11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 1873cd49ecSMiklos Szeredi #include <linux/idr.h> 1957f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 20d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 21d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 22d10577a8SAl Viro #include <linux/uaccess.h> 230bb80f24SDavid Howells #include <linux/proc_ns.h> 2420b4fb48SLinus Torvalds #include <linux/magic.h> 250818bf27SAl Viro #include <linux/bootmem.h> 269ea459e1SAl Viro #include <linux/task_work.h> 2707b20889SRam Pai #include "pnode.h" 28948730b0SAdrian Bunk #include "internal.h" 291da177e4SLinus Torvalds 300818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 310818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 320818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 330818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 340818bf27SAl Viro 350818bf27SAl Viro static __initdata unsigned long mhash_entries; 360818bf27SAl Viro static int __init set_mhash_entries(char *str) 370818bf27SAl Viro { 380818bf27SAl Viro if (!str) 390818bf27SAl Viro return 0; 400818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 410818bf27SAl Viro return 1; 420818bf27SAl Viro } 430818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 440818bf27SAl Viro 450818bf27SAl Viro static __initdata unsigned long mphash_entries; 460818bf27SAl Viro static int __init set_mphash_entries(char *str) 470818bf27SAl Viro { 480818bf27SAl Viro if (!str) 490818bf27SAl Viro return 0; 500818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 510818bf27SAl Viro return 1; 520818bf27SAl Viro } 530818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 5413f14b4dSEric Dumazet 55c7999c36SAl Viro static u64 event; 5673cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 57719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 5899b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 59f21f6220SAl Viro static int mnt_id_start = 0; 60f21f6220SAl Viro static int mnt_group_start = 1; 615addc5ddSAl Viro 6238129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 630818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 64e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 6559aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 661da177e4SLinus Torvalds 67f87fd4c2SMiklos Szeredi /* /sys/fs */ 6800d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 6900d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 70f87fd4c2SMiklos Szeredi 7199b7db7bSNick Piggin /* 7299b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7399b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 7499b7db7bSNick Piggin * up the tree. 7599b7db7bSNick Piggin * 7699b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 7799b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 7899b7db7bSNick Piggin */ 7948a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8099b7db7bSNick Piggin 8138129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 821da177e4SLinus Torvalds { 831da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 841da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 850818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 860818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 870818bf27SAl Viro } 880818bf27SAl Viro 890818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 900818bf27SAl Viro { 910818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 920818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 930818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 941da177e4SLinus Torvalds } 951da177e4SLinus Torvalds 9699b7db7bSNick Piggin /* 9799b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 9899b7db7bSNick Piggin * serialize with freeing. 9999b7db7bSNick Piggin */ 100b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10173cd49ecSMiklos Szeredi { 10273cd49ecSMiklos Szeredi int res; 10373cd49ecSMiklos Szeredi 10473cd49ecSMiklos Szeredi retry: 10573cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 10699b7db7bSNick Piggin spin_lock(&mnt_id_lock); 10715169fe7SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 108f21f6220SAl Viro if (!res) 10915169fe7SAl Viro mnt_id_start = mnt->mnt_id + 1; 11099b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 11173cd49ecSMiklos Szeredi if (res == -EAGAIN) 11273cd49ecSMiklos Szeredi goto retry; 11373cd49ecSMiklos Szeredi 11473cd49ecSMiklos Szeredi return res; 11573cd49ecSMiklos Szeredi } 11673cd49ecSMiklos Szeredi 117b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11873cd49ecSMiklos Szeredi { 11915169fe7SAl Viro int id = mnt->mnt_id; 12099b7db7bSNick Piggin spin_lock(&mnt_id_lock); 121f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 122f21f6220SAl Viro if (mnt_id_start > id) 123f21f6220SAl Viro mnt_id_start = id; 12499b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 12573cd49ecSMiklos Szeredi } 12673cd49ecSMiklos Szeredi 127719f5d7fSMiklos Szeredi /* 128719f5d7fSMiklos Szeredi * Allocate a new peer group ID 129719f5d7fSMiklos Szeredi * 130719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 131719f5d7fSMiklos Szeredi */ 1324b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 133719f5d7fSMiklos Szeredi { 134f21f6220SAl Viro int res; 135f21f6220SAl Viro 136719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 137719f5d7fSMiklos Szeredi return -ENOMEM; 138719f5d7fSMiklos Szeredi 139f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 140f21f6220SAl Viro mnt_group_start, 14115169fe7SAl Viro &mnt->mnt_group_id); 142f21f6220SAl Viro if (!res) 14315169fe7SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 144f21f6220SAl Viro 145f21f6220SAl Viro return res; 146719f5d7fSMiklos Szeredi } 147719f5d7fSMiklos Szeredi 148719f5d7fSMiklos Szeredi /* 149719f5d7fSMiklos Szeredi * Release a peer group ID 150719f5d7fSMiklos Szeredi */ 1514b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 152719f5d7fSMiklos Szeredi { 15315169fe7SAl Viro int id = mnt->mnt_group_id; 154f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 155f21f6220SAl Viro if (mnt_group_start > id) 156f21f6220SAl Viro mnt_group_start = id; 15715169fe7SAl Viro mnt->mnt_group_id = 0; 158719f5d7fSMiklos Szeredi } 159719f5d7fSMiklos Szeredi 160b3e19d92SNick Piggin /* 161b3e19d92SNick Piggin * vfsmount lock must be held for read 162b3e19d92SNick Piggin */ 16383adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 164b3e19d92SNick Piggin { 165b3e19d92SNick Piggin #ifdef CONFIG_SMP 16668e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 167b3e19d92SNick Piggin #else 168b3e19d92SNick Piggin preempt_disable(); 16968e8a9feSAl Viro mnt->mnt_count += n; 170b3e19d92SNick Piggin preempt_enable(); 171b3e19d92SNick Piggin #endif 172b3e19d92SNick Piggin } 173b3e19d92SNick Piggin 174b3e19d92SNick Piggin /* 175b3e19d92SNick Piggin * vfsmount lock must be held for write 176b3e19d92SNick Piggin */ 17783adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 178b3e19d92SNick Piggin { 179b3e19d92SNick Piggin #ifdef CONFIG_SMP 180f03c6599SAl Viro unsigned int count = 0; 181b3e19d92SNick Piggin int cpu; 182b3e19d92SNick Piggin 183b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 18468e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 185b3e19d92SNick Piggin } 186b3e19d92SNick Piggin 187b3e19d92SNick Piggin return count; 188b3e19d92SNick Piggin #else 18968e8a9feSAl Viro return mnt->mnt_count; 190b3e19d92SNick Piggin #endif 191b3e19d92SNick Piggin } 192b3e19d92SNick Piggin 19387b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 19487b95ce0SAl Viro { 19587b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 19687b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 19787b95ce0SAl Viro pin_remove(p); 19887b95ce0SAl Viro mntput(&m->mnt); 19987b95ce0SAl Viro } 20087b95ce0SAl Viro 201b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 2021da177e4SLinus Torvalds { 203c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 204c63181e6SAl Viro if (mnt) { 20573cd49ecSMiklos Szeredi int err; 20673cd49ecSMiklos Szeredi 207c63181e6SAl Viro err = mnt_alloc_id(mnt); 20888b38782SLi Zefan if (err) 20988b38782SLi Zefan goto out_free_cache; 21088b38782SLi Zefan 21188b38782SLi Zefan if (name) { 212fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 213c63181e6SAl Viro if (!mnt->mnt_devname) 21488b38782SLi Zefan goto out_free_id; 21573cd49ecSMiklos Szeredi } 21673cd49ecSMiklos Szeredi 217b3e19d92SNick Piggin #ifdef CONFIG_SMP 218c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 219c63181e6SAl Viro if (!mnt->mnt_pcp) 220b3e19d92SNick Piggin goto out_free_devname; 221b3e19d92SNick Piggin 222c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 223b3e19d92SNick Piggin #else 224c63181e6SAl Viro mnt->mnt_count = 1; 225c63181e6SAl Viro mnt->mnt_writers = 0; 226b3e19d92SNick Piggin #endif 227b3e19d92SNick Piggin 22838129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 229c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 230c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 231c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 232c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 233c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 234c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 235c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2360a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 2372504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 2382504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 2392504c5d6SAndreas Gruenbacher #endif 24087b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2411da177e4SLinus Torvalds } 242c63181e6SAl Viro return mnt; 24388b38782SLi Zefan 244d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 245d3ef3d73Snpiggin@suse.de out_free_devname: 246fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 247d3ef3d73Snpiggin@suse.de #endif 24888b38782SLi Zefan out_free_id: 249c63181e6SAl Viro mnt_free_id(mnt); 25088b38782SLi Zefan out_free_cache: 251c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 25288b38782SLi Zefan return NULL; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds 2558366025eSDave Hansen /* 2568366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2578366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2588366025eSDave Hansen * We must keep track of when those operations start 2598366025eSDave Hansen * (for permission checks) and when they end, so that 2608366025eSDave Hansen * we can determine when writes are able to occur to 2618366025eSDave Hansen * a filesystem. 2628366025eSDave Hansen */ 2633d733633SDave Hansen /* 2643d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2653d733633SDave Hansen * @mnt: the mount to check for its write status 2663d733633SDave Hansen * 2673d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2683d733633SDave Hansen * It does not guarantee that the filesystem will stay 2693d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2703d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2713d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2723d733633SDave Hansen * r/w. 2733d733633SDave Hansen */ 2743d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2753d733633SDave Hansen { 2762e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2772e4b7fcdSDave Hansen return 1; 2782e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2792e4b7fcdSDave Hansen return 1; 2802e4b7fcdSDave Hansen return 0; 2813d733633SDave Hansen } 2823d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2833d733633SDave Hansen 28483adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2853d733633SDave Hansen { 286d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 28768e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 288d3ef3d73Snpiggin@suse.de #else 28968e8a9feSAl Viro mnt->mnt_writers++; 290d3ef3d73Snpiggin@suse.de #endif 2913d733633SDave Hansen } 2923d733633SDave Hansen 29383adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2943d733633SDave Hansen { 295d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 29668e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 297d3ef3d73Snpiggin@suse.de #else 29868e8a9feSAl Viro mnt->mnt_writers--; 299d3ef3d73Snpiggin@suse.de #endif 300d3ef3d73Snpiggin@suse.de } 301d3ef3d73Snpiggin@suse.de 30283adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 303d3ef3d73Snpiggin@suse.de { 304d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 305d3ef3d73Snpiggin@suse.de unsigned int count = 0; 3063d733633SDave Hansen int cpu; 3073d733633SDave Hansen 3083d733633SDave Hansen for_each_possible_cpu(cpu) { 30968e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 3103d733633SDave Hansen } 3113d733633SDave Hansen 312d3ef3d73Snpiggin@suse.de return count; 313d3ef3d73Snpiggin@suse.de #else 314d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 315d3ef3d73Snpiggin@suse.de #endif 3163d733633SDave Hansen } 3173d733633SDave Hansen 3184ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3194ed5e82fSMiklos Szeredi { 3204ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3214ed5e82fSMiklos Szeredi return 1; 3224ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3234ed5e82fSMiklos Szeredi smp_rmb(); 3244ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3254ed5e82fSMiklos Szeredi } 3264ed5e82fSMiklos Szeredi 3273d733633SDave Hansen /* 328eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 329eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 330eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 331eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3323d733633SDave Hansen */ 3338366025eSDave Hansen /** 334eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 33583adc753SAl Viro * @m: the mount on which to take a write 3368366025eSDave Hansen * 337eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 338eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 339eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 340eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 341eb04c282SJan Kara * called. This is effectively a refcount. 3428366025eSDave Hansen */ 343eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3448366025eSDave Hansen { 34583adc753SAl Viro struct mount *mnt = real_mount(m); 3463d733633SDave Hansen int ret = 0; 3473d733633SDave Hansen 348d3ef3d73Snpiggin@suse.de preempt_disable(); 349c6653a83SNick Piggin mnt_inc_writers(mnt); 350d3ef3d73Snpiggin@suse.de /* 351c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 352d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 353d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 354d3ef3d73Snpiggin@suse.de */ 355d3ef3d73Snpiggin@suse.de smp_mb(); 3561e75529eSMiao Xie while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 357d3ef3d73Snpiggin@suse.de cpu_relax(); 358d3ef3d73Snpiggin@suse.de /* 359d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 360d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 361d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 362d3ef3d73Snpiggin@suse.de */ 363d3ef3d73Snpiggin@suse.de smp_rmb(); 3644ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 365c6653a83SNick Piggin mnt_dec_writers(mnt); 3663d733633SDave Hansen ret = -EROFS; 3673d733633SDave Hansen } 368d3ef3d73Snpiggin@suse.de preempt_enable(); 369eb04c282SJan Kara 370eb04c282SJan Kara return ret; 371eb04c282SJan Kara } 372eb04c282SJan Kara 373eb04c282SJan Kara /** 374eb04c282SJan Kara * mnt_want_write - get write access to a mount 375eb04c282SJan Kara * @m: the mount on which to take a write 376eb04c282SJan Kara * 377eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 378eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 379eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 380eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 381eb04c282SJan Kara */ 382eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 383eb04c282SJan Kara { 384eb04c282SJan Kara int ret; 385eb04c282SJan Kara 386eb04c282SJan Kara sb_start_write(m->mnt_sb); 387eb04c282SJan Kara ret = __mnt_want_write(m); 388eb04c282SJan Kara if (ret) 389eb04c282SJan Kara sb_end_write(m->mnt_sb); 3903d733633SDave Hansen return ret; 3918366025eSDave Hansen } 3928366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3938366025eSDave Hansen 3948366025eSDave Hansen /** 39596029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 39696029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 39796029c4eSnpiggin@suse.de * 39896029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 39996029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 40096029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 40196029c4eSnpiggin@suse.de * on it. This allows some optimisation. 40296029c4eSnpiggin@suse.de * 40396029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 40496029c4eSnpiggin@suse.de * drop the reference. 40596029c4eSnpiggin@suse.de */ 40696029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 40796029c4eSnpiggin@suse.de { 40896029c4eSnpiggin@suse.de /* superblock may be r/o */ 40996029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 41096029c4eSnpiggin@suse.de return -EROFS; 41196029c4eSnpiggin@suse.de preempt_disable(); 41283adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 41396029c4eSnpiggin@suse.de preempt_enable(); 41496029c4eSnpiggin@suse.de return 0; 41596029c4eSnpiggin@suse.de } 41696029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 41796029c4eSnpiggin@suse.de 41896029c4eSnpiggin@suse.de /** 419eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 420eb04c282SJan Kara * @file: the file who's mount on which to take a write 421eb04c282SJan Kara * 422eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 423eb04c282SJan Kara * do some optimisations if the file is open for write already 424eb04c282SJan Kara */ 425eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 426eb04c282SJan Kara { 42783f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 428eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 429eb04c282SJan Kara else 430eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 431eb04c282SJan Kara } 432eb04c282SJan Kara 433eb04c282SJan Kara /** 43496029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 43596029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 43696029c4eSnpiggin@suse.de * 43796029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 43896029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 43996029c4eSnpiggin@suse.de */ 44096029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 44196029c4eSnpiggin@suse.de { 442eb04c282SJan Kara int ret; 443eb04c282SJan Kara 444eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 445eb04c282SJan Kara ret = __mnt_want_write_file(file); 446eb04c282SJan Kara if (ret) 447eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 448eb04c282SJan Kara return ret; 44996029c4eSnpiggin@suse.de } 45096029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 45196029c4eSnpiggin@suse.de 45296029c4eSnpiggin@suse.de /** 453eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4548366025eSDave Hansen * @mnt: the mount on which to give up write access 4558366025eSDave Hansen * 4568366025eSDave Hansen * Tells the low-level filesystem that we are done 4578366025eSDave Hansen * performing writes to it. Must be matched with 458eb04c282SJan Kara * __mnt_want_write() call above. 4598366025eSDave Hansen */ 460eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4618366025eSDave Hansen { 462d3ef3d73Snpiggin@suse.de preempt_disable(); 46383adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 464d3ef3d73Snpiggin@suse.de preempt_enable(); 4658366025eSDave Hansen } 466eb04c282SJan Kara 467eb04c282SJan Kara /** 468eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 469eb04c282SJan Kara * @mnt: the mount on which to give up write access 470eb04c282SJan Kara * 471eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 472eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 473eb04c282SJan Kara * mnt_want_write() call above. 474eb04c282SJan Kara */ 475eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 476eb04c282SJan Kara { 477eb04c282SJan Kara __mnt_drop_write(mnt); 478eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 479eb04c282SJan Kara } 4808366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4818366025eSDave Hansen 482eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 483eb04c282SJan Kara { 484eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 485eb04c282SJan Kara } 486eb04c282SJan Kara 4872a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4882a79f17eSAl Viro { 4892a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4902a79f17eSAl Viro } 4912a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4922a79f17eSAl Viro 49383adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4948366025eSDave Hansen { 4953d733633SDave Hansen int ret = 0; 4963d733633SDave Hansen 497719ea2fbSAl Viro lock_mount_hash(); 49883adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 499d3ef3d73Snpiggin@suse.de /* 500d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 501d3ef3d73Snpiggin@suse.de * should be visible before we do. 502d3ef3d73Snpiggin@suse.de */ 503d3ef3d73Snpiggin@suse.de smp_mb(); 504d3ef3d73Snpiggin@suse.de 505d3ef3d73Snpiggin@suse.de /* 506d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 507d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 508d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 509d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 510d3ef3d73Snpiggin@suse.de * 511d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 512d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 513d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 514d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 515d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 516d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 517d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 518d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 519d3ef3d73Snpiggin@suse.de * we're counting up here. 520d3ef3d73Snpiggin@suse.de */ 521c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 522d3ef3d73Snpiggin@suse.de ret = -EBUSY; 523d3ef3d73Snpiggin@suse.de else 52483adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 525d3ef3d73Snpiggin@suse.de /* 526d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 527d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 528d3ef3d73Snpiggin@suse.de */ 529d3ef3d73Snpiggin@suse.de smp_wmb(); 53083adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 531719ea2fbSAl Viro unlock_mount_hash(); 5323d733633SDave Hansen return ret; 5333d733633SDave Hansen } 5348366025eSDave Hansen 53583adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5362e4b7fcdSDave Hansen { 537719ea2fbSAl Viro lock_mount_hash(); 53883adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 539719ea2fbSAl Viro unlock_mount_hash(); 5402e4b7fcdSDave Hansen } 5412e4b7fcdSDave Hansen 5424ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5434ed5e82fSMiklos Szeredi { 5444ed5e82fSMiklos Szeredi struct mount *mnt; 5454ed5e82fSMiklos Szeredi int err = 0; 5464ed5e82fSMiklos Szeredi 5478e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5488e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5498e8b8796SMiklos Szeredi return -EBUSY; 5508e8b8796SMiklos Szeredi 551719ea2fbSAl Viro lock_mount_hash(); 5524ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5534ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5544ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5554ed5e82fSMiklos Szeredi smp_mb(); 5564ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5574ed5e82fSMiklos Szeredi err = -EBUSY; 5584ed5e82fSMiklos Szeredi break; 5594ed5e82fSMiklos Szeredi } 5604ed5e82fSMiklos Szeredi } 5614ed5e82fSMiklos Szeredi } 5628e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5638e8b8796SMiklos Szeredi err = -EBUSY; 5648e8b8796SMiklos Szeredi 5654ed5e82fSMiklos Szeredi if (!err) { 5664ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5674ed5e82fSMiklos Szeredi smp_wmb(); 5684ed5e82fSMiklos Szeredi } 5694ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5704ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5714ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5724ed5e82fSMiklos Szeredi } 573719ea2fbSAl Viro unlock_mount_hash(); 5744ed5e82fSMiklos Szeredi 5754ed5e82fSMiklos Szeredi return err; 5764ed5e82fSMiklos Szeredi } 5774ed5e82fSMiklos Szeredi 578b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5791da177e4SLinus Torvalds { 580fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 581d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 58268e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 583d3ef3d73Snpiggin@suse.de #endif 584b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5851da177e4SLinus Torvalds } 5861da177e4SLinus Torvalds 5878ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5888ffcb32eSDavid Howells { 5898ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5908ffcb32eSDavid Howells } 5918ffcb32eSDavid Howells 59248a066e7SAl Viro /* call under rcu_read_lock */ 593294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 59448a066e7SAl Viro { 59548a066e7SAl Viro struct mount *mnt; 59648a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 597294d71ffSAl Viro return 1; 59848a066e7SAl Viro if (bastard == NULL) 599294d71ffSAl Viro return 0; 60048a066e7SAl Viro mnt = real_mount(bastard); 60148a066e7SAl Viro mnt_add_count(mnt, 1); 60248a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 603294d71ffSAl Viro return 0; 60448a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 60548a066e7SAl Viro mnt_add_count(mnt, -1); 606294d71ffSAl Viro return 1; 60748a066e7SAl Viro } 608294d71ffSAl Viro return -1; 609294d71ffSAl Viro } 610294d71ffSAl Viro 611294d71ffSAl Viro /* call under rcu_read_lock */ 612294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 613294d71ffSAl Viro { 614294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 615294d71ffSAl Viro if (likely(!res)) 616294d71ffSAl Viro return true; 617294d71ffSAl Viro if (unlikely(res < 0)) { 61848a066e7SAl Viro rcu_read_unlock(); 61948a066e7SAl Viro mntput(bastard); 62048a066e7SAl Viro rcu_read_lock(); 621294d71ffSAl Viro } 62248a066e7SAl Viro return false; 62348a066e7SAl Viro } 62448a066e7SAl Viro 6251da177e4SLinus Torvalds /* 626474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 62748a066e7SAl Viro * call under rcu_read_lock() 6281da177e4SLinus Torvalds */ 629474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6301da177e4SLinus Torvalds { 63138129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 632474279dcSAl Viro struct mount *p; 6331da177e4SLinus Torvalds 63438129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 635474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 636474279dcSAl Viro return p; 637474279dcSAl Viro return NULL; 6381da177e4SLinus Torvalds } 639474279dcSAl Viro 640474279dcSAl Viro /* 641474279dcSAl Viro * find the last mount at @dentry on vfsmount @mnt. 64248a066e7SAl Viro * mount_lock must be held. 643474279dcSAl Viro */ 644474279dcSAl Viro struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) 645474279dcSAl Viro { 646411a938bSEric W. Biederman struct mount *p, *res = NULL; 647411a938bSEric W. Biederman p = __lookup_mnt(mnt, dentry); 64838129a13SAl Viro if (!p) 64938129a13SAl Viro goto out; 650411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 651411a938bSEric W. Biederman res = p; 65238129a13SAl Viro hlist_for_each_entry_continue(p, mnt_hash) { 6531d6a32acSAl Viro if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) 6541d6a32acSAl Viro break; 655411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 6561d6a32acSAl Viro res = p; 6571d6a32acSAl Viro } 65838129a13SAl Viro out: 6591d6a32acSAl Viro return res; 6601da177e4SLinus Torvalds } 6611da177e4SLinus Torvalds 662a05964f3SRam Pai /* 663f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 664f015f126SDavid Howells * 665f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 666f015f126SDavid Howells * following mounts: 667f015f126SDavid Howells * 668f015f126SDavid Howells * mount /dev/sda1 /mnt 669f015f126SDavid Howells * mount /dev/sda2 /mnt 670f015f126SDavid Howells * mount /dev/sda3 /mnt 671f015f126SDavid Howells * 672f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 673f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 674f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 675f015f126SDavid Howells * 676f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 677a05964f3SRam Pai */ 6781c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 679a05964f3SRam Pai { 680c7105365SAl Viro struct mount *child_mnt; 68148a066e7SAl Viro struct vfsmount *m; 68248a066e7SAl Viro unsigned seq; 68399b7db7bSNick Piggin 68448a066e7SAl Viro rcu_read_lock(); 68548a066e7SAl Viro do { 68648a066e7SAl Viro seq = read_seqbegin(&mount_lock); 687474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 68848a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 68948a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 69048a066e7SAl Viro rcu_read_unlock(); 69148a066e7SAl Viro return m; 692a05964f3SRam Pai } 693a05964f3SRam Pai 6947af1364fSEric W. Biederman /* 6957af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6967af1364fSEric W. Biederman * current mount namespace. 6977af1364fSEric W. Biederman * 6987af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6997af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 7007af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 7017af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 7027af1364fSEric W. Biederman * is a mountpoint. 7037af1364fSEric W. Biederman * 7047af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 7057af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 7067af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 7077af1364fSEric W. Biederman * parent mount. 7087af1364fSEric W. Biederman */ 7097af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 7107af1364fSEric W. Biederman { 7117af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7127af1364fSEric W. Biederman struct mount *mnt; 7137af1364fSEric W. Biederman bool is_covered = false; 7147af1364fSEric W. Biederman 7157af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 7167af1364fSEric W. Biederman goto out; 7177af1364fSEric W. Biederman 7187af1364fSEric W. Biederman down_read(&namespace_sem); 7197af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7207af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7217af1364fSEric W. Biederman if (is_covered) 7227af1364fSEric W. Biederman break; 7237af1364fSEric W. Biederman } 7247af1364fSEric W. Biederman up_read(&namespace_sem); 7257af1364fSEric W. Biederman out: 7267af1364fSEric W. Biederman return is_covered; 7277af1364fSEric W. Biederman } 7287af1364fSEric W. Biederman 729e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 73084d17192SAl Viro { 7310818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 73284d17192SAl Viro struct mountpoint *mp; 73384d17192SAl Viro 7340818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 73584d17192SAl Viro if (mp->m_dentry == dentry) { 73684d17192SAl Viro /* might be worth a WARN_ON() */ 73784d17192SAl Viro if (d_unlinked(dentry)) 73884d17192SAl Viro return ERR_PTR(-ENOENT); 73984d17192SAl Viro mp->m_count++; 74084d17192SAl Viro return mp; 74184d17192SAl Viro } 74284d17192SAl Viro } 743e2dfa935SEric W. Biederman return NULL; 744e2dfa935SEric W. Biederman } 745e2dfa935SEric W. Biederman 746e2dfa935SEric W. Biederman static struct mountpoint *new_mountpoint(struct dentry *dentry) 747e2dfa935SEric W. Biederman { 748e2dfa935SEric W. Biederman struct hlist_head *chain = mp_hash(dentry); 749e2dfa935SEric W. Biederman struct mountpoint *mp; 750e2dfa935SEric W. Biederman int ret; 75184d17192SAl Viro 75284d17192SAl Viro mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 75384d17192SAl Viro if (!mp) 75484d17192SAl Viro return ERR_PTR(-ENOMEM); 75584d17192SAl Viro 756eed81007SMiklos Szeredi ret = d_set_mounted(dentry); 757eed81007SMiklos Szeredi if (ret) { 75884d17192SAl Viro kfree(mp); 759eed81007SMiklos Szeredi return ERR_PTR(ret); 76084d17192SAl Viro } 761eed81007SMiklos Szeredi 76284d17192SAl Viro mp->m_dentry = dentry; 76384d17192SAl Viro mp->m_count = 1; 7640818bf27SAl Viro hlist_add_head(&mp->m_hash, chain); 7650a5eb7c8SEric W. Biederman INIT_HLIST_HEAD(&mp->m_list); 76684d17192SAl Viro return mp; 76784d17192SAl Viro } 76884d17192SAl Viro 76984d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 77084d17192SAl Viro { 77184d17192SAl Viro if (!--mp->m_count) { 77284d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7730a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 77484d17192SAl Viro spin_lock(&dentry->d_lock); 77584d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 77684d17192SAl Viro spin_unlock(&dentry->d_lock); 7770818bf27SAl Viro hlist_del(&mp->m_hash); 77884d17192SAl Viro kfree(mp); 77984d17192SAl Viro } 78084d17192SAl Viro } 78184d17192SAl Viro 782143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7831da177e4SLinus Torvalds { 7846b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7851da177e4SLinus Torvalds } 7861da177e4SLinus Torvalds 78799b7db7bSNick Piggin /* 78899b7db7bSNick Piggin * vfsmount lock must be held for write 78999b7db7bSNick Piggin */ 7906b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7915addc5ddSAl Viro { 7925addc5ddSAl Viro if (ns) { 7935addc5ddSAl Viro ns->event = ++event; 7945addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7955addc5ddSAl Viro } 7965addc5ddSAl Viro } 7975addc5ddSAl Viro 79899b7db7bSNick Piggin /* 79999b7db7bSNick Piggin * vfsmount lock must be held for write 80099b7db7bSNick Piggin */ 8016b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8025addc5ddSAl Viro { 8035addc5ddSAl Viro if (ns && ns->event != event) { 8045addc5ddSAl Viro ns->event = event; 8055addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8065addc5ddSAl Viro } 8075addc5ddSAl Viro } 8085addc5ddSAl Viro 80999b7db7bSNick Piggin /* 81099b7db7bSNick Piggin * vfsmount lock must be held for write 81199b7db7bSNick Piggin */ 8127bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 8131da177e4SLinus Torvalds { 8140714a533SAl Viro mnt->mnt_parent = mnt; 815a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8166b41d536SAl Viro list_del_init(&mnt->mnt_child); 81738129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8180a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 81984d17192SAl Viro put_mountpoint(mnt->mnt_mp); 82084d17192SAl Viro mnt->mnt_mp = NULL; 8211da177e4SLinus Torvalds } 8221da177e4SLinus Torvalds 82399b7db7bSNick Piggin /* 82499b7db7bSNick Piggin * vfsmount lock must be held for write 82599b7db7bSNick Piggin */ 8267bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8277bdb11deSEric W. Biederman { 8287bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8297bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8307bdb11deSEric W. Biederman unhash_mnt(mnt); 8317bdb11deSEric W. Biederman } 8327bdb11deSEric W. Biederman 8337bdb11deSEric W. Biederman /* 8347bdb11deSEric W. Biederman * vfsmount lock must be held for write 8357bdb11deSEric W. Biederman */ 8366a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8376a46c573SEric W. Biederman { 8386a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8396a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8406a46c573SEric W. Biederman unhash_mnt(mnt); 8416a46c573SEric W. Biederman } 8426a46c573SEric W. Biederman 8436a46c573SEric W. Biederman /* 8446a46c573SEric W. Biederman * vfsmount lock must be held for write 8456a46c573SEric W. Biederman */ 84684d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 84784d17192SAl Viro struct mountpoint *mp, 84844d964d6SAl Viro struct mount *child_mnt) 849b90fa9aeSRam Pai { 85084d17192SAl Viro mp->m_count++; 8513a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 85284d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8533a2393d7SAl Viro child_mnt->mnt_parent = mnt; 85484d17192SAl Viro child_mnt->mnt_mp = mp; 8550a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 856b90fa9aeSRam Pai } 857b90fa9aeSRam Pai 85899b7db7bSNick Piggin /* 85999b7db7bSNick Piggin * vfsmount lock must be held for write 86099b7db7bSNick Piggin */ 86184d17192SAl Viro static void attach_mnt(struct mount *mnt, 86284d17192SAl Viro struct mount *parent, 86384d17192SAl Viro struct mountpoint *mp) 8641da177e4SLinus Torvalds { 86584d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 86638129a13SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); 86784d17192SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 868b90fa9aeSRam Pai } 869b90fa9aeSRam Pai 87012a5b529SAl Viro static void attach_shadowed(struct mount *mnt, 87112a5b529SAl Viro struct mount *parent, 87212a5b529SAl Viro struct mount *shadows) 87312a5b529SAl Viro { 87412a5b529SAl Viro if (shadows) { 875f6f99332SLinus Torvalds hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); 87612a5b529SAl Viro list_add(&mnt->mnt_child, &shadows->mnt_child); 87712a5b529SAl Viro } else { 87812a5b529SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, 87912a5b529SAl Viro m_hash(&parent->mnt, mnt->mnt_mountpoint)); 88012a5b529SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 88112a5b529SAl Viro } 88212a5b529SAl Viro } 88312a5b529SAl Viro 884b90fa9aeSRam Pai /* 88599b7db7bSNick Piggin * vfsmount lock must be held for write 886b90fa9aeSRam Pai */ 8871d6a32acSAl Viro static void commit_tree(struct mount *mnt, struct mount *shadows) 888b90fa9aeSRam Pai { 8890714a533SAl Viro struct mount *parent = mnt->mnt_parent; 89083adc753SAl Viro struct mount *m; 891b90fa9aeSRam Pai LIST_HEAD(head); 892143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 893b90fa9aeSRam Pai 8940714a533SAl Viro BUG_ON(parent == mnt); 895b90fa9aeSRam Pai 8961a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 897f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 898143c8c91SAl Viro m->mnt_ns = n; 899f03c6599SAl Viro 900b90fa9aeSRam Pai list_splice(&head, n->list.prev); 901b90fa9aeSRam Pai 90212a5b529SAl Viro attach_shadowed(mnt, parent, shadows); 9036b3286edSKirill Korotaev touch_mnt_namespace(n); 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds 906909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9071da177e4SLinus Torvalds { 9086b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9096b41d536SAl Viro if (next == &p->mnt_mounts) { 9101da177e4SLinus Torvalds while (1) { 911909b0a88SAl Viro if (p == root) 9121da177e4SLinus Torvalds return NULL; 9136b41d536SAl Viro next = p->mnt_child.next; 9146b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9151da177e4SLinus Torvalds break; 9160714a533SAl Viro p = p->mnt_parent; 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds } 9196b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9201da177e4SLinus Torvalds } 9211da177e4SLinus Torvalds 922315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9239676f0c6SRam Pai { 9246b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9256b41d536SAl Viro while (prev != &p->mnt_mounts) { 9266b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9276b41d536SAl Viro prev = p->mnt_mounts.prev; 9289676f0c6SRam Pai } 9299676f0c6SRam Pai return p; 9309676f0c6SRam Pai } 9319676f0c6SRam Pai 9329d412a43SAl Viro struct vfsmount * 9339d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 9349d412a43SAl Viro { 935b105e270SAl Viro struct mount *mnt; 9369d412a43SAl Viro struct dentry *root; 9379d412a43SAl Viro 9389d412a43SAl Viro if (!type) 9399d412a43SAl Viro return ERR_PTR(-ENODEV); 9409d412a43SAl Viro 9419d412a43SAl Viro mnt = alloc_vfsmnt(name); 9429d412a43SAl Viro if (!mnt) 9439d412a43SAl Viro return ERR_PTR(-ENOMEM); 9449d412a43SAl Viro 9459d412a43SAl Viro if (flags & MS_KERNMOUNT) 946b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9479d412a43SAl Viro 9489d412a43SAl Viro root = mount_fs(type, flags, name, data); 9499d412a43SAl Viro if (IS_ERR(root)) { 9508ffcb32eSDavid Howells mnt_free_id(mnt); 9519d412a43SAl Viro free_vfsmnt(mnt); 9529d412a43SAl Viro return ERR_CAST(root); 9539d412a43SAl Viro } 9549d412a43SAl Viro 955b105e270SAl Viro mnt->mnt.mnt_root = root; 956b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 957a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9580714a533SAl Viro mnt->mnt_parent = mnt; 959719ea2fbSAl Viro lock_mount_hash(); 96039f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 961719ea2fbSAl Viro unlock_mount_hash(); 962b105e270SAl Viro return &mnt->mnt; 9639d412a43SAl Viro } 9649d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 9659d412a43SAl Viro 96687129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 96736341f64SRam Pai int flag) 9681da177e4SLinus Torvalds { 96987129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 970be34d1a3SDavid Howells struct mount *mnt; 971be34d1a3SDavid Howells int err; 9721da177e4SLinus Torvalds 973be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 974be34d1a3SDavid Howells if (!mnt) 975be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 976be34d1a3SDavid Howells 9777a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 97815169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 979719f5d7fSMiklos Szeredi else 98015169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 981719f5d7fSMiklos Szeredi 98215169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 983be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 984719f5d7fSMiklos Szeredi if (err) 985719f5d7fSMiklos Szeredi goto out_free; 986719f5d7fSMiklos Szeredi } 987719f5d7fSMiklos Szeredi 988f2ebb3a9SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); 989132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 9909566d674SEric W. Biederman if (flag & CL_UNPRIVILEGED) { 9919566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; 9929566d674SEric W. Biederman 9939566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_READONLY) 994132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 995132c94e3SEric W. Biederman 9969566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NODEV) 9979566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; 9989566d674SEric W. Biederman 9999566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOSUID) 10009566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; 10019566d674SEric W. Biederman 10029566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOEXEC) 10039566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; 10049566d674SEric W. Biederman } 10059566d674SEric W. Biederman 10065ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 1007381cacb1SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && 1008381cacb1SEric W. Biederman (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) 10095ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 10105ff9d8a6SEric W. Biederman 10111da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1012b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1013b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1014a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10150714a533SAl Viro mnt->mnt_parent = mnt; 1016719ea2fbSAl Viro lock_mount_hash(); 101739f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1018719ea2fbSAl Viro unlock_mount_hash(); 1019b90fa9aeSRam Pai 10207a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10217a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10226776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 102332301920SAl Viro mnt->mnt_master = old; 1024fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10258aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1026fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10276776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1028d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10296776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1030d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10315afe0022SRam Pai } 1032b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10330f0afb1dSAl Viro set_mnt_shared(mnt); 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10361da177e4SLinus Torvalds * as the original if that was on one */ 103736341f64SRam Pai if (flag & CL_EXPIRE) { 10386776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10396776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10401da177e4SLinus Torvalds } 1041be34d1a3SDavid Howells 1042cb338d06SAl Viro return mnt; 1043719f5d7fSMiklos Szeredi 1044719f5d7fSMiklos Szeredi out_free: 10458ffcb32eSDavid Howells mnt_free_id(mnt); 1046719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1047be34d1a3SDavid Howells return ERR_PTR(err); 10481da177e4SLinus Torvalds } 10491da177e4SLinus Torvalds 10509ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10519ea459e1SAl Viro { 10529ea459e1SAl Viro /* 10539ea459e1SAl Viro * This probably indicates that somebody messed 10549ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 10559ea459e1SAl Viro * happens, the filesystem was probably unable 10569ea459e1SAl Viro * to make r/w->r/o transitions. 10579ea459e1SAl Viro */ 10589ea459e1SAl Viro /* 10599ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 10609ea459e1SAl Viro * so mnt_get_writers() below is safe. 10619ea459e1SAl Viro */ 10629ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 10639ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 10649ea459e1SAl Viro mnt_pin_kill(mnt); 10659ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 10669ea459e1SAl Viro dput(mnt->mnt.mnt_root); 10679ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 10689ea459e1SAl Viro mnt_free_id(mnt); 10699ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 10709ea459e1SAl Viro } 10719ea459e1SAl Viro 10729ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 10739ea459e1SAl Viro { 10749ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 10759ea459e1SAl Viro } 10769ea459e1SAl Viro 10779ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 10789ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 10799ea459e1SAl Viro { 10809ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 10819ea459e1SAl Viro struct llist_node *next; 10829ea459e1SAl Viro 10839ea459e1SAl Viro for (; node; node = next) { 10849ea459e1SAl Viro next = llist_next(node); 10859ea459e1SAl Viro cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); 10869ea459e1SAl Viro } 10879ea459e1SAl Viro } 10889ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 10899ea459e1SAl Viro 1090900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 10917b7b1aceSAl Viro { 109248a066e7SAl Viro rcu_read_lock(); 1093aa9c0e07SAl Viro mnt_add_count(mnt, -1); 109448a066e7SAl Viro if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ 109548a066e7SAl Viro rcu_read_unlock(); 109699b7db7bSNick Piggin return; 1097b3e19d92SNick Piggin } 1098719ea2fbSAl Viro lock_mount_hash(); 1099b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 110048a066e7SAl Viro rcu_read_unlock(); 1101719ea2fbSAl Viro unlock_mount_hash(); 110299b7db7bSNick Piggin return; 110399b7db7bSNick Piggin } 110448a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 110548a066e7SAl Viro rcu_read_unlock(); 110648a066e7SAl Viro unlock_mount_hash(); 110748a066e7SAl Viro return; 110848a066e7SAl Viro } 110948a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 111048a066e7SAl Viro rcu_read_unlock(); 1111962830dfSAndi Kleen 111239f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1113ce07d891SEric W. Biederman 1114ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1115ce07d891SEric W. Biederman struct mount *p, *tmp; 1116ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1117ce07d891SEric W. Biederman umount_mnt(p); 1118ce07d891SEric W. Biederman } 1119ce07d891SEric W. Biederman } 1120719ea2fbSAl Viro unlock_mount_hash(); 1121649a795aSAl Viro 11229ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11239ea459e1SAl Viro struct task_struct *task = current; 11249ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11259ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11269ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11279ea459e1SAl Viro return; 11289ea459e1SAl Viro } 11299ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11309ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11319ea459e1SAl Viro return; 11329ea459e1SAl Viro } 11339ea459e1SAl Viro cleanup_mnt(mnt); 1134b3e19d92SNick Piggin } 1135b3e19d92SNick Piggin 1136b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1137b3e19d92SNick Piggin { 1138b3e19d92SNick Piggin if (mnt) { 1139863d684fSAl Viro struct mount *m = real_mount(mnt); 1140b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1141863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1142863d684fSAl Viro m->mnt_expiry_mark = 0; 1143863d684fSAl Viro mntput_no_expire(m); 1144b3e19d92SNick Piggin } 1145b3e19d92SNick Piggin } 1146b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1147b3e19d92SNick Piggin 1148b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1149b3e19d92SNick Piggin { 1150b3e19d92SNick Piggin if (mnt) 115183adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1152b3e19d92SNick Piggin return mnt; 1153b3e19d92SNick Piggin } 1154b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1155b3e19d92SNick Piggin 11563064c356SAl Viro struct vfsmount *mnt_clone_internal(struct path *path) 11577b7b1aceSAl Viro { 11583064c356SAl Viro struct mount *p; 11593064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 11603064c356SAl Viro if (IS_ERR(p)) 11613064c356SAl Viro return ERR_CAST(p); 11623064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 11633064c356SAl Viro return &p->mnt; 11647b7b1aceSAl Viro } 11651da177e4SLinus Torvalds 1166b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 1167b3b304a2SMiklos Szeredi { 1168b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 1169b3b304a2SMiklos Szeredi } 1170b3b304a2SMiklos Szeredi 1171b3b304a2SMiklos Szeredi /* 1172b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 1173b3b304a2SMiklos Szeredi * implement more complex mount option showing. 1174b3b304a2SMiklos Szeredi * 1175b3b304a2SMiklos Szeredi * See also save_mount_options(). 1176b3b304a2SMiklos Szeredi */ 117734c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 1178b3b304a2SMiklos Szeredi { 11792a32cebdSAl Viro const char *options; 11802a32cebdSAl Viro 11812a32cebdSAl Viro rcu_read_lock(); 118234c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 1183b3b304a2SMiklos Szeredi 1184b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 1185b3b304a2SMiklos Szeredi seq_putc(m, ','); 1186b3b304a2SMiklos Szeredi mangle(m, options); 1187b3b304a2SMiklos Szeredi } 11882a32cebdSAl Viro rcu_read_unlock(); 1189b3b304a2SMiklos Szeredi 1190b3b304a2SMiklos Szeredi return 0; 1191b3b304a2SMiklos Szeredi } 1192b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 1193b3b304a2SMiklos Szeredi 1194b3b304a2SMiklos Szeredi /* 1195b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 1196b3b304a2SMiklos Szeredi * called from the fill_super() callback. 1197b3b304a2SMiklos Szeredi * 1198b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 1199b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1200b3b304a2SMiklos Szeredi * remount fails. 1201b3b304a2SMiklos Szeredi * 1202b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1203b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1204b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1205b3b304a2SMiklos Szeredi * any more. 1206b3b304a2SMiklos Szeredi */ 1207b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1208b3b304a2SMiklos Szeredi { 12092a32cebdSAl Viro BUG_ON(sb->s_options); 12102a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1211b3b304a2SMiklos Szeredi } 1212b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1213b3b304a2SMiklos Szeredi 12142a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 12152a32cebdSAl Viro { 12162a32cebdSAl Viro char *old = sb->s_options; 12172a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 12182a32cebdSAl Viro if (old) { 12192a32cebdSAl Viro synchronize_rcu(); 12202a32cebdSAl Viro kfree(old); 12212a32cebdSAl Viro } 12222a32cebdSAl Viro } 12232a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 12242a32cebdSAl Viro 1225a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12260226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12271da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12281da177e4SLinus Torvalds { 1229ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12301da177e4SLinus Torvalds 1231390c6843SRam Pai down_read(&namespace_sem); 1232c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1233c7999c36SAl Viro void *v = p->cached_mount; 1234c7999c36SAl Viro if (*pos == p->cached_index) 1235c7999c36SAl Viro return v; 1236c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1237c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1238c7999c36SAl Viro return p->cached_mount = v; 1239c7999c36SAl Viro } 1240c7999c36SAl Viro } 1241c7999c36SAl Viro 1242c7999c36SAl Viro p->cached_event = p->ns->event; 1243c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1244c7999c36SAl Viro p->cached_index = *pos; 1245c7999c36SAl Viro return p->cached_mount; 12461da177e4SLinus Torvalds } 12471da177e4SLinus Torvalds 12481da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12491da177e4SLinus Torvalds { 1250ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1251b0765fb8SPavel Emelianov 1252c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1253c7999c36SAl Viro p->cached_index = *pos; 1254c7999c36SAl Viro return p->cached_mount; 12551da177e4SLinus Torvalds } 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12581da177e4SLinus Torvalds { 1259390c6843SRam Pai up_read(&namespace_sem); 12601da177e4SLinus Torvalds } 12611da177e4SLinus Torvalds 12620226f492SAl Viro static int m_show(struct seq_file *m, void *v) 12639f5596afSAl Viro { 1264ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12651a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 12660226f492SAl Viro return p->show(m, &r->mnt); 12671da177e4SLinus Torvalds } 12681da177e4SLinus Torvalds 1269a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 12701da177e4SLinus Torvalds .start = m_start, 12711da177e4SLinus Torvalds .next = m_next, 12721da177e4SLinus Torvalds .stop = m_stop, 12730226f492SAl Viro .show = m_show, 1274b4629fe2SChuck Lever }; 1275a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1276b4629fe2SChuck Lever 12771da177e4SLinus Torvalds /** 12781da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 12791da177e4SLinus Torvalds * @mnt: root of mount tree 12801da177e4SLinus Torvalds * 12811da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 12821da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 12831da177e4SLinus Torvalds * busy. 12841da177e4SLinus Torvalds */ 1285909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 12861da177e4SLinus Torvalds { 1287909b0a88SAl Viro struct mount *mnt = real_mount(m); 128836341f64SRam Pai int actual_refs = 0; 128936341f64SRam Pai int minimum_refs = 0; 1290315fc83eSAl Viro struct mount *p; 1291909b0a88SAl Viro BUG_ON(!m); 12921da177e4SLinus Torvalds 1293b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1294719ea2fbSAl Viro lock_mount_hash(); 1295909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 129683adc753SAl Viro actual_refs += mnt_get_count(p); 12971da177e4SLinus Torvalds minimum_refs += 2; 12981da177e4SLinus Torvalds } 1299719ea2fbSAl Viro unlock_mount_hash(); 13001da177e4SLinus Torvalds 13011da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13021da177e4SLinus Torvalds return 0; 1303e3474a8eSIan Kent 1304e3474a8eSIan Kent return 1; 13051da177e4SLinus Torvalds } 13061da177e4SLinus Torvalds 13071da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13081da177e4SLinus Torvalds 13091da177e4SLinus Torvalds /** 13101da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13111da177e4SLinus Torvalds * @mnt: root of mount 13121da177e4SLinus Torvalds * 13131da177e4SLinus Torvalds * This is called to check if a mount point has any 13141da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13151da177e4SLinus Torvalds * mount has sub mounts this will return busy 13161da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13171da177e4SLinus Torvalds * 13181da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13191da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13201da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13211da177e4SLinus Torvalds */ 13221da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13231da177e4SLinus Torvalds { 1324e3474a8eSIan Kent int ret = 1; 13258ad08d8aSAl Viro down_read(&namespace_sem); 1326719ea2fbSAl Viro lock_mount_hash(); 13271ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1328e3474a8eSIan Kent ret = 0; 1329719ea2fbSAl Viro unlock_mount_hash(); 13308ad08d8aSAl Viro up_read(&namespace_sem); 1331a05964f3SRam Pai return ret; 13321da177e4SLinus Torvalds } 13331da177e4SLinus Torvalds 13341da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13351da177e4SLinus Torvalds 133638129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1337e3197d83SAl Viro 133897216be0SAl Viro static void namespace_unlock(void) 13391da177e4SLinus Torvalds { 1340a3b3c562SEric W. Biederman struct hlist_head head; 134197216be0SAl Viro 1342a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1343a3b3c562SEric W. Biederman 134497216be0SAl Viro up_write(&namespace_sem); 1345a3b3c562SEric W. Biederman 1346a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 134797216be0SAl Viro return; 134897216be0SAl Viro 134948a066e7SAl Viro synchronize_rcu(); 135048a066e7SAl Viro 135187b95ce0SAl Viro group_pin_kill(&head); 135270fbcdf4SRam Pai } 135370fbcdf4SRam Pai 135497216be0SAl Viro static inline void namespace_lock(void) 1355e3197d83SAl Viro { 135697216be0SAl Viro down_write(&namespace_sem); 1357e3197d83SAl Viro } 1358e3197d83SAl Viro 1359e819f152SEric W. Biederman enum umount_tree_flags { 1360e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1361e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1362e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1363e819f152SEric W. Biederman }; 136499b7db7bSNick Piggin /* 136548a066e7SAl Viro * mount_lock must be held 136699b7db7bSNick Piggin * namespace_sem must be held for write 136799b7db7bSNick Piggin */ 1368e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 136970fbcdf4SRam Pai { 1370c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1371315fc83eSAl Viro struct mount *p; 137270fbcdf4SRam Pai 13735d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 13745d88457eSEric W. Biederman propagate_mount_unlock(mnt); 13755d88457eSEric W. Biederman 1376c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1377590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1378590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1379c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1380590ce4bcSEric W. Biederman } 1381c003b26fSEric W. Biederman 1382411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1383c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1384c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 138538129a13SAl Viro } 138670fbcdf4SRam Pai 1387c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1388e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 13897b8a53fdSAl Viro propagate_umount(&tmp_list); 1390a05964f3SRam Pai 1391c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1392ce07d891SEric W. Biederman bool disconnect; 1393c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 13946776db3dSAl Viro list_del_init(&p->mnt_expire); 13951a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1396143c8c91SAl Viro __touch_mnt_namespace(p->mnt_ns); 139763d37a84SAl Viro p->mnt_ns = NULL; 1398e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 139948a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 140087b95ce0SAl Viro 1401e0c9c0afSEric W. Biederman disconnect = !(((how & UMOUNT_CONNECTED) && 1402e0c9c0afSEric W. Biederman mnt_has_parent(p) && 1403e0c9c0afSEric W. Biederman (p->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) || 1404e0c9c0afSEric W. Biederman IS_MNT_LOCKED_AND_LAZY(p)); 1405ce07d891SEric W. Biederman 1406ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1407ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1408676da58dSAl Viro if (mnt_has_parent(p)) { 140981b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1410ce07d891SEric W. Biederman if (!disconnect) { 1411ce07d891SEric W. Biederman /* Don't forget about p */ 1412ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1413ce07d891SEric W. Biederman } else { 14146a46c573SEric W. Biederman umount_mnt(p); 14157c4b93d8SAl Viro } 1416ce07d891SEric W. Biederman } 14170f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 141838129a13SAl Viro } 14191da177e4SLinus Torvalds } 14201da177e4SLinus Torvalds 1421b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1422c35038beSAl Viro 14231ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 14241da177e4SLinus Torvalds { 14251ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 14261da177e4SLinus Torvalds int retval; 14271da177e4SLinus Torvalds 14281ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 14291da177e4SLinus Torvalds if (retval) 14301da177e4SLinus Torvalds return retval; 14311da177e4SLinus Torvalds 14321da177e4SLinus Torvalds /* 14331da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 14341da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 14351da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 14361da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 14371da177e4SLinus Torvalds */ 14381da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 14391ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 14401da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 14411da177e4SLinus Torvalds return -EINVAL; 14421da177e4SLinus Torvalds 1443b3e19d92SNick Piggin /* 1444b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1445b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1446b3e19d92SNick Piggin */ 1447719ea2fbSAl Viro lock_mount_hash(); 144883adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1449719ea2fbSAl Viro unlock_mount_hash(); 14501da177e4SLinus Torvalds return -EBUSY; 1451b3e19d92SNick Piggin } 1452719ea2fbSAl Viro unlock_mount_hash(); 14531da177e4SLinus Torvalds 1454863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 14551da177e4SLinus Torvalds return -EAGAIN; 14561da177e4SLinus Torvalds } 14571da177e4SLinus Torvalds 14581da177e4SLinus Torvalds /* 14591da177e4SLinus Torvalds * If we may have to abort operations to get out of this 14601da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 14611da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 14621da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 14631da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 14641da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 14651da177e4SLinus Torvalds * about for the moment. 14661da177e4SLinus Torvalds */ 14671da177e4SLinus Torvalds 146842faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 146942faad99SAl Viro sb->s_op->umount_begin(sb); 147042faad99SAl Viro } 14711da177e4SLinus Torvalds 14721da177e4SLinus Torvalds /* 14731da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 14741da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 14751da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 14761da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 14771da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 14781da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 14791da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 14801da177e4SLinus Torvalds */ 14811ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 14821da177e4SLinus Torvalds /* 14831da177e4SLinus Torvalds * Special case for "unmounting" root ... 14841da177e4SLinus Torvalds * we just try to remount it readonly. 14851da177e4SLinus Torvalds */ 1486a1480dccSAndy Lutomirski if (!capable(CAP_SYS_ADMIN)) 1487a1480dccSAndy Lutomirski return -EPERM; 14881da177e4SLinus Torvalds down_write(&sb->s_umount); 14894aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 14901da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 14911da177e4SLinus Torvalds up_write(&sb->s_umount); 14921da177e4SLinus Torvalds return retval; 14931da177e4SLinus Torvalds } 14941da177e4SLinus Torvalds 149597216be0SAl Viro namespace_lock(); 1496719ea2fbSAl Viro lock_mount_hash(); 14975addc5ddSAl Viro event++; 14981da177e4SLinus Torvalds 149948a066e7SAl Viro if (flags & MNT_DETACH) { 150048a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1501e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 150248a066e7SAl Viro retval = 0; 150348a066e7SAl Viro } else { 1504b54b9be7SAl Viro shrink_submounts(mnt); 15051da177e4SLinus Torvalds retval = -EBUSY; 150648a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 15071a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1508e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 15091da177e4SLinus Torvalds retval = 0; 15101da177e4SLinus Torvalds } 151148a066e7SAl Viro } 1512719ea2fbSAl Viro unlock_mount_hash(); 1513e3197d83SAl Viro namespace_unlock(); 15141da177e4SLinus Torvalds return retval; 15151da177e4SLinus Torvalds } 15161da177e4SLinus Torvalds 15171da177e4SLinus Torvalds /* 151880b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 151980b5dce8SEric W. Biederman * 152080b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 152180b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 152280b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 152380b5dce8SEric W. Biederman * leaking them. 152480b5dce8SEric W. Biederman * 152580b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 152680b5dce8SEric W. Biederman */ 152780b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 152880b5dce8SEric W. Biederman { 152980b5dce8SEric W. Biederman struct mountpoint *mp; 153080b5dce8SEric W. Biederman struct mount *mnt; 153180b5dce8SEric W. Biederman 153280b5dce8SEric W. Biederman namespace_lock(); 153380b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1534f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 153580b5dce8SEric W. Biederman goto out_unlock; 153680b5dce8SEric W. Biederman 153780b5dce8SEric W. Biederman lock_mount_hash(); 153880b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 153980b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1540ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1541ce07d891SEric W. Biederman struct mount *p, *tmp; 1542ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1543ce07d891SEric W. Biederman hlist_add_head(&p->mnt_umount.s_list, &unmounted); 1544ce07d891SEric W. Biederman umount_mnt(p); 1545ce07d891SEric W. Biederman } 1546ce07d891SEric W. Biederman } 1547e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 154880b5dce8SEric W. Biederman } 154980b5dce8SEric W. Biederman unlock_mount_hash(); 155080b5dce8SEric W. Biederman put_mountpoint(mp); 155180b5dce8SEric W. Biederman out_unlock: 155280b5dce8SEric W. Biederman namespace_unlock(); 155380b5dce8SEric W. Biederman } 155480b5dce8SEric W. Biederman 155580b5dce8SEric W. Biederman /* 15569b40bc90SAl Viro * Is the caller allowed to modify his namespace? 15579b40bc90SAl Viro */ 15589b40bc90SAl Viro static inline bool may_mount(void) 15599b40bc90SAl Viro { 15609b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 15619b40bc90SAl Viro } 15629b40bc90SAl Viro 15639b40bc90SAl Viro /* 15641da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 15651da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 15661da177e4SLinus Torvalds * 15671da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 15681da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 15691da177e4SLinus Torvalds */ 15701da177e4SLinus Torvalds 1571bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 15721da177e4SLinus Torvalds { 15732d8f3038SAl Viro struct path path; 1574900148dcSAl Viro struct mount *mnt; 15751da177e4SLinus Torvalds int retval; 1576db1f05bbSMiklos Szeredi int lookup_flags = 0; 15771da177e4SLinus Torvalds 1578db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1579db1f05bbSMiklos Szeredi return -EINVAL; 1580db1f05bbSMiklos Szeredi 15819b40bc90SAl Viro if (!may_mount()) 15829b40bc90SAl Viro return -EPERM; 15839b40bc90SAl Viro 1584db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1585db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1586db1f05bbSMiklos Szeredi 1587197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 15881da177e4SLinus Torvalds if (retval) 15891da177e4SLinus Torvalds goto out; 1590900148dcSAl Viro mnt = real_mount(path.mnt); 15911da177e4SLinus Torvalds retval = -EINVAL; 15922d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 15931da177e4SLinus Torvalds goto dput_and_out; 1594143c8c91SAl Viro if (!check_mnt(mnt)) 15951da177e4SLinus Torvalds goto dput_and_out; 15965ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 15975ff9d8a6SEric W. Biederman goto dput_and_out; 1598b2f5d4dcSEric W. Biederman retval = -EPERM; 1599b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1600b2f5d4dcSEric W. Biederman goto dput_and_out; 16011da177e4SLinus Torvalds 1602900148dcSAl Viro retval = do_umount(mnt, flags); 16031da177e4SLinus Torvalds dput_and_out: 1604429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 16052d8f3038SAl Viro dput(path.dentry); 1606900148dcSAl Viro mntput_no_expire(mnt); 16071da177e4SLinus Torvalds out: 16081da177e4SLinus Torvalds return retval; 16091da177e4SLinus Torvalds } 16101da177e4SLinus Torvalds 16111da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 16121da177e4SLinus Torvalds 16131da177e4SLinus Torvalds /* 16141da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 16151da177e4SLinus Torvalds */ 1616bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 16171da177e4SLinus Torvalds { 16181da177e4SLinus Torvalds return sys_umount(name, 0); 16191da177e4SLinus Torvalds } 16201da177e4SLinus Torvalds 16211da177e4SLinus Torvalds #endif 16221da177e4SLinus Torvalds 16234ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 16248823c079SEric W. Biederman { 16254ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1626e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1627e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 16284ce5d2b1SEric W. Biederman } 16294ce5d2b1SEric W. Biederman 163058be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 163158be2825SAl Viro { 163258be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 163358be2825SAl Viro } 163458be2825SAl Viro 16354ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 16364ce5d2b1SEric W. Biederman { 16374ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 16384ce5d2b1SEric W. Biederman * mount namespace loop? 16394ce5d2b1SEric W. Biederman */ 16404ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 16414ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 16424ce5d2b1SEric W. Biederman return false; 16434ce5d2b1SEric W. Biederman 1644f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 16458823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 16468823c079SEric W. Biederman } 16478823c079SEric W. Biederman 164887129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 164936341f64SRam Pai int flag) 16501da177e4SLinus Torvalds { 165184d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 16521da177e4SLinus Torvalds 16534ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 16544ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 16554ce5d2b1SEric W. Biederman 16564ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1657be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 16589676f0c6SRam Pai 165936341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1660be34d1a3SDavid Howells if (IS_ERR(q)) 1661be34d1a3SDavid Howells return q; 1662be34d1a3SDavid Howells 1663a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds p = mnt; 16666b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1667315fc83eSAl Viro struct mount *s; 16687ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 16691da177e4SLinus Torvalds continue; 16701da177e4SLinus Torvalds 1671909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 167212a5b529SAl Viro struct mount *t = NULL; 16734ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 16744ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 16754ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 16764ce5d2b1SEric W. Biederman continue; 16774ce5d2b1SEric W. Biederman } 16784ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 16794ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 16809676f0c6SRam Pai s = skip_mnt_tree(s); 16819676f0c6SRam Pai continue; 16829676f0c6SRam Pai } 16830714a533SAl Viro while (p != s->mnt_parent) { 16840714a533SAl Viro p = p->mnt_parent; 16850714a533SAl Viro q = q->mnt_parent; 16861da177e4SLinus Torvalds } 168787129cc0SAl Viro p = s; 168884d17192SAl Viro parent = q; 168987129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1690be34d1a3SDavid Howells if (IS_ERR(q)) 1691be34d1a3SDavid Howells goto out; 1692719ea2fbSAl Viro lock_mount_hash(); 16931a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 169412a5b529SAl Viro mnt_set_mountpoint(parent, p->mnt_mp, q); 169512a5b529SAl Viro if (!list_empty(&parent->mnt_mounts)) { 169612a5b529SAl Viro t = list_last_entry(&parent->mnt_mounts, 169712a5b529SAl Viro struct mount, mnt_child); 169812a5b529SAl Viro if (t->mnt_mp != p->mnt_mp) 169912a5b529SAl Viro t = NULL; 170012a5b529SAl Viro } 170112a5b529SAl Viro attach_shadowed(q, parent, t); 1702719ea2fbSAl Viro unlock_mount_hash(); 17031da177e4SLinus Torvalds } 17041da177e4SLinus Torvalds } 17051da177e4SLinus Torvalds return res; 1706be34d1a3SDavid Howells out: 17071da177e4SLinus Torvalds if (res) { 1708719ea2fbSAl Viro lock_mount_hash(); 1709e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1710719ea2fbSAl Viro unlock_mount_hash(); 17111da177e4SLinus Torvalds } 1712be34d1a3SDavid Howells return q; 17131da177e4SLinus Torvalds } 17141da177e4SLinus Torvalds 1715be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1716be34d1a3SDavid Howells 1717589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 17188aec0809SAl Viro { 1719cb338d06SAl Viro struct mount *tree; 172097216be0SAl Viro namespace_lock(); 1721cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1722cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1723cd4a4017SEric W. Biederman else 172487129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 172587129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1726328e6d90SAl Viro namespace_unlock(); 1727be34d1a3SDavid Howells if (IS_ERR(tree)) 172852e220d3SDan Carpenter return ERR_CAST(tree); 1729be34d1a3SDavid Howells return &tree->mnt; 17308aec0809SAl Viro } 17318aec0809SAl Viro 17328aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 17338aec0809SAl Viro { 173497216be0SAl Viro namespace_lock(); 1735719ea2fbSAl Viro lock_mount_hash(); 1736e819f152SEric W. Biederman umount_tree(real_mount(mnt), UMOUNT_SYNC); 1737719ea2fbSAl Viro unlock_mount_hash(); 17383ab6abeeSAl Viro namespace_unlock(); 17398aec0809SAl Viro } 17408aec0809SAl Viro 1741c771d683SMiklos Szeredi /** 1742c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1743c771d683SMiklos Szeredi * 1744c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1745c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1746c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1747c771d683SMiklos Szeredi * 1748c771d683SMiklos Szeredi * Release with mntput(). 1749c771d683SMiklos Szeredi */ 1750c771d683SMiklos Szeredi struct vfsmount *clone_private_mount(struct path *path) 1751c771d683SMiklos Szeredi { 1752c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1753c771d683SMiklos Szeredi struct mount *new_mnt; 1754c771d683SMiklos Szeredi 1755c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1756c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1757c771d683SMiklos Szeredi 1758c771d683SMiklos Szeredi down_read(&namespace_sem); 1759c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1760c771d683SMiklos Szeredi up_read(&namespace_sem); 1761c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1762c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1763c771d683SMiklos Szeredi 1764c771d683SMiklos Szeredi return &new_mnt->mnt; 1765c771d683SMiklos Szeredi } 1766c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1767c771d683SMiklos Szeredi 17681f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 17691f707137SAl Viro struct vfsmount *root) 17701f707137SAl Viro { 17711a4eeaf2SAl Viro struct mount *mnt; 17721f707137SAl Viro int res = f(root, arg); 17731f707137SAl Viro if (res) 17741f707137SAl Viro return res; 17751a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 17761a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 17771f707137SAl Viro if (res) 17781f707137SAl Viro return res; 17791f707137SAl Viro } 17801f707137SAl Viro return 0; 17811f707137SAl Viro } 17821f707137SAl Viro 17834b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1784719f5d7fSMiklos Szeredi { 1785315fc83eSAl Viro struct mount *p; 1786719f5d7fSMiklos Szeredi 1787909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1788fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 17894b8b21f4SAl Viro mnt_release_group_id(p); 1790719f5d7fSMiklos Szeredi } 1791719f5d7fSMiklos Szeredi } 1792719f5d7fSMiklos Szeredi 17934b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1794719f5d7fSMiklos Szeredi { 1795315fc83eSAl Viro struct mount *p; 1796719f5d7fSMiklos Szeredi 1797909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1798fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 17994b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1800719f5d7fSMiklos Szeredi if (err) { 18014b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1802719f5d7fSMiklos Szeredi return err; 1803719f5d7fSMiklos Szeredi } 1804719f5d7fSMiklos Szeredi } 1805719f5d7fSMiklos Szeredi } 1806719f5d7fSMiklos Szeredi 1807719f5d7fSMiklos Szeredi return 0; 1808719f5d7fSMiklos Szeredi } 1809719f5d7fSMiklos Szeredi 1810b90fa9aeSRam Pai /* 1811b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1812b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 181321444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 181421444403SRam Pai * store the parent mount and mountpoint dentry. 181521444403SRam Pai * (done when source_mnt is moved) 1816b90fa9aeSRam Pai * 1817b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1818b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 18199676f0c6SRam Pai * --------------------------------------------------------------------------- 1820b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 18219676f0c6SRam Pai * |************************************************************************** 18229676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 18239676f0c6SRam Pai * | dest | | | | | 18249676f0c6SRam Pai * | | | | | | | 18259676f0c6SRam Pai * | v | | | | | 18269676f0c6SRam Pai * |************************************************************************** 18279676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 18285afe0022SRam Pai * | | | | | | 18299676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 18309676f0c6SRam Pai * *************************************************************************** 1831b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1832b90fa9aeSRam Pai * destination mount. 1833b90fa9aeSRam Pai * 1834b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1835b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1836b90fa9aeSRam Pai * the peer group of the source mount. 1837b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1838b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1839b90fa9aeSRam Pai * mount. 18405afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 18415afe0022SRam Pai * of the destination mount and the cloned mount is made slave 18425afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 18435afe0022SRam Pai * is marked as 'shared and slave'. 18445afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 18455afe0022SRam Pai * source mount. 18465afe0022SRam Pai * 18479676f0c6SRam Pai * --------------------------------------------------------------------------- 184821444403SRam Pai * | MOVE MOUNT OPERATION | 18499676f0c6SRam Pai * |************************************************************************** 18509676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 18519676f0c6SRam Pai * | dest | | | | | 18529676f0c6SRam Pai * | | | | | | | 18539676f0c6SRam Pai * | v | | | | | 18549676f0c6SRam Pai * |************************************************************************** 18559676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 18565afe0022SRam Pai * | | | | | | 18579676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 18589676f0c6SRam Pai * *************************************************************************** 18595afe0022SRam Pai * 18605afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 18615afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 186221444403SRam Pai * (+*) the mount is moved to the destination. 18635afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 18645afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 18655afe0022SRam Pai * the mount is marked as 'shared and slave'. 18665afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1867b90fa9aeSRam Pai * 1868b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1869b90fa9aeSRam Pai * applied to each mount in the tree. 1870b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1871b90fa9aeSRam Pai * in allocations. 1872b90fa9aeSRam Pai */ 18730fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 187484d17192SAl Viro struct mount *dest_mnt, 187584d17192SAl Viro struct mountpoint *dest_mp, 187684d17192SAl Viro struct path *parent_path) 1877b90fa9aeSRam Pai { 187838129a13SAl Viro HLIST_HEAD(tree_list); 1879315fc83eSAl Viro struct mount *child, *p; 188038129a13SAl Viro struct hlist_node *n; 1881719f5d7fSMiklos Szeredi int err; 1882b90fa9aeSRam Pai 1883fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 18840fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1885719f5d7fSMiklos Szeredi if (err) 1886719f5d7fSMiklos Szeredi goto out; 188784d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1888f2ebb3a9SAl Viro lock_mount_hash(); 1889719f5d7fSMiklos Szeredi if (err) 1890719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1891909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 18920f0afb1dSAl Viro set_mnt_shared(p); 18930b1b901bSAl Viro } else { 18940b1b901bSAl Viro lock_mount_hash(); 1895b90fa9aeSRam Pai } 18961a390689SAl Viro if (parent_path) { 18970fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 189884d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1899143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 190021444403SRam Pai } else { 190184d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 19021d6a32acSAl Viro commit_tree(source_mnt, NULL); 190321444403SRam Pai } 1904b90fa9aeSRam Pai 190538129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 19061d6a32acSAl Viro struct mount *q; 190738129a13SAl Viro hlist_del_init(&child->mnt_hash); 19081d6a32acSAl Viro q = __lookup_mnt_last(&child->mnt_parent->mnt, 19091d6a32acSAl Viro child->mnt_mountpoint); 19101d6a32acSAl Viro commit_tree(child, q); 1911b90fa9aeSRam Pai } 1912719ea2fbSAl Viro unlock_mount_hash(); 191399b7db7bSNick Piggin 1914b90fa9aeSRam Pai return 0; 1915719f5d7fSMiklos Szeredi 1916719f5d7fSMiklos Szeredi out_cleanup_ids: 1917f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 1918f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 1919e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 1920f2ebb3a9SAl Viro } 1921f2ebb3a9SAl Viro unlock_mount_hash(); 19220fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1923719f5d7fSMiklos Szeredi out: 1924719f5d7fSMiklos Szeredi return err; 1925b90fa9aeSRam Pai } 1926b90fa9aeSRam Pai 192784d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 1928b12cea91SAl Viro { 1929b12cea91SAl Viro struct vfsmount *mnt; 193084d17192SAl Viro struct dentry *dentry = path->dentry; 1931b12cea91SAl Viro retry: 193284d17192SAl Viro mutex_lock(&dentry->d_inode->i_mutex); 193384d17192SAl Viro if (unlikely(cant_mount(dentry))) { 193484d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 193584d17192SAl Viro return ERR_PTR(-ENOENT); 1936b12cea91SAl Viro } 193797216be0SAl Viro namespace_lock(); 1938b12cea91SAl Viro mnt = lookup_mnt(path); 193984d17192SAl Viro if (likely(!mnt)) { 1940e2dfa935SEric W. Biederman struct mountpoint *mp = lookup_mountpoint(dentry); 1941e2dfa935SEric W. Biederman if (!mp) 1942e2dfa935SEric W. Biederman mp = new_mountpoint(dentry); 194384d17192SAl Viro if (IS_ERR(mp)) { 194497216be0SAl Viro namespace_unlock(); 194584d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 194684d17192SAl Viro return mp; 194784d17192SAl Viro } 194884d17192SAl Viro return mp; 194984d17192SAl Viro } 195097216be0SAl Viro namespace_unlock(); 1951b12cea91SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 1952b12cea91SAl Viro path_put(path); 1953b12cea91SAl Viro path->mnt = mnt; 195484d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 1955b12cea91SAl Viro goto retry; 1956b12cea91SAl Viro } 1957b12cea91SAl Viro 195884d17192SAl Viro static void unlock_mount(struct mountpoint *where) 1959b12cea91SAl Viro { 196084d17192SAl Viro struct dentry *dentry = where->m_dentry; 196184d17192SAl Viro put_mountpoint(where); 1962328e6d90SAl Viro namespace_unlock(); 196384d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 1964b12cea91SAl Viro } 1965b12cea91SAl Viro 196684d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 19671da177e4SLinus Torvalds { 196895bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 19691da177e4SLinus Torvalds return -EINVAL; 19701da177e4SLinus Torvalds 1971e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 1972e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 19731da177e4SLinus Torvalds return -ENOTDIR; 19741da177e4SLinus Torvalds 197584d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 19761da177e4SLinus Torvalds } 19771da177e4SLinus Torvalds 19781da177e4SLinus Torvalds /* 19797a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 19807a2e8a8fSValerie Aurora */ 19817a2e8a8fSValerie Aurora 19827a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 19837a2e8a8fSValerie Aurora { 19847c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 19857a2e8a8fSValerie Aurora 19867a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 19877a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 19887a2e8a8fSValerie Aurora return 0; 19897a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 19907a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 19917a2e8a8fSValerie Aurora return 0; 19927a2e8a8fSValerie Aurora return type; 19937a2e8a8fSValerie Aurora } 19947a2e8a8fSValerie Aurora 19957a2e8a8fSValerie Aurora /* 199607b20889SRam Pai * recursively change the type of the mountpoint. 199707b20889SRam Pai */ 19980a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 199907b20889SRam Pai { 2000315fc83eSAl Viro struct mount *m; 20014b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 200207b20889SRam Pai int recurse = flag & MS_REC; 20037a2e8a8fSValerie Aurora int type; 2004719f5d7fSMiklos Szeredi int err = 0; 200507b20889SRam Pai 20062d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 200707b20889SRam Pai return -EINVAL; 200807b20889SRam Pai 20097a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 20107a2e8a8fSValerie Aurora if (!type) 20117a2e8a8fSValerie Aurora return -EINVAL; 20127a2e8a8fSValerie Aurora 201397216be0SAl Viro namespace_lock(); 2014719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2015719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2016719f5d7fSMiklos Szeredi if (err) 2017719f5d7fSMiklos Szeredi goto out_unlock; 2018719f5d7fSMiklos Szeredi } 2019719f5d7fSMiklos Szeredi 2020719ea2fbSAl Viro lock_mount_hash(); 2021909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 20220f0afb1dSAl Viro change_mnt_propagation(m, type); 2023719ea2fbSAl Viro unlock_mount_hash(); 2024719f5d7fSMiklos Szeredi 2025719f5d7fSMiklos Szeredi out_unlock: 202697216be0SAl Viro namespace_unlock(); 2027719f5d7fSMiklos Szeredi return err; 202807b20889SRam Pai } 202907b20889SRam Pai 20305ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 20315ff9d8a6SEric W. Biederman { 20325ff9d8a6SEric W. Biederman struct mount *child; 20335ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 20345ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 20355ff9d8a6SEric W. Biederman continue; 20365ff9d8a6SEric W. Biederman 20375ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 20385ff9d8a6SEric W. Biederman return true; 20395ff9d8a6SEric W. Biederman } 20405ff9d8a6SEric W. Biederman return false; 20415ff9d8a6SEric W. Biederman } 20425ff9d8a6SEric W. Biederman 204307b20889SRam Pai /* 20441da177e4SLinus Torvalds * do loopback mount. 20451da177e4SLinus Torvalds */ 2046808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 20472dafe1c4SEric Sandeen int recurse) 20481da177e4SLinus Torvalds { 20492d92ab3cSAl Viro struct path old_path; 205084d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 205184d17192SAl Viro struct mountpoint *mp; 205257eccb83SAl Viro int err; 20531da177e4SLinus Torvalds if (!old_name || !*old_name) 20541da177e4SLinus Torvalds return -EINVAL; 2055815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 20561da177e4SLinus Torvalds if (err) 20571da177e4SLinus Torvalds return err; 20581da177e4SLinus Torvalds 20598823c079SEric W. Biederman err = -EINVAL; 20604ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 20618823c079SEric W. Biederman goto out; 20628823c079SEric W. Biederman 206384d17192SAl Viro mp = lock_mount(path); 206484d17192SAl Viro err = PTR_ERR(mp); 206584d17192SAl Viro if (IS_ERR(mp)) 20669676f0c6SRam Pai goto out; 20679676f0c6SRam Pai 206887129cc0SAl Viro old = real_mount(old_path.mnt); 206984d17192SAl Viro parent = real_mount(path->mnt); 207087129cc0SAl Viro 2071b12cea91SAl Viro err = -EINVAL; 2072fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2073b12cea91SAl Viro goto out2; 2074b12cea91SAl Viro 2075e149ed2bSAl Viro if (!check_mnt(parent)) 2076e149ed2bSAl Viro goto out2; 2077e149ed2bSAl Viro 2078e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2079b12cea91SAl Viro goto out2; 2080ccd48bc7SAl Viro 20815ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 20825ff9d8a6SEric W. Biederman goto out2; 20835ff9d8a6SEric W. Biederman 20841da177e4SLinus Torvalds if (recurse) 20854ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 20861da177e4SLinus Torvalds else 208787129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 20881da177e4SLinus Torvalds 2089be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2090be34d1a3SDavid Howells err = PTR_ERR(mnt); 2091e9c5d8a5SAndrey Vagin goto out2; 2092be34d1a3SDavid Howells } 2093ccd48bc7SAl Viro 20945ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 20955ff9d8a6SEric W. Biederman 209684d17192SAl Viro err = graft_tree(mnt, parent, mp); 20971da177e4SLinus Torvalds if (err) { 2098719ea2fbSAl Viro lock_mount_hash(); 2099e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2100719ea2fbSAl Viro unlock_mount_hash(); 21015b83d2c5SRam Pai } 2102b12cea91SAl Viro out2: 210384d17192SAl Viro unlock_mount(mp); 2104ccd48bc7SAl Viro out: 21052d92ab3cSAl Viro path_put(&old_path); 21061da177e4SLinus Torvalds return err; 21071da177e4SLinus Torvalds } 21081da177e4SLinus Torvalds 21092e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 21102e4b7fcdSDave Hansen { 21112e4b7fcdSDave Hansen int error = 0; 21122e4b7fcdSDave Hansen int readonly_request = 0; 21132e4b7fcdSDave Hansen 21142e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 21152e4b7fcdSDave Hansen readonly_request = 1; 21162e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 21172e4b7fcdSDave Hansen return 0; 21182e4b7fcdSDave Hansen 21192e4b7fcdSDave Hansen if (readonly_request) 212083adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 21212e4b7fcdSDave Hansen else 212283adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 21232e4b7fcdSDave Hansen return error; 21242e4b7fcdSDave Hansen } 21252e4b7fcdSDave Hansen 21261da177e4SLinus Torvalds /* 21271da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 21281da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 21291da177e4SLinus Torvalds * on it - tough luck. 21301da177e4SLinus Torvalds */ 21310a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 21321da177e4SLinus Torvalds void *data) 21331da177e4SLinus Torvalds { 21341da177e4SLinus Torvalds int err; 21352d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2136143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 21371da177e4SLinus Torvalds 2138143c8c91SAl Viro if (!check_mnt(mnt)) 21391da177e4SLinus Torvalds return -EINVAL; 21401da177e4SLinus Torvalds 21412d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 21421da177e4SLinus Torvalds return -EINVAL; 21431da177e4SLinus Torvalds 214407b64558SEric W. Biederman /* Don't allow changing of locked mnt flags. 214507b64558SEric W. Biederman * 214607b64558SEric W. Biederman * No locks need to be held here while testing the various 214707b64558SEric W. Biederman * MNT_LOCK flags because those flags can never be cleared 214807b64558SEric W. Biederman * once they are set. 214907b64558SEric W. Biederman */ 215007b64558SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 215107b64558SEric W. Biederman !(mnt_flags & MNT_READONLY)) { 215207b64558SEric W. Biederman return -EPERM; 215307b64558SEric W. Biederman } 21549566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 21559566d674SEric W. Biederman !(mnt_flags & MNT_NODEV)) { 21563e186641SEric W. Biederman /* Was the nodev implicitly added in mount? */ 21573e186641SEric W. Biederman if ((mnt->mnt_ns->user_ns != &init_user_ns) && 21583e186641SEric W. Biederman !(sb->s_type->fs_flags & FS_USERNS_DEV_MOUNT)) { 21593e186641SEric W. Biederman mnt_flags |= MNT_NODEV; 21603e186641SEric W. Biederman } else { 21619566d674SEric W. Biederman return -EPERM; 21629566d674SEric W. Biederman } 21633e186641SEric W. Biederman } 21649566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && 21659566d674SEric W. Biederman !(mnt_flags & MNT_NOSUID)) { 21669566d674SEric W. Biederman return -EPERM; 21679566d674SEric W. Biederman } 21689566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && 21699566d674SEric W. Biederman !(mnt_flags & MNT_NOEXEC)) { 21709566d674SEric W. Biederman return -EPERM; 21719566d674SEric W. Biederman } 21729566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 21739566d674SEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { 21749566d674SEric W. Biederman return -EPERM; 21759566d674SEric W. Biederman } 21769566d674SEric W. Biederman 2177ff36fe2cSEric Paris err = security_sb_remount(sb, data); 2178ff36fe2cSEric Paris if (err) 2179ff36fe2cSEric Paris return err; 2180ff36fe2cSEric Paris 21811da177e4SLinus Torvalds down_write(&sb->s_umount); 21822e4b7fcdSDave Hansen if (flags & MS_BIND) 21832d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 218457eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 218557eccb83SAl Viro err = -EPERM; 21864aa98cf7SAl Viro else 21871da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 21887b43a79fSAl Viro if (!err) { 2189719ea2fbSAl Viro lock_mount_hash(); 2190a6138db8SEric W. Biederman mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 2191143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 2192143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2193719ea2fbSAl Viro unlock_mount_hash(); 21940e55a7ccSDan Williams } 21956339dab8SAl Viro up_write(&sb->s_umount); 21961da177e4SLinus Torvalds return err; 21971da177e4SLinus Torvalds } 21981da177e4SLinus Torvalds 2199cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 22009676f0c6SRam Pai { 2201315fc83eSAl Viro struct mount *p; 2202909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2203fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 22049676f0c6SRam Pai return 1; 22059676f0c6SRam Pai } 22069676f0c6SRam Pai return 0; 22079676f0c6SRam Pai } 22089676f0c6SRam Pai 2209808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 22101da177e4SLinus Torvalds { 22112d92ab3cSAl Viro struct path old_path, parent_path; 2212676da58dSAl Viro struct mount *p; 22130fb54e50SAl Viro struct mount *old; 221484d17192SAl Viro struct mountpoint *mp; 221557eccb83SAl Viro int err; 22161da177e4SLinus Torvalds if (!old_name || !*old_name) 22171da177e4SLinus Torvalds return -EINVAL; 22182d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 22191da177e4SLinus Torvalds if (err) 22201da177e4SLinus Torvalds return err; 22211da177e4SLinus Torvalds 222284d17192SAl Viro mp = lock_mount(path); 222384d17192SAl Viro err = PTR_ERR(mp); 222484d17192SAl Viro if (IS_ERR(mp)) 2225cc53ce53SDavid Howells goto out; 2226cc53ce53SDavid Howells 2227143c8c91SAl Viro old = real_mount(old_path.mnt); 2228fc7be130SAl Viro p = real_mount(path->mnt); 2229143c8c91SAl Viro 22301da177e4SLinus Torvalds err = -EINVAL; 2231fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 22321da177e4SLinus Torvalds goto out1; 22331da177e4SLinus Torvalds 22345ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 22355ff9d8a6SEric W. Biederman goto out1; 22365ff9d8a6SEric W. Biederman 22371da177e4SLinus Torvalds err = -EINVAL; 22382d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 223921444403SRam Pai goto out1; 22401da177e4SLinus Torvalds 2241676da58dSAl Viro if (!mnt_has_parent(old)) 224221444403SRam Pai goto out1; 22431da177e4SLinus Torvalds 2244e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2245e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 224621444403SRam Pai goto out1; 224721444403SRam Pai /* 224821444403SRam Pai * Don't move a mount residing in a shared parent. 224921444403SRam Pai */ 2250fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 225121444403SRam Pai goto out1; 22529676f0c6SRam Pai /* 22539676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 22549676f0c6SRam Pai * mount which is shared. 22559676f0c6SRam Pai */ 2256fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 22579676f0c6SRam Pai goto out1; 22581da177e4SLinus Torvalds err = -ELOOP; 2259fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2260676da58dSAl Viro if (p == old) 226121444403SRam Pai goto out1; 22621da177e4SLinus Torvalds 226384d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 22644ac91378SJan Blunck if (err) 226521444403SRam Pai goto out1; 22661da177e4SLinus Torvalds 22671da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 22681da177e4SLinus Torvalds * automatically */ 22696776db3dSAl Viro list_del_init(&old->mnt_expire); 22701da177e4SLinus Torvalds out1: 227184d17192SAl Viro unlock_mount(mp); 22721da177e4SLinus Torvalds out: 22731da177e4SLinus Torvalds if (!err) 22741a390689SAl Viro path_put(&parent_path); 22752d92ab3cSAl Viro path_put(&old_path); 22761da177e4SLinus Torvalds return err; 22771da177e4SLinus Torvalds } 22781da177e4SLinus Torvalds 22799d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 22809d412a43SAl Viro { 22819d412a43SAl Viro int err; 22829d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 22839d412a43SAl Viro if (subtype) { 22849d412a43SAl Viro subtype++; 22859d412a43SAl Viro err = -EINVAL; 22869d412a43SAl Viro if (!subtype[0]) 22879d412a43SAl Viro goto err; 22889d412a43SAl Viro } else 22899d412a43SAl Viro subtype = ""; 22909d412a43SAl Viro 22919d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 22929d412a43SAl Viro err = -ENOMEM; 22939d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 22949d412a43SAl Viro goto err; 22959d412a43SAl Viro return mnt; 22969d412a43SAl Viro 22979d412a43SAl Viro err: 22989d412a43SAl Viro mntput(mnt); 22999d412a43SAl Viro return ERR_PTR(err); 23009d412a43SAl Viro } 23019d412a43SAl Viro 23029d412a43SAl Viro /* 23039d412a43SAl Viro * add a mount into a namespace's mount tree 23049d412a43SAl Viro */ 230595bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 23069d412a43SAl Viro { 230784d17192SAl Viro struct mountpoint *mp; 230884d17192SAl Viro struct mount *parent; 23099d412a43SAl Viro int err; 23109d412a43SAl Viro 2311f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 23129d412a43SAl Viro 231384d17192SAl Viro mp = lock_mount(path); 231484d17192SAl Viro if (IS_ERR(mp)) 231584d17192SAl Viro return PTR_ERR(mp); 23169d412a43SAl Viro 231784d17192SAl Viro parent = real_mount(path->mnt); 23189d412a43SAl Viro err = -EINVAL; 231984d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2320156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2321156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 23229d412a43SAl Viro goto unlock; 2323156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 232484d17192SAl Viro if (!parent->mnt_ns) 2325156cacb1SAl Viro goto unlock; 2326156cacb1SAl Viro } 23279d412a43SAl Viro 23289d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 23299d412a43SAl Viro err = -EBUSY; 233095bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 23319d412a43SAl Viro path->mnt->mnt_root == path->dentry) 23329d412a43SAl Viro goto unlock; 23339d412a43SAl Viro 23349d412a43SAl Viro err = -EINVAL; 2335e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 23369d412a43SAl Viro goto unlock; 23379d412a43SAl Viro 233895bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 233984d17192SAl Viro err = graft_tree(newmnt, parent, mp); 23409d412a43SAl Viro 23419d412a43SAl Viro unlock: 234284d17192SAl Viro unlock_mount(mp); 23439d412a43SAl Viro return err; 23449d412a43SAl Viro } 2345b1e75df4SAl Viro 23468c6cf9ccSEric W. Biederman static bool fs_fully_visible(struct file_system_type *fs_type, int *new_mnt_flags); 23471b852bceSEric W. Biederman 23481da177e4SLinus Torvalds /* 23491da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 23501da177e4SLinus Torvalds * namespace's tree 23511da177e4SLinus Torvalds */ 23520c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2353808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 23541da177e4SLinus Torvalds { 23550c55cfc4SEric W. Biederman struct file_system_type *type; 23569b40bc90SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 23571da177e4SLinus Torvalds struct vfsmount *mnt; 235815f9a3f3SAl Viro int err; 23591da177e4SLinus Torvalds 23600c55cfc4SEric W. Biederman if (!fstype) 23611da177e4SLinus Torvalds return -EINVAL; 23621da177e4SLinus Torvalds 23630c55cfc4SEric W. Biederman type = get_fs_type(fstype); 23640c55cfc4SEric W. Biederman if (!type) 23650c55cfc4SEric W. Biederman return -ENODEV; 23660c55cfc4SEric W. Biederman 23670c55cfc4SEric W. Biederman if (user_ns != &init_user_ns) { 23680c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_MOUNT)) { 23690c55cfc4SEric W. Biederman put_filesystem(type); 23700c55cfc4SEric W. Biederman return -EPERM; 23710c55cfc4SEric W. Biederman } 23720c55cfc4SEric W. Biederman /* Only in special cases allow devices from mounts 23730c55cfc4SEric W. Biederman * created outside the initial user namespace. 23740c55cfc4SEric W. Biederman */ 23750c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { 23760c55cfc4SEric W. Biederman flags |= MS_NODEV; 23779566d674SEric W. Biederman mnt_flags |= MNT_NODEV | MNT_LOCK_NODEV; 23780c55cfc4SEric W. Biederman } 23791b852bceSEric W. Biederman if (type->fs_flags & FS_USERNS_VISIBLE) { 23808c6cf9ccSEric W. Biederman if (!fs_fully_visible(type, &mnt_flags)) 23811b852bceSEric W. Biederman return -EPERM; 23821b852bceSEric W. Biederman } 23830c55cfc4SEric W. Biederman } 23840c55cfc4SEric W. Biederman 23850c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 23860c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 23870c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 23880c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 23890c55cfc4SEric W. Biederman 23900c55cfc4SEric W. Biederman put_filesystem(type); 23911da177e4SLinus Torvalds if (IS_ERR(mnt)) 23921da177e4SLinus Torvalds return PTR_ERR(mnt); 23931da177e4SLinus Torvalds 239495bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 239515f9a3f3SAl Viro if (err) 239615f9a3f3SAl Viro mntput(mnt); 239715f9a3f3SAl Viro return err; 23981da177e4SLinus Torvalds } 23991da177e4SLinus Torvalds 240019a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 240119a167afSAl Viro { 24026776db3dSAl Viro struct mount *mnt = real_mount(m); 240319a167afSAl Viro int err; 240419a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 240519a167afSAl Viro * expired before we get a chance to add it 240619a167afSAl Viro */ 24076776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 240819a167afSAl Viro 240919a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 241019a167afSAl Viro m->mnt_root == path->dentry) { 2411b1e75df4SAl Viro err = -ELOOP; 2412b1e75df4SAl Viro goto fail; 241319a167afSAl Viro } 241419a167afSAl Viro 241595bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2416b1e75df4SAl Viro if (!err) 2417b1e75df4SAl Viro return 0; 2418b1e75df4SAl Viro fail: 2419b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 24206776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 242197216be0SAl Viro namespace_lock(); 24226776db3dSAl Viro list_del_init(&mnt->mnt_expire); 242397216be0SAl Viro namespace_unlock(); 242419a167afSAl Viro } 2425b1e75df4SAl Viro mntput(m); 2426b1e75df4SAl Viro mntput(m); 242719a167afSAl Viro return err; 242819a167afSAl Viro } 242919a167afSAl Viro 2430ea5b778aSDavid Howells /** 2431ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2432ea5b778aSDavid Howells * @mnt: The mount to list. 2433ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2434ea5b778aSDavid Howells */ 2435ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2436ea5b778aSDavid Howells { 243797216be0SAl Viro namespace_lock(); 2438ea5b778aSDavid Howells 24396776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2440ea5b778aSDavid Howells 244197216be0SAl Viro namespace_unlock(); 2442ea5b778aSDavid Howells } 2443ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2444ea5b778aSDavid Howells 2445ea5b778aSDavid Howells /* 24461da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 24471da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 24481da177e4SLinus Torvalds * here 24491da177e4SLinus Torvalds */ 24501da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 24511da177e4SLinus Torvalds { 2452761d5c38SAl Viro struct mount *mnt, *next; 24531da177e4SLinus Torvalds LIST_HEAD(graveyard); 24541da177e4SLinus Torvalds 24551da177e4SLinus Torvalds if (list_empty(mounts)) 24561da177e4SLinus Torvalds return; 24571da177e4SLinus Torvalds 245897216be0SAl Viro namespace_lock(); 2459719ea2fbSAl Viro lock_mount_hash(); 24601da177e4SLinus Torvalds 24611da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 24621da177e4SLinus Torvalds * following criteria: 24631da177e4SLinus Torvalds * - only referenced by its parent vfsmount 24641da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 24651da177e4SLinus Torvalds * cleared by mntput()) 24661da177e4SLinus Torvalds */ 24676776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2468863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 24691ab59738SAl Viro propagate_mount_busy(mnt, 1)) 24701da177e4SLinus Torvalds continue; 24716776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 24721da177e4SLinus Torvalds } 2473bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 24746776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2475143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2476e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2477bcc5c7d2SAl Viro } 2478719ea2fbSAl Viro unlock_mount_hash(); 24793ab6abeeSAl Viro namespace_unlock(); 24801da177e4SLinus Torvalds } 24811da177e4SLinus Torvalds 24821da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 24831da177e4SLinus Torvalds 24841da177e4SLinus Torvalds /* 24855528f911STrond Myklebust * Ripoff of 'select_parent()' 24865528f911STrond Myklebust * 24875528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 24885528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 24895528f911STrond Myklebust */ 2490692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 24915528f911STrond Myklebust { 2492692afc31SAl Viro struct mount *this_parent = parent; 24935528f911STrond Myklebust struct list_head *next; 24945528f911STrond Myklebust int found = 0; 24955528f911STrond Myklebust 24965528f911STrond Myklebust repeat: 24976b41d536SAl Viro next = this_parent->mnt_mounts.next; 24985528f911STrond Myklebust resume: 24996b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 25005528f911STrond Myklebust struct list_head *tmp = next; 25016b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 25025528f911STrond Myklebust 25035528f911STrond Myklebust next = tmp->next; 2504692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 25055528f911STrond Myklebust continue; 25065528f911STrond Myklebust /* 25075528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 25085528f911STrond Myklebust */ 25096b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 25105528f911STrond Myklebust this_parent = mnt; 25115528f911STrond Myklebust goto repeat; 25125528f911STrond Myklebust } 25135528f911STrond Myklebust 25141ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 25156776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 25165528f911STrond Myklebust found++; 25175528f911STrond Myklebust } 25185528f911STrond Myklebust } 25195528f911STrond Myklebust /* 25205528f911STrond Myklebust * All done at this level ... ascend and resume the search 25215528f911STrond Myklebust */ 25225528f911STrond Myklebust if (this_parent != parent) { 25236b41d536SAl Viro next = this_parent->mnt_child.next; 25240714a533SAl Viro this_parent = this_parent->mnt_parent; 25255528f911STrond Myklebust goto resume; 25265528f911STrond Myklebust } 25275528f911STrond Myklebust return found; 25285528f911STrond Myklebust } 25295528f911STrond Myklebust 25305528f911STrond Myklebust /* 25315528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 25325528f911STrond Myklebust * submounts of a specific parent mountpoint 253399b7db7bSNick Piggin * 253448a066e7SAl Viro * mount_lock must be held for write 25355528f911STrond Myklebust */ 2536b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 25375528f911STrond Myklebust { 25385528f911STrond Myklebust LIST_HEAD(graveyard); 2539761d5c38SAl Viro struct mount *m; 25405528f911STrond Myklebust 25415528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2542c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2543bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2544761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 25456776db3dSAl Viro mnt_expire); 2546143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2547e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2548bcc5c7d2SAl Viro } 2549bcc5c7d2SAl Viro } 25505528f911STrond Myklebust } 25515528f911STrond Myklebust 25525528f911STrond Myklebust /* 25531da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 25541da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 25551da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 25561da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 25571da177e4SLinus Torvalds */ 2558b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2559b58fed8bSRam Pai unsigned long n) 25601da177e4SLinus Torvalds { 25611da177e4SLinus Torvalds char *t = to; 25621da177e4SLinus Torvalds const char __user *f = from; 25631da177e4SLinus Torvalds char c; 25641da177e4SLinus Torvalds 25651da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 25661da177e4SLinus Torvalds return n; 25671da177e4SLinus Torvalds 25681da177e4SLinus Torvalds while (n) { 25691da177e4SLinus Torvalds if (__get_user(c, f)) { 25701da177e4SLinus Torvalds memset(t, 0, n); 25711da177e4SLinus Torvalds break; 25721da177e4SLinus Torvalds } 25731da177e4SLinus Torvalds *t++ = c; 25741da177e4SLinus Torvalds f++; 25751da177e4SLinus Torvalds n--; 25761da177e4SLinus Torvalds } 25771da177e4SLinus Torvalds return n; 25781da177e4SLinus Torvalds } 25791da177e4SLinus Torvalds 25801da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 25811da177e4SLinus Torvalds { 25821da177e4SLinus Torvalds int i; 25831da177e4SLinus Torvalds unsigned long page; 25841da177e4SLinus Torvalds unsigned long size; 25851da177e4SLinus Torvalds 25861da177e4SLinus Torvalds *where = 0; 25871da177e4SLinus Torvalds if (!data) 25881da177e4SLinus Torvalds return 0; 25891da177e4SLinus Torvalds 25901da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 25911da177e4SLinus Torvalds return -ENOMEM; 25921da177e4SLinus Torvalds 25931da177e4SLinus Torvalds /* We only care that *some* data at the address the user 25941da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 25951da177e4SLinus Torvalds * the remainder of the page. 25961da177e4SLinus Torvalds */ 25971da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 25981da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 25991da177e4SLinus Torvalds if (size > PAGE_SIZE) 26001da177e4SLinus Torvalds size = PAGE_SIZE; 26011da177e4SLinus Torvalds 26021da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 26031da177e4SLinus Torvalds if (!i) { 26041da177e4SLinus Torvalds free_page(page); 26051da177e4SLinus Torvalds return -EFAULT; 26061da177e4SLinus Torvalds } 26071da177e4SLinus Torvalds if (i != PAGE_SIZE) 26081da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 26091da177e4SLinus Torvalds *where = page; 26101da177e4SLinus Torvalds return 0; 26111da177e4SLinus Torvalds } 26121da177e4SLinus Torvalds 2613b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2614eca6f534SVegard Nossum { 2615b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2616eca6f534SVegard Nossum } 2617eca6f534SVegard Nossum 26181da177e4SLinus Torvalds /* 26191da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 26201da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 26211da177e4SLinus Torvalds * 26221da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 26231da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 26241da177e4SLinus Torvalds * information (or be NULL). 26251da177e4SLinus Torvalds * 26261da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 26271da177e4SLinus Torvalds * When the flags word was introduced its top half was required 26281da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 26291da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 26301da177e4SLinus Torvalds * and must be discarded. 26311da177e4SLinus Torvalds */ 26325e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2633808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 26341da177e4SLinus Torvalds { 26352d92ab3cSAl Viro struct path path; 26361da177e4SLinus Torvalds int retval = 0; 26371da177e4SLinus Torvalds int mnt_flags = 0; 26381da177e4SLinus Torvalds 26391da177e4SLinus Torvalds /* Discard magic */ 26401da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 26411da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 26421da177e4SLinus Torvalds 26431da177e4SLinus Torvalds /* Basic sanity checks */ 26441da177e4SLinus Torvalds if (data_page) 26451da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 26461da177e4SLinus Torvalds 2647a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 26485e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2649a27ab9f2STetsuo Handa if (retval) 2650a27ab9f2STetsuo Handa return retval; 2651a27ab9f2STetsuo Handa 2652a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2653a27ab9f2STetsuo Handa type_page, flags, data_page); 26540d5cadb8SAl Viro if (!retval && !may_mount()) 26550d5cadb8SAl Viro retval = -EPERM; 2656a27ab9f2STetsuo Handa if (retval) 2657a27ab9f2STetsuo Handa goto dput_out; 2658a27ab9f2STetsuo Handa 2659613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2660613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 26610a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 26620a1c01c9SMatthew Garrett 26631da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 26641da177e4SLinus Torvalds if (flags & MS_NOSUID) 26651da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 26661da177e4SLinus Torvalds if (flags & MS_NODEV) 26671da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 26681da177e4SLinus Torvalds if (flags & MS_NOEXEC) 26691da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2670fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2671fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2672fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2673fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2674d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2675d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 26762e4b7fcdSDave Hansen if (flags & MS_RDONLY) 26772e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2678fc33a7bbSChristoph Hellwig 2679ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2680ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2681ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2682ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2683ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2684ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2685ffbc6f0eSEric W. Biederman } 2686ffbc6f0eSEric W. Biederman 26877a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2688d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2689d0adde57SMatthew Garrett MS_STRICTATIME); 26901da177e4SLinus Torvalds 26911da177e4SLinus Torvalds if (flags & MS_REMOUNT) 26922d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 26931da177e4SLinus Torvalds data_page); 26941da177e4SLinus Torvalds else if (flags & MS_BIND) 26952d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 26969676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 26972d92ab3cSAl Viro retval = do_change_type(&path, flags); 26981da177e4SLinus Torvalds else if (flags & MS_MOVE) 26992d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 27001da177e4SLinus Torvalds else 27012d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 27021da177e4SLinus Torvalds dev_name, data_page); 27031da177e4SLinus Torvalds dput_out: 27042d92ab3cSAl Viro path_put(&path); 27051da177e4SLinus Torvalds return retval; 27061da177e4SLinus Torvalds } 27071da177e4SLinus Torvalds 2708771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2709771b1371SEric W. Biederman { 27106344c433SAl Viro ns_free_inum(&ns->ns); 2711771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2712771b1371SEric W. Biederman kfree(ns); 2713771b1371SEric W. Biederman } 2714771b1371SEric W. Biederman 27158823c079SEric W. Biederman /* 27168823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 27178823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 27188823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 27198823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 27208823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 27218823c079SEric W. Biederman */ 27228823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 27238823c079SEric W. Biederman 2724771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2725cf8d2c11STrond Myklebust { 2726cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 272798f842e6SEric W. Biederman int ret; 2728cf8d2c11STrond Myklebust 2729cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2730cf8d2c11STrond Myklebust if (!new_ns) 2731cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 27326344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 273398f842e6SEric W. Biederman if (ret) { 273498f842e6SEric W. Biederman kfree(new_ns); 273598f842e6SEric W. Biederman return ERR_PTR(ret); 273698f842e6SEric W. Biederman } 273733c42940SAl Viro new_ns->ns.ops = &mntns_operations; 27388823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2739cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2740cf8d2c11STrond Myklebust new_ns->root = NULL; 2741cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2742cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2743cf8d2c11STrond Myklebust new_ns->event = 0; 2744771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2745cf8d2c11STrond Myklebust return new_ns; 2746cf8d2c11STrond Myklebust } 2747cf8d2c11STrond Myklebust 27489559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 27499559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 27501da177e4SLinus Torvalds { 27516b3286edSKirill Korotaev struct mnt_namespace *new_ns; 27527f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2753315fc83eSAl Viro struct mount *p, *q; 27549559f689SAl Viro struct mount *old; 2755cb338d06SAl Viro struct mount *new; 27567a472ef4SEric W. Biederman int copy_flags; 27571da177e4SLinus Torvalds 27589559f689SAl Viro BUG_ON(!ns); 27599559f689SAl Viro 27609559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 27619559f689SAl Viro get_mnt_ns(ns); 27629559f689SAl Viro return ns; 27639559f689SAl Viro } 27649559f689SAl Viro 27659559f689SAl Viro old = ns->root; 27669559f689SAl Viro 2767771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2768cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2769cf8d2c11STrond Myklebust return new_ns; 27701da177e4SLinus Torvalds 277197216be0SAl Viro namespace_lock(); 27721da177e4SLinus Torvalds /* First pass: copy the tree topology */ 27734ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 27749559f689SAl Viro if (user_ns != ns->user_ns) 2775132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 27767a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2777be34d1a3SDavid Howells if (IS_ERR(new)) { 2778328e6d90SAl Viro namespace_unlock(); 2779771b1371SEric W. Biederman free_mnt_ns(new_ns); 2780be34d1a3SDavid Howells return ERR_CAST(new); 27811da177e4SLinus Torvalds } 2782be08d6d2SAl Viro new_ns->root = new; 27831a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 27841da177e4SLinus Torvalds 27851da177e4SLinus Torvalds /* 27861da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 27871da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 27881da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 27891da177e4SLinus Torvalds */ 2790909b0a88SAl Viro p = old; 2791cb338d06SAl Viro q = new; 27921da177e4SLinus Torvalds while (p) { 2793143c8c91SAl Viro q->mnt_ns = new_ns; 27949559f689SAl Viro if (new_fs) { 27959559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 27969559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2797315fc83eSAl Viro rootmnt = &p->mnt; 27981da177e4SLinus Torvalds } 27999559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 28009559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2801315fc83eSAl Viro pwdmnt = &p->mnt; 28021da177e4SLinus Torvalds } 28031da177e4SLinus Torvalds } 2804909b0a88SAl Viro p = next_mnt(p, old); 2805909b0a88SAl Viro q = next_mnt(q, new); 28064ce5d2b1SEric W. Biederman if (!q) 28074ce5d2b1SEric W. Biederman break; 28084ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 28094ce5d2b1SEric W. Biederman p = next_mnt(p, old); 28101da177e4SLinus Torvalds } 2811328e6d90SAl Viro namespace_unlock(); 28121da177e4SLinus Torvalds 28131da177e4SLinus Torvalds if (rootmnt) 2814f03c6599SAl Viro mntput(rootmnt); 28151da177e4SLinus Torvalds if (pwdmnt) 2816f03c6599SAl Viro mntput(pwdmnt); 28171da177e4SLinus Torvalds 2818741a2951SJANAK DESAI return new_ns; 2819741a2951SJANAK DESAI } 2820741a2951SJANAK DESAI 2821cf8d2c11STrond Myklebust /** 2822cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2823cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2824cf8d2c11STrond Myklebust */ 28251a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2826cf8d2c11STrond Myklebust { 2827771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2828cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 28291a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 28301a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2831be08d6d2SAl Viro new_ns->root = mnt; 2832b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2833c1334495SAl Viro } else { 28341a4eeaf2SAl Viro mntput(m); 2835cf8d2c11STrond Myklebust } 2836cf8d2c11STrond Myklebust return new_ns; 2837cf8d2c11STrond Myklebust } 2838cf8d2c11STrond Myklebust 2839ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2840ea441d11SAl Viro { 2841ea441d11SAl Viro struct mnt_namespace *ns; 2842d31da0f0SAl Viro struct super_block *s; 2843ea441d11SAl Viro struct path path; 2844ea441d11SAl Viro int err; 2845ea441d11SAl Viro 2846ea441d11SAl Viro ns = create_mnt_ns(mnt); 2847ea441d11SAl Viro if (IS_ERR(ns)) 2848ea441d11SAl Viro return ERR_CAST(ns); 2849ea441d11SAl Viro 2850ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2851ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2852ea441d11SAl Viro 2853ea441d11SAl Viro put_mnt_ns(ns); 2854ea441d11SAl Viro 2855ea441d11SAl Viro if (err) 2856ea441d11SAl Viro return ERR_PTR(err); 2857ea441d11SAl Viro 2858ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2859d31da0f0SAl Viro s = path.mnt->mnt_sb; 2860d31da0f0SAl Viro atomic_inc(&s->s_active); 2861ea441d11SAl Viro mntput(path.mnt); 2862ea441d11SAl Viro /* lock the sucker */ 2863d31da0f0SAl Viro down_write(&s->s_umount); 2864ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2865ea441d11SAl Viro return path.dentry; 2866ea441d11SAl Viro } 2867ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2868ea441d11SAl Viro 2869bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2870bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 28711da177e4SLinus Torvalds { 2872eca6f534SVegard Nossum int ret; 2873eca6f534SVegard Nossum char *kernel_type; 2874eca6f534SVegard Nossum char *kernel_dev; 28751da177e4SLinus Torvalds unsigned long data_page; 28761da177e4SLinus Torvalds 2877b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 2878b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 2879b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 2880eca6f534SVegard Nossum goto out_type; 28811da177e4SLinus Torvalds 2882b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 2883b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 2884b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 2885eca6f534SVegard Nossum goto out_dev; 28861da177e4SLinus Torvalds 2887eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2888eca6f534SVegard Nossum if (ret < 0) 2889eca6f534SVegard Nossum goto out_data; 28901da177e4SLinus Torvalds 28915e6123f3SSeunghun Lee ret = do_mount(kernel_dev, dir_name, kernel_type, flags, 2892eca6f534SVegard Nossum (void *) data_page); 2893eca6f534SVegard Nossum 28941da177e4SLinus Torvalds free_page(data_page); 2895eca6f534SVegard Nossum out_data: 2896eca6f534SVegard Nossum kfree(kernel_dev); 2897eca6f534SVegard Nossum out_dev: 2898eca6f534SVegard Nossum kfree(kernel_type); 2899eca6f534SVegard Nossum out_type: 2900eca6f534SVegard Nossum return ret; 29011da177e4SLinus Torvalds } 29021da177e4SLinus Torvalds 29031da177e4SLinus Torvalds /* 2904afac7cbaSAl Viro * Return true if path is reachable from root 2905afac7cbaSAl Viro * 290648a066e7SAl Viro * namespace_sem or mount_lock is held 2907afac7cbaSAl Viro */ 2908643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2909afac7cbaSAl Viro const struct path *root) 2910afac7cbaSAl Viro { 2911643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2912a73324daSAl Viro dentry = mnt->mnt_mountpoint; 29130714a533SAl Viro mnt = mnt->mnt_parent; 2914afac7cbaSAl Viro } 2915643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2916afac7cbaSAl Viro } 2917afac7cbaSAl Viro 2918afac7cbaSAl Viro int path_is_under(struct path *path1, struct path *path2) 2919afac7cbaSAl Viro { 2920afac7cbaSAl Viro int res; 292148a066e7SAl Viro read_seqlock_excl(&mount_lock); 2922643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 292348a066e7SAl Viro read_sequnlock_excl(&mount_lock); 2924afac7cbaSAl Viro return res; 2925afac7cbaSAl Viro } 2926afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 2927afac7cbaSAl Viro 2928afac7cbaSAl Viro /* 29291da177e4SLinus Torvalds * pivot_root Semantics: 29301da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 29311da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 29321da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 29331da177e4SLinus Torvalds * 29341da177e4SLinus Torvalds * Restrictions: 29351da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 29361da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 29371da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 29381da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 29391da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 29401da177e4SLinus Torvalds * 29414a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 29424a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 29434a0d11faSNeil Brown * in this situation. 29444a0d11faSNeil Brown * 29451da177e4SLinus Torvalds * Notes: 29461da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 29471da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 29481da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 29491da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 29501da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 29511da177e4SLinus Torvalds * first. 29521da177e4SLinus Torvalds */ 29533480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 29543480b257SHeiko Carstens const char __user *, put_old) 29551da177e4SLinus Torvalds { 29562d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 295784d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 295884d17192SAl Viro struct mountpoint *old_mp, *root_mp; 29591da177e4SLinus Torvalds int error; 29601da177e4SLinus Torvalds 29619b40bc90SAl Viro if (!may_mount()) 29621da177e4SLinus Torvalds return -EPERM; 29631da177e4SLinus Torvalds 29642d8f3038SAl Viro error = user_path_dir(new_root, &new); 29651da177e4SLinus Torvalds if (error) 29661da177e4SLinus Torvalds goto out0; 29671da177e4SLinus Torvalds 29682d8f3038SAl Viro error = user_path_dir(put_old, &old); 29691da177e4SLinus Torvalds if (error) 29701da177e4SLinus Torvalds goto out1; 29711da177e4SLinus Torvalds 29722d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 2973b12cea91SAl Viro if (error) 2974b12cea91SAl Viro goto out2; 29751da177e4SLinus Torvalds 2976f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 297784d17192SAl Viro old_mp = lock_mount(&old); 297884d17192SAl Viro error = PTR_ERR(old_mp); 297984d17192SAl Viro if (IS_ERR(old_mp)) 2980b12cea91SAl Viro goto out3; 2981b12cea91SAl Viro 29821da177e4SLinus Torvalds error = -EINVAL; 2983419148daSAl Viro new_mnt = real_mount(new.mnt); 2984419148daSAl Viro root_mnt = real_mount(root.mnt); 298584d17192SAl Viro old_mnt = real_mount(old.mnt); 298684d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 2987fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 2988fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 2989b12cea91SAl Viro goto out4; 2990143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 2991b12cea91SAl Viro goto out4; 29925ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 29935ff9d8a6SEric W. Biederman goto out4; 29941da177e4SLinus Torvalds error = -ENOENT; 2995f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 2996b12cea91SAl Viro goto out4; 29971da177e4SLinus Torvalds error = -EBUSY; 299884d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 2999b12cea91SAl Viro goto out4; /* loop, on the same file system */ 30001da177e4SLinus Torvalds error = -EINVAL; 30018c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3002b12cea91SAl Viro goto out4; /* not a mountpoint */ 3003676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3004b12cea91SAl Viro goto out4; /* not attached */ 300584d17192SAl Viro root_mp = root_mnt->mnt_mp; 30062d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3007b12cea91SAl Viro goto out4; /* not a mountpoint */ 3008676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3009b12cea91SAl Viro goto out4; /* not attached */ 30104ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 301184d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3012b12cea91SAl Viro goto out4; 30130d082601SEric W. Biederman /* make certain new is below the root */ 30140d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 30150d082601SEric W. Biederman goto out4; 301684d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3017719ea2fbSAl Viro lock_mount_hash(); 3018419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3019419148daSAl Viro detach_mnt(root_mnt, &root_parent); 30205ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 30215ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 30225ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 30235ff9d8a6SEric W. Biederman } 30244ac91378SJan Blunck /* mount old root on put_old */ 302584d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 30264ac91378SJan Blunck /* mount new_root on / */ 302784d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 30286b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 30294fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 30304fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 3031719ea2fbSAl Viro unlock_mount_hash(); 30322d8f3038SAl Viro chroot_fs_refs(&root, &new); 303384d17192SAl Viro put_mountpoint(root_mp); 30341da177e4SLinus Torvalds error = 0; 3035b12cea91SAl Viro out4: 303684d17192SAl Viro unlock_mount(old_mp); 3037b12cea91SAl Viro if (!error) { 30381a390689SAl Viro path_put(&root_parent); 30391a390689SAl Viro path_put(&parent_path); 3040b12cea91SAl Viro } 3041b12cea91SAl Viro out3: 30428c3ee42eSAl Viro path_put(&root); 3043b12cea91SAl Viro out2: 30442d8f3038SAl Viro path_put(&old); 30451da177e4SLinus Torvalds out1: 30462d8f3038SAl Viro path_put(&new); 30471da177e4SLinus Torvalds out0: 30481da177e4SLinus Torvalds return error; 30491da177e4SLinus Torvalds } 30501da177e4SLinus Torvalds 30511da177e4SLinus Torvalds static void __init init_mount_tree(void) 30521da177e4SLinus Torvalds { 30531da177e4SLinus Torvalds struct vfsmount *mnt; 30546b3286edSKirill Korotaev struct mnt_namespace *ns; 3055ac748a09SJan Blunck struct path root; 30560c55cfc4SEric W. Biederman struct file_system_type *type; 30571da177e4SLinus Torvalds 30580c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 30590c55cfc4SEric W. Biederman if (!type) 30600c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 30610c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 30620c55cfc4SEric W. Biederman put_filesystem(type); 30631da177e4SLinus Torvalds if (IS_ERR(mnt)) 30641da177e4SLinus Torvalds panic("Can't create rootfs"); 3065b3e19d92SNick Piggin 30663b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 30673b22edc5STrond Myklebust if (IS_ERR(ns)) 30681da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 30691da177e4SLinus Torvalds 30706b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 30716b3286edSKirill Korotaev get_mnt_ns(ns); 30721da177e4SLinus Torvalds 3073be08d6d2SAl Viro root.mnt = mnt; 3074be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3075da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3076ac748a09SJan Blunck 3077ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3078ac748a09SJan Blunck set_fs_root(current->fs, &root); 30791da177e4SLinus Torvalds } 30801da177e4SLinus Torvalds 308174bf17cfSDenis Cheng void __init mnt_init(void) 30821da177e4SLinus Torvalds { 308313f14b4dSEric Dumazet unsigned u; 308415a67dd8SRandy Dunlap int err; 30851da177e4SLinus Torvalds 30867d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 308720c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 30881da177e4SLinus Torvalds 30890818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 309038129a13SAl Viro sizeof(struct hlist_head), 30910818bf27SAl Viro mhash_entries, 19, 30920818bf27SAl Viro 0, 30930818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 30940818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 30950818bf27SAl Viro sizeof(struct hlist_head), 30960818bf27SAl Viro mphash_entries, 19, 30970818bf27SAl Viro 0, 30980818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 30991da177e4SLinus Torvalds 310084d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 31011da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 31021da177e4SLinus Torvalds 31030818bf27SAl Viro for (u = 0; u <= m_hash_mask; u++) 310438129a13SAl Viro INIT_HLIST_HEAD(&mount_hashtable[u]); 31050818bf27SAl Viro for (u = 0; u <= mp_hash_mask; u++) 31060818bf27SAl Viro INIT_HLIST_HEAD(&mountpoint_hashtable[u]); 31071da177e4SLinus Torvalds 31084b93dc9bSTejun Heo kernfs_init(); 31094b93dc9bSTejun Heo 311015a67dd8SRandy Dunlap err = sysfs_init(); 311115a67dd8SRandy Dunlap if (err) 311215a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 31138e24eea7SHarvey Harrison __func__, err); 311400d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 311500d26666SGreg Kroah-Hartman if (!fs_kobj) 31168e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 31171da177e4SLinus Torvalds init_rootfs(); 31181da177e4SLinus Torvalds init_mount_tree(); 31191da177e4SLinus Torvalds } 31201da177e4SLinus Torvalds 3121616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 31221da177e4SLinus Torvalds { 3123d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3124616511d0STrond Myklebust return; 31257b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3126771b1371SEric W. Biederman free_mnt_ns(ns); 31271da177e4SLinus Torvalds } 31289d412a43SAl Viro 31299d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 31309d412a43SAl Viro { 3131423e0ab0STim Chen struct vfsmount *mnt; 3132423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 3133423e0ab0STim Chen if (!IS_ERR(mnt)) { 3134423e0ab0STim Chen /* 3135423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3136423e0ab0STim Chen * we unmount before file sys is unregistered 3137423e0ab0STim Chen */ 3138f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3139423e0ab0STim Chen } 3140423e0ab0STim Chen return mnt; 31419d412a43SAl Viro } 31429d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3143423e0ab0STim Chen 3144423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3145423e0ab0STim Chen { 3146423e0ab0STim Chen /* release long term mount so mount point can be released */ 3147423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3148f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 314948a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3150423e0ab0STim Chen mntput(mnt); 3151423e0ab0STim Chen } 3152423e0ab0STim Chen } 3153423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 315402125a82SAl Viro 315502125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 315602125a82SAl Viro { 3157143c8c91SAl Viro return check_mnt(real_mount(mnt)); 315802125a82SAl Viro } 31598823c079SEric W. Biederman 31603151527eSEric W. Biederman bool current_chrooted(void) 31613151527eSEric W. Biederman { 31623151527eSEric W. Biederman /* Does the current process have a non-standard root */ 31633151527eSEric W. Biederman struct path ns_root; 31643151527eSEric W. Biederman struct path fs_root; 31653151527eSEric W. Biederman bool chrooted; 31663151527eSEric W. Biederman 31673151527eSEric W. Biederman /* Find the namespace root */ 31683151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 31693151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 31703151527eSEric W. Biederman path_get(&ns_root); 31713151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 31723151527eSEric W. Biederman ; 31733151527eSEric W. Biederman 31743151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 31753151527eSEric W. Biederman 31763151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 31773151527eSEric W. Biederman 31783151527eSEric W. Biederman path_put(&fs_root); 31793151527eSEric W. Biederman path_put(&ns_root); 31803151527eSEric W. Biederman 31813151527eSEric W. Biederman return chrooted; 31823151527eSEric W. Biederman } 31833151527eSEric W. Biederman 31848c6cf9ccSEric W. Biederman static bool fs_fully_visible(struct file_system_type *type, int *new_mnt_flags) 318587a8ebd6SEric W. Biederman { 318687a8ebd6SEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 31878c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 318887a8ebd6SEric W. Biederman struct mount *mnt; 3189e51db735SEric W. Biederman bool visible = false; 319087a8ebd6SEric W. Biederman 3191e51db735SEric W. Biederman if (unlikely(!ns)) 3192e51db735SEric W. Biederman return false; 3193e51db735SEric W. Biederman 319444bb4385SAl Viro down_read(&namespace_sem); 319587a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3196e51db735SEric W. Biederman struct mount *child; 319777b1a97dSEric W. Biederman int mnt_flags; 319877b1a97dSEric W. Biederman 3199e51db735SEric W. Biederman if (mnt->mnt.mnt_sb->s_type != type) 3200e51db735SEric W. Biederman continue; 3201e51db735SEric W. Biederman 32027e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 32037e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 32047e96c1b0SEric W. Biederman */ 32057e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 32067e96c1b0SEric W. Biederman continue; 32077e96c1b0SEric W. Biederman 320877b1a97dSEric W. Biederman /* Read the mount flags and filter out flags that 320977b1a97dSEric W. Biederman * may safely be ignored. 321077b1a97dSEric W. Biederman */ 321177b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 321277b1a97dSEric W. Biederman if (mnt->mnt.mnt_sb->s_iflags & SB_I_NOEXEC) 321377b1a97dSEric W. Biederman mnt_flags &= ~(MNT_LOCK_NOSUID | MNT_LOCK_NOEXEC); 321477b1a97dSEric W. Biederman 32158c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 32168c6cf9ccSEric W. Biederman * than the proposed new mount. 32178c6cf9ccSEric W. Biederman */ 321877b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 32198c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 32208c6cf9ccSEric W. Biederman continue; 322177b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_NODEV) && 32228c6cf9ccSEric W. Biederman !(new_flags & MNT_NODEV)) 32238c6cf9ccSEric W. Biederman continue; 322477b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_NOSUID) && 322577b1a97dSEric W. Biederman !(new_flags & MNT_NOSUID)) 322677b1a97dSEric W. Biederman continue; 322777b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_NOEXEC) && 322877b1a97dSEric W. Biederman !(new_flags & MNT_NOEXEC)) 322977b1a97dSEric W. Biederman continue; 323077b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 323177b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 32328c6cf9ccSEric W. Biederman continue; 32338c6cf9ccSEric W. Biederman 3234ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3235ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3236ceeb0e5dSEric W. Biederman * empty directories. 3237e51db735SEric W. Biederman */ 3238e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3239e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3240ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 324177b1a97dSEric W. Biederman if (!(mnt_flags & MNT_LOCKED)) 3242ceeb0e5dSEric W. Biederman continue; 32437236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 32447236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3245e51db735SEric W. Biederman goto next; 324687a8ebd6SEric W. Biederman } 32478c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 324877b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 32498c6cf9ccSEric W. Biederman MNT_LOCK_NODEV | \ 325077b1a97dSEric W. Biederman MNT_LOCK_NOSUID | \ 325177b1a97dSEric W. Biederman MNT_LOCK_NOEXEC | \ 32528c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3253e51db735SEric W. Biederman visible = true; 3254e51db735SEric W. Biederman goto found; 3255e51db735SEric W. Biederman next: ; 325687a8ebd6SEric W. Biederman } 3257e51db735SEric W. Biederman found: 325844bb4385SAl Viro up_read(&namespace_sem); 3259e51db735SEric W. Biederman return visible; 326087a8ebd6SEric W. Biederman } 326187a8ebd6SEric W. Biederman 326264964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 32638823c079SEric W. Biederman { 326458be2825SAl Viro struct ns_common *ns = NULL; 32658823c079SEric W. Biederman struct nsproxy *nsproxy; 32668823c079SEric W. Biederman 3267728dba3aSEric W. Biederman task_lock(task); 3268728dba3aSEric W. Biederman nsproxy = task->nsproxy; 32698823c079SEric W. Biederman if (nsproxy) { 327058be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 327158be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 32728823c079SEric W. Biederman } 3273728dba3aSEric W. Biederman task_unlock(task); 32748823c079SEric W. Biederman 32758823c079SEric W. Biederman return ns; 32768823c079SEric W. Biederman } 32778823c079SEric W. Biederman 327864964528SAl Viro static void mntns_put(struct ns_common *ns) 32798823c079SEric W. Biederman { 328058be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 32818823c079SEric W. Biederman } 32828823c079SEric W. Biederman 328364964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 32848823c079SEric W. Biederman { 32858823c079SEric W. Biederman struct fs_struct *fs = current->fs; 328658be2825SAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns); 32878823c079SEric W. Biederman struct path root; 32888823c079SEric W. Biederman 32890c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3290c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3291c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3292ae11e0f1SZhao Hongjiang return -EPERM; 32938823c079SEric W. Biederman 32948823c079SEric W. Biederman if (fs->users != 1) 32958823c079SEric W. Biederman return -EINVAL; 32968823c079SEric W. Biederman 32978823c079SEric W. Biederman get_mnt_ns(mnt_ns); 32988823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 32998823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 33008823c079SEric W. Biederman 33018823c079SEric W. Biederman /* Find the root */ 33028823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 33038823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 33048823c079SEric W. Biederman path_get(&root); 33058823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 33068823c079SEric W. Biederman ; 33078823c079SEric W. Biederman 33088823c079SEric W. Biederman /* Update the pwd and root */ 33098823c079SEric W. Biederman set_fs_pwd(fs, &root); 33108823c079SEric W. Biederman set_fs_root(fs, &root); 33118823c079SEric W. Biederman 33128823c079SEric W. Biederman path_put(&root); 33138823c079SEric W. Biederman return 0; 33148823c079SEric W. Biederman } 33158823c079SEric W. Biederman 33168823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 33178823c079SEric W. Biederman .name = "mnt", 33188823c079SEric W. Biederman .type = CLONE_NEWNS, 33198823c079SEric W. Biederman .get = mntns_get, 33208823c079SEric W. Biederman .put = mntns_put, 33218823c079SEric W. Biederman .install = mntns_install, 33228823c079SEric W. Biederman }; 3323