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 */ 59348a066e7SAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 59448a066e7SAl Viro { 59548a066e7SAl Viro struct mount *mnt; 59648a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 59748a066e7SAl Viro return false; 59848a066e7SAl Viro if (bastard == NULL) 59948a066e7SAl Viro return true; 60048a066e7SAl Viro mnt = real_mount(bastard); 60148a066e7SAl Viro mnt_add_count(mnt, 1); 60248a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 60348a066e7SAl Viro return true; 60448a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 60548a066e7SAl Viro mnt_add_count(mnt, -1); 60648a066e7SAl Viro return false; 60748a066e7SAl Viro } 60848a066e7SAl Viro rcu_read_unlock(); 60948a066e7SAl Viro mntput(bastard); 61048a066e7SAl Viro rcu_read_lock(); 61148a066e7SAl Viro return false; 61248a066e7SAl Viro } 61348a066e7SAl Viro 6141da177e4SLinus Torvalds /* 615474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 61648a066e7SAl Viro * call under rcu_read_lock() 6171da177e4SLinus Torvalds */ 618474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6191da177e4SLinus Torvalds { 62038129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 621474279dcSAl Viro struct mount *p; 6221da177e4SLinus Torvalds 62338129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 624474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 625474279dcSAl Viro return p; 626474279dcSAl Viro return NULL; 6271da177e4SLinus Torvalds } 628474279dcSAl Viro 629474279dcSAl Viro /* 630474279dcSAl Viro * find the last mount at @dentry on vfsmount @mnt. 63148a066e7SAl Viro * mount_lock must be held. 632474279dcSAl Viro */ 633474279dcSAl Viro struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) 634474279dcSAl Viro { 635411a938bSEric W. Biederman struct mount *p, *res = NULL; 636411a938bSEric W. Biederman p = __lookup_mnt(mnt, dentry); 63738129a13SAl Viro if (!p) 63838129a13SAl Viro goto out; 639411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 640411a938bSEric W. Biederman res = p; 64138129a13SAl Viro hlist_for_each_entry_continue(p, mnt_hash) { 6421d6a32acSAl Viro if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) 6431d6a32acSAl Viro break; 644411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 6451d6a32acSAl Viro res = p; 6461d6a32acSAl Viro } 64738129a13SAl Viro out: 6481d6a32acSAl Viro return res; 6491da177e4SLinus Torvalds } 6501da177e4SLinus Torvalds 651a05964f3SRam Pai /* 652f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 653f015f126SDavid Howells * 654f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 655f015f126SDavid Howells * following mounts: 656f015f126SDavid Howells * 657f015f126SDavid Howells * mount /dev/sda1 /mnt 658f015f126SDavid Howells * mount /dev/sda2 /mnt 659f015f126SDavid Howells * mount /dev/sda3 /mnt 660f015f126SDavid Howells * 661f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 662f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 663f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 664f015f126SDavid Howells * 665f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 666a05964f3SRam Pai */ 6671c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 668a05964f3SRam Pai { 669c7105365SAl Viro struct mount *child_mnt; 67048a066e7SAl Viro struct vfsmount *m; 67148a066e7SAl Viro unsigned seq; 67299b7db7bSNick Piggin 67348a066e7SAl Viro rcu_read_lock(); 67448a066e7SAl Viro do { 67548a066e7SAl Viro seq = read_seqbegin(&mount_lock); 676474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 67748a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 67848a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 67948a066e7SAl Viro rcu_read_unlock(); 68048a066e7SAl Viro return m; 681a05964f3SRam Pai } 682a05964f3SRam Pai 6837af1364fSEric W. Biederman /* 6847af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6857af1364fSEric W. Biederman * current mount namespace. 6867af1364fSEric W. Biederman * 6877af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6887af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6897af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6907af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6917af1364fSEric W. Biederman * is a mountpoint. 6927af1364fSEric W. Biederman * 6937af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6947af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6957af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6967af1364fSEric W. Biederman * parent mount. 6977af1364fSEric W. Biederman */ 6987af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6997af1364fSEric W. Biederman { 7007af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7017af1364fSEric W. Biederman struct mount *mnt; 7027af1364fSEric W. Biederman bool is_covered = false; 7037af1364fSEric W. Biederman 7047af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 7057af1364fSEric W. Biederman goto out; 7067af1364fSEric W. Biederman 7077af1364fSEric W. Biederman down_read(&namespace_sem); 7087af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7097af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7107af1364fSEric W. Biederman if (is_covered) 7117af1364fSEric W. Biederman break; 7127af1364fSEric W. Biederman } 7137af1364fSEric W. Biederman up_read(&namespace_sem); 7147af1364fSEric W. Biederman out: 7157af1364fSEric W. Biederman return is_covered; 7167af1364fSEric W. Biederman } 7177af1364fSEric W. Biederman 718e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 71984d17192SAl Viro { 7200818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 72184d17192SAl Viro struct mountpoint *mp; 72284d17192SAl Viro 7230818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 72484d17192SAl Viro if (mp->m_dentry == dentry) { 72584d17192SAl Viro /* might be worth a WARN_ON() */ 72684d17192SAl Viro if (d_unlinked(dentry)) 72784d17192SAl Viro return ERR_PTR(-ENOENT); 72884d17192SAl Viro mp->m_count++; 72984d17192SAl Viro return mp; 73084d17192SAl Viro } 73184d17192SAl Viro } 732e2dfa935SEric W. Biederman return NULL; 733e2dfa935SEric W. Biederman } 734e2dfa935SEric W. Biederman 735e2dfa935SEric W. Biederman static struct mountpoint *new_mountpoint(struct dentry *dentry) 736e2dfa935SEric W. Biederman { 737e2dfa935SEric W. Biederman struct hlist_head *chain = mp_hash(dentry); 738e2dfa935SEric W. Biederman struct mountpoint *mp; 739e2dfa935SEric W. Biederman int ret; 74084d17192SAl Viro 74184d17192SAl Viro mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 74284d17192SAl Viro if (!mp) 74384d17192SAl Viro return ERR_PTR(-ENOMEM); 74484d17192SAl Viro 745eed81007SMiklos Szeredi ret = d_set_mounted(dentry); 746eed81007SMiklos Szeredi if (ret) { 74784d17192SAl Viro kfree(mp); 748eed81007SMiklos Szeredi return ERR_PTR(ret); 74984d17192SAl Viro } 750eed81007SMiklos Szeredi 75184d17192SAl Viro mp->m_dentry = dentry; 75284d17192SAl Viro mp->m_count = 1; 7530818bf27SAl Viro hlist_add_head(&mp->m_hash, chain); 7540a5eb7c8SEric W. Biederman INIT_HLIST_HEAD(&mp->m_list); 75584d17192SAl Viro return mp; 75684d17192SAl Viro } 75784d17192SAl Viro 75884d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 75984d17192SAl Viro { 76084d17192SAl Viro if (!--mp->m_count) { 76184d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7620a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 76384d17192SAl Viro spin_lock(&dentry->d_lock); 76484d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 76584d17192SAl Viro spin_unlock(&dentry->d_lock); 7660818bf27SAl Viro hlist_del(&mp->m_hash); 76784d17192SAl Viro kfree(mp); 76884d17192SAl Viro } 76984d17192SAl Viro } 77084d17192SAl Viro 771143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7721da177e4SLinus Torvalds { 7736b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7741da177e4SLinus Torvalds } 7751da177e4SLinus Torvalds 77699b7db7bSNick Piggin /* 77799b7db7bSNick Piggin * vfsmount lock must be held for write 77899b7db7bSNick Piggin */ 7796b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7805addc5ddSAl Viro { 7815addc5ddSAl Viro if (ns) { 7825addc5ddSAl Viro ns->event = ++event; 7835addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7845addc5ddSAl Viro } 7855addc5ddSAl Viro } 7865addc5ddSAl Viro 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 && ns->event != event) { 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 */ 8017bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 8021da177e4SLinus Torvalds { 8030714a533SAl Viro mnt->mnt_parent = mnt; 804a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8056b41d536SAl Viro list_del_init(&mnt->mnt_child); 80638129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8070a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 80884d17192SAl Viro put_mountpoint(mnt->mnt_mp); 80984d17192SAl Viro mnt->mnt_mp = NULL; 8101da177e4SLinus Torvalds } 8111da177e4SLinus Torvalds 81299b7db7bSNick Piggin /* 81399b7db7bSNick Piggin * vfsmount lock must be held for write 81499b7db7bSNick Piggin */ 8157bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8167bdb11deSEric W. Biederman { 8177bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8187bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8197bdb11deSEric W. Biederman unhash_mnt(mnt); 8207bdb11deSEric W. Biederman } 8217bdb11deSEric W. Biederman 8227bdb11deSEric W. Biederman /* 8237bdb11deSEric W. Biederman * vfsmount lock must be held for write 8247bdb11deSEric W. Biederman */ 8256a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8266a46c573SEric W. Biederman { 8276a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8286a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8296a46c573SEric W. Biederman unhash_mnt(mnt); 8306a46c573SEric W. Biederman } 8316a46c573SEric W. Biederman 8326a46c573SEric W. Biederman /* 8336a46c573SEric W. Biederman * vfsmount lock must be held for write 8346a46c573SEric W. Biederman */ 83584d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 83684d17192SAl Viro struct mountpoint *mp, 83744d964d6SAl Viro struct mount *child_mnt) 838b90fa9aeSRam Pai { 83984d17192SAl Viro mp->m_count++; 8403a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 84184d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8423a2393d7SAl Viro child_mnt->mnt_parent = mnt; 84384d17192SAl Viro child_mnt->mnt_mp = mp; 8440a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 845b90fa9aeSRam Pai } 846b90fa9aeSRam Pai 84799b7db7bSNick Piggin /* 84899b7db7bSNick Piggin * vfsmount lock must be held for write 84999b7db7bSNick Piggin */ 85084d17192SAl Viro static void attach_mnt(struct mount *mnt, 85184d17192SAl Viro struct mount *parent, 85284d17192SAl Viro struct mountpoint *mp) 8531da177e4SLinus Torvalds { 85484d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 85538129a13SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); 85684d17192SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 857b90fa9aeSRam Pai } 858b90fa9aeSRam Pai 85912a5b529SAl Viro static void attach_shadowed(struct mount *mnt, 86012a5b529SAl Viro struct mount *parent, 86112a5b529SAl Viro struct mount *shadows) 86212a5b529SAl Viro { 86312a5b529SAl Viro if (shadows) { 864f6f99332SLinus Torvalds hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); 86512a5b529SAl Viro list_add(&mnt->mnt_child, &shadows->mnt_child); 86612a5b529SAl Viro } else { 86712a5b529SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, 86812a5b529SAl Viro m_hash(&parent->mnt, mnt->mnt_mountpoint)); 86912a5b529SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 87012a5b529SAl Viro } 87112a5b529SAl Viro } 87212a5b529SAl Viro 873b90fa9aeSRam Pai /* 87499b7db7bSNick Piggin * vfsmount lock must be held for write 875b90fa9aeSRam Pai */ 8761d6a32acSAl Viro static void commit_tree(struct mount *mnt, struct mount *shadows) 877b90fa9aeSRam Pai { 8780714a533SAl Viro struct mount *parent = mnt->mnt_parent; 87983adc753SAl Viro struct mount *m; 880b90fa9aeSRam Pai LIST_HEAD(head); 881143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 882b90fa9aeSRam Pai 8830714a533SAl Viro BUG_ON(parent == mnt); 884b90fa9aeSRam Pai 8851a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 886f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 887143c8c91SAl Viro m->mnt_ns = n; 888f03c6599SAl Viro 889b90fa9aeSRam Pai list_splice(&head, n->list.prev); 890b90fa9aeSRam Pai 89112a5b529SAl Viro attach_shadowed(mnt, parent, shadows); 8926b3286edSKirill Korotaev touch_mnt_namespace(n); 8931da177e4SLinus Torvalds } 8941da177e4SLinus Torvalds 895909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 8961da177e4SLinus Torvalds { 8976b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 8986b41d536SAl Viro if (next == &p->mnt_mounts) { 8991da177e4SLinus Torvalds while (1) { 900909b0a88SAl Viro if (p == root) 9011da177e4SLinus Torvalds return NULL; 9026b41d536SAl Viro next = p->mnt_child.next; 9036b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9041da177e4SLinus Torvalds break; 9050714a533SAl Viro p = p->mnt_parent; 9061da177e4SLinus Torvalds } 9071da177e4SLinus Torvalds } 9086b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9091da177e4SLinus Torvalds } 9101da177e4SLinus Torvalds 911315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9129676f0c6SRam Pai { 9136b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9146b41d536SAl Viro while (prev != &p->mnt_mounts) { 9156b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9166b41d536SAl Viro prev = p->mnt_mounts.prev; 9179676f0c6SRam Pai } 9189676f0c6SRam Pai return p; 9199676f0c6SRam Pai } 9209676f0c6SRam Pai 9219d412a43SAl Viro struct vfsmount * 9229d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 9239d412a43SAl Viro { 924b105e270SAl Viro struct mount *mnt; 9259d412a43SAl Viro struct dentry *root; 9269d412a43SAl Viro 9279d412a43SAl Viro if (!type) 9289d412a43SAl Viro return ERR_PTR(-ENODEV); 9299d412a43SAl Viro 9309d412a43SAl Viro mnt = alloc_vfsmnt(name); 9319d412a43SAl Viro if (!mnt) 9329d412a43SAl Viro return ERR_PTR(-ENOMEM); 9339d412a43SAl Viro 9349d412a43SAl Viro if (flags & MS_KERNMOUNT) 935b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9369d412a43SAl Viro 9379d412a43SAl Viro root = mount_fs(type, flags, name, data); 9389d412a43SAl Viro if (IS_ERR(root)) { 9398ffcb32eSDavid Howells mnt_free_id(mnt); 9409d412a43SAl Viro free_vfsmnt(mnt); 9419d412a43SAl Viro return ERR_CAST(root); 9429d412a43SAl Viro } 9439d412a43SAl Viro 944b105e270SAl Viro mnt->mnt.mnt_root = root; 945b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 946a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9470714a533SAl Viro mnt->mnt_parent = mnt; 948719ea2fbSAl Viro lock_mount_hash(); 94939f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 950719ea2fbSAl Viro unlock_mount_hash(); 951b105e270SAl Viro return &mnt->mnt; 9529d412a43SAl Viro } 9539d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 9549d412a43SAl Viro 95587129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 95636341f64SRam Pai int flag) 9571da177e4SLinus Torvalds { 95887129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 959be34d1a3SDavid Howells struct mount *mnt; 960be34d1a3SDavid Howells int err; 9611da177e4SLinus Torvalds 962be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 963be34d1a3SDavid Howells if (!mnt) 964be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 965be34d1a3SDavid Howells 9667a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 96715169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 968719f5d7fSMiklos Szeredi else 96915169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 970719f5d7fSMiklos Szeredi 97115169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 972be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 973719f5d7fSMiklos Szeredi if (err) 974719f5d7fSMiklos Szeredi goto out_free; 975719f5d7fSMiklos Szeredi } 976719f5d7fSMiklos Szeredi 977f2ebb3a9SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); 978132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 9799566d674SEric W. Biederman if (flag & CL_UNPRIVILEGED) { 9809566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; 9819566d674SEric W. Biederman 9829566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_READONLY) 983132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 984132c94e3SEric W. Biederman 9859566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NODEV) 9869566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; 9879566d674SEric W. Biederman 9889566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOSUID) 9899566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; 9909566d674SEric W. Biederman 9919566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOEXEC) 9929566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; 9939566d674SEric W. Biederman } 9949566d674SEric W. Biederman 9955ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 996381cacb1SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && 997381cacb1SEric W. Biederman (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) 9985ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 9995ff9d8a6SEric W. Biederman 10001da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1001b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1002b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1003a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10040714a533SAl Viro mnt->mnt_parent = mnt; 1005719ea2fbSAl Viro lock_mount_hash(); 100639f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1007719ea2fbSAl Viro unlock_mount_hash(); 1008b90fa9aeSRam Pai 10097a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10107a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10116776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 101232301920SAl Viro mnt->mnt_master = old; 1013fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10148aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1015fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10166776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1017d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10186776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1019d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10205afe0022SRam Pai } 1021b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10220f0afb1dSAl Viro set_mnt_shared(mnt); 10231da177e4SLinus Torvalds 10241da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10251da177e4SLinus Torvalds * as the original if that was on one */ 102636341f64SRam Pai if (flag & CL_EXPIRE) { 10276776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10286776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10291da177e4SLinus Torvalds } 1030be34d1a3SDavid Howells 1031cb338d06SAl Viro return mnt; 1032719f5d7fSMiklos Szeredi 1033719f5d7fSMiklos Szeredi out_free: 10348ffcb32eSDavid Howells mnt_free_id(mnt); 1035719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1036be34d1a3SDavid Howells return ERR_PTR(err); 10371da177e4SLinus Torvalds } 10381da177e4SLinus Torvalds 10399ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10409ea459e1SAl Viro { 10419ea459e1SAl Viro /* 10429ea459e1SAl Viro * This probably indicates that somebody messed 10439ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 10449ea459e1SAl Viro * happens, the filesystem was probably unable 10459ea459e1SAl Viro * to make r/w->r/o transitions. 10469ea459e1SAl Viro */ 10479ea459e1SAl Viro /* 10489ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 10499ea459e1SAl Viro * so mnt_get_writers() below is safe. 10509ea459e1SAl Viro */ 10519ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 10529ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 10539ea459e1SAl Viro mnt_pin_kill(mnt); 10549ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 10559ea459e1SAl Viro dput(mnt->mnt.mnt_root); 10569ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 10579ea459e1SAl Viro mnt_free_id(mnt); 10589ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 10599ea459e1SAl Viro } 10609ea459e1SAl Viro 10619ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 10629ea459e1SAl Viro { 10639ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 10649ea459e1SAl Viro } 10659ea459e1SAl Viro 10669ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 10679ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 10689ea459e1SAl Viro { 10699ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 10709ea459e1SAl Viro struct llist_node *next; 10719ea459e1SAl Viro 10729ea459e1SAl Viro for (; node; node = next) { 10739ea459e1SAl Viro next = llist_next(node); 10749ea459e1SAl Viro cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); 10759ea459e1SAl Viro } 10769ea459e1SAl Viro } 10779ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 10789ea459e1SAl Viro 1079900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 10807b7b1aceSAl Viro { 108148a066e7SAl Viro rcu_read_lock(); 1082aa9c0e07SAl Viro mnt_add_count(mnt, -1); 108348a066e7SAl Viro if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ 108448a066e7SAl Viro rcu_read_unlock(); 108599b7db7bSNick Piggin return; 1086b3e19d92SNick Piggin } 1087719ea2fbSAl Viro lock_mount_hash(); 1088b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 108948a066e7SAl Viro rcu_read_unlock(); 1090719ea2fbSAl Viro unlock_mount_hash(); 109199b7db7bSNick Piggin return; 109299b7db7bSNick Piggin } 109348a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 109448a066e7SAl Viro rcu_read_unlock(); 109548a066e7SAl Viro unlock_mount_hash(); 109648a066e7SAl Viro return; 109748a066e7SAl Viro } 109848a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 109948a066e7SAl Viro rcu_read_unlock(); 1100962830dfSAndi Kleen 110139f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1102ce07d891SEric W. Biederman 1103ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1104ce07d891SEric W. Biederman struct mount *p, *tmp; 1105ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1106ce07d891SEric W. Biederman umount_mnt(p); 1107ce07d891SEric W. Biederman } 1108ce07d891SEric W. Biederman } 1109719ea2fbSAl Viro unlock_mount_hash(); 1110649a795aSAl Viro 11119ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11129ea459e1SAl Viro struct task_struct *task = current; 11139ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11149ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11159ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11169ea459e1SAl Viro return; 11179ea459e1SAl Viro } 11189ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11199ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11209ea459e1SAl Viro return; 11219ea459e1SAl Viro } 11229ea459e1SAl Viro cleanup_mnt(mnt); 1123b3e19d92SNick Piggin } 1124b3e19d92SNick Piggin 1125b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1126b3e19d92SNick Piggin { 1127b3e19d92SNick Piggin if (mnt) { 1128863d684fSAl Viro struct mount *m = real_mount(mnt); 1129b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1130863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1131863d684fSAl Viro m->mnt_expiry_mark = 0; 1132863d684fSAl Viro mntput_no_expire(m); 1133b3e19d92SNick Piggin } 1134b3e19d92SNick Piggin } 1135b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1136b3e19d92SNick Piggin 1137b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1138b3e19d92SNick Piggin { 1139b3e19d92SNick Piggin if (mnt) 114083adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1141b3e19d92SNick Piggin return mnt; 1142b3e19d92SNick Piggin } 1143b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1144b3e19d92SNick Piggin 11453064c356SAl Viro struct vfsmount *mnt_clone_internal(struct path *path) 11467b7b1aceSAl Viro { 11473064c356SAl Viro struct mount *p; 11483064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 11493064c356SAl Viro if (IS_ERR(p)) 11503064c356SAl Viro return ERR_CAST(p); 11513064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 11523064c356SAl Viro return &p->mnt; 11537b7b1aceSAl Viro } 11541da177e4SLinus Torvalds 1155b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 1156b3b304a2SMiklos Szeredi { 1157b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 1158b3b304a2SMiklos Szeredi } 1159b3b304a2SMiklos Szeredi 1160b3b304a2SMiklos Szeredi /* 1161b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 1162b3b304a2SMiklos Szeredi * implement more complex mount option showing. 1163b3b304a2SMiklos Szeredi * 1164b3b304a2SMiklos Szeredi * See also save_mount_options(). 1165b3b304a2SMiklos Szeredi */ 116634c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 1167b3b304a2SMiklos Szeredi { 11682a32cebdSAl Viro const char *options; 11692a32cebdSAl Viro 11702a32cebdSAl Viro rcu_read_lock(); 117134c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 1172b3b304a2SMiklos Szeredi 1173b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 1174b3b304a2SMiklos Szeredi seq_putc(m, ','); 1175b3b304a2SMiklos Szeredi mangle(m, options); 1176b3b304a2SMiklos Szeredi } 11772a32cebdSAl Viro rcu_read_unlock(); 1178b3b304a2SMiklos Szeredi 1179b3b304a2SMiklos Szeredi return 0; 1180b3b304a2SMiklos Szeredi } 1181b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 1182b3b304a2SMiklos Szeredi 1183b3b304a2SMiklos Szeredi /* 1184b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 1185b3b304a2SMiklos Szeredi * called from the fill_super() callback. 1186b3b304a2SMiklos Szeredi * 1187b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 1188b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1189b3b304a2SMiklos Szeredi * remount fails. 1190b3b304a2SMiklos Szeredi * 1191b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1192b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1193b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1194b3b304a2SMiklos Szeredi * any more. 1195b3b304a2SMiklos Szeredi */ 1196b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1197b3b304a2SMiklos Szeredi { 11982a32cebdSAl Viro BUG_ON(sb->s_options); 11992a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1200b3b304a2SMiklos Szeredi } 1201b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1202b3b304a2SMiklos Szeredi 12032a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 12042a32cebdSAl Viro { 12052a32cebdSAl Viro char *old = sb->s_options; 12062a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 12072a32cebdSAl Viro if (old) { 12082a32cebdSAl Viro synchronize_rcu(); 12092a32cebdSAl Viro kfree(old); 12102a32cebdSAl Viro } 12112a32cebdSAl Viro } 12122a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 12132a32cebdSAl Viro 1214a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12150226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12161da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12171da177e4SLinus Torvalds { 12186ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 12191da177e4SLinus Torvalds 1220390c6843SRam Pai down_read(&namespace_sem); 1221c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1222c7999c36SAl Viro void *v = p->cached_mount; 1223c7999c36SAl Viro if (*pos == p->cached_index) 1224c7999c36SAl Viro return v; 1225c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1226c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1227c7999c36SAl Viro return p->cached_mount = v; 1228c7999c36SAl Viro } 1229c7999c36SAl Viro } 1230c7999c36SAl Viro 1231c7999c36SAl Viro p->cached_event = p->ns->event; 1232c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1233c7999c36SAl Viro p->cached_index = *pos; 1234c7999c36SAl Viro return p->cached_mount; 12351da177e4SLinus Torvalds } 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12381da177e4SLinus Torvalds { 12396ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 1240b0765fb8SPavel Emelianov 1241c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1242c7999c36SAl Viro p->cached_index = *pos; 1243c7999c36SAl Viro return p->cached_mount; 12441da177e4SLinus Torvalds } 12451da177e4SLinus Torvalds 12461da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12471da177e4SLinus Torvalds { 1248390c6843SRam Pai up_read(&namespace_sem); 12491da177e4SLinus Torvalds } 12501da177e4SLinus Torvalds 12510226f492SAl Viro static int m_show(struct seq_file *m, void *v) 12529f5596afSAl Viro { 12536ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 12541a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 12550226f492SAl Viro return p->show(m, &r->mnt); 12561da177e4SLinus Torvalds } 12571da177e4SLinus Torvalds 1258a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 12591da177e4SLinus Torvalds .start = m_start, 12601da177e4SLinus Torvalds .next = m_next, 12611da177e4SLinus Torvalds .stop = m_stop, 12620226f492SAl Viro .show = m_show, 1263b4629fe2SChuck Lever }; 1264a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1265b4629fe2SChuck Lever 12661da177e4SLinus Torvalds /** 12671da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 12681da177e4SLinus Torvalds * @mnt: root of mount tree 12691da177e4SLinus Torvalds * 12701da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 12711da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 12721da177e4SLinus Torvalds * busy. 12731da177e4SLinus Torvalds */ 1274909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 12751da177e4SLinus Torvalds { 1276909b0a88SAl Viro struct mount *mnt = real_mount(m); 127736341f64SRam Pai int actual_refs = 0; 127836341f64SRam Pai int minimum_refs = 0; 1279315fc83eSAl Viro struct mount *p; 1280909b0a88SAl Viro BUG_ON(!m); 12811da177e4SLinus Torvalds 1282b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1283719ea2fbSAl Viro lock_mount_hash(); 1284909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 128583adc753SAl Viro actual_refs += mnt_get_count(p); 12861da177e4SLinus Torvalds minimum_refs += 2; 12871da177e4SLinus Torvalds } 1288719ea2fbSAl Viro unlock_mount_hash(); 12891da177e4SLinus Torvalds 12901da177e4SLinus Torvalds if (actual_refs > minimum_refs) 12911da177e4SLinus Torvalds return 0; 1292e3474a8eSIan Kent 1293e3474a8eSIan Kent return 1; 12941da177e4SLinus Torvalds } 12951da177e4SLinus Torvalds 12961da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds /** 12991da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13001da177e4SLinus Torvalds * @mnt: root of mount 13011da177e4SLinus Torvalds * 13021da177e4SLinus Torvalds * This is called to check if a mount point has any 13031da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13041da177e4SLinus Torvalds * mount has sub mounts this will return busy 13051da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13061da177e4SLinus Torvalds * 13071da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13081da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13091da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13101da177e4SLinus Torvalds */ 13111da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13121da177e4SLinus Torvalds { 1313e3474a8eSIan Kent int ret = 1; 13148ad08d8aSAl Viro down_read(&namespace_sem); 1315719ea2fbSAl Viro lock_mount_hash(); 13161ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1317e3474a8eSIan Kent ret = 0; 1318719ea2fbSAl Viro unlock_mount_hash(); 13198ad08d8aSAl Viro up_read(&namespace_sem); 1320a05964f3SRam Pai return ret; 13211da177e4SLinus Torvalds } 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13241da177e4SLinus Torvalds 132538129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1326e3197d83SAl Viro 132797216be0SAl Viro static void namespace_unlock(void) 13281da177e4SLinus Torvalds { 1329a3b3c562SEric W. Biederman struct hlist_head head; 133097216be0SAl Viro 1331a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1332a3b3c562SEric W. Biederman 133397216be0SAl Viro up_write(&namespace_sem); 1334a3b3c562SEric W. Biederman 1335a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 133697216be0SAl Viro return; 133797216be0SAl Viro 133848a066e7SAl Viro synchronize_rcu(); 133948a066e7SAl Viro 134087b95ce0SAl Viro group_pin_kill(&head); 134170fbcdf4SRam Pai } 134270fbcdf4SRam Pai 134397216be0SAl Viro static inline void namespace_lock(void) 1344e3197d83SAl Viro { 134597216be0SAl Viro down_write(&namespace_sem); 1346e3197d83SAl Viro } 1347e3197d83SAl Viro 1348e819f152SEric W. Biederman enum umount_tree_flags { 1349e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1350e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1351e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1352e819f152SEric W. Biederman }; 135399b7db7bSNick Piggin /* 135448a066e7SAl Viro * mount_lock must be held 135599b7db7bSNick Piggin * namespace_sem must be held for write 135699b7db7bSNick Piggin */ 1357e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 135870fbcdf4SRam Pai { 1359c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1360315fc83eSAl Viro struct mount *p; 136170fbcdf4SRam Pai 13625d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 13635d88457eSEric W. Biederman propagate_mount_unlock(mnt); 13645d88457eSEric W. Biederman 1365c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1366590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1367590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1368c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1369590ce4bcSEric W. Biederman } 1370c003b26fSEric W. Biederman 1371411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1372c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1373c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 137438129a13SAl Viro } 137570fbcdf4SRam Pai 1376c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1377e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 13787b8a53fdSAl Viro propagate_umount(&tmp_list); 1379a05964f3SRam Pai 1380c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1381ce07d891SEric W. Biederman bool disconnect; 1382c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 13836776db3dSAl Viro list_del_init(&p->mnt_expire); 13841a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1385143c8c91SAl Viro __touch_mnt_namespace(p->mnt_ns); 138663d37a84SAl Viro p->mnt_ns = NULL; 1387e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 138848a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 138987b95ce0SAl Viro 1390e0c9c0afSEric W. Biederman disconnect = !(((how & UMOUNT_CONNECTED) && 1391e0c9c0afSEric W. Biederman mnt_has_parent(p) && 1392e0c9c0afSEric W. Biederman (p->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) || 1393e0c9c0afSEric W. Biederman IS_MNT_LOCKED_AND_LAZY(p)); 1394ce07d891SEric W. Biederman 1395ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1396ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1397676da58dSAl Viro if (mnt_has_parent(p)) { 139881b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1399ce07d891SEric W. Biederman if (!disconnect) { 1400ce07d891SEric W. Biederman /* Don't forget about p */ 1401ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1402ce07d891SEric W. Biederman } else { 14036a46c573SEric W. Biederman umount_mnt(p); 14047c4b93d8SAl Viro } 1405ce07d891SEric W. Biederman } 14060f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 140738129a13SAl Viro } 14081da177e4SLinus Torvalds } 14091da177e4SLinus Torvalds 1410b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1411c35038beSAl Viro 14121ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 14131da177e4SLinus Torvalds { 14141ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 14151da177e4SLinus Torvalds int retval; 14161da177e4SLinus Torvalds 14171ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 14181da177e4SLinus Torvalds if (retval) 14191da177e4SLinus Torvalds return retval; 14201da177e4SLinus Torvalds 14211da177e4SLinus Torvalds /* 14221da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 14231da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 14241da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 14251da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 14261da177e4SLinus Torvalds */ 14271da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 14281ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 14291da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 14301da177e4SLinus Torvalds return -EINVAL; 14311da177e4SLinus Torvalds 1432b3e19d92SNick Piggin /* 1433b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1434b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1435b3e19d92SNick Piggin */ 1436719ea2fbSAl Viro lock_mount_hash(); 143783adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1438719ea2fbSAl Viro unlock_mount_hash(); 14391da177e4SLinus Torvalds return -EBUSY; 1440b3e19d92SNick Piggin } 1441719ea2fbSAl Viro unlock_mount_hash(); 14421da177e4SLinus Torvalds 1443863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 14441da177e4SLinus Torvalds return -EAGAIN; 14451da177e4SLinus Torvalds } 14461da177e4SLinus Torvalds 14471da177e4SLinus Torvalds /* 14481da177e4SLinus Torvalds * If we may have to abort operations to get out of this 14491da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 14501da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 14511da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 14521da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 14531da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 14541da177e4SLinus Torvalds * about for the moment. 14551da177e4SLinus Torvalds */ 14561da177e4SLinus Torvalds 145742faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 145842faad99SAl Viro sb->s_op->umount_begin(sb); 145942faad99SAl Viro } 14601da177e4SLinus Torvalds 14611da177e4SLinus Torvalds /* 14621da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 14631da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 14641da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 14651da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 14661da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 14671da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 14681da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 14691da177e4SLinus Torvalds */ 14701ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 14711da177e4SLinus Torvalds /* 14721da177e4SLinus Torvalds * Special case for "unmounting" root ... 14731da177e4SLinus Torvalds * we just try to remount it readonly. 14741da177e4SLinus Torvalds */ 1475a1480dccSAndy Lutomirski if (!capable(CAP_SYS_ADMIN)) 1476a1480dccSAndy Lutomirski return -EPERM; 14771da177e4SLinus Torvalds down_write(&sb->s_umount); 14784aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 14791da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 14801da177e4SLinus Torvalds up_write(&sb->s_umount); 14811da177e4SLinus Torvalds return retval; 14821da177e4SLinus Torvalds } 14831da177e4SLinus Torvalds 148497216be0SAl Viro namespace_lock(); 1485719ea2fbSAl Viro lock_mount_hash(); 14865addc5ddSAl Viro event++; 14871da177e4SLinus Torvalds 148848a066e7SAl Viro if (flags & MNT_DETACH) { 148948a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1490e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 149148a066e7SAl Viro retval = 0; 149248a066e7SAl Viro } else { 1493b54b9be7SAl Viro shrink_submounts(mnt); 14941da177e4SLinus Torvalds retval = -EBUSY; 149548a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 14961a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1497e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 14981da177e4SLinus Torvalds retval = 0; 14991da177e4SLinus Torvalds } 150048a066e7SAl Viro } 1501719ea2fbSAl Viro unlock_mount_hash(); 1502e3197d83SAl Viro namespace_unlock(); 15031da177e4SLinus Torvalds return retval; 15041da177e4SLinus Torvalds } 15051da177e4SLinus Torvalds 15061da177e4SLinus Torvalds /* 150780b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 150880b5dce8SEric W. Biederman * 150980b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 151080b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 151180b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 151280b5dce8SEric W. Biederman * leaking them. 151380b5dce8SEric W. Biederman * 151480b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 151580b5dce8SEric W. Biederman */ 151680b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 151780b5dce8SEric W. Biederman { 151880b5dce8SEric W. Biederman struct mountpoint *mp; 151980b5dce8SEric W. Biederman struct mount *mnt; 152080b5dce8SEric W. Biederman 152180b5dce8SEric W. Biederman namespace_lock(); 152280b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1523f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 152480b5dce8SEric W. Biederman goto out_unlock; 152580b5dce8SEric W. Biederman 152680b5dce8SEric W. Biederman lock_mount_hash(); 152780b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 152880b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1529ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1530ce07d891SEric W. Biederman struct mount *p, *tmp; 1531ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1532ce07d891SEric W. Biederman hlist_add_head(&p->mnt_umount.s_list, &unmounted); 1533ce07d891SEric W. Biederman umount_mnt(p); 1534ce07d891SEric W. Biederman } 1535ce07d891SEric W. Biederman } 1536e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 153780b5dce8SEric W. Biederman } 153880b5dce8SEric W. Biederman unlock_mount_hash(); 153980b5dce8SEric W. Biederman put_mountpoint(mp); 154080b5dce8SEric W. Biederman out_unlock: 154180b5dce8SEric W. Biederman namespace_unlock(); 154280b5dce8SEric W. Biederman } 154380b5dce8SEric W. Biederman 154480b5dce8SEric W. Biederman /* 15459b40bc90SAl Viro * Is the caller allowed to modify his namespace? 15469b40bc90SAl Viro */ 15479b40bc90SAl Viro static inline bool may_mount(void) 15489b40bc90SAl Viro { 15499b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 15509b40bc90SAl Viro } 15519b40bc90SAl Viro 15529b40bc90SAl Viro /* 15531da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 15541da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 15551da177e4SLinus Torvalds * 15561da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 15571da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 15581da177e4SLinus Torvalds */ 15591da177e4SLinus Torvalds 1560bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 15611da177e4SLinus Torvalds { 15622d8f3038SAl Viro struct path path; 1563900148dcSAl Viro struct mount *mnt; 15641da177e4SLinus Torvalds int retval; 1565db1f05bbSMiklos Szeredi int lookup_flags = 0; 15661da177e4SLinus Torvalds 1567db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1568db1f05bbSMiklos Szeredi return -EINVAL; 1569db1f05bbSMiklos Szeredi 15709b40bc90SAl Viro if (!may_mount()) 15719b40bc90SAl Viro return -EPERM; 15729b40bc90SAl Viro 1573db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1574db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1575db1f05bbSMiklos Szeredi 1576197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 15771da177e4SLinus Torvalds if (retval) 15781da177e4SLinus Torvalds goto out; 1579900148dcSAl Viro mnt = real_mount(path.mnt); 15801da177e4SLinus Torvalds retval = -EINVAL; 15812d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 15821da177e4SLinus Torvalds goto dput_and_out; 1583143c8c91SAl Viro if (!check_mnt(mnt)) 15841da177e4SLinus Torvalds goto dput_and_out; 15855ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 15865ff9d8a6SEric W. Biederman goto dput_and_out; 1587b2f5d4dcSEric W. Biederman retval = -EPERM; 1588b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1589b2f5d4dcSEric W. Biederman goto dput_and_out; 15901da177e4SLinus Torvalds 1591900148dcSAl Viro retval = do_umount(mnt, flags); 15921da177e4SLinus Torvalds dput_and_out: 1593429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 15942d8f3038SAl Viro dput(path.dentry); 1595900148dcSAl Viro mntput_no_expire(mnt); 15961da177e4SLinus Torvalds out: 15971da177e4SLinus Torvalds return retval; 15981da177e4SLinus Torvalds } 15991da177e4SLinus Torvalds 16001da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 16011da177e4SLinus Torvalds 16021da177e4SLinus Torvalds /* 16031da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 16041da177e4SLinus Torvalds */ 1605bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 16061da177e4SLinus Torvalds { 16071da177e4SLinus Torvalds return sys_umount(name, 0); 16081da177e4SLinus Torvalds } 16091da177e4SLinus Torvalds 16101da177e4SLinus Torvalds #endif 16111da177e4SLinus Torvalds 16124ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 16138823c079SEric W. Biederman { 16144ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1615e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1616e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 16174ce5d2b1SEric W. Biederman } 16184ce5d2b1SEric W. Biederman 161958be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 162058be2825SAl Viro { 162158be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 162258be2825SAl Viro } 162358be2825SAl Viro 16244ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 16254ce5d2b1SEric W. Biederman { 16264ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 16274ce5d2b1SEric W. Biederman * mount namespace loop? 16284ce5d2b1SEric W. Biederman */ 16294ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 16304ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 16314ce5d2b1SEric W. Biederman return false; 16324ce5d2b1SEric W. Biederman 1633f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 16348823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 16358823c079SEric W. Biederman } 16368823c079SEric W. Biederman 163787129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 163836341f64SRam Pai int flag) 16391da177e4SLinus Torvalds { 164084d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 16411da177e4SLinus Torvalds 16424ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 16434ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 16444ce5d2b1SEric W. Biederman 16454ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1646be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 16479676f0c6SRam Pai 164836341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1649be34d1a3SDavid Howells if (IS_ERR(q)) 1650be34d1a3SDavid Howells return q; 1651be34d1a3SDavid Howells 1652a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 16531da177e4SLinus Torvalds 16541da177e4SLinus Torvalds p = mnt; 16556b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1656315fc83eSAl Viro struct mount *s; 16577ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 16581da177e4SLinus Torvalds continue; 16591da177e4SLinus Torvalds 1660909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 166112a5b529SAl Viro struct mount *t = NULL; 16624ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 16634ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 16644ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 16654ce5d2b1SEric W. Biederman continue; 16664ce5d2b1SEric W. Biederman } 16674ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 16684ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 16699676f0c6SRam Pai s = skip_mnt_tree(s); 16709676f0c6SRam Pai continue; 16719676f0c6SRam Pai } 16720714a533SAl Viro while (p != s->mnt_parent) { 16730714a533SAl Viro p = p->mnt_parent; 16740714a533SAl Viro q = q->mnt_parent; 16751da177e4SLinus Torvalds } 167687129cc0SAl Viro p = s; 167784d17192SAl Viro parent = q; 167887129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1679be34d1a3SDavid Howells if (IS_ERR(q)) 1680be34d1a3SDavid Howells goto out; 1681719ea2fbSAl Viro lock_mount_hash(); 16821a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 168312a5b529SAl Viro mnt_set_mountpoint(parent, p->mnt_mp, q); 168412a5b529SAl Viro if (!list_empty(&parent->mnt_mounts)) { 168512a5b529SAl Viro t = list_last_entry(&parent->mnt_mounts, 168612a5b529SAl Viro struct mount, mnt_child); 168712a5b529SAl Viro if (t->mnt_mp != p->mnt_mp) 168812a5b529SAl Viro t = NULL; 168912a5b529SAl Viro } 169012a5b529SAl Viro attach_shadowed(q, parent, t); 1691719ea2fbSAl Viro unlock_mount_hash(); 16921da177e4SLinus Torvalds } 16931da177e4SLinus Torvalds } 16941da177e4SLinus Torvalds return res; 1695be34d1a3SDavid Howells out: 16961da177e4SLinus Torvalds if (res) { 1697719ea2fbSAl Viro lock_mount_hash(); 1698e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1699719ea2fbSAl Viro unlock_mount_hash(); 17001da177e4SLinus Torvalds } 1701be34d1a3SDavid Howells return q; 17021da177e4SLinus Torvalds } 17031da177e4SLinus Torvalds 1704be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1705be34d1a3SDavid Howells 1706589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 17078aec0809SAl Viro { 1708cb338d06SAl Viro struct mount *tree; 170997216be0SAl Viro namespace_lock(); 1710cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1711cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1712cd4a4017SEric W. Biederman else 171387129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 171487129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1715328e6d90SAl Viro namespace_unlock(); 1716be34d1a3SDavid Howells if (IS_ERR(tree)) 171752e220d3SDan Carpenter return ERR_CAST(tree); 1718be34d1a3SDavid Howells return &tree->mnt; 17198aec0809SAl Viro } 17208aec0809SAl Viro 17218aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 17228aec0809SAl Viro { 172397216be0SAl Viro namespace_lock(); 1724719ea2fbSAl Viro lock_mount_hash(); 1725e819f152SEric W. Biederman umount_tree(real_mount(mnt), UMOUNT_SYNC); 1726719ea2fbSAl Viro unlock_mount_hash(); 17273ab6abeeSAl Viro namespace_unlock(); 17288aec0809SAl Viro } 17298aec0809SAl Viro 1730c771d683SMiklos Szeredi /** 1731c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1732c771d683SMiklos Szeredi * 1733c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1734c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1735c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1736c771d683SMiklos Szeredi * 1737c771d683SMiklos Szeredi * Release with mntput(). 1738c771d683SMiklos Szeredi */ 1739c771d683SMiklos Szeredi struct vfsmount *clone_private_mount(struct path *path) 1740c771d683SMiklos Szeredi { 1741c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1742c771d683SMiklos Szeredi struct mount *new_mnt; 1743c771d683SMiklos Szeredi 1744c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1745c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1746c771d683SMiklos Szeredi 1747c771d683SMiklos Szeredi down_read(&namespace_sem); 1748c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1749c771d683SMiklos Szeredi up_read(&namespace_sem); 1750c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1751c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1752c771d683SMiklos Szeredi 1753c771d683SMiklos Szeredi return &new_mnt->mnt; 1754c771d683SMiklos Szeredi } 1755c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1756c771d683SMiklos Szeredi 17571f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 17581f707137SAl Viro struct vfsmount *root) 17591f707137SAl Viro { 17601a4eeaf2SAl Viro struct mount *mnt; 17611f707137SAl Viro int res = f(root, arg); 17621f707137SAl Viro if (res) 17631f707137SAl Viro return res; 17641a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 17651a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 17661f707137SAl Viro if (res) 17671f707137SAl Viro return res; 17681f707137SAl Viro } 17691f707137SAl Viro return 0; 17701f707137SAl Viro } 17711f707137SAl Viro 17724b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1773719f5d7fSMiklos Szeredi { 1774315fc83eSAl Viro struct mount *p; 1775719f5d7fSMiklos Szeredi 1776909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1777fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 17784b8b21f4SAl Viro mnt_release_group_id(p); 1779719f5d7fSMiklos Szeredi } 1780719f5d7fSMiklos Szeredi } 1781719f5d7fSMiklos Szeredi 17824b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1783719f5d7fSMiklos Szeredi { 1784315fc83eSAl Viro struct mount *p; 1785719f5d7fSMiklos Szeredi 1786909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1787fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 17884b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1789719f5d7fSMiklos Szeredi if (err) { 17904b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1791719f5d7fSMiklos Szeredi return err; 1792719f5d7fSMiklos Szeredi } 1793719f5d7fSMiklos Szeredi } 1794719f5d7fSMiklos Szeredi } 1795719f5d7fSMiklos Szeredi 1796719f5d7fSMiklos Szeredi return 0; 1797719f5d7fSMiklos Szeredi } 1798719f5d7fSMiklos Szeredi 1799b90fa9aeSRam Pai /* 1800b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1801b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 180221444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 180321444403SRam Pai * store the parent mount and mountpoint dentry. 180421444403SRam Pai * (done when source_mnt is moved) 1805b90fa9aeSRam Pai * 1806b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1807b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 18089676f0c6SRam Pai * --------------------------------------------------------------------------- 1809b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 18109676f0c6SRam Pai * |************************************************************************** 18119676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 18129676f0c6SRam Pai * | dest | | | | | 18139676f0c6SRam Pai * | | | | | | | 18149676f0c6SRam Pai * | v | | | | | 18159676f0c6SRam Pai * |************************************************************************** 18169676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 18175afe0022SRam Pai * | | | | | | 18189676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 18199676f0c6SRam Pai * *************************************************************************** 1820b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1821b90fa9aeSRam Pai * destination mount. 1822b90fa9aeSRam Pai * 1823b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1824b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1825b90fa9aeSRam Pai * the peer group of the source mount. 1826b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1827b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1828b90fa9aeSRam Pai * mount. 18295afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 18305afe0022SRam Pai * of the destination mount and the cloned mount is made slave 18315afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 18325afe0022SRam Pai * is marked as 'shared and slave'. 18335afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 18345afe0022SRam Pai * source mount. 18355afe0022SRam Pai * 18369676f0c6SRam Pai * --------------------------------------------------------------------------- 183721444403SRam Pai * | MOVE MOUNT OPERATION | 18389676f0c6SRam Pai * |************************************************************************** 18399676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 18409676f0c6SRam Pai * | dest | | | | | 18419676f0c6SRam Pai * | | | | | | | 18429676f0c6SRam Pai * | v | | | | | 18439676f0c6SRam Pai * |************************************************************************** 18449676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 18455afe0022SRam Pai * | | | | | | 18469676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 18479676f0c6SRam Pai * *************************************************************************** 18485afe0022SRam Pai * 18495afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 18505afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 185121444403SRam Pai * (+*) the mount is moved to the destination. 18525afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 18535afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 18545afe0022SRam Pai * the mount is marked as 'shared and slave'. 18555afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1856b90fa9aeSRam Pai * 1857b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1858b90fa9aeSRam Pai * applied to each mount in the tree. 1859b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1860b90fa9aeSRam Pai * in allocations. 1861b90fa9aeSRam Pai */ 18620fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 186384d17192SAl Viro struct mount *dest_mnt, 186484d17192SAl Viro struct mountpoint *dest_mp, 186584d17192SAl Viro struct path *parent_path) 1866b90fa9aeSRam Pai { 186738129a13SAl Viro HLIST_HEAD(tree_list); 1868315fc83eSAl Viro struct mount *child, *p; 186938129a13SAl Viro struct hlist_node *n; 1870719f5d7fSMiklos Szeredi int err; 1871b90fa9aeSRam Pai 1872fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 18730fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1874719f5d7fSMiklos Szeredi if (err) 1875719f5d7fSMiklos Szeredi goto out; 187684d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1877f2ebb3a9SAl Viro lock_mount_hash(); 1878719f5d7fSMiklos Szeredi if (err) 1879719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1880909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 18810f0afb1dSAl Viro set_mnt_shared(p); 18820b1b901bSAl Viro } else { 18830b1b901bSAl Viro lock_mount_hash(); 1884b90fa9aeSRam Pai } 18851a390689SAl Viro if (parent_path) { 18860fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 188784d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1888143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 188921444403SRam Pai } else { 189084d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 18911d6a32acSAl Viro commit_tree(source_mnt, NULL); 189221444403SRam Pai } 1893b90fa9aeSRam Pai 189438129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 18951d6a32acSAl Viro struct mount *q; 189638129a13SAl Viro hlist_del_init(&child->mnt_hash); 18971d6a32acSAl Viro q = __lookup_mnt_last(&child->mnt_parent->mnt, 18981d6a32acSAl Viro child->mnt_mountpoint); 18991d6a32acSAl Viro commit_tree(child, q); 1900b90fa9aeSRam Pai } 1901719ea2fbSAl Viro unlock_mount_hash(); 190299b7db7bSNick Piggin 1903b90fa9aeSRam Pai return 0; 1904719f5d7fSMiklos Szeredi 1905719f5d7fSMiklos Szeredi out_cleanup_ids: 1906f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 1907f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 1908e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 1909f2ebb3a9SAl Viro } 1910f2ebb3a9SAl Viro unlock_mount_hash(); 19110fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1912719f5d7fSMiklos Szeredi out: 1913719f5d7fSMiklos Szeredi return err; 1914b90fa9aeSRam Pai } 1915b90fa9aeSRam Pai 191684d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 1917b12cea91SAl Viro { 1918b12cea91SAl Viro struct vfsmount *mnt; 191984d17192SAl Viro struct dentry *dentry = path->dentry; 1920b12cea91SAl Viro retry: 192184d17192SAl Viro mutex_lock(&dentry->d_inode->i_mutex); 192284d17192SAl Viro if (unlikely(cant_mount(dentry))) { 192384d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 192484d17192SAl Viro return ERR_PTR(-ENOENT); 1925b12cea91SAl Viro } 192697216be0SAl Viro namespace_lock(); 1927b12cea91SAl Viro mnt = lookup_mnt(path); 192884d17192SAl Viro if (likely(!mnt)) { 1929e2dfa935SEric W. Biederman struct mountpoint *mp = lookup_mountpoint(dentry); 1930e2dfa935SEric W. Biederman if (!mp) 1931e2dfa935SEric W. Biederman mp = new_mountpoint(dentry); 193284d17192SAl Viro if (IS_ERR(mp)) { 193397216be0SAl Viro namespace_unlock(); 193484d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 193584d17192SAl Viro return mp; 193684d17192SAl Viro } 193784d17192SAl Viro return mp; 193884d17192SAl Viro } 193997216be0SAl Viro namespace_unlock(); 1940b12cea91SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 1941b12cea91SAl Viro path_put(path); 1942b12cea91SAl Viro path->mnt = mnt; 194384d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 1944b12cea91SAl Viro goto retry; 1945b12cea91SAl Viro } 1946b12cea91SAl Viro 194784d17192SAl Viro static void unlock_mount(struct mountpoint *where) 1948b12cea91SAl Viro { 194984d17192SAl Viro struct dentry *dentry = where->m_dentry; 195084d17192SAl Viro put_mountpoint(where); 1951328e6d90SAl Viro namespace_unlock(); 195284d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 1953b12cea91SAl Viro } 1954b12cea91SAl Viro 195584d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 19561da177e4SLinus Torvalds { 195795bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 19581da177e4SLinus Torvalds return -EINVAL; 19591da177e4SLinus Torvalds 1960e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 1961e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 19621da177e4SLinus Torvalds return -ENOTDIR; 19631da177e4SLinus Torvalds 196484d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 19651da177e4SLinus Torvalds } 19661da177e4SLinus Torvalds 19671da177e4SLinus Torvalds /* 19687a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 19697a2e8a8fSValerie Aurora */ 19707a2e8a8fSValerie Aurora 19717a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 19727a2e8a8fSValerie Aurora { 19737c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 19747a2e8a8fSValerie Aurora 19757a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 19767a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 19777a2e8a8fSValerie Aurora return 0; 19787a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 19797a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 19807a2e8a8fSValerie Aurora return 0; 19817a2e8a8fSValerie Aurora return type; 19827a2e8a8fSValerie Aurora } 19837a2e8a8fSValerie Aurora 19847a2e8a8fSValerie Aurora /* 198507b20889SRam Pai * recursively change the type of the mountpoint. 198607b20889SRam Pai */ 19870a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 198807b20889SRam Pai { 1989315fc83eSAl Viro struct mount *m; 19904b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 199107b20889SRam Pai int recurse = flag & MS_REC; 19927a2e8a8fSValerie Aurora int type; 1993719f5d7fSMiklos Szeredi int err = 0; 199407b20889SRam Pai 19952d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 199607b20889SRam Pai return -EINVAL; 199707b20889SRam Pai 19987a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 19997a2e8a8fSValerie Aurora if (!type) 20007a2e8a8fSValerie Aurora return -EINVAL; 20017a2e8a8fSValerie Aurora 200297216be0SAl Viro namespace_lock(); 2003719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2004719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2005719f5d7fSMiklos Szeredi if (err) 2006719f5d7fSMiklos Szeredi goto out_unlock; 2007719f5d7fSMiklos Szeredi } 2008719f5d7fSMiklos Szeredi 2009719ea2fbSAl Viro lock_mount_hash(); 2010909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 20110f0afb1dSAl Viro change_mnt_propagation(m, type); 2012719ea2fbSAl Viro unlock_mount_hash(); 2013719f5d7fSMiklos Szeredi 2014719f5d7fSMiklos Szeredi out_unlock: 201597216be0SAl Viro namespace_unlock(); 2016719f5d7fSMiklos Szeredi return err; 201707b20889SRam Pai } 201807b20889SRam Pai 20195ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 20205ff9d8a6SEric W. Biederman { 20215ff9d8a6SEric W. Biederman struct mount *child; 20225ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 20235ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 20245ff9d8a6SEric W. Biederman continue; 20255ff9d8a6SEric W. Biederman 20265ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 20275ff9d8a6SEric W. Biederman return true; 20285ff9d8a6SEric W. Biederman } 20295ff9d8a6SEric W. Biederman return false; 20305ff9d8a6SEric W. Biederman } 20315ff9d8a6SEric W. Biederman 203207b20889SRam Pai /* 20331da177e4SLinus Torvalds * do loopback mount. 20341da177e4SLinus Torvalds */ 2035808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 20362dafe1c4SEric Sandeen int recurse) 20371da177e4SLinus Torvalds { 20382d92ab3cSAl Viro struct path old_path; 203984d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 204084d17192SAl Viro struct mountpoint *mp; 204157eccb83SAl Viro int err; 20421da177e4SLinus Torvalds if (!old_name || !*old_name) 20431da177e4SLinus Torvalds return -EINVAL; 2044815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 20451da177e4SLinus Torvalds if (err) 20461da177e4SLinus Torvalds return err; 20471da177e4SLinus Torvalds 20488823c079SEric W. Biederman err = -EINVAL; 20494ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 20508823c079SEric W. Biederman goto out; 20518823c079SEric W. Biederman 205284d17192SAl Viro mp = lock_mount(path); 205384d17192SAl Viro err = PTR_ERR(mp); 205484d17192SAl Viro if (IS_ERR(mp)) 20559676f0c6SRam Pai goto out; 20569676f0c6SRam Pai 205787129cc0SAl Viro old = real_mount(old_path.mnt); 205884d17192SAl Viro parent = real_mount(path->mnt); 205987129cc0SAl Viro 2060b12cea91SAl Viro err = -EINVAL; 2061fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2062b12cea91SAl Viro goto out2; 2063b12cea91SAl Viro 2064e149ed2bSAl Viro if (!check_mnt(parent)) 2065e149ed2bSAl Viro goto out2; 2066e149ed2bSAl Viro 2067e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2068b12cea91SAl Viro goto out2; 2069ccd48bc7SAl Viro 20705ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 20715ff9d8a6SEric W. Biederman goto out2; 20725ff9d8a6SEric W. Biederman 20731da177e4SLinus Torvalds if (recurse) 20744ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 20751da177e4SLinus Torvalds else 207687129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 20771da177e4SLinus Torvalds 2078be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2079be34d1a3SDavid Howells err = PTR_ERR(mnt); 2080e9c5d8a5SAndrey Vagin goto out2; 2081be34d1a3SDavid Howells } 2082ccd48bc7SAl Viro 20835ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 20845ff9d8a6SEric W. Biederman 208584d17192SAl Viro err = graft_tree(mnt, parent, mp); 20861da177e4SLinus Torvalds if (err) { 2087719ea2fbSAl Viro lock_mount_hash(); 2088e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2089719ea2fbSAl Viro unlock_mount_hash(); 20905b83d2c5SRam Pai } 2091b12cea91SAl Viro out2: 209284d17192SAl Viro unlock_mount(mp); 2093ccd48bc7SAl Viro out: 20942d92ab3cSAl Viro path_put(&old_path); 20951da177e4SLinus Torvalds return err; 20961da177e4SLinus Torvalds } 20971da177e4SLinus Torvalds 20982e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 20992e4b7fcdSDave Hansen { 21002e4b7fcdSDave Hansen int error = 0; 21012e4b7fcdSDave Hansen int readonly_request = 0; 21022e4b7fcdSDave Hansen 21032e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 21042e4b7fcdSDave Hansen readonly_request = 1; 21052e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 21062e4b7fcdSDave Hansen return 0; 21072e4b7fcdSDave Hansen 21082e4b7fcdSDave Hansen if (readonly_request) 210983adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 21102e4b7fcdSDave Hansen else 211183adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 21122e4b7fcdSDave Hansen return error; 21132e4b7fcdSDave Hansen } 21142e4b7fcdSDave Hansen 21151da177e4SLinus Torvalds /* 21161da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 21171da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 21181da177e4SLinus Torvalds * on it - tough luck. 21191da177e4SLinus Torvalds */ 21200a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 21211da177e4SLinus Torvalds void *data) 21221da177e4SLinus Torvalds { 21231da177e4SLinus Torvalds int err; 21242d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2125143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 21261da177e4SLinus Torvalds 2127143c8c91SAl Viro if (!check_mnt(mnt)) 21281da177e4SLinus Torvalds return -EINVAL; 21291da177e4SLinus Torvalds 21302d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 21311da177e4SLinus Torvalds return -EINVAL; 21321da177e4SLinus Torvalds 213307b64558SEric W. Biederman /* Don't allow changing of locked mnt flags. 213407b64558SEric W. Biederman * 213507b64558SEric W. Biederman * No locks need to be held here while testing the various 213607b64558SEric W. Biederman * MNT_LOCK flags because those flags can never be cleared 213707b64558SEric W. Biederman * once they are set. 213807b64558SEric W. Biederman */ 213907b64558SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 214007b64558SEric W. Biederman !(mnt_flags & MNT_READONLY)) { 214107b64558SEric W. Biederman return -EPERM; 214207b64558SEric W. Biederman } 21439566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 21449566d674SEric W. Biederman !(mnt_flags & MNT_NODEV)) { 21453e186641SEric W. Biederman /* Was the nodev implicitly added in mount? */ 21463e186641SEric W. Biederman if ((mnt->mnt_ns->user_ns != &init_user_ns) && 21473e186641SEric W. Biederman !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) { 21483e186641SEric W. Biederman mnt_flags |= MNT_NODEV; 21493e186641SEric W. Biederman } else { 21509566d674SEric W. Biederman return -EPERM; 21519566d674SEric W. Biederman } 21523e186641SEric W. Biederman } 21539566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && 21549566d674SEric W. Biederman !(mnt_flags & MNT_NOSUID)) { 21559566d674SEric W. Biederman return -EPERM; 21569566d674SEric W. Biederman } 21579566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && 21589566d674SEric W. Biederman !(mnt_flags & MNT_NOEXEC)) { 21599566d674SEric W. Biederman return -EPERM; 21609566d674SEric W. Biederman } 21619566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 21629566d674SEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { 21639566d674SEric W. Biederman return -EPERM; 21649566d674SEric W. Biederman } 21659566d674SEric W. Biederman 2166ff36fe2cSEric Paris err = security_sb_remount(sb, data); 2167ff36fe2cSEric Paris if (err) 2168ff36fe2cSEric Paris return err; 2169ff36fe2cSEric Paris 21701da177e4SLinus Torvalds down_write(&sb->s_umount); 21712e4b7fcdSDave Hansen if (flags & MS_BIND) 21722d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 217357eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 217457eccb83SAl Viro err = -EPERM; 21754aa98cf7SAl Viro else 21761da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 21777b43a79fSAl Viro if (!err) { 2178719ea2fbSAl Viro lock_mount_hash(); 2179a6138db8SEric W. Biederman mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 2180143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 2181143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2182719ea2fbSAl Viro unlock_mount_hash(); 21830e55a7ccSDan Williams } 21846339dab8SAl Viro up_write(&sb->s_umount); 21851da177e4SLinus Torvalds return err; 21861da177e4SLinus Torvalds } 21871da177e4SLinus Torvalds 2188cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 21899676f0c6SRam Pai { 2190315fc83eSAl Viro struct mount *p; 2191909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2192fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 21939676f0c6SRam Pai return 1; 21949676f0c6SRam Pai } 21959676f0c6SRam Pai return 0; 21969676f0c6SRam Pai } 21979676f0c6SRam Pai 2198808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 21991da177e4SLinus Torvalds { 22002d92ab3cSAl Viro struct path old_path, parent_path; 2201676da58dSAl Viro struct mount *p; 22020fb54e50SAl Viro struct mount *old; 220384d17192SAl Viro struct mountpoint *mp; 220457eccb83SAl Viro int err; 22051da177e4SLinus Torvalds if (!old_name || !*old_name) 22061da177e4SLinus Torvalds return -EINVAL; 22072d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 22081da177e4SLinus Torvalds if (err) 22091da177e4SLinus Torvalds return err; 22101da177e4SLinus Torvalds 221184d17192SAl Viro mp = lock_mount(path); 221284d17192SAl Viro err = PTR_ERR(mp); 221384d17192SAl Viro if (IS_ERR(mp)) 2214cc53ce53SDavid Howells goto out; 2215cc53ce53SDavid Howells 2216143c8c91SAl Viro old = real_mount(old_path.mnt); 2217fc7be130SAl Viro p = real_mount(path->mnt); 2218143c8c91SAl Viro 22191da177e4SLinus Torvalds err = -EINVAL; 2220fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 22211da177e4SLinus Torvalds goto out1; 22221da177e4SLinus Torvalds 22235ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 22245ff9d8a6SEric W. Biederman goto out1; 22255ff9d8a6SEric W. Biederman 22261da177e4SLinus Torvalds err = -EINVAL; 22272d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 222821444403SRam Pai goto out1; 22291da177e4SLinus Torvalds 2230676da58dSAl Viro if (!mnt_has_parent(old)) 223121444403SRam Pai goto out1; 22321da177e4SLinus Torvalds 2233e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2234e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 223521444403SRam Pai goto out1; 223621444403SRam Pai /* 223721444403SRam Pai * Don't move a mount residing in a shared parent. 223821444403SRam Pai */ 2239fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 224021444403SRam Pai goto out1; 22419676f0c6SRam Pai /* 22429676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 22439676f0c6SRam Pai * mount which is shared. 22449676f0c6SRam Pai */ 2245fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 22469676f0c6SRam Pai goto out1; 22471da177e4SLinus Torvalds err = -ELOOP; 2248fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2249676da58dSAl Viro if (p == old) 225021444403SRam Pai goto out1; 22511da177e4SLinus Torvalds 225284d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 22534ac91378SJan Blunck if (err) 225421444403SRam Pai goto out1; 22551da177e4SLinus Torvalds 22561da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 22571da177e4SLinus Torvalds * automatically */ 22586776db3dSAl Viro list_del_init(&old->mnt_expire); 22591da177e4SLinus Torvalds out1: 226084d17192SAl Viro unlock_mount(mp); 22611da177e4SLinus Torvalds out: 22621da177e4SLinus Torvalds if (!err) 22631a390689SAl Viro path_put(&parent_path); 22642d92ab3cSAl Viro path_put(&old_path); 22651da177e4SLinus Torvalds return err; 22661da177e4SLinus Torvalds } 22671da177e4SLinus Torvalds 22689d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 22699d412a43SAl Viro { 22709d412a43SAl Viro int err; 22719d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 22729d412a43SAl Viro if (subtype) { 22739d412a43SAl Viro subtype++; 22749d412a43SAl Viro err = -EINVAL; 22759d412a43SAl Viro if (!subtype[0]) 22769d412a43SAl Viro goto err; 22779d412a43SAl Viro } else 22789d412a43SAl Viro subtype = ""; 22799d412a43SAl Viro 22809d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 22819d412a43SAl Viro err = -ENOMEM; 22829d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 22839d412a43SAl Viro goto err; 22849d412a43SAl Viro return mnt; 22859d412a43SAl Viro 22869d412a43SAl Viro err: 22879d412a43SAl Viro mntput(mnt); 22889d412a43SAl Viro return ERR_PTR(err); 22899d412a43SAl Viro } 22909d412a43SAl Viro 22919d412a43SAl Viro /* 22929d412a43SAl Viro * add a mount into a namespace's mount tree 22939d412a43SAl Viro */ 229495bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 22959d412a43SAl Viro { 229684d17192SAl Viro struct mountpoint *mp; 229784d17192SAl Viro struct mount *parent; 22989d412a43SAl Viro int err; 22999d412a43SAl Viro 2300f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 23019d412a43SAl Viro 230284d17192SAl Viro mp = lock_mount(path); 230384d17192SAl Viro if (IS_ERR(mp)) 230484d17192SAl Viro return PTR_ERR(mp); 23059d412a43SAl Viro 230684d17192SAl Viro parent = real_mount(path->mnt); 23079d412a43SAl Viro err = -EINVAL; 230884d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2309156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2310156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 23119d412a43SAl Viro goto unlock; 2312156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 231384d17192SAl Viro if (!parent->mnt_ns) 2314156cacb1SAl Viro goto unlock; 2315156cacb1SAl Viro } 23169d412a43SAl Viro 23179d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 23189d412a43SAl Viro err = -EBUSY; 231995bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 23209d412a43SAl Viro path->mnt->mnt_root == path->dentry) 23219d412a43SAl Viro goto unlock; 23229d412a43SAl Viro 23239d412a43SAl Viro err = -EINVAL; 2324e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 23259d412a43SAl Viro goto unlock; 23269d412a43SAl Viro 232795bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 232884d17192SAl Viro err = graft_tree(newmnt, parent, mp); 23299d412a43SAl Viro 23309d412a43SAl Viro unlock: 233184d17192SAl Viro unlock_mount(mp); 23329d412a43SAl Viro return err; 23339d412a43SAl Viro } 2334b1e75df4SAl Viro 23358c6cf9ccSEric W. Biederman static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags); 23361b852bceSEric W. Biederman 23371da177e4SLinus Torvalds /* 23381da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 23391da177e4SLinus Torvalds * namespace's tree 23401da177e4SLinus Torvalds */ 23410c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2342808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 23431da177e4SLinus Torvalds { 23440c55cfc4SEric W. Biederman struct file_system_type *type; 23459b40bc90SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 23461da177e4SLinus Torvalds struct vfsmount *mnt; 234715f9a3f3SAl Viro int err; 23481da177e4SLinus Torvalds 23490c55cfc4SEric W. Biederman if (!fstype) 23501da177e4SLinus Torvalds return -EINVAL; 23511da177e4SLinus Torvalds 23520c55cfc4SEric W. Biederman type = get_fs_type(fstype); 23530c55cfc4SEric W. Biederman if (!type) 23540c55cfc4SEric W. Biederman return -ENODEV; 23550c55cfc4SEric W. Biederman 23560c55cfc4SEric W. Biederman if (user_ns != &init_user_ns) { 23570c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_MOUNT)) { 23580c55cfc4SEric W. Biederman put_filesystem(type); 23590c55cfc4SEric W. Biederman return -EPERM; 23600c55cfc4SEric W. Biederman } 23610c55cfc4SEric W. Biederman /* Only in special cases allow devices from mounts 23620c55cfc4SEric W. Biederman * created outside the initial user namespace. 23630c55cfc4SEric W. Biederman */ 23640c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { 23650c55cfc4SEric W. Biederman flags |= MS_NODEV; 23669566d674SEric W. Biederman mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; 23670c55cfc4SEric W. Biederman } 23681b852bceSEric W. Biederman if (type->fs_flags & FS_USERNS_VISIBLE) { 23698c6cf9ccSEric W. Biederman if (!fs_fully_visible(type, &mnt_flags)) 23701b852bceSEric W. Biederman return -EPERM; 23711b852bceSEric W. Biederman } 23720c55cfc4SEric W. Biederman } 23730c55cfc4SEric W. Biederman 23740c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 23750c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 23760c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 23770c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 23780c55cfc4SEric W. Biederman 23790c55cfc4SEric W. Biederman put_filesystem(type); 23801da177e4SLinus Torvalds if (IS_ERR(mnt)) 23811da177e4SLinus Torvalds return PTR_ERR(mnt); 23821da177e4SLinus Torvalds 238395bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 238415f9a3f3SAl Viro if (err) 238515f9a3f3SAl Viro mntput(mnt); 238615f9a3f3SAl Viro return err; 23871da177e4SLinus Torvalds } 23881da177e4SLinus Torvalds 238919a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 239019a167afSAl Viro { 23916776db3dSAl Viro struct mount *mnt = real_mount(m); 239219a167afSAl Viro int err; 239319a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 239419a167afSAl Viro * expired before we get a chance to add it 239519a167afSAl Viro */ 23966776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 239719a167afSAl Viro 239819a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 239919a167afSAl Viro m->mnt_root == path->dentry) { 2400b1e75df4SAl Viro err = -ELOOP; 2401b1e75df4SAl Viro goto fail; 240219a167afSAl Viro } 240319a167afSAl Viro 240495bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2405b1e75df4SAl Viro if (!err) 2406b1e75df4SAl Viro return 0; 2407b1e75df4SAl Viro fail: 2408b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 24096776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 241097216be0SAl Viro namespace_lock(); 24116776db3dSAl Viro list_del_init(&mnt->mnt_expire); 241297216be0SAl Viro namespace_unlock(); 241319a167afSAl Viro } 2414b1e75df4SAl Viro mntput(m); 2415b1e75df4SAl Viro mntput(m); 241619a167afSAl Viro return err; 241719a167afSAl Viro } 241819a167afSAl Viro 2419ea5b778aSDavid Howells /** 2420ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2421ea5b778aSDavid Howells * @mnt: The mount to list. 2422ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2423ea5b778aSDavid Howells */ 2424ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2425ea5b778aSDavid Howells { 242697216be0SAl Viro namespace_lock(); 2427ea5b778aSDavid Howells 24286776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2429ea5b778aSDavid Howells 243097216be0SAl Viro namespace_unlock(); 2431ea5b778aSDavid Howells } 2432ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2433ea5b778aSDavid Howells 2434ea5b778aSDavid Howells /* 24351da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 24361da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 24371da177e4SLinus Torvalds * here 24381da177e4SLinus Torvalds */ 24391da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 24401da177e4SLinus Torvalds { 2441761d5c38SAl Viro struct mount *mnt, *next; 24421da177e4SLinus Torvalds LIST_HEAD(graveyard); 24431da177e4SLinus Torvalds 24441da177e4SLinus Torvalds if (list_empty(mounts)) 24451da177e4SLinus Torvalds return; 24461da177e4SLinus Torvalds 244797216be0SAl Viro namespace_lock(); 2448719ea2fbSAl Viro lock_mount_hash(); 24491da177e4SLinus Torvalds 24501da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 24511da177e4SLinus Torvalds * following criteria: 24521da177e4SLinus Torvalds * - only referenced by its parent vfsmount 24531da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 24541da177e4SLinus Torvalds * cleared by mntput()) 24551da177e4SLinus Torvalds */ 24566776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2457863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 24581ab59738SAl Viro propagate_mount_busy(mnt, 1)) 24591da177e4SLinus Torvalds continue; 24606776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 24611da177e4SLinus Torvalds } 2462bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 24636776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2464143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2465e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2466bcc5c7d2SAl Viro } 2467719ea2fbSAl Viro unlock_mount_hash(); 24683ab6abeeSAl Viro namespace_unlock(); 24691da177e4SLinus Torvalds } 24701da177e4SLinus Torvalds 24711da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 24721da177e4SLinus Torvalds 24731da177e4SLinus Torvalds /* 24745528f911STrond Myklebust * Ripoff of 'select_parent()' 24755528f911STrond Myklebust * 24765528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 24775528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 24785528f911STrond Myklebust */ 2479692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 24805528f911STrond Myklebust { 2481692afc31SAl Viro struct mount *this_parent = parent; 24825528f911STrond Myklebust struct list_head *next; 24835528f911STrond Myklebust int found = 0; 24845528f911STrond Myklebust 24855528f911STrond Myklebust repeat: 24866b41d536SAl Viro next = this_parent->mnt_mounts.next; 24875528f911STrond Myklebust resume: 24886b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 24895528f911STrond Myklebust struct list_head *tmp = next; 24906b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 24915528f911STrond Myklebust 24925528f911STrond Myklebust next = tmp->next; 2493692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 24945528f911STrond Myklebust continue; 24955528f911STrond Myklebust /* 24965528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 24975528f911STrond Myklebust */ 24986b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 24995528f911STrond Myklebust this_parent = mnt; 25005528f911STrond Myklebust goto repeat; 25015528f911STrond Myklebust } 25025528f911STrond Myklebust 25031ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 25046776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 25055528f911STrond Myklebust found++; 25065528f911STrond Myklebust } 25075528f911STrond Myklebust } 25085528f911STrond Myklebust /* 25095528f911STrond Myklebust * All done at this level ... ascend and resume the search 25105528f911STrond Myklebust */ 25115528f911STrond Myklebust if (this_parent != parent) { 25126b41d536SAl Viro next = this_parent->mnt_child.next; 25130714a533SAl Viro this_parent = this_parent->mnt_parent; 25145528f911STrond Myklebust goto resume; 25155528f911STrond Myklebust } 25165528f911STrond Myklebust return found; 25175528f911STrond Myklebust } 25185528f911STrond Myklebust 25195528f911STrond Myklebust /* 25205528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 25215528f911STrond Myklebust * submounts of a specific parent mountpoint 252299b7db7bSNick Piggin * 252348a066e7SAl Viro * mount_lock must be held for write 25245528f911STrond Myklebust */ 2525b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 25265528f911STrond Myklebust { 25275528f911STrond Myklebust LIST_HEAD(graveyard); 2528761d5c38SAl Viro struct mount *m; 25295528f911STrond Myklebust 25305528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2531c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2532bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2533761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 25346776db3dSAl Viro mnt_expire); 2535143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2536e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2537bcc5c7d2SAl Viro } 2538bcc5c7d2SAl Viro } 25395528f911STrond Myklebust } 25405528f911STrond Myklebust 25415528f911STrond Myklebust /* 25421da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 25431da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 25441da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 25451da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 25461da177e4SLinus Torvalds */ 2547b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2548b58fed8bSRam Pai unsigned long n) 25491da177e4SLinus Torvalds { 25501da177e4SLinus Torvalds char *t = to; 25511da177e4SLinus Torvalds const char __user *f = from; 25521da177e4SLinus Torvalds char c; 25531da177e4SLinus Torvalds 25541da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 25551da177e4SLinus Torvalds return n; 25561da177e4SLinus Torvalds 25571da177e4SLinus Torvalds while (n) { 25581da177e4SLinus Torvalds if (__get_user(c, f)) { 25591da177e4SLinus Torvalds memset(t, 0, n); 25601da177e4SLinus Torvalds break; 25611da177e4SLinus Torvalds } 25621da177e4SLinus Torvalds *t++ = c; 25631da177e4SLinus Torvalds f++; 25641da177e4SLinus Torvalds n--; 25651da177e4SLinus Torvalds } 25661da177e4SLinus Torvalds return n; 25671da177e4SLinus Torvalds } 25681da177e4SLinus Torvalds 25691da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 25701da177e4SLinus Torvalds { 25711da177e4SLinus Torvalds int i; 25721da177e4SLinus Torvalds unsigned long page; 25731da177e4SLinus Torvalds unsigned long size; 25741da177e4SLinus Torvalds 25751da177e4SLinus Torvalds *where = 0; 25761da177e4SLinus Torvalds if (!data) 25771da177e4SLinus Torvalds return 0; 25781da177e4SLinus Torvalds 25791da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 25801da177e4SLinus Torvalds return -ENOMEM; 25811da177e4SLinus Torvalds 25821da177e4SLinus Torvalds /* We only care that *some* data at the address the user 25831da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 25841da177e4SLinus Torvalds * the remainder of the page. 25851da177e4SLinus Torvalds */ 25861da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 25871da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 25881da177e4SLinus Torvalds if (size > PAGE_SIZE) 25891da177e4SLinus Torvalds size = PAGE_SIZE; 25901da177e4SLinus Torvalds 25911da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 25921da177e4SLinus Torvalds if (!i) { 25931da177e4SLinus Torvalds free_page(page); 25941da177e4SLinus Torvalds return -EFAULT; 25951da177e4SLinus Torvalds } 25961da177e4SLinus Torvalds if (i != PAGE_SIZE) 25971da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 25981da177e4SLinus Torvalds *where = page; 25991da177e4SLinus Torvalds return 0; 26001da177e4SLinus Torvalds } 26011da177e4SLinus Torvalds 2602b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2603eca6f534SVegard Nossum { 2604b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2605eca6f534SVegard Nossum } 2606eca6f534SVegard Nossum 26071da177e4SLinus Torvalds /* 26081da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 26091da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 26101da177e4SLinus Torvalds * 26111da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 26121da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 26131da177e4SLinus Torvalds * information (or be NULL). 26141da177e4SLinus Torvalds * 26151da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 26161da177e4SLinus Torvalds * When the flags word was introduced its top half was required 26171da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 26181da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 26191da177e4SLinus Torvalds * and must be discarded. 26201da177e4SLinus Torvalds */ 26215e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2622808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 26231da177e4SLinus Torvalds { 26242d92ab3cSAl Viro struct path path; 26251da177e4SLinus Torvalds int retval = 0; 26261da177e4SLinus Torvalds int mnt_flags = 0; 26271da177e4SLinus Torvalds 26281da177e4SLinus Torvalds /* Discard magic */ 26291da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 26301da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 26311da177e4SLinus Torvalds 26321da177e4SLinus Torvalds /* Basic sanity checks */ 26331da177e4SLinus Torvalds if (data_page) 26341da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 26351da177e4SLinus Torvalds 2636a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 26375e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2638a27ab9f2STetsuo Handa if (retval) 2639a27ab9f2STetsuo Handa return retval; 2640a27ab9f2STetsuo Handa 2641a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2642a27ab9f2STetsuo Handa type_page, flags, data_page); 26430d5cadb8SAl Viro if (!retval && !may_mount()) 26440d5cadb8SAl Viro retval = -EPERM; 2645a27ab9f2STetsuo Handa if (retval) 2646a27ab9f2STetsuo Handa goto dput_out; 2647a27ab9f2STetsuo Handa 2648613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2649613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 26500a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 26510a1c01c9SMatthew Garrett 26521da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 26531da177e4SLinus Torvalds if (flags & MS_NOSUID) 26541da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 26551da177e4SLinus Torvalds if (flags & MS_NODEV) 26561da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 26571da177e4SLinus Torvalds if (flags & MS_NOEXEC) 26581da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2659fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2660fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2661fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2662fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2663d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2664d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 26652e4b7fcdSDave Hansen if (flags & MS_RDONLY) 26662e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2667fc33a7bbSChristoph Hellwig 2668ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2669ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2670ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2671ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2672ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2673ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2674ffbc6f0eSEric W. Biederman } 2675ffbc6f0eSEric W. Biederman 26767a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2677d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2678d0adde57SMatthew Garrett MS_STRICTATIME); 26791da177e4SLinus Torvalds 26801da177e4SLinus Torvalds if (flags & MS_REMOUNT) 26812d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 26821da177e4SLinus Torvalds data_page); 26831da177e4SLinus Torvalds else if (flags & MS_BIND) 26842d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 26859676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 26862d92ab3cSAl Viro retval = do_change_type(&path, flags); 26871da177e4SLinus Torvalds else if (flags & MS_MOVE) 26882d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 26891da177e4SLinus Torvalds else 26902d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 26911da177e4SLinus Torvalds dev_name, data_page); 26921da177e4SLinus Torvalds dput_out: 26932d92ab3cSAl Viro path_put(&path); 26941da177e4SLinus Torvalds return retval; 26951da177e4SLinus Torvalds } 26961da177e4SLinus Torvalds 2697771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2698771b1371SEric W. Biederman { 26996344c433SAl Viro ns_free_inum(&ns->ns); 2700771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2701771b1371SEric W. Biederman kfree(ns); 2702771b1371SEric W. Biederman } 2703771b1371SEric W. Biederman 27048823c079SEric W. Biederman /* 27058823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 27068823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 27078823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 27088823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 27098823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 27108823c079SEric W. Biederman */ 27118823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 27128823c079SEric W. Biederman 2713771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2714cf8d2c11STrond Myklebust { 2715cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 271698f842e6SEric W. Biederman int ret; 2717cf8d2c11STrond Myklebust 2718cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2719cf8d2c11STrond Myklebust if (!new_ns) 2720cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 27216344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 272298f842e6SEric W. Biederman if (ret) { 272398f842e6SEric W. Biederman kfree(new_ns); 272498f842e6SEric W. Biederman return ERR_PTR(ret); 272598f842e6SEric W. Biederman } 272633c42940SAl Viro new_ns->ns.ops = &mntns_operations; 27278823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2728cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2729cf8d2c11STrond Myklebust new_ns->root = NULL; 2730cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2731cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2732cf8d2c11STrond Myklebust new_ns->event = 0; 2733771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2734cf8d2c11STrond Myklebust return new_ns; 2735cf8d2c11STrond Myklebust } 2736cf8d2c11STrond Myklebust 27379559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 27389559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 27391da177e4SLinus Torvalds { 27406b3286edSKirill Korotaev struct mnt_namespace *new_ns; 27417f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2742315fc83eSAl Viro struct mount *p, *q; 27439559f689SAl Viro struct mount *old; 2744cb338d06SAl Viro struct mount *new; 27457a472ef4SEric W. Biederman int copy_flags; 27461da177e4SLinus Torvalds 27479559f689SAl Viro BUG_ON(!ns); 27489559f689SAl Viro 27499559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 27509559f689SAl Viro get_mnt_ns(ns); 27519559f689SAl Viro return ns; 27529559f689SAl Viro } 27539559f689SAl Viro 27549559f689SAl Viro old = ns->root; 27559559f689SAl Viro 2756771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2757cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2758cf8d2c11STrond Myklebust return new_ns; 27591da177e4SLinus Torvalds 276097216be0SAl Viro namespace_lock(); 27611da177e4SLinus Torvalds /* First pass: copy the tree topology */ 27624ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 27639559f689SAl Viro if (user_ns != ns->user_ns) 2764132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 27657a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2766be34d1a3SDavid Howells if (IS_ERR(new)) { 2767328e6d90SAl Viro namespace_unlock(); 2768771b1371SEric W. Biederman free_mnt_ns(new_ns); 2769be34d1a3SDavid Howells return ERR_CAST(new); 27701da177e4SLinus Torvalds } 2771be08d6d2SAl Viro new_ns->root = new; 27721a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 27731da177e4SLinus Torvalds 27741da177e4SLinus Torvalds /* 27751da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 27761da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 27771da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 27781da177e4SLinus Torvalds */ 2779909b0a88SAl Viro p = old; 2780cb338d06SAl Viro q = new; 27811da177e4SLinus Torvalds while (p) { 2782143c8c91SAl Viro q->mnt_ns = new_ns; 27839559f689SAl Viro if (new_fs) { 27849559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 27859559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2786315fc83eSAl Viro rootmnt = &p->mnt; 27871da177e4SLinus Torvalds } 27889559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 27899559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2790315fc83eSAl Viro pwdmnt = &p->mnt; 27911da177e4SLinus Torvalds } 27921da177e4SLinus Torvalds } 2793909b0a88SAl Viro p = next_mnt(p, old); 2794909b0a88SAl Viro q = next_mnt(q, new); 27954ce5d2b1SEric W. Biederman if (!q) 27964ce5d2b1SEric W. Biederman break; 27974ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 27984ce5d2b1SEric W. Biederman p = next_mnt(p, old); 27991da177e4SLinus Torvalds } 2800328e6d90SAl Viro namespace_unlock(); 28011da177e4SLinus Torvalds 28021da177e4SLinus Torvalds if (rootmnt) 2803f03c6599SAl Viro mntput(rootmnt); 28041da177e4SLinus Torvalds if (pwdmnt) 2805f03c6599SAl Viro mntput(pwdmnt); 28061da177e4SLinus Torvalds 2807741a2951SJANAK DESAI return new_ns; 2808741a2951SJANAK DESAI } 2809741a2951SJANAK DESAI 2810cf8d2c11STrond Myklebust /** 2811cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2812cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2813cf8d2c11STrond Myklebust */ 28141a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2815cf8d2c11STrond Myklebust { 2816771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2817cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 28181a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 28191a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2820be08d6d2SAl Viro new_ns->root = mnt; 2821b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2822c1334495SAl Viro } else { 28231a4eeaf2SAl Viro mntput(m); 2824cf8d2c11STrond Myklebust } 2825cf8d2c11STrond Myklebust return new_ns; 2826cf8d2c11STrond Myklebust } 2827cf8d2c11STrond Myklebust 2828ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2829ea441d11SAl Viro { 2830ea441d11SAl Viro struct mnt_namespace *ns; 2831d31da0f0SAl Viro struct super_block *s; 2832ea441d11SAl Viro struct path path; 2833ea441d11SAl Viro int err; 2834ea441d11SAl Viro 2835ea441d11SAl Viro ns = create_mnt_ns(mnt); 2836ea441d11SAl Viro if (IS_ERR(ns)) 2837ea441d11SAl Viro return ERR_CAST(ns); 2838ea441d11SAl Viro 2839ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2840ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2841ea441d11SAl Viro 2842ea441d11SAl Viro put_mnt_ns(ns); 2843ea441d11SAl Viro 2844ea441d11SAl Viro if (err) 2845ea441d11SAl Viro return ERR_PTR(err); 2846ea441d11SAl Viro 2847ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2848d31da0f0SAl Viro s = path.mnt->mnt_sb; 2849d31da0f0SAl Viro atomic_inc(&s->s_active); 2850ea441d11SAl Viro mntput(path.mnt); 2851ea441d11SAl Viro /* lock the sucker */ 2852d31da0f0SAl Viro down_write(&s->s_umount); 2853ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2854ea441d11SAl Viro return path.dentry; 2855ea441d11SAl Viro } 2856ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2857ea441d11SAl Viro 2858bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2859bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 28601da177e4SLinus Torvalds { 2861eca6f534SVegard Nossum int ret; 2862eca6f534SVegard Nossum char *kernel_type; 2863eca6f534SVegard Nossum char *kernel_dev; 28641da177e4SLinus Torvalds unsigned long data_page; 28651da177e4SLinus Torvalds 2866b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 2867b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 2868b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 2869eca6f534SVegard Nossum goto out_type; 28701da177e4SLinus Torvalds 2871b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 2872b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 2873b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 2874eca6f534SVegard Nossum goto out_dev; 28751da177e4SLinus Torvalds 2876eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2877eca6f534SVegard Nossum if (ret < 0) 2878eca6f534SVegard Nossum goto out_data; 28791da177e4SLinus Torvalds 28805e6123f3SSeunghun Lee ret = do_mount(kernel_dev, dir_name, kernel_type, flags, 2881eca6f534SVegard Nossum (void *) data_page); 2882eca6f534SVegard Nossum 28831da177e4SLinus Torvalds free_page(data_page); 2884eca6f534SVegard Nossum out_data: 2885eca6f534SVegard Nossum kfree(kernel_dev); 2886eca6f534SVegard Nossum out_dev: 2887eca6f534SVegard Nossum kfree(kernel_type); 2888eca6f534SVegard Nossum out_type: 2889eca6f534SVegard Nossum return ret; 28901da177e4SLinus Torvalds } 28911da177e4SLinus Torvalds 28921da177e4SLinus Torvalds /* 2893afac7cbaSAl Viro * Return true if path is reachable from root 2894afac7cbaSAl Viro * 289548a066e7SAl Viro * namespace_sem or mount_lock is held 2896afac7cbaSAl Viro */ 2897643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2898afac7cbaSAl Viro const struct path *root) 2899afac7cbaSAl Viro { 2900643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2901a73324daSAl Viro dentry = mnt->mnt_mountpoint; 29020714a533SAl Viro mnt = mnt->mnt_parent; 2903afac7cbaSAl Viro } 2904643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2905afac7cbaSAl Viro } 2906afac7cbaSAl Viro 2907afac7cbaSAl Viro int path_is_under(struct path *path1, struct path *path2) 2908afac7cbaSAl Viro { 2909afac7cbaSAl Viro int res; 291048a066e7SAl Viro read_seqlock_excl(&mount_lock); 2911643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 291248a066e7SAl Viro read_sequnlock_excl(&mount_lock); 2913afac7cbaSAl Viro return res; 2914afac7cbaSAl Viro } 2915afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 2916afac7cbaSAl Viro 2917afac7cbaSAl Viro /* 29181da177e4SLinus Torvalds * pivot_root Semantics: 29191da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 29201da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 29211da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 29221da177e4SLinus Torvalds * 29231da177e4SLinus Torvalds * Restrictions: 29241da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 29251da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 29261da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 29271da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 29281da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 29291da177e4SLinus Torvalds * 29304a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 29314a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 29324a0d11faSNeil Brown * in this situation. 29334a0d11faSNeil Brown * 29341da177e4SLinus Torvalds * Notes: 29351da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 29361da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 29371da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 29381da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 29391da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 29401da177e4SLinus Torvalds * first. 29411da177e4SLinus Torvalds */ 29423480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 29433480b257SHeiko Carstens const char __user *, put_old) 29441da177e4SLinus Torvalds { 29452d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 294684d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 294784d17192SAl Viro struct mountpoint *old_mp, *root_mp; 29481da177e4SLinus Torvalds int error; 29491da177e4SLinus Torvalds 29509b40bc90SAl Viro if (!may_mount()) 29511da177e4SLinus Torvalds return -EPERM; 29521da177e4SLinus Torvalds 29532d8f3038SAl Viro error = user_path_dir(new_root, &new); 29541da177e4SLinus Torvalds if (error) 29551da177e4SLinus Torvalds goto out0; 29561da177e4SLinus Torvalds 29572d8f3038SAl Viro error = user_path_dir(put_old, &old); 29581da177e4SLinus Torvalds if (error) 29591da177e4SLinus Torvalds goto out1; 29601da177e4SLinus Torvalds 29612d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 2962b12cea91SAl Viro if (error) 2963b12cea91SAl Viro goto out2; 29641da177e4SLinus Torvalds 2965f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 296684d17192SAl Viro old_mp = lock_mount(&old); 296784d17192SAl Viro error = PTR_ERR(old_mp); 296884d17192SAl Viro if (IS_ERR(old_mp)) 2969b12cea91SAl Viro goto out3; 2970b12cea91SAl Viro 29711da177e4SLinus Torvalds error = -EINVAL; 2972419148daSAl Viro new_mnt = real_mount(new.mnt); 2973419148daSAl Viro root_mnt = real_mount(root.mnt); 297484d17192SAl Viro old_mnt = real_mount(old.mnt); 297584d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 2976fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 2977fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 2978b12cea91SAl Viro goto out4; 2979143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 2980b12cea91SAl Viro goto out4; 29815ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 29825ff9d8a6SEric W. Biederman goto out4; 29831da177e4SLinus Torvalds error = -ENOENT; 2984f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 2985b12cea91SAl Viro goto out4; 29861da177e4SLinus Torvalds error = -EBUSY; 298784d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 2988b12cea91SAl Viro goto out4; /* loop, on the same file system */ 29891da177e4SLinus Torvalds error = -EINVAL; 29908c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 2991b12cea91SAl Viro goto out4; /* not a mountpoint */ 2992676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 2993b12cea91SAl Viro goto out4; /* not attached */ 299484d17192SAl Viro root_mp = root_mnt->mnt_mp; 29952d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 2996b12cea91SAl Viro goto out4; /* not a mountpoint */ 2997676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 2998b12cea91SAl Viro goto out4; /* not attached */ 29994ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 300084d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3001b12cea91SAl Viro goto out4; 30020d082601SEric W. Biederman /* make certain new is below the root */ 30030d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 30040d082601SEric W. Biederman goto out4; 300584d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3006719ea2fbSAl Viro lock_mount_hash(); 3007419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3008419148daSAl Viro detach_mnt(root_mnt, &root_parent); 30095ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 30105ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 30115ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 30125ff9d8a6SEric W. Biederman } 30134ac91378SJan Blunck /* mount old root on put_old */ 301484d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 30154ac91378SJan Blunck /* mount new_root on / */ 301684d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 30176b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 30184fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 30194fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 3020719ea2fbSAl Viro unlock_mount_hash(); 30212d8f3038SAl Viro chroot_fs_refs(&root, &new); 302284d17192SAl Viro put_mountpoint(root_mp); 30231da177e4SLinus Torvalds error = 0; 3024b12cea91SAl Viro out4: 302584d17192SAl Viro unlock_mount(old_mp); 3026b12cea91SAl Viro if (!error) { 30271a390689SAl Viro path_put(&root_parent); 30281a390689SAl Viro path_put(&parent_path); 3029b12cea91SAl Viro } 3030b12cea91SAl Viro out3: 30318c3ee42eSAl Viro path_put(&root); 3032b12cea91SAl Viro out2: 30332d8f3038SAl Viro path_put(&old); 30341da177e4SLinus Torvalds out1: 30352d8f3038SAl Viro path_put(&new); 30361da177e4SLinus Torvalds out0: 30371da177e4SLinus Torvalds return error; 30381da177e4SLinus Torvalds } 30391da177e4SLinus Torvalds 30401da177e4SLinus Torvalds static void __init init_mount_tree(void) 30411da177e4SLinus Torvalds { 30421da177e4SLinus Torvalds struct vfsmount *mnt; 30436b3286edSKirill Korotaev struct mnt_namespace *ns; 3044ac748a09SJan Blunck struct path root; 30450c55cfc4SEric W. Biederman struct file_system_type *type; 30461da177e4SLinus Torvalds 30470c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 30480c55cfc4SEric W. Biederman if (!type) 30490c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 30500c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 30510c55cfc4SEric W. Biederman put_filesystem(type); 30521da177e4SLinus Torvalds if (IS_ERR(mnt)) 30531da177e4SLinus Torvalds panic("Can't create rootfs"); 3054b3e19d92SNick Piggin 30553b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 30563b22edc5STrond Myklebust if (IS_ERR(ns)) 30571da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 30581da177e4SLinus Torvalds 30596b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 30606b3286edSKirill Korotaev get_mnt_ns(ns); 30611da177e4SLinus Torvalds 3062be08d6d2SAl Viro root.mnt = mnt; 3063be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3064da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3065ac748a09SJan Blunck 3066ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3067ac748a09SJan Blunck set_fs_root(current->fs, &root); 30681da177e4SLinus Torvalds } 30691da177e4SLinus Torvalds 307074bf17cfSDenis Cheng void __init mnt_init(void) 30711da177e4SLinus Torvalds { 307213f14b4dSEric Dumazet unsigned u; 307315a67dd8SRandy Dunlap int err; 30741da177e4SLinus Torvalds 30757d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 307620c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 30771da177e4SLinus Torvalds 30780818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 307938129a13SAl Viro sizeof(struct hlist_head), 30800818bf27SAl Viro mhash_entries, 19, 30810818bf27SAl Viro 0, 30820818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 30830818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 30840818bf27SAl Viro sizeof(struct hlist_head), 30850818bf27SAl Viro mphash_entries, 19, 30860818bf27SAl Viro 0, 30870818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 30881da177e4SLinus Torvalds 308984d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 30901da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 30911da177e4SLinus Torvalds 30920818bf27SAl Viro for (u = 0; u <= m_hash_mask; u++) 309338129a13SAl Viro INIT_HLIST_HEAD(&mount_hashtable[u]); 30940818bf27SAl Viro for (u = 0; u <= mp_hash_mask; u++) 30950818bf27SAl Viro INIT_HLIST_HEAD(&mountpoint_hashtable[u]); 30961da177e4SLinus Torvalds 30974b93dc9bSTejun Heo kernfs_init(); 30984b93dc9bSTejun Heo 309915a67dd8SRandy Dunlap err = sysfs_init(); 310015a67dd8SRandy Dunlap if (err) 310115a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 31028e24eea7SHarvey Harrison __func__, err); 310300d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 310400d26666SGreg Kroah-Hartman if (!fs_kobj) 31058e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 31061da177e4SLinus Torvalds init_rootfs(); 31071da177e4SLinus Torvalds init_mount_tree(); 31081da177e4SLinus Torvalds } 31091da177e4SLinus Torvalds 3110616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 31111da177e4SLinus Torvalds { 3112d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3113616511d0STrond Myklebust return; 31147b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3115771b1371SEric W. Biederman free_mnt_ns(ns); 31161da177e4SLinus Torvalds } 31179d412a43SAl Viro 31189d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 31199d412a43SAl Viro { 3120423e0ab0STim Chen struct vfsmount *mnt; 3121423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 3122423e0ab0STim Chen if (!IS_ERR(mnt)) { 3123423e0ab0STim Chen /* 3124423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3125423e0ab0STim Chen * we unmount before file sys is unregistered 3126423e0ab0STim Chen */ 3127f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3128423e0ab0STim Chen } 3129423e0ab0STim Chen return mnt; 31309d412a43SAl Viro } 31319d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3132423e0ab0STim Chen 3133423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3134423e0ab0STim Chen { 3135423e0ab0STim Chen /* release long term mount so mount point can be released */ 3136423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3137f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 313848a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3139423e0ab0STim Chen mntput(mnt); 3140423e0ab0STim Chen } 3141423e0ab0STim Chen } 3142423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 314302125a82SAl Viro 314402125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 314502125a82SAl Viro { 3146143c8c91SAl Viro return check_mnt(real_mount(mnt)); 314702125a82SAl Viro } 31488823c079SEric W. Biederman 31493151527eSEric W. Biederman bool current_chrooted(void) 31503151527eSEric W. Biederman { 31513151527eSEric W. Biederman /* Does the current process have a non-standard root */ 31523151527eSEric W. Biederman struct path ns_root; 31533151527eSEric W. Biederman struct path fs_root; 31543151527eSEric W. Biederman bool chrooted; 31553151527eSEric W. Biederman 31563151527eSEric W. Biederman /* Find the namespace root */ 31573151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 31583151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 31593151527eSEric W. Biederman path_get(&ns_root); 31603151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 31613151527eSEric W. Biederman ; 31623151527eSEric W. Biederman 31633151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 31643151527eSEric W. Biederman 31653151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 31663151527eSEric W. Biederman 31673151527eSEric W. Biederman path_put(&fs_root); 31683151527eSEric W. Biederman path_put(&ns_root); 31693151527eSEric W. Biederman 31703151527eSEric W. Biederman return chrooted; 31713151527eSEric W. Biederman } 31723151527eSEric W. Biederman 31738c6cf9ccSEric W. Biederman static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags) 317487a8ebd6SEric W. Biederman { 317587a8ebd6SEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 31768c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 317787a8ebd6SEric W. Biederman struct mount *mnt; 3178e51db735SEric W. Biederman bool visible = false; 317987a8ebd6SEric W. Biederman 3180e51db735SEric W. Biederman if (unlikely(!ns)) 3181e51db735SEric W. Biederman return false; 3182e51db735SEric W. Biederman 318344bb4385SAl Viro down_read(&namespace_sem); 318487a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3185e51db735SEric W. Biederman struct mount *child; 3186e51db735SEric W. Biederman if (mnt->mnt.mnt_sb->s_type != type) 3187e51db735SEric W. Biederman continue; 3188e51db735SEric W. Biederman 31897e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 31907e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 31917e96c1b0SEric W. Biederman */ 31927e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 31937e96c1b0SEric W. Biederman continue; 31947e96c1b0SEric W. Biederman 31958c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 31968c6cf9ccSEric W. Biederman * than the proposed new mount. 31978c6cf9ccSEric W. Biederman */ 31988c6cf9ccSEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 31998c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 32008c6cf9ccSEric W. Biederman continue; 32018c6cf9ccSEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 32028c6cf9ccSEric W. Biederman !(new_flags & MNT_NODEV)) 32038c6cf9ccSEric W. Biederman continue; 32048c6cf9ccSEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 32058c6cf9ccSEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 32068c6cf9ccSEric W. Biederman continue; 32078c6cf9ccSEric W. Biederman 3208e51db735SEric W. Biederman /* This mount is not fully visible if there are any child mounts 3209e51db735SEric W. Biederman * that cover anything except for empty directories. 3210e51db735SEric W. Biederman */ 3211e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3212e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3213e51db735SEric W. Biederman if (!S_ISDIR(inode->i_mode)) 3214e51db735SEric W. Biederman goto next; 321541301ae7SEric W. Biederman if (inode->i_nlink > 2) 3216e51db735SEric W. Biederman goto next; 321787a8ebd6SEric W. Biederman } 32188c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 32198c6cf9ccSEric W. Biederman *new_mnt_flags |= mnt->mnt.mnt_flags & (MNT_LOCK_READONLY | \ 32208c6cf9ccSEric W. Biederman MNT_LOCK_NODEV | \ 32218c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3222e51db735SEric W. Biederman visible = true; 3223e51db735SEric W. Biederman goto found; 3224e51db735SEric W. Biederman next: ; 322587a8ebd6SEric W. Biederman } 3226e51db735SEric W. Biederman found: 322744bb4385SAl Viro up_read(&namespace_sem); 3228e51db735SEric W. Biederman return visible; 322987a8ebd6SEric W. Biederman } 323087a8ebd6SEric W. Biederman 323164964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 32328823c079SEric W. Biederman { 323358be2825SAl Viro struct ns_common *ns = NULL; 32348823c079SEric W. Biederman struct nsproxy *nsproxy; 32358823c079SEric W. Biederman 3236728dba3aSEric W. Biederman task_lock(task); 3237728dba3aSEric W. Biederman nsproxy = task->nsproxy; 32388823c079SEric W. Biederman if (nsproxy) { 323958be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 324058be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 32418823c079SEric W. Biederman } 3242728dba3aSEric W. Biederman task_unlock(task); 32438823c079SEric W. Biederman 32448823c079SEric W. Biederman return ns; 32458823c079SEric W. Biederman } 32468823c079SEric W. Biederman 324764964528SAl Viro static void mntns_put(struct ns_common *ns) 32488823c079SEric W. Biederman { 324958be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 32508823c079SEric W. Biederman } 32518823c079SEric W. Biederman 325264964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 32538823c079SEric W. Biederman { 32548823c079SEric W. Biederman struct fs_struct *fs = current->fs; 325558be2825SAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns); 32568823c079SEric W. Biederman struct path root; 32578823c079SEric W. Biederman 32580c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3259c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3260c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3261ae11e0f1SZhao Hongjiang return -EPERM; 32628823c079SEric W. Biederman 32638823c079SEric W. Biederman if (fs->users != 1) 32648823c079SEric W. Biederman return -EINVAL; 32658823c079SEric W. Biederman 32668823c079SEric W. Biederman get_mnt_ns(mnt_ns); 32678823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 32688823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 32698823c079SEric W. Biederman 32708823c079SEric W. Biederman /* Find the root */ 32718823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 32728823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 32738823c079SEric W. Biederman path_get(&root); 32748823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 32758823c079SEric W. Biederman ; 32768823c079SEric W. Biederman 32778823c079SEric W. Biederman /* Update the pwd and root */ 32788823c079SEric W. Biederman set_fs_pwd(fs, &root); 32798823c079SEric W. Biederman set_fs_root(fs, &root); 32808823c079SEric W. Biederman 32818823c079SEric W. Biederman path_put(&root); 32828823c079SEric W. Biederman return 0; 32838823c079SEric W. Biederman } 32848823c079SEric W. Biederman 32858823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 32868823c079SEric W. Biederman .name = "mnt", 32878823c079SEric W. Biederman .type = CLONE_NEWNS, 32888823c079SEric W. Biederman .get = mntns_get, 32898823c079SEric W. Biederman .put = mntns_put, 32908823c079SEric W. Biederman .install = mntns_install, 32918823c079SEric W. Biederman }; 3292