric W. Biederman 32113151527eSEric W. Biederman return chrooted; 32123151527eSEric W. Biederman } 32133151527eSEric W. Biederman 32148654df4eSEric W. Biederman static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, 32158654df4eSEric W. Biederman int *new_mnt_flags) 321687a8ebd6SEric W. Biederman { 32178c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 321887a8ebd6SEric W. Biederman struct mount *mnt; 3219e51db735SEric W. Biederman bool visible = false; 322087a8ebd6SEric W. Biederman 322144bb4385SAl Viro down_read(&namespace_sem); 322287a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3223e51db735SEric W. Biederman struct mount *child; 322477b1a97dSEric W. Biederman int mnt_flags; 322577b1a97dSEric W. Biederman 32268654df4eSEric W. Biederman if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) 3227e51db735SEric W. Biederman continue; 3228e51db735SEric W. Biederman 32297e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 32307e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 32317e96c1b0SEric W. Biederman */ 32327e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 32337e96c1b0SEric W. Biederman continue; 32347e96c1b0SEric W. Biederman 3235a1935c17SEric W. Biederman /* A local view of the mount flags */ 323677b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 323777b1a97dSEric W. Biederman 3238695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3239695e9df0SEric W. Biederman if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY) 3240695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3241695e9df0SEric W. Biederman 32428c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 32438c6cf9ccSEric W. Biederman * than the proposed new mount. 32448c6cf9ccSEric W. Biederman */ 324577b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 32468c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 32478c6cf9ccSEric W. Biederman continue; 324877b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 324977b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 32508c6cf9ccSEric W. Biederman continue; 32518c6cf9ccSEric W. Biederman 3252ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3253ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3254ceeb0e5dSEric W. Biederman * empty directories. 3255e51db735SEric W. Biederman */ 3256e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3257e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3258ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3259d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3260ceeb0e5dSEric W. Biederman continue; 32617236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 32627236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3263e51db735SEric W. Biederman goto next; 326487a8ebd6SEric W. Biederman } 32658c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 326677b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 32678c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3268e51db735SEric W. Biederman visible = true; 3269e51db735SEric W. Biederman goto found; 3270e51db735SEric W. Biederman next: ; 327187a8ebd6SEric W. Biederman } 3272e51db735SEric W. Biederman found: 327344bb4385SAl Viro up_read(&namespace_sem); 3274e51db735SEric W. Biederman return visible; 327587a8ebd6SEric W. Biederman } 327687a8ebd6SEric W. Biederman 32778654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) 32788654df4eSEric W. Biederman { 3279a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 32808654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 32818654df4eSEric W. Biederman unsigned long s_iflags; 32828654df4eSEric W. Biederman 32838654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 32848654df4eSEric W. Biederman return false; 32858654df4eSEric W. Biederman 32868654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 32878654df4eSEric W. Biederman s_iflags = mnt->mnt_sb->s_iflags; 32888654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 32898654df4eSEric W. Biederman return false; 32908654df4eSEric W. Biederman 3291a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3292a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3293a1935c17SEric W. Biederman required_iflags); 3294a1935c17SEric W. Biederman return true; 3295a1935c17SEric W. Biederman } 3296a1935c17SEric W. Biederman 32978654df4eSEric W. Biederman return !mnt_already_visible(ns, mnt, new_mnt_flags); 32988654df4eSEric W. Biederman } 32998654df4eSEric W. Biederman 330064964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 33018823c079SEric W. Biederman { 330258be2825SAl Viro struct ns_common *ns = NULL; 33038823c079SEric W. Biederman struct nsproxy *nsproxy; 33048823c079SEric W. Biederman 3305728dba3aSEric W. Biederman task_lock(task); 3306728dba3aSEric W. Biederman nsproxy = task->nsproxy; 33078823c079SEric W. Biederman if (nsproxy) { 330858be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 330958be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 33108823c079SEric W. Biederman } 3311728dba3aSEric W. Biederman task_unlock(task); 33128823c079SEric W. Biederman 33138823c079SEric W. Biederman return ns; 33148823c079SEric W. Biederman } 33158823c079SEric W. Biederman 331664964528SAl Viro static void mntns_put(struct ns_common *ns) 33178823c079SEric W. Biederman { 331858be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 33198823c079SEric W. Biederman } 33208823c079SEric W. Biederman 332164964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 33228823c079SEric W. Biederman { 33238823c079SEric W. Biederman struct fs_struct *fs = current->fs; 332458be2825SAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns); 33258823c079SEric W. Biederman struct path root; 33268823c079SEric W. Biederman 33270c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3328c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3329c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3330ae11e0f1SZhao Hongjiang return -EPERM; 33318823c079SEric W. Biederman 33328823c079SEric W. Biederman if (fs->users != 1) 33338823c079SEric W. Biederman return -EINVAL; 33348823c079SEric W. Biederman 33358823c079SEric W. Biederman get_mnt_ns(mnt_ns); 33368823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 33378823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 33388823c079SEric W. Biederman 33398823c079SEric W. Biederman /* Find the root */ 33408823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 33418823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 33428823c079SEric W. Biederman path_get(&root); 33438823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 33448823c079SEric W. Biederman ; 33458823c079SEric W. Biederman 33468823c079SEric W. Biederman /* Update the pwd and root */ 33478823c079SEric W. Biederman set_fs_pwd(fs, &root); 33488823c079SEric W. Biederman set_fs_root(fs, &root); 33498823c079SEric W. Biederman 33508823c079SEric W. Biederman path_put(&root); 33518823c079SEric W. Biederman return 0; 33528823c079SEric W. Biederman } 33538823c079SEric W. Biederman 33548823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 33558823c079SEric W. Biederman .name = "mnt", 33568823c079SEric W. Biederman .type = CLONE_NEWNS, 33578823c079SEric W. Biederman .get = mntns_get, 33588823c079SEric W. Biederman .put = mntns_put, 33598823c079SEric W. Biederman .install = mntns_install, 33608823c079SEric W. Biederman }; 3361