11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 1873cd49ecSMiklos Szeredi #include <linux/idr.h> 1957f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 20d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 21d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 22d10577a8SAl Viro #include <linux/uaccess.h> 230bb80f24SDavid Howells #include <linux/proc_ns.h> 2420b4fb48SLinus Torvalds #include <linux/magic.h> 250818bf27SAl Viro #include <linux/bootmem.h> 269ea459e1SAl Viro #include <linux/task_work.h> 2707b20889SRam Pai #include "pnode.h" 28948730b0SAdrian Bunk #include "internal.h" 291da177e4SLinus Torvalds 30d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 31d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 32d2921684SEric W. Biederman 330818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 340818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 350818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 360818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 370818bf27SAl Viro 380818bf27SAl Viro static __initdata unsigned long mhash_entries; 390818bf27SAl Viro static int __init set_mhash_entries(char *str) 400818bf27SAl Viro { 410818bf27SAl Viro if (!str) 420818bf27SAl Viro return 0; 430818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 440818bf27SAl Viro return 1; 450818bf27SAl Viro } 460818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 470818bf27SAl Viro 480818bf27SAl Viro static __initdata unsigned long mphash_entries; 490818bf27SAl Viro static int __init set_mphash_entries(char *str) 500818bf27SAl Viro { 510818bf27SAl Viro if (!str) 520818bf27SAl Viro return 0; 530818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 540818bf27SAl Viro return 1; 550818bf27SAl Viro } 560818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 5713f14b4dSEric Dumazet 58c7999c36SAl Viro static u64 event; 5973cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 60719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 6199b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 62f21f6220SAl Viro static int mnt_id_start = 0; 63f21f6220SAl Viro static int mnt_group_start = 1; 645addc5ddSAl Viro 6538129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 660818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 67e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 6859aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 691da177e4SLinus Torvalds 70f87fd4c2SMiklos Szeredi /* /sys/fs */ 7100d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7200d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 73f87fd4c2SMiklos Szeredi 7499b7db7bSNick Piggin /* 7599b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7699b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 7799b7db7bSNick Piggin * up the tree. 7899b7db7bSNick Piggin * 7999b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8099b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8199b7db7bSNick Piggin */ 8248a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8399b7db7bSNick Piggin 8438129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 851da177e4SLinus Torvalds { 861da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 871da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 880818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 890818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 900818bf27SAl Viro } 910818bf27SAl Viro 920818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 930818bf27SAl Viro { 940818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 950818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 960818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 971da177e4SLinus Torvalds } 981da177e4SLinus Torvalds 99b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10073cd49ecSMiklos Szeredi { 10173cd49ecSMiklos Szeredi int res; 10273cd49ecSMiklos Szeredi 10373cd49ecSMiklos Szeredi retry: 10473cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 10599b7db7bSNick Piggin spin_lock(&mnt_id_lock); 10615169fe7SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 107f21f6220SAl Viro if (!res) 10815169fe7SAl Viro mnt_id_start = mnt->mnt_id + 1; 10999b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 11073cd49ecSMiklos Szeredi if (res == -EAGAIN) 11173cd49ecSMiklos Szeredi goto retry; 11273cd49ecSMiklos Szeredi 11373cd49ecSMiklos Szeredi return res; 11473cd49ecSMiklos Szeredi } 11573cd49ecSMiklos Szeredi 116b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11773cd49ecSMiklos Szeredi { 11815169fe7SAl Viro int id = mnt->mnt_id; 11999b7db7bSNick Piggin spin_lock(&mnt_id_lock); 120f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 121f21f6220SAl Viro if (mnt_id_start > id) 122f21f6220SAl Viro mnt_id_start = id; 12399b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 12473cd49ecSMiklos Szeredi } 12573cd49ecSMiklos Szeredi 126719f5d7fSMiklos Szeredi /* 127719f5d7fSMiklos Szeredi * Allocate a new peer group ID 128719f5d7fSMiklos Szeredi * 129719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 130719f5d7fSMiklos Szeredi */ 1314b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 132719f5d7fSMiklos Szeredi { 133f21f6220SAl Viro int res; 134f21f6220SAl Viro 135719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 136719f5d7fSMiklos Szeredi return -ENOMEM; 137719f5d7fSMiklos Szeredi 138f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 139f21f6220SAl Viro mnt_group_start, 14015169fe7SAl Viro &mnt->mnt_group_id); 141f21f6220SAl Viro if (!res) 14215169fe7SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 143f21f6220SAl Viro 144f21f6220SAl Viro return res; 145719f5d7fSMiklos Szeredi } 146719f5d7fSMiklos Szeredi 147719f5d7fSMiklos Szeredi /* 148719f5d7fSMiklos Szeredi * Release a peer group ID 149719f5d7fSMiklos Szeredi */ 1504b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 151719f5d7fSMiklos Szeredi { 15215169fe7SAl Viro int id = mnt->mnt_group_id; 153f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 154f21f6220SAl Viro if (mnt_group_start > id) 155f21f6220SAl Viro mnt_group_start = id; 15615169fe7SAl Viro mnt->mnt_group_id = 0; 157719f5d7fSMiklos Szeredi } 158719f5d7fSMiklos Szeredi 159b3e19d92SNick Piggin /* 160b3e19d92SNick Piggin * vfsmount lock must be held for read 161b3e19d92SNick Piggin */ 16283adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 163b3e19d92SNick Piggin { 164b3e19d92SNick Piggin #ifdef CONFIG_SMP 16568e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 166b3e19d92SNick Piggin #else 167b3e19d92SNick Piggin preempt_disable(); 16868e8a9feSAl Viro mnt->mnt_count += n; 169b3e19d92SNick Piggin preempt_enable(); 170b3e19d92SNick Piggin #endif 171b3e19d92SNick Piggin } 172b3e19d92SNick Piggin 173b3e19d92SNick Piggin /* 174b3e19d92SNick Piggin * vfsmount lock must be held for write 175b3e19d92SNick Piggin */ 17683adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 177b3e19d92SNick Piggin { 178b3e19d92SNick Piggin #ifdef CONFIG_SMP 179f03c6599SAl Viro unsigned int count = 0; 180b3e19d92SNick Piggin int cpu; 181b3e19d92SNick Piggin 182b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 18368e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 184b3e19d92SNick Piggin } 185b3e19d92SNick Piggin 186b3e19d92SNick Piggin return count; 187b3e19d92SNick Piggin #else 18868e8a9feSAl Viro return mnt->mnt_count; 189b3e19d92SNick Piggin #endif 190b3e19d92SNick Piggin } 191b3e19d92SNick Piggin 19287b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 19387b95ce0SAl Viro { 19487b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 19587b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 19687b95ce0SAl Viro pin_remove(p); 19787b95ce0SAl Viro mntput(&m->mnt); 19887b95ce0SAl Viro } 19987b95ce0SAl Viro 200b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 2011da177e4SLinus Torvalds { 202c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 203c63181e6SAl Viro if (mnt) { 20473cd49ecSMiklos Szeredi int err; 20573cd49ecSMiklos Szeredi 206c63181e6SAl Viro err = mnt_alloc_id(mnt); 20788b38782SLi Zefan if (err) 20888b38782SLi Zefan goto out_free_cache; 20988b38782SLi Zefan 21088b38782SLi Zefan if (name) { 211fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 212c63181e6SAl Viro if (!mnt->mnt_devname) 21388b38782SLi Zefan goto out_free_id; 21473cd49ecSMiklos Szeredi } 21573cd49ecSMiklos Szeredi 216b3e19d92SNick Piggin #ifdef CONFIG_SMP 217c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 218c63181e6SAl Viro if (!mnt->mnt_pcp) 219b3e19d92SNick Piggin goto out_free_devname; 220b3e19d92SNick Piggin 221c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 222b3e19d92SNick Piggin #else 223c63181e6SAl Viro mnt->mnt_count = 1; 224c63181e6SAl Viro mnt->mnt_writers = 0; 225b3e19d92SNick Piggin #endif 226b3e19d92SNick Piggin 22738129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 228c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 229c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 230c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 231c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 232c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 233c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 234c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2350a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 2362504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 2372504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 2382504c5d6SAndreas Gruenbacher #endif 23987b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2401da177e4SLinus Torvalds } 241c63181e6SAl Viro return mnt; 24288b38782SLi Zefan 243d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 244d3ef3d73Snpiggin@suse.de out_free_devname: 245fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 246d3ef3d73Snpiggin@suse.de #endif 24788b38782SLi Zefan out_free_id: 248c63181e6SAl Viro mnt_free_id(mnt); 24988b38782SLi Zefan out_free_cache: 250c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 25188b38782SLi Zefan return NULL; 2521da177e4SLinus Torvalds } 2531da177e4SLinus Torvalds 2548366025eSDave Hansen /* 2558366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2568366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2578366025eSDave Hansen * We must keep track of when those operations start 2588366025eSDave Hansen * (for permission checks) and when they end, so that 2598366025eSDave Hansen * we can determine when writes are able to occur to 2608366025eSDave Hansen * a filesystem. 2618366025eSDave Hansen */ 2623d733633SDave Hansen /* 2633d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2643d733633SDave Hansen * @mnt: the mount to check for its write status 2653d733633SDave Hansen * 2663d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2673d733633SDave Hansen * It does not guarantee that the filesystem will stay 2683d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2693d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2703d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2713d733633SDave Hansen * r/w. 2723d733633SDave Hansen */ 2733d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2743d733633SDave Hansen { 2752e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2762e4b7fcdSDave Hansen return 1; 2772e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2782e4b7fcdSDave Hansen return 1; 2792e4b7fcdSDave Hansen return 0; 2803d733633SDave Hansen } 2813d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2823d733633SDave Hansen 28383adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2843d733633SDave Hansen { 285d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 28668e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 287d3ef3d73Snpiggin@suse.de #else 28868e8a9feSAl Viro mnt->mnt_writers++; 289d3ef3d73Snpiggin@suse.de #endif 2903d733633SDave Hansen } 2913d733633SDave Hansen 29283adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2933d733633SDave Hansen { 294d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 29568e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 296d3ef3d73Snpiggin@suse.de #else 29768e8a9feSAl Viro mnt->mnt_writers--; 298d3ef3d73Snpiggin@suse.de #endif 299d3ef3d73Snpiggin@suse.de } 300d3ef3d73Snpiggin@suse.de 30183adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 302d3ef3d73Snpiggin@suse.de { 303d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 304d3ef3d73Snpiggin@suse.de unsigned int count = 0; 3053d733633SDave Hansen int cpu; 3063d733633SDave Hansen 3073d733633SDave Hansen for_each_possible_cpu(cpu) { 30868e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 3093d733633SDave Hansen } 3103d733633SDave Hansen 311d3ef3d73Snpiggin@suse.de return count; 312d3ef3d73Snpiggin@suse.de #else 313d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 314d3ef3d73Snpiggin@suse.de #endif 3153d733633SDave Hansen } 3163d733633SDave Hansen 3174ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3184ed5e82fSMiklos Szeredi { 3194ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3204ed5e82fSMiklos Szeredi return 1; 3214ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3224ed5e82fSMiklos Szeredi smp_rmb(); 3234ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3244ed5e82fSMiklos Szeredi } 3254ed5e82fSMiklos Szeredi 3263d733633SDave Hansen /* 327eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 328eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 329eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 330eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3313d733633SDave Hansen */ 3328366025eSDave Hansen /** 333eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 33483adc753SAl Viro * @m: the mount on which to take a write 3358366025eSDave Hansen * 336eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 337eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 338eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 339eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 340eb04c282SJan Kara * called. This is effectively a refcount. 3418366025eSDave Hansen */ 342eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3438366025eSDave Hansen { 34483adc753SAl Viro struct mount *mnt = real_mount(m); 3453d733633SDave Hansen int ret = 0; 3463d733633SDave Hansen 347d3ef3d73Snpiggin@suse.de preempt_disable(); 348c6653a83SNick Piggin mnt_inc_writers(mnt); 349d3ef3d73Snpiggin@suse.de /* 350c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 351d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 352d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 353d3ef3d73Snpiggin@suse.de */ 354d3ef3d73Snpiggin@suse.de smp_mb(); 3551e75529eSMiao Xie while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 356d3ef3d73Snpiggin@suse.de cpu_relax(); 357d3ef3d73Snpiggin@suse.de /* 358d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 359d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 360d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 361d3ef3d73Snpiggin@suse.de */ 362d3ef3d73Snpiggin@suse.de smp_rmb(); 3634ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 364c6653a83SNick Piggin mnt_dec_writers(mnt); 3653d733633SDave Hansen ret = -EROFS; 3663d733633SDave Hansen } 367d3ef3d73Snpiggin@suse.de preempt_enable(); 368eb04c282SJan Kara 369eb04c282SJan Kara return ret; 370eb04c282SJan Kara } 371eb04c282SJan Kara 372eb04c282SJan Kara /** 373eb04c282SJan Kara * mnt_want_write - get write access to a mount 374eb04c282SJan Kara * @m: the mount on which to take a write 375eb04c282SJan Kara * 376eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 377eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 378eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 379eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 380eb04c282SJan Kara */ 381eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 382eb04c282SJan Kara { 383eb04c282SJan Kara int ret; 384eb04c282SJan Kara 385eb04c282SJan Kara sb_start_write(m->mnt_sb); 386eb04c282SJan Kara ret = __mnt_want_write(m); 387eb04c282SJan Kara if (ret) 388eb04c282SJan Kara sb_end_write(m->mnt_sb); 3893d733633SDave Hansen return ret; 3908366025eSDave Hansen } 3918366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3928366025eSDave Hansen 3938366025eSDave Hansen /** 39496029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 39596029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 39696029c4eSnpiggin@suse.de * 39796029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 39896029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 39996029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 40096029c4eSnpiggin@suse.de * on it. This allows some optimisation. 40196029c4eSnpiggin@suse.de * 40296029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 40396029c4eSnpiggin@suse.de * drop the reference. 40496029c4eSnpiggin@suse.de */ 40596029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 40696029c4eSnpiggin@suse.de { 40796029c4eSnpiggin@suse.de /* superblock may be r/o */ 40896029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 40996029c4eSnpiggin@suse.de return -EROFS; 41096029c4eSnpiggin@suse.de preempt_disable(); 41183adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 41296029c4eSnpiggin@suse.de preempt_enable(); 41396029c4eSnpiggin@suse.de return 0; 41496029c4eSnpiggin@suse.de } 41596029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 41696029c4eSnpiggin@suse.de 41796029c4eSnpiggin@suse.de /** 418eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 419eb04c282SJan Kara * @file: the file who's mount on which to take a write 420eb04c282SJan Kara * 421eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 422eb04c282SJan Kara * do some optimisations if the file is open for write already 423eb04c282SJan Kara */ 424eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 425eb04c282SJan Kara { 42683f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 427eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 428eb04c282SJan Kara else 429eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 430eb04c282SJan Kara } 431eb04c282SJan Kara 432eb04c282SJan Kara /** 43396029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 43496029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 43596029c4eSnpiggin@suse.de * 43696029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 43796029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 43896029c4eSnpiggin@suse.de */ 43996029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 44096029c4eSnpiggin@suse.de { 441eb04c282SJan Kara int ret; 442eb04c282SJan Kara 443eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 444eb04c282SJan Kara ret = __mnt_want_write_file(file); 445eb04c282SJan Kara if (ret) 446eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 447eb04c282SJan Kara return ret; 44896029c4eSnpiggin@suse.de } 44996029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 45096029c4eSnpiggin@suse.de 45196029c4eSnpiggin@suse.de /** 452eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4538366025eSDave Hansen * @mnt: the mount on which to give up write access 4548366025eSDave Hansen * 4558366025eSDave Hansen * Tells the low-level filesystem that we are done 4568366025eSDave Hansen * performing writes to it. Must be matched with 457eb04c282SJan Kara * __mnt_want_write() call above. 4588366025eSDave Hansen */ 459eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4608366025eSDave Hansen { 461d3ef3d73Snpiggin@suse.de preempt_disable(); 46283adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 463d3ef3d73Snpiggin@suse.de preempt_enable(); 4648366025eSDave Hansen } 465eb04c282SJan Kara 466eb04c282SJan Kara /** 467eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 468eb04c282SJan Kara * @mnt: the mount on which to give up write access 469eb04c282SJan Kara * 470eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 471eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 472eb04c282SJan Kara * mnt_want_write() call above. 473eb04c282SJan Kara */ 474eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 475eb04c282SJan Kara { 476eb04c282SJan Kara __mnt_drop_write(mnt); 477eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 478eb04c282SJan Kara } 4798366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4808366025eSDave Hansen 481eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 482eb04c282SJan Kara { 483eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 484eb04c282SJan Kara } 485eb04c282SJan Kara 4862a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4872a79f17eSAl Viro { 4882a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4892a79f17eSAl Viro } 4902a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4912a79f17eSAl Viro 49283adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4938366025eSDave Hansen { 4943d733633SDave Hansen int ret = 0; 4953d733633SDave Hansen 496719ea2fbSAl Viro lock_mount_hash(); 49783adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 498d3ef3d73Snpiggin@suse.de /* 499d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 500d3ef3d73Snpiggin@suse.de * should be visible before we do. 501d3ef3d73Snpiggin@suse.de */ 502d3ef3d73Snpiggin@suse.de smp_mb(); 503d3ef3d73Snpiggin@suse.de 504d3ef3d73Snpiggin@suse.de /* 505d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 506d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 507d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 508d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 509d3ef3d73Snpiggin@suse.de * 510d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 511d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 512d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 513d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 514d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 515d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 516d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 517d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 518d3ef3d73Snpiggin@suse.de * we're counting up here. 519d3ef3d73Snpiggin@suse.de */ 520c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 521d3ef3d73Snpiggin@suse.de ret = -EBUSY; 522d3ef3d73Snpiggin@suse.de else 52383adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 524d3ef3d73Snpiggin@suse.de /* 525d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 526d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 527d3ef3d73Snpiggin@suse.de */ 528d3ef3d73Snpiggin@suse.de smp_wmb(); 52983adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 530719ea2fbSAl Viro unlock_mount_hash(); 5313d733633SDave Hansen return ret; 5323d733633SDave Hansen } 5338366025eSDave Hansen 53483adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5352e4b7fcdSDave Hansen { 536719ea2fbSAl Viro lock_mount_hash(); 53783adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 538719ea2fbSAl Viro unlock_mount_hash(); 5392e4b7fcdSDave Hansen } 5402e4b7fcdSDave Hansen 5414ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5424ed5e82fSMiklos Szeredi { 5434ed5e82fSMiklos Szeredi struct mount *mnt; 5444ed5e82fSMiklos Szeredi int err = 0; 5454ed5e82fSMiklos Szeredi 5468e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5478e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5488e8b8796SMiklos Szeredi return -EBUSY; 5498e8b8796SMiklos Szeredi 550719ea2fbSAl Viro lock_mount_hash(); 5514ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5524ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5534ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5544ed5e82fSMiklos Szeredi smp_mb(); 5554ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5564ed5e82fSMiklos Szeredi err = -EBUSY; 5574ed5e82fSMiklos Szeredi break; 5584ed5e82fSMiklos Szeredi } 5594ed5e82fSMiklos Szeredi } 5604ed5e82fSMiklos Szeredi } 5618e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5628e8b8796SMiklos Szeredi err = -EBUSY; 5638e8b8796SMiklos Szeredi 5644ed5e82fSMiklos Szeredi if (!err) { 5654ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5664ed5e82fSMiklos Szeredi smp_wmb(); 5674ed5e82fSMiklos Szeredi } 5684ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5694ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5704ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5714ed5e82fSMiklos Szeredi } 572719ea2fbSAl Viro unlock_mount_hash(); 5734ed5e82fSMiklos Szeredi 5744ed5e82fSMiklos Szeredi return err; 5754ed5e82fSMiklos Szeredi } 5764ed5e82fSMiklos Szeredi 577b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5781da177e4SLinus Torvalds { 579fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 580d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 58168e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 582d3ef3d73Snpiggin@suse.de #endif 583b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5841da177e4SLinus Torvalds } 5851da177e4SLinus Torvalds 5868ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5878ffcb32eSDavid Howells { 5888ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5898ffcb32eSDavid Howells } 5908ffcb32eSDavid Howells 59148a066e7SAl Viro /* call under rcu_read_lock */ 592294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 59348a066e7SAl Viro { 59448a066e7SAl Viro struct mount *mnt; 59548a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 596294d71ffSAl Viro return 1; 59748a066e7SAl Viro if (bastard == NULL) 598294d71ffSAl Viro return 0; 59948a066e7SAl Viro mnt = real_mount(bastard); 60048a066e7SAl Viro mnt_add_count(mnt, 1); 60148a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 602294d71ffSAl Viro return 0; 60348a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 60448a066e7SAl Viro mnt_add_count(mnt, -1); 605294d71ffSAl Viro return 1; 60648a066e7SAl Viro } 607294d71ffSAl Viro return -1; 608294d71ffSAl Viro } 609294d71ffSAl Viro 610294d71ffSAl Viro /* call under rcu_read_lock */ 611294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 612294d71ffSAl Viro { 613294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 614294d71ffSAl Viro if (likely(!res)) 615294d71ffSAl Viro return true; 616294d71ffSAl Viro if (unlikely(res < 0)) { 61748a066e7SAl Viro rcu_read_unlock(); 61848a066e7SAl Viro mntput(bastard); 61948a066e7SAl Viro rcu_read_lock(); 620294d71ffSAl Viro } 62148a066e7SAl Viro return false; 62248a066e7SAl Viro } 62348a066e7SAl Viro 6241da177e4SLinus Torvalds /* 625474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 62648a066e7SAl Viro * call under rcu_read_lock() 6271da177e4SLinus Torvalds */ 628474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6291da177e4SLinus Torvalds { 63038129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 631474279dcSAl Viro struct mount *p; 6321da177e4SLinus Torvalds 63338129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 634474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 635474279dcSAl Viro return p; 636474279dcSAl Viro return NULL; 6371da177e4SLinus Torvalds } 638474279dcSAl Viro 639474279dcSAl Viro /* 640474279dcSAl Viro * find the last mount at @dentry on vfsmount @mnt. 64148a066e7SAl Viro * mount_lock must be held. 642474279dcSAl Viro */ 643474279dcSAl Viro struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) 644474279dcSAl Viro { 645411a938bSEric W. Biederman struct mount *p, *res = NULL; 646411a938bSEric W. Biederman p = __lookup_mnt(mnt, dentry); 64738129a13SAl Viro if (!p) 64838129a13SAl Viro goto out; 649411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 650411a938bSEric W. Biederman res = p; 65138129a13SAl Viro hlist_for_each_entry_continue(p, mnt_hash) { 6521d6a32acSAl Viro if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) 6531d6a32acSAl Viro break; 654411a938bSEric W. Biederman if (!(p->mnt.mnt_flags & MNT_UMOUNT)) 6551d6a32acSAl Viro res = p; 6561d6a32acSAl Viro } 65738129a13SAl Viro out: 6581d6a32acSAl Viro return res; 6591da177e4SLinus Torvalds } 6601da177e4SLinus Torvalds 661a05964f3SRam Pai /* 662f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 663f015f126SDavid Howells * 664f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 665f015f126SDavid Howells * following mounts: 666f015f126SDavid Howells * 667f015f126SDavid Howells * mount /dev/sda1 /mnt 668f015f126SDavid Howells * mount /dev/sda2 /mnt 669f015f126SDavid Howells * mount /dev/sda3 /mnt 670f015f126SDavid Howells * 671f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 672f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 673f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 674f015f126SDavid Howells * 675f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 676a05964f3SRam Pai */ 6771c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 678a05964f3SRam Pai { 679c7105365SAl Viro struct mount *child_mnt; 68048a066e7SAl Viro struct vfsmount *m; 68148a066e7SAl Viro unsigned seq; 68299b7db7bSNick Piggin 68348a066e7SAl Viro rcu_read_lock(); 68448a066e7SAl Viro do { 68548a066e7SAl Viro seq = read_seqbegin(&mount_lock); 686474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 68748a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 68848a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 68948a066e7SAl Viro rcu_read_unlock(); 69048a066e7SAl Viro return m; 691a05964f3SRam Pai } 692a05964f3SRam Pai 6937af1364fSEric W. Biederman /* 6947af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6957af1364fSEric W. Biederman * current mount namespace. 6967af1364fSEric W. Biederman * 6977af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6987af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6997af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 7007af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 7017af1364fSEric W. Biederman * is a mountpoint. 7027af1364fSEric W. Biederman * 7037af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 7047af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 7057af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 7067af1364fSEric W. Biederman * parent mount. 7077af1364fSEric W. Biederman */ 7087af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 7097af1364fSEric W. Biederman { 7107af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 7117af1364fSEric W. Biederman struct mount *mnt; 7127af1364fSEric W. Biederman bool is_covered = false; 7137af1364fSEric W. Biederman 7147af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 7157af1364fSEric W. Biederman goto out; 7167af1364fSEric W. Biederman 7177af1364fSEric W. Biederman down_read(&namespace_sem); 7187af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 7197af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 7207af1364fSEric W. Biederman if (is_covered) 7217af1364fSEric W. Biederman break; 7227af1364fSEric W. Biederman } 7237af1364fSEric W. Biederman up_read(&namespace_sem); 7247af1364fSEric W. Biederman out: 7257af1364fSEric W. Biederman return is_covered; 7267af1364fSEric W. Biederman } 7277af1364fSEric W. Biederman 728e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 72984d17192SAl Viro { 7300818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 73184d17192SAl Viro struct mountpoint *mp; 73284d17192SAl Viro 7330818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 73484d17192SAl Viro if (mp->m_dentry == dentry) { 73584d17192SAl Viro /* might be worth a WARN_ON() */ 73684d17192SAl Viro if (d_unlinked(dentry)) 73784d17192SAl Viro return ERR_PTR(-ENOENT); 73884d17192SAl Viro mp->m_count++; 73984d17192SAl Viro return mp; 74084d17192SAl Viro } 74184d17192SAl Viro } 742e2dfa935SEric W. Biederman return NULL; 743e2dfa935SEric W. Biederman } 744e2dfa935SEric W. Biederman 745e2dfa935SEric W. Biederman static struct mountpoint *new_mountpoint(struct dentry *dentry) 746e2dfa935SEric W. Biederman { 747e2dfa935SEric W. Biederman struct hlist_head *chain = mp_hash(dentry); 748e2dfa935SEric W. Biederman struct mountpoint *mp; 749e2dfa935SEric W. Biederman int ret; 75084d17192SAl Viro 75184d17192SAl Viro mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 75284d17192SAl Viro if (!mp) 75384d17192SAl Viro return ERR_PTR(-ENOMEM); 75484d17192SAl Viro 755eed81007SMiklos Szeredi ret = d_set_mounted(dentry); 756eed81007SMiklos Szeredi if (ret) { 75784d17192SAl Viro kfree(mp); 758eed81007SMiklos Szeredi return ERR_PTR(ret); 75984d17192SAl Viro } 760eed81007SMiklos Szeredi 76184d17192SAl Viro mp->m_dentry = dentry; 76284d17192SAl Viro mp->m_count = 1; 7630818bf27SAl Viro hlist_add_head(&mp->m_hash, chain); 7640a5eb7c8SEric W. Biederman INIT_HLIST_HEAD(&mp->m_list); 76584d17192SAl Viro return mp; 76684d17192SAl Viro } 76784d17192SAl Viro 76884d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 76984d17192SAl Viro { 77084d17192SAl Viro if (!--mp->m_count) { 77184d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7720a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 77384d17192SAl Viro spin_lock(&dentry->d_lock); 77484d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 77584d17192SAl Viro spin_unlock(&dentry->d_lock); 7760818bf27SAl Viro hlist_del(&mp->m_hash); 77784d17192SAl Viro kfree(mp); 77884d17192SAl Viro } 77984d17192SAl Viro } 78084d17192SAl Viro 781143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7821da177e4SLinus Torvalds { 7836b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7841da177e4SLinus Torvalds } 7851da177e4SLinus Torvalds 78699b7db7bSNick Piggin /* 78799b7db7bSNick Piggin * vfsmount lock must be held for write 78899b7db7bSNick Piggin */ 7896b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7905addc5ddSAl Viro { 7915addc5ddSAl Viro if (ns) { 7925addc5ddSAl Viro ns->event = ++event; 7935addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7945addc5ddSAl Viro } 7955addc5ddSAl Viro } 7965addc5ddSAl Viro 79799b7db7bSNick Piggin /* 79899b7db7bSNick Piggin * vfsmount lock must be held for write 79999b7db7bSNick Piggin */ 8006b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 8015addc5ddSAl Viro { 8025addc5ddSAl Viro if (ns && ns->event != event) { 8035addc5ddSAl Viro ns->event = event; 8045addc5ddSAl Viro wake_up_interruptible(&ns->poll); 8055addc5ddSAl Viro } 8065addc5ddSAl Viro } 8075addc5ddSAl Viro 80899b7db7bSNick Piggin /* 80999b7db7bSNick Piggin * vfsmount lock must be held for write 81099b7db7bSNick Piggin */ 8117bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 8121da177e4SLinus Torvalds { 8130714a533SAl Viro mnt->mnt_parent = mnt; 814a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8156b41d536SAl Viro list_del_init(&mnt->mnt_child); 81638129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8170a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 81884d17192SAl Viro put_mountpoint(mnt->mnt_mp); 81984d17192SAl Viro mnt->mnt_mp = NULL; 8201da177e4SLinus Torvalds } 8211da177e4SLinus Torvalds 82299b7db7bSNick Piggin /* 82399b7db7bSNick Piggin * vfsmount lock must be held for write 82499b7db7bSNick Piggin */ 8257bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8267bdb11deSEric W. Biederman { 8277bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8287bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8297bdb11deSEric W. Biederman unhash_mnt(mnt); 8307bdb11deSEric W. Biederman } 8317bdb11deSEric W. Biederman 8327bdb11deSEric W. Biederman /* 8337bdb11deSEric W. Biederman * vfsmount lock must be held for write 8347bdb11deSEric W. Biederman */ 8356a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8366a46c573SEric W. Biederman { 8376a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8386a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8396a46c573SEric W. Biederman unhash_mnt(mnt); 8406a46c573SEric W. Biederman } 8416a46c573SEric W. Biederman 8426a46c573SEric W. Biederman /* 8436a46c573SEric W. Biederman * vfsmount lock must be held for write 8446a46c573SEric W. Biederman */ 84584d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 84684d17192SAl Viro struct mountpoint *mp, 84744d964d6SAl Viro struct mount *child_mnt) 848b90fa9aeSRam Pai { 84984d17192SAl Viro mp->m_count++; 8503a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 85184d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8523a2393d7SAl Viro child_mnt->mnt_parent = mnt; 85384d17192SAl Viro child_mnt->mnt_mp = mp; 8540a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 855b90fa9aeSRam Pai } 856b90fa9aeSRam Pai 85799b7db7bSNick Piggin /* 85899b7db7bSNick Piggin * vfsmount lock must be held for write 85999b7db7bSNick Piggin */ 86084d17192SAl Viro static void attach_mnt(struct mount *mnt, 86184d17192SAl Viro struct mount *parent, 86284d17192SAl Viro struct mountpoint *mp) 8631da177e4SLinus Torvalds { 86484d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 86538129a13SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); 86684d17192SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 867b90fa9aeSRam Pai } 868b90fa9aeSRam Pai 86912a5b529SAl Viro static void attach_shadowed(struct mount *mnt, 87012a5b529SAl Viro struct mount *parent, 87112a5b529SAl Viro struct mount *shadows) 87212a5b529SAl Viro { 87312a5b529SAl Viro if (shadows) { 874f6f99332SLinus Torvalds hlist_add_behind_rcu(&mnt->mnt_hash, &shadows->mnt_hash); 87512a5b529SAl Viro list_add(&mnt->mnt_child, &shadows->mnt_child); 87612a5b529SAl Viro } else { 87712a5b529SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, 87812a5b529SAl Viro m_hash(&parent->mnt, mnt->mnt_mountpoint)); 87912a5b529SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 88012a5b529SAl Viro } 88112a5b529SAl Viro } 88212a5b529SAl Viro 883b90fa9aeSRam Pai /* 88499b7db7bSNick Piggin * vfsmount lock must be held for write 885b90fa9aeSRam Pai */ 8861d6a32acSAl Viro static void commit_tree(struct mount *mnt, struct mount *shadows) 887b90fa9aeSRam Pai { 8880714a533SAl Viro struct mount *parent = mnt->mnt_parent; 88983adc753SAl Viro struct mount *m; 890b90fa9aeSRam Pai LIST_HEAD(head); 891143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 892b90fa9aeSRam Pai 8930714a533SAl Viro BUG_ON(parent == mnt); 894b90fa9aeSRam Pai 8951a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 896f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 897143c8c91SAl Viro m->mnt_ns = n; 898f03c6599SAl Viro 899b90fa9aeSRam Pai list_splice(&head, n->list.prev); 900b90fa9aeSRam Pai 901d2921684SEric W. Biederman n->mounts += n->pending_mounts; 902d2921684SEric W. Biederman n->pending_mounts = 0; 903d2921684SEric W. Biederman 90412a5b529SAl Viro attach_shadowed(mnt, parent, shadows); 9056b3286edSKirill Korotaev touch_mnt_namespace(n); 9061da177e4SLinus Torvalds } 9071da177e4SLinus Torvalds 908909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9091da177e4SLinus Torvalds { 9106b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9116b41d536SAl Viro if (next == &p->mnt_mounts) { 9121da177e4SLinus Torvalds while (1) { 913909b0a88SAl Viro if (p == root) 9141da177e4SLinus Torvalds return NULL; 9156b41d536SAl Viro next = p->mnt_child.next; 9166b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9171da177e4SLinus Torvalds break; 9180714a533SAl Viro p = p->mnt_parent; 9191da177e4SLinus Torvalds } 9201da177e4SLinus Torvalds } 9216b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9221da177e4SLinus Torvalds } 9231da177e4SLinus Torvalds 924315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9259676f0c6SRam Pai { 9266b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9276b41d536SAl Viro while (prev != &p->mnt_mounts) { 9286b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9296b41d536SAl Viro prev = p->mnt_mounts.prev; 9309676f0c6SRam Pai } 9319676f0c6SRam Pai return p; 9329676f0c6SRam Pai } 9339676f0c6SRam Pai 9349d412a43SAl Viro struct vfsmount * 9359d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 9369d412a43SAl Viro { 937b105e270SAl Viro struct mount *mnt; 9389d412a43SAl Viro struct dentry *root; 9399d412a43SAl Viro 9409d412a43SAl Viro if (!type) 9419d412a43SAl Viro return ERR_PTR(-ENODEV); 9429d412a43SAl Viro 9439d412a43SAl Viro mnt = alloc_vfsmnt(name); 9449d412a43SAl Viro if (!mnt) 9459d412a43SAl Viro return ERR_PTR(-ENOMEM); 9469d412a43SAl Viro 9479d412a43SAl Viro if (flags & MS_KERNMOUNT) 948b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9499d412a43SAl Viro 9509d412a43SAl Viro root = mount_fs(type, flags, name, data); 9519d412a43SAl Viro if (IS_ERR(root)) { 9528ffcb32eSDavid Howells mnt_free_id(mnt); 9539d412a43SAl Viro free_vfsmnt(mnt); 9549d412a43SAl Viro return ERR_CAST(root); 9559d412a43SAl Viro } 9569d412a43SAl Viro 957b105e270SAl Viro mnt->mnt.mnt_root = root; 958b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 959a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9600714a533SAl Viro mnt->mnt_parent = mnt; 961719ea2fbSAl Viro lock_mount_hash(); 96239f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 963719ea2fbSAl Viro unlock_mount_hash(); 964b105e270SAl Viro return &mnt->mnt; 9659d412a43SAl Viro } 9669d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 9679d412a43SAl Viro 96887129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 96936341f64SRam Pai int flag) 9701da177e4SLinus Torvalds { 97187129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 972be34d1a3SDavid Howells struct mount *mnt; 973be34d1a3SDavid Howells int err; 9741da177e4SLinus Torvalds 975be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 976be34d1a3SDavid Howells if (!mnt) 977be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 978be34d1a3SDavid Howells 9797a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 98015169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 981719f5d7fSMiklos Szeredi else 98215169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 983719f5d7fSMiklos Szeredi 98415169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 985be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 986719f5d7fSMiklos Szeredi if (err) 987719f5d7fSMiklos Szeredi goto out_free; 988719f5d7fSMiklos Szeredi } 989719f5d7fSMiklos Szeredi 990f2ebb3a9SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); 991132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 9929566d674SEric W. Biederman if (flag & CL_UNPRIVILEGED) { 9939566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_ATIME; 9949566d674SEric W. Biederman 9959566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_READONLY) 996132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 997132c94e3SEric W. Biederman 9989566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NODEV) 9999566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NODEV; 10009566d674SEric W. Biederman 10019566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOSUID) 10029566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOSUID; 10039566d674SEric W. Biederman 10049566d674SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_NOEXEC) 10059566d674SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_NOEXEC; 10069566d674SEric W. Biederman } 10079566d674SEric W. Biederman 10085ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 1009381cacb1SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && 1010381cacb1SEric W. Biederman (!(flag & CL_EXPIRE) || list_empty(&old->mnt_expire))) 10115ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 10125ff9d8a6SEric W. Biederman 10131da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1014b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1015b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1016a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10170714a533SAl Viro mnt->mnt_parent = mnt; 1018719ea2fbSAl Viro lock_mount_hash(); 101939f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1020719ea2fbSAl Viro unlock_mount_hash(); 1021b90fa9aeSRam Pai 10227a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10237a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10246776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 102532301920SAl Viro mnt->mnt_master = old; 1026fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10278aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1028fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10296776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1030d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10316776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1032d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10335235d448SAl Viro } else { 10345235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10355afe0022SRam Pai } 1036b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10370f0afb1dSAl Viro set_mnt_shared(mnt); 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10401da177e4SLinus Torvalds * as the original if that was on one */ 104136341f64SRam Pai if (flag & CL_EXPIRE) { 10426776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10436776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10441da177e4SLinus Torvalds } 1045be34d1a3SDavid Howells 1046cb338d06SAl Viro return mnt; 1047719f5d7fSMiklos Szeredi 1048719f5d7fSMiklos Szeredi out_free: 10498ffcb32eSDavid Howells mnt_free_id(mnt); 1050719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1051be34d1a3SDavid Howells return ERR_PTR(err); 10521da177e4SLinus Torvalds } 10531da177e4SLinus Torvalds 10549ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 10559ea459e1SAl Viro { 10569ea459e1SAl Viro /* 10579ea459e1SAl Viro * This probably indicates that somebody messed 10589ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 10599ea459e1SAl Viro * happens, the filesystem was probably unable 10609ea459e1SAl Viro * to make r/w->r/o transitions. 10619ea459e1SAl Viro */ 10629ea459e1SAl Viro /* 10639ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 10649ea459e1SAl Viro * so mnt_get_writers() below is safe. 10659ea459e1SAl Viro */ 10669ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 10679ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 10689ea459e1SAl Viro mnt_pin_kill(mnt); 10699ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 10709ea459e1SAl Viro dput(mnt->mnt.mnt_root); 10719ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 10729ea459e1SAl Viro mnt_free_id(mnt); 10739ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 10749ea459e1SAl Viro } 10759ea459e1SAl Viro 10769ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 10779ea459e1SAl Viro { 10789ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 10799ea459e1SAl Viro } 10809ea459e1SAl Viro 10819ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 10829ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 10839ea459e1SAl Viro { 10849ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 10859ea459e1SAl Viro struct llist_node *next; 10869ea459e1SAl Viro 10879ea459e1SAl Viro for (; node; node = next) { 10889ea459e1SAl Viro next = llist_next(node); 10899ea459e1SAl Viro cleanup_mnt(llist_entry(node, struct mount, mnt_llist)); 10909ea459e1SAl Viro } 10919ea459e1SAl Viro } 10929ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 10939ea459e1SAl Viro 1094900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 10957b7b1aceSAl Viro { 109648a066e7SAl Viro rcu_read_lock(); 1097aa9c0e07SAl Viro mnt_add_count(mnt, -1); 109848a066e7SAl Viro if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ 109948a066e7SAl Viro rcu_read_unlock(); 110099b7db7bSNick Piggin return; 1101b3e19d92SNick Piggin } 1102719ea2fbSAl Viro lock_mount_hash(); 1103b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 110448a066e7SAl Viro rcu_read_unlock(); 1105719ea2fbSAl Viro unlock_mount_hash(); 110699b7db7bSNick Piggin return; 110799b7db7bSNick Piggin } 110848a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 110948a066e7SAl Viro rcu_read_unlock(); 111048a066e7SAl Viro unlock_mount_hash(); 111148a066e7SAl Viro return; 111248a066e7SAl Viro } 111348a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 111448a066e7SAl Viro rcu_read_unlock(); 1115962830dfSAndi Kleen 111639f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1117ce07d891SEric W. Biederman 1118ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1119ce07d891SEric W. Biederman struct mount *p, *tmp; 1120ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1121ce07d891SEric W. Biederman umount_mnt(p); 1122ce07d891SEric W. Biederman } 1123ce07d891SEric W. Biederman } 1124719ea2fbSAl Viro unlock_mount_hash(); 1125649a795aSAl Viro 11269ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11279ea459e1SAl Viro struct task_struct *task = current; 11289ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11299ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11309ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11319ea459e1SAl Viro return; 11329ea459e1SAl Viro } 11339ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11349ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11359ea459e1SAl Viro return; 11369ea459e1SAl Viro } 11379ea459e1SAl Viro cleanup_mnt(mnt); 1138b3e19d92SNick Piggin } 1139b3e19d92SNick Piggin 1140b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1141b3e19d92SNick Piggin { 1142b3e19d92SNick Piggin if (mnt) { 1143863d684fSAl Viro struct mount *m = real_mount(mnt); 1144b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1145863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1146863d684fSAl Viro m->mnt_expiry_mark = 0; 1147863d684fSAl Viro mntput_no_expire(m); 1148b3e19d92SNick Piggin } 1149b3e19d92SNick Piggin } 1150b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1151b3e19d92SNick Piggin 1152b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1153b3e19d92SNick Piggin { 1154b3e19d92SNick Piggin if (mnt) 115583adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1156b3e19d92SNick Piggin return mnt; 1157b3e19d92SNick Piggin } 1158b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1159b3e19d92SNick Piggin 11603064c356SAl Viro struct vfsmount *mnt_clone_internal(struct path *path) 11617b7b1aceSAl Viro { 11623064c356SAl Viro struct mount *p; 11633064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 11643064c356SAl Viro if (IS_ERR(p)) 11653064c356SAl Viro return ERR_CAST(p); 11663064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 11673064c356SAl Viro return &p->mnt; 11687b7b1aceSAl Viro } 11691da177e4SLinus Torvalds 1170b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 1171b3b304a2SMiklos Szeredi { 1172b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 1173b3b304a2SMiklos Szeredi } 1174b3b304a2SMiklos Szeredi 1175b3b304a2SMiklos Szeredi /* 1176b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 1177b3b304a2SMiklos Szeredi * implement more complex mount option showing. 1178b3b304a2SMiklos Szeredi * 1179b3b304a2SMiklos Szeredi * See also save_mount_options(). 1180b3b304a2SMiklos Szeredi */ 118134c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 1182b3b304a2SMiklos Szeredi { 11832a32cebdSAl Viro const char *options; 11842a32cebdSAl Viro 11852a32cebdSAl Viro rcu_read_lock(); 118634c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 1187b3b304a2SMiklos Szeredi 1188b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 1189b3b304a2SMiklos Szeredi seq_putc(m, ','); 1190b3b304a2SMiklos Szeredi mangle(m, options); 1191b3b304a2SMiklos Szeredi } 11922a32cebdSAl Viro rcu_read_unlock(); 1193b3b304a2SMiklos Szeredi 1194b3b304a2SMiklos Szeredi return 0; 1195b3b304a2SMiklos Szeredi } 1196b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 1197b3b304a2SMiklos Szeredi 1198b3b304a2SMiklos Szeredi /* 1199b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 1200b3b304a2SMiklos Szeredi * called from the fill_super() callback. 1201b3b304a2SMiklos Szeredi * 1202b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 1203b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1204b3b304a2SMiklos Szeredi * remount fails. 1205b3b304a2SMiklos Szeredi * 1206b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1207b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1208b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1209b3b304a2SMiklos Szeredi * any more. 1210b3b304a2SMiklos Szeredi */ 1211b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1212b3b304a2SMiklos Szeredi { 12132a32cebdSAl Viro BUG_ON(sb->s_options); 12142a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1215b3b304a2SMiklos Szeredi } 1216b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1217b3b304a2SMiklos Szeredi 12182a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 12192a32cebdSAl Viro { 12202a32cebdSAl Viro char *old = sb->s_options; 12212a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 12222a32cebdSAl Viro if (old) { 12232a32cebdSAl Viro synchronize_rcu(); 12242a32cebdSAl Viro kfree(old); 12252a32cebdSAl Viro } 12262a32cebdSAl Viro } 12272a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 12282a32cebdSAl Viro 1229a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12300226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12311da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12321da177e4SLinus Torvalds { 1233ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12341da177e4SLinus Torvalds 1235390c6843SRam Pai down_read(&namespace_sem); 1236c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1237c7999c36SAl Viro void *v = p->cached_mount; 1238c7999c36SAl Viro if (*pos == p->cached_index) 1239c7999c36SAl Viro return v; 1240c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1241c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1242c7999c36SAl Viro return p->cached_mount = v; 1243c7999c36SAl Viro } 1244c7999c36SAl Viro } 1245c7999c36SAl Viro 1246c7999c36SAl Viro p->cached_event = p->ns->event; 1247c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1248c7999c36SAl Viro p->cached_index = *pos; 1249c7999c36SAl Viro return p->cached_mount; 12501da177e4SLinus Torvalds } 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12531da177e4SLinus Torvalds { 1254ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1255b0765fb8SPavel Emelianov 1256c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1257c7999c36SAl Viro p->cached_index = *pos; 1258c7999c36SAl Viro return p->cached_mount; 12591da177e4SLinus Torvalds } 12601da177e4SLinus Torvalds 12611da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12621da177e4SLinus Torvalds { 1263390c6843SRam Pai up_read(&namespace_sem); 12641da177e4SLinus Torvalds } 12651da177e4SLinus Torvalds 12660226f492SAl Viro static int m_show(struct seq_file *m, void *v) 12679f5596afSAl Viro { 1268ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12691a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 12700226f492SAl Viro return p->show(m, &r->mnt); 12711da177e4SLinus Torvalds } 12721da177e4SLinus Torvalds 1273a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 12741da177e4SLinus Torvalds .start = m_start, 12751da177e4SLinus Torvalds .next = m_next, 12761da177e4SLinus Torvalds .stop = m_stop, 12770226f492SAl Viro .show = m_show, 1278b4629fe2SChuck Lever }; 1279a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1280b4629fe2SChuck Lever 12811da177e4SLinus Torvalds /** 12821da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 12831da177e4SLinus Torvalds * @mnt: root of mount tree 12841da177e4SLinus Torvalds * 12851da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 12861da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 12871da177e4SLinus Torvalds * busy. 12881da177e4SLinus Torvalds */ 1289909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 12901da177e4SLinus Torvalds { 1291909b0a88SAl Viro struct mount *mnt = real_mount(m); 129236341f64SRam Pai int actual_refs = 0; 129336341f64SRam Pai int minimum_refs = 0; 1294315fc83eSAl Viro struct mount *p; 1295909b0a88SAl Viro BUG_ON(!m); 12961da177e4SLinus Torvalds 1297b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1298719ea2fbSAl Viro lock_mount_hash(); 1299909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 130083adc753SAl Viro actual_refs += mnt_get_count(p); 13011da177e4SLinus Torvalds minimum_refs += 2; 13021da177e4SLinus Torvalds } 1303719ea2fbSAl Viro unlock_mount_hash(); 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13061da177e4SLinus Torvalds return 0; 1307e3474a8eSIan Kent 1308e3474a8eSIan Kent return 1; 13091da177e4SLinus Torvalds } 13101da177e4SLinus Torvalds 13111da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds /** 13141da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13151da177e4SLinus Torvalds * @mnt: root of mount 13161da177e4SLinus Torvalds * 13171da177e4SLinus Torvalds * This is called to check if a mount point has any 13181da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13191da177e4SLinus Torvalds * mount has sub mounts this will return busy 13201da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13211da177e4SLinus Torvalds * 13221da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13231da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13241da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13251da177e4SLinus Torvalds */ 13261da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13271da177e4SLinus Torvalds { 1328e3474a8eSIan Kent int ret = 1; 13298ad08d8aSAl Viro down_read(&namespace_sem); 1330719ea2fbSAl Viro lock_mount_hash(); 13311ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1332e3474a8eSIan Kent ret = 0; 1333719ea2fbSAl Viro unlock_mount_hash(); 13348ad08d8aSAl Viro up_read(&namespace_sem); 1335a05964f3SRam Pai return ret; 13361da177e4SLinus Torvalds } 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13391da177e4SLinus Torvalds 134038129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1341e3197d83SAl Viro 134297216be0SAl Viro static void namespace_unlock(void) 13431da177e4SLinus Torvalds { 1344a3b3c562SEric W. Biederman struct hlist_head head; 134597216be0SAl Viro 1346a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1347a3b3c562SEric W. Biederman 134897216be0SAl Viro up_write(&namespace_sem); 1349a3b3c562SEric W. Biederman 1350a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 135197216be0SAl Viro return; 135297216be0SAl Viro 135348a066e7SAl Viro synchronize_rcu(); 135448a066e7SAl Viro 135587b95ce0SAl Viro group_pin_kill(&head); 135670fbcdf4SRam Pai } 135770fbcdf4SRam Pai 135897216be0SAl Viro static inline void namespace_lock(void) 1359e3197d83SAl Viro { 136097216be0SAl Viro down_write(&namespace_sem); 1361e3197d83SAl Viro } 1362e3197d83SAl Viro 1363e819f152SEric W. Biederman enum umount_tree_flags { 1364e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1365e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1366e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1367e819f152SEric W. Biederman }; 1368f2d0a123SEric W. Biederman 1369f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1370f2d0a123SEric W. Biederman { 1371f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1372f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1373f2d0a123SEric W. Biederman return true; 1374f2d0a123SEric W. Biederman 1375f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1376f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1377f2d0a123SEric W. Biederman return true; 1378f2d0a123SEric W. Biederman 1379f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1380f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1381f2d0a123SEric W. Biederman * connected to mounted mounts. 1382f2d0a123SEric W. Biederman */ 1383f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1384f2d0a123SEric W. Biederman return true; 1385f2d0a123SEric W. Biederman 1386f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1387f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1388f2d0a123SEric W. Biederman return false; 1389f2d0a123SEric W. Biederman 1390f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1391f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1392f2d0a123SEric W. Biederman return false; 1393f2d0a123SEric W. Biederman 1394f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1395f2d0a123SEric W. Biederman return true; 1396f2d0a123SEric W. Biederman } 1397f2d0a123SEric W. Biederman 139899b7db7bSNick Piggin /* 139948a066e7SAl Viro * mount_lock must be held 140099b7db7bSNick Piggin * namespace_sem must be held for write 140199b7db7bSNick Piggin */ 1402e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 140370fbcdf4SRam Pai { 1404c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1405315fc83eSAl Viro struct mount *p; 140670fbcdf4SRam Pai 14075d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14085d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14095d88457eSEric W. Biederman 1410c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1411590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1412590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1413c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1414590ce4bcSEric W. Biederman } 1415c003b26fSEric W. Biederman 1416411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1417c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1418c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 141938129a13SAl Viro } 142070fbcdf4SRam Pai 1421c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1422e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14237b8a53fdSAl Viro propagate_umount(&tmp_list); 1424a05964f3SRam Pai 1425c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1426d2921684SEric W. Biederman struct mnt_namespace *ns; 1427ce07d891SEric W. Biederman bool disconnect; 1428c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 14296776db3dSAl Viro list_del_init(&p->mnt_expire); 14301a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1431d2921684SEric W. Biederman ns = p->mnt_ns; 1432d2921684SEric W. Biederman if (ns) { 1433d2921684SEric W. Biederman ns->mounts--; 1434d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1435d2921684SEric W. Biederman } 143663d37a84SAl Viro p->mnt_ns = NULL; 1437e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 143848a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 143987b95ce0SAl Viro 1440f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1441ce07d891SEric W. Biederman 1442ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1443ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1444676da58dSAl Viro if (mnt_has_parent(p)) { 144581b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1446ce07d891SEric W. Biederman if (!disconnect) { 1447ce07d891SEric W. Biederman /* Don't forget about p */ 1448ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1449ce07d891SEric W. Biederman } else { 14506a46c573SEric W. Biederman umount_mnt(p); 14517c4b93d8SAl Viro } 1452ce07d891SEric W. Biederman } 14530f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 145438129a13SAl Viro } 14551da177e4SLinus Torvalds } 14561da177e4SLinus Torvalds 1457b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1458c35038beSAl Viro 14591ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 14601da177e4SLinus Torvalds { 14611ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 14621da177e4SLinus Torvalds int retval; 14631da177e4SLinus Torvalds 14641ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 14651da177e4SLinus Torvalds if (retval) 14661da177e4SLinus Torvalds return retval; 14671da177e4SLinus Torvalds 14681da177e4SLinus Torvalds /* 14691da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 14701da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 14711da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 14721da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 14731da177e4SLinus Torvalds */ 14741da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 14751ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 14761da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 14771da177e4SLinus Torvalds return -EINVAL; 14781da177e4SLinus Torvalds 1479b3e19d92SNick Piggin /* 1480b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1481b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1482b3e19d92SNick Piggin */ 1483719ea2fbSAl Viro lock_mount_hash(); 148483adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1485719ea2fbSAl Viro unlock_mount_hash(); 14861da177e4SLinus Torvalds return -EBUSY; 1487b3e19d92SNick Piggin } 1488719ea2fbSAl Viro unlock_mount_hash(); 14891da177e4SLinus Torvalds 1490863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 14911da177e4SLinus Torvalds return -EAGAIN; 14921da177e4SLinus Torvalds } 14931da177e4SLinus Torvalds 14941da177e4SLinus Torvalds /* 14951da177e4SLinus Torvalds * If we may have to abort operations to get out of this 14961da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 14971da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 14981da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 14991da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 15001da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 15011da177e4SLinus Torvalds * about for the moment. 15021da177e4SLinus Torvalds */ 15031da177e4SLinus Torvalds 150442faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 150542faad99SAl Viro sb->s_op->umount_begin(sb); 150642faad99SAl Viro } 15071da177e4SLinus Torvalds 15081da177e4SLinus Torvalds /* 15091da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15101da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15111da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15121da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15131da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15141da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15151da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15161da177e4SLinus Torvalds */ 15171ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15181da177e4SLinus Torvalds /* 15191da177e4SLinus Torvalds * Special case for "unmounting" root ... 15201da177e4SLinus Torvalds * we just try to remount it readonly. 15211da177e4SLinus Torvalds */ 1522a1480dccSAndy Lutomirski if (!capable(CAP_SYS_ADMIN)) 1523a1480dccSAndy Lutomirski return -EPERM; 15241da177e4SLinus Torvalds down_write(&sb->s_umount); 15254aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 15261da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 15271da177e4SLinus Torvalds up_write(&sb->s_umount); 15281da177e4SLinus Torvalds return retval; 15291da177e4SLinus Torvalds } 15301da177e4SLinus Torvalds 153197216be0SAl Viro namespace_lock(); 1532719ea2fbSAl Viro lock_mount_hash(); 15335addc5ddSAl Viro event++; 15341da177e4SLinus Torvalds 153548a066e7SAl Viro if (flags & MNT_DETACH) { 153648a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1537e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 153848a066e7SAl Viro retval = 0; 153948a066e7SAl Viro } else { 1540b54b9be7SAl Viro shrink_submounts(mnt); 15411da177e4SLinus Torvalds retval = -EBUSY; 154248a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 15431a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1544e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 15451da177e4SLinus Torvalds retval = 0; 15461da177e4SLinus Torvalds } 154748a066e7SAl Viro } 1548719ea2fbSAl Viro unlock_mount_hash(); 1549e3197d83SAl Viro namespace_unlock(); 15501da177e4SLinus Torvalds return retval; 15511da177e4SLinus Torvalds } 15521da177e4SLinus Torvalds 15531da177e4SLinus Torvalds /* 155480b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 155580b5dce8SEric W. Biederman * 155680b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 155780b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 155880b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 155980b5dce8SEric W. Biederman * leaking them. 156080b5dce8SEric W. Biederman * 156180b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 156280b5dce8SEric W. Biederman */ 156380b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 156480b5dce8SEric W. Biederman { 156580b5dce8SEric W. Biederman struct mountpoint *mp; 156680b5dce8SEric W. Biederman struct mount *mnt; 156780b5dce8SEric W. Biederman 156880b5dce8SEric W. Biederman namespace_lock(); 156980b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1570f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 157180b5dce8SEric W. Biederman goto out_unlock; 157280b5dce8SEric W. Biederman 157380b5dce8SEric W. Biederman lock_mount_hash(); 1574e06b933eSAndrey Ulanov event++; 157580b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 157680b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1577ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1578fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1579fe78fcc8SEric W. Biederman umount_mnt(mnt); 1580ce07d891SEric W. Biederman } 1581e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 158280b5dce8SEric W. Biederman } 158380b5dce8SEric W. Biederman unlock_mount_hash(); 158480b5dce8SEric W. Biederman put_mountpoint(mp); 158580b5dce8SEric W. Biederman out_unlock: 158680b5dce8SEric W. Biederman namespace_unlock(); 158780b5dce8SEric W. Biederman } 158880b5dce8SEric W. Biederman 158980b5dce8SEric W. Biederman /* 15909b40bc90SAl Viro * Is the caller allowed to modify his namespace? 15919b40bc90SAl Viro */ 15929b40bc90SAl Viro static inline bool may_mount(void) 15939b40bc90SAl Viro { 15949b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 15959b40bc90SAl Viro } 15969b40bc90SAl Viro 15979e8925b6SJeff Layton static inline bool may_mandlock(void) 15989e8925b6SJeff Layton { 15999e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 16009e8925b6SJeff Layton return false; 16019e8925b6SJeff Layton #endif 160295ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 16039e8925b6SJeff Layton } 16049e8925b6SJeff Layton 16059b40bc90SAl Viro /* 16061da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16071da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16081da177e4SLinus Torvalds * 16091da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16101da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16111da177e4SLinus Torvalds */ 16121da177e4SLinus Torvalds 1613bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 16141da177e4SLinus Torvalds { 16152d8f3038SAl Viro struct path path; 1616900148dcSAl Viro struct mount *mnt; 16171da177e4SLinus Torvalds int retval; 1618db1f05bbSMiklos Szeredi int lookup_flags = 0; 16191da177e4SLinus Torvalds 1620db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1621db1f05bbSMiklos Szeredi return -EINVAL; 1622db1f05bbSMiklos Szeredi 16239b40bc90SAl Viro if (!may_mount()) 16249b40bc90SAl Viro return -EPERM; 16259b40bc90SAl Viro 1626db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1627db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1628db1f05bbSMiklos Szeredi 1629197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 16301da177e4SLinus Torvalds if (retval) 16311da177e4SLinus Torvalds goto out; 1632900148dcSAl Viro mnt = real_mount(path.mnt); 16331da177e4SLinus Torvalds retval = -EINVAL; 16342d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 16351da177e4SLinus Torvalds goto dput_and_out; 1636143c8c91SAl Viro if (!check_mnt(mnt)) 16371da177e4SLinus Torvalds goto dput_and_out; 16385ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 16395ff9d8a6SEric W. Biederman goto dput_and_out; 1640b2f5d4dcSEric W. Biederman retval = -EPERM; 1641b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1642b2f5d4dcSEric W. Biederman goto dput_and_out; 16431da177e4SLinus Torvalds 1644900148dcSAl Viro retval = do_umount(mnt, flags); 16451da177e4SLinus Torvalds dput_and_out: 1646429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 16472d8f3038SAl Viro dput(path.dentry); 1648900148dcSAl Viro mntput_no_expire(mnt); 16491da177e4SLinus Torvalds out: 16501da177e4SLinus Torvalds return retval; 16511da177e4SLinus Torvalds } 16521da177e4SLinus Torvalds 16531da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds /* 16561da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 16571da177e4SLinus Torvalds */ 1658bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 16591da177e4SLinus Torvalds { 16601da177e4SLinus Torvalds return sys_umount(name, 0); 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds #endif 16641da177e4SLinus Torvalds 16654ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 16668823c079SEric W. Biederman { 16674ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1668e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1669e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 16704ce5d2b1SEric W. Biederman } 16714ce5d2b1SEric W. Biederman 167258be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 167358be2825SAl Viro { 167458be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 167558be2825SAl Viro } 167658be2825SAl Viro 16774ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 16784ce5d2b1SEric W. Biederman { 16794ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 16804ce5d2b1SEric W. Biederman * mount namespace loop? 16814ce5d2b1SEric W. Biederman */ 16824ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 16834ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 16844ce5d2b1SEric W. Biederman return false; 16854ce5d2b1SEric W. Biederman 1686f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 16878823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 16888823c079SEric W. Biederman } 16898823c079SEric W. Biederman 169087129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 169136341f64SRam Pai int flag) 16921da177e4SLinus Torvalds { 169384d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 16941da177e4SLinus Torvalds 16954ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 16964ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 16974ce5d2b1SEric W. Biederman 16984ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1699be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 17009676f0c6SRam Pai 170136341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1702be34d1a3SDavid Howells if (IS_ERR(q)) 1703be34d1a3SDavid Howells return q; 1704be34d1a3SDavid Howells 1705a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17061da177e4SLinus Torvalds 17071da177e4SLinus Torvalds p = mnt; 17086b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1709315fc83eSAl Viro struct mount *s; 17107ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17111da177e4SLinus Torvalds continue; 17121da177e4SLinus Torvalds 1713909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 171412a5b529SAl Viro struct mount *t = NULL; 17154ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17164ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 17174ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17184ce5d2b1SEric W. Biederman continue; 17194ce5d2b1SEric W. Biederman } 17204ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17214ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17229676f0c6SRam Pai s = skip_mnt_tree(s); 17239676f0c6SRam Pai continue; 17249676f0c6SRam Pai } 17250714a533SAl Viro while (p != s->mnt_parent) { 17260714a533SAl Viro p = p->mnt_parent; 17270714a533SAl Viro q = q->mnt_parent; 17281da177e4SLinus Torvalds } 172987129cc0SAl Viro p = s; 173084d17192SAl Viro parent = q; 173187129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1732be34d1a3SDavid Howells if (IS_ERR(q)) 1733be34d1a3SDavid Howells goto out; 1734719ea2fbSAl Viro lock_mount_hash(); 17351a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 173612a5b529SAl Viro mnt_set_mountpoint(parent, p->mnt_mp, q); 173712a5b529SAl Viro if (!list_empty(&parent->mnt_mounts)) { 173812a5b529SAl Viro t = list_last_entry(&parent->mnt_mounts, 173912a5b529SAl Viro struct mount, mnt_child); 174012a5b529SAl Viro if (t->mnt_mp != p->mnt_mp) 174112a5b529SAl Viro t = NULL; 174212a5b529SAl Viro } 174312a5b529SAl Viro attach_shadowed(q, parent, t); 1744719ea2fbSAl Viro unlock_mount_hash(); 17451da177e4SLinus Torvalds } 17461da177e4SLinus Torvalds } 17471da177e4SLinus Torvalds return res; 1748be34d1a3SDavid Howells out: 17491da177e4SLinus Torvalds if (res) { 1750719ea2fbSAl Viro lock_mount_hash(); 1751e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1752719ea2fbSAl Viro unlock_mount_hash(); 17531da177e4SLinus Torvalds } 1754be34d1a3SDavid Howells return q; 17551da177e4SLinus Torvalds } 17561da177e4SLinus Torvalds 1757be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1758be34d1a3SDavid Howells 1759589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 17608aec0809SAl Viro { 1761cb338d06SAl Viro struct mount *tree; 176297216be0SAl Viro namespace_lock(); 1763cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1764cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1765cd4a4017SEric W. Biederman else 176687129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 176787129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1768328e6d90SAl Viro namespace_unlock(); 1769be34d1a3SDavid Howells if (IS_ERR(tree)) 177052e220d3SDan Carpenter return ERR_CAST(tree); 1771be34d1a3SDavid Howells return &tree->mnt; 17728aec0809SAl Viro } 17738aec0809SAl Viro 17748aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 17758aec0809SAl Viro { 177697216be0SAl Viro namespace_lock(); 1777719ea2fbSAl Viro lock_mount_hash(); 1778e819f152SEric W. Biederman umount_tree(real_mount(mnt), UMOUNT_SYNC); 1779719ea2fbSAl Viro unlock_mount_hash(); 17803ab6abeeSAl Viro namespace_unlock(); 17818aec0809SAl Viro } 17828aec0809SAl Viro 1783c771d683SMiklos Szeredi /** 1784c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1785c771d683SMiklos Szeredi * 1786c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1787c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1788c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1789c771d683SMiklos Szeredi * 1790c771d683SMiklos Szeredi * Release with mntput(). 1791c771d683SMiklos Szeredi */ 1792c771d683SMiklos Szeredi struct vfsmount *clone_private_mount(struct path *path) 1793c771d683SMiklos Szeredi { 1794c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1795c771d683SMiklos Szeredi struct mount *new_mnt; 1796c771d683SMiklos Szeredi 1797c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1798c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1799c771d683SMiklos Szeredi 1800c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1801c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1802c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1803c771d683SMiklos Szeredi 1804c771d683SMiklos Szeredi return &new_mnt->mnt; 1805c771d683SMiklos Szeredi } 1806c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1807c771d683SMiklos Szeredi 18081f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18091f707137SAl Viro struct vfsmount *root) 18101f707137SAl Viro { 18111a4eeaf2SAl Viro struct mount *mnt; 18121f707137SAl Viro int res = f(root, arg); 18131f707137SAl Viro if (res) 18141f707137SAl Viro return res; 18151a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18161a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18171f707137SAl Viro if (res) 18181f707137SAl Viro return res; 18191f707137SAl Viro } 18201f707137SAl Viro return 0; 18211f707137SAl Viro } 18221f707137SAl Viro 18234b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1824719f5d7fSMiklos Szeredi { 1825315fc83eSAl Viro struct mount *p; 1826719f5d7fSMiklos Szeredi 1827909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1828fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 18294b8b21f4SAl Viro mnt_release_group_id(p); 1830719f5d7fSMiklos Szeredi } 1831719f5d7fSMiklos Szeredi } 1832719f5d7fSMiklos Szeredi 18334b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1834719f5d7fSMiklos Szeredi { 1835315fc83eSAl Viro struct mount *p; 1836719f5d7fSMiklos Szeredi 1837909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1838fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 18394b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1840719f5d7fSMiklos Szeredi if (err) { 18414b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1842719f5d7fSMiklos Szeredi return err; 1843719f5d7fSMiklos Szeredi } 1844719f5d7fSMiklos Szeredi } 1845719f5d7fSMiklos Szeredi } 1846719f5d7fSMiklos Szeredi 1847719f5d7fSMiklos Szeredi return 0; 1848719f5d7fSMiklos Szeredi } 1849719f5d7fSMiklos Szeredi 1850d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1851d2921684SEric W. Biederman { 1852d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1853d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1854d2921684SEric W. Biederman struct mount *p; 1855d2921684SEric W. Biederman 1856d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1857d2921684SEric W. Biederman mounts++; 1858d2921684SEric W. Biederman 1859d2921684SEric W. Biederman old = ns->mounts; 1860d2921684SEric W. Biederman pending = ns->pending_mounts; 1861d2921684SEric W. Biederman sum = old + pending; 1862d2921684SEric W. Biederman if ((old > sum) || 1863d2921684SEric W. Biederman (pending > sum) || 1864d2921684SEric W. Biederman (max < sum) || 1865d2921684SEric W. Biederman (mounts > (max - sum))) 1866d2921684SEric W. Biederman return -ENOSPC; 1867d2921684SEric W. Biederman 1868d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1869d2921684SEric W. Biederman return 0; 1870d2921684SEric W. Biederman } 1871d2921684SEric W. Biederman 1872b90fa9aeSRam Pai /* 1873b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1874b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 187521444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 187621444403SRam Pai * store the parent mount and mountpoint dentry. 187721444403SRam Pai * (done when source_mnt is moved) 1878b90fa9aeSRam Pai * 1879b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1880b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 18819676f0c6SRam Pai * --------------------------------------------------------------------------- 1882b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 18839676f0c6SRam Pai * |************************************************************************** 18849676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 18859676f0c6SRam Pai * | dest | | | | | 18869676f0c6SRam Pai * | | | | | | | 18879676f0c6SRam Pai * | v | | | | | 18889676f0c6SRam Pai * |************************************************************************** 18899676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 18905afe0022SRam Pai * | | | | | | 18919676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 18929676f0c6SRam Pai * *************************************************************************** 1893b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1894b90fa9aeSRam Pai * destination mount. 1895b90fa9aeSRam Pai * 1896b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1897b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1898b90fa9aeSRam Pai * the peer group of the source mount. 1899b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1900b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1901b90fa9aeSRam Pai * mount. 19025afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 19035afe0022SRam Pai * of the destination mount and the cloned mount is made slave 19045afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 19055afe0022SRam Pai * is marked as 'shared and slave'. 19065afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 19075afe0022SRam Pai * source mount. 19085afe0022SRam Pai * 19099676f0c6SRam Pai * --------------------------------------------------------------------------- 191021444403SRam Pai * | MOVE MOUNT OPERATION | 19119676f0c6SRam Pai * |************************************************************************** 19129676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19139676f0c6SRam Pai * | dest | | | | | 19149676f0c6SRam Pai * | | | | | | | 19159676f0c6SRam Pai * | v | | | | | 19169676f0c6SRam Pai * |************************************************************************** 19179676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 19185afe0022SRam Pai * | | | | | | 19199676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 19209676f0c6SRam Pai * *************************************************************************** 19215afe0022SRam Pai * 19225afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 19235afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 192421444403SRam Pai * (+*) the mount is moved to the destination. 19255afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 19265afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 19275afe0022SRam Pai * the mount is marked as 'shared and slave'. 19285afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1929b90fa9aeSRam Pai * 1930b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1931b90fa9aeSRam Pai * applied to each mount in the tree. 1932b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1933b90fa9aeSRam Pai * in allocations. 1934b90fa9aeSRam Pai */ 19350fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 193684d17192SAl Viro struct mount *dest_mnt, 193784d17192SAl Viro struct mountpoint *dest_mp, 193884d17192SAl Viro struct path *parent_path) 1939b90fa9aeSRam Pai { 194038129a13SAl Viro HLIST_HEAD(tree_list); 1941d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 1942315fc83eSAl Viro struct mount *child, *p; 194338129a13SAl Viro struct hlist_node *n; 1944719f5d7fSMiklos Szeredi int err; 1945b90fa9aeSRam Pai 1946d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 1947d2921684SEric W. Biederman if (!parent_path) { 1948d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 1949d2921684SEric W. Biederman if (err) 1950d2921684SEric W. Biederman goto out; 1951d2921684SEric W. Biederman } 1952d2921684SEric W. Biederman 1953fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 19540fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1955719f5d7fSMiklos Szeredi if (err) 1956719f5d7fSMiklos Szeredi goto out; 195784d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1958f2ebb3a9SAl Viro lock_mount_hash(); 1959719f5d7fSMiklos Szeredi if (err) 1960719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1961909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 19620f0afb1dSAl Viro set_mnt_shared(p); 19630b1b901bSAl Viro } else { 19640b1b901bSAl Viro lock_mount_hash(); 1965b90fa9aeSRam Pai } 19661a390689SAl Viro if (parent_path) { 19670fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 196884d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1969143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 197021444403SRam Pai } else { 197184d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 19721d6a32acSAl Viro commit_tree(source_mnt, NULL); 197321444403SRam Pai } 1974b90fa9aeSRam Pai 197538129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 19761d6a32acSAl Viro struct mount *q; 197738129a13SAl Viro hlist_del_init(&child->mnt_hash); 19781d6a32acSAl Viro q = __lookup_mnt_last(&child->mnt_parent->mnt, 19791d6a32acSAl Viro child->mnt_mountpoint); 19801d6a32acSAl Viro commit_tree(child, q); 1981b90fa9aeSRam Pai } 1982719ea2fbSAl Viro unlock_mount_hash(); 198399b7db7bSNick Piggin 1984b90fa9aeSRam Pai return 0; 1985719f5d7fSMiklos Szeredi 1986719f5d7fSMiklos Szeredi out_cleanup_ids: 1987f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 1988f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 1989d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 1990e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 1991f2ebb3a9SAl Viro } 1992f2ebb3a9SAl Viro unlock_mount_hash(); 19930fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1994719f5d7fSMiklos Szeredi out: 1995d2921684SEric W. Biederman ns->pending_mounts = 0; 1996719f5d7fSMiklos Szeredi return err; 1997b90fa9aeSRam Pai } 1998b90fa9aeSRam Pai 199984d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2000b12cea91SAl Viro { 2001b12cea91SAl Viro struct vfsmount *mnt; 200284d17192SAl Viro struct dentry *dentry = path->dentry; 2003b12cea91SAl Viro retry: 20045955102cSAl Viro inode_lock(dentry->d_inode); 200584d17192SAl Viro if (unlikely(cant_mount(dentry))) { 20065955102cSAl Viro inode_unlock(dentry->d_inode); 200784d17192SAl Viro return ERR_PTR(-ENOENT); 2008b12cea91SAl Viro } 200997216be0SAl Viro namespace_lock(); 2010b12cea91SAl Viro mnt = lookup_mnt(path); 201184d17192SAl Viro if (likely(!mnt)) { 2012e2dfa935SEric W. Biederman struct mountpoint *mp = lookup_mountpoint(dentry); 2013e2dfa935SEric W. Biederman if (!mp) 2014e2dfa935SEric W. Biederman mp = new_mountpoint(dentry); 201584d17192SAl Viro if (IS_ERR(mp)) { 201697216be0SAl Viro namespace_unlock(); 20175955102cSAl Viro inode_unlock(dentry->d_inode); 201884d17192SAl Viro return mp; 201984d17192SAl Viro } 202084d17192SAl Viro return mp; 202184d17192SAl Viro } 202297216be0SAl Viro namespace_unlock(); 20235955102cSAl Viro inode_unlock(path->dentry->d_inode); 2024b12cea91SAl Viro path_put(path); 2025b12cea91SAl Viro path->mnt = mnt; 202684d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2027b12cea91SAl Viro goto retry; 2028b12cea91SAl Viro } 2029b12cea91SAl Viro 203084d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2031b12cea91SAl Viro { 203284d17192SAl Viro struct dentry *dentry = where->m_dentry; 203384d17192SAl Viro put_mountpoint(where); 2034328e6d90SAl Viro namespace_unlock(); 20355955102cSAl Viro inode_unlock(dentry->d_inode); 2036b12cea91SAl Viro } 2037b12cea91SAl Viro 203884d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 20391da177e4SLinus Torvalds { 204095bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 20411da177e4SLinus Torvalds return -EINVAL; 20421da177e4SLinus Torvalds 2043e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2044e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 20451da177e4SLinus Torvalds return -ENOTDIR; 20461da177e4SLinus Torvalds 204784d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 20481da177e4SLinus Torvalds } 20491da177e4SLinus Torvalds 20501da177e4SLinus Torvalds /* 20517a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 20527a2e8a8fSValerie Aurora */ 20537a2e8a8fSValerie Aurora 20547a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 20557a2e8a8fSValerie Aurora { 20567c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 20577a2e8a8fSValerie Aurora 20587a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 20597a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 20607a2e8a8fSValerie Aurora return 0; 20617a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 20627a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 20637a2e8a8fSValerie Aurora return 0; 20647a2e8a8fSValerie Aurora return type; 20657a2e8a8fSValerie Aurora } 20667a2e8a8fSValerie Aurora 20677a2e8a8fSValerie Aurora /* 206807b20889SRam Pai * recursively change the type of the mountpoint. 206907b20889SRam Pai */ 20700a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 207107b20889SRam Pai { 2072315fc83eSAl Viro struct mount *m; 20734b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 207407b20889SRam Pai int recurse = flag & MS_REC; 20757a2e8a8fSValerie Aurora int type; 2076719f5d7fSMiklos Szeredi int err = 0; 207707b20889SRam Pai 20782d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 207907b20889SRam Pai return -EINVAL; 208007b20889SRam Pai 20817a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 20827a2e8a8fSValerie Aurora if (!type) 20837a2e8a8fSValerie Aurora return -EINVAL; 20847a2e8a8fSValerie Aurora 208597216be0SAl Viro namespace_lock(); 2086719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2087719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2088719f5d7fSMiklos Szeredi if (err) 2089719f5d7fSMiklos Szeredi goto out_unlock; 2090719f5d7fSMiklos Szeredi } 2091719f5d7fSMiklos Szeredi 2092719ea2fbSAl Viro lock_mount_hash(); 2093909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 20940f0afb1dSAl Viro change_mnt_propagation(m, type); 2095719ea2fbSAl Viro unlock_mount_hash(); 2096719f5d7fSMiklos Szeredi 2097719f5d7fSMiklos Szeredi out_unlock: 209897216be0SAl Viro namespace_unlock(); 2099719f5d7fSMiklos Szeredi return err; 210007b20889SRam Pai } 210107b20889SRam Pai 21025ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 21035ff9d8a6SEric W. Biederman { 21045ff9d8a6SEric W. Biederman struct mount *child; 21055ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 21065ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 21075ff9d8a6SEric W. Biederman continue; 21085ff9d8a6SEric W. Biederman 21095ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 21105ff9d8a6SEric W. Biederman return true; 21115ff9d8a6SEric W. Biederman } 21125ff9d8a6SEric W. Biederman return false; 21135ff9d8a6SEric W. Biederman } 21145ff9d8a6SEric W. Biederman 211507b20889SRam Pai /* 21161da177e4SLinus Torvalds * do loopback mount. 21171da177e4SLinus Torvalds */ 2118808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 21192dafe1c4SEric Sandeen int recurse) 21201da177e4SLinus Torvalds { 21212d92ab3cSAl Viro struct path old_path; 212284d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 212384d17192SAl Viro struct mountpoint *mp; 212457eccb83SAl Viro int err; 21251da177e4SLinus Torvalds if (!old_name || !*old_name) 21261da177e4SLinus Torvalds return -EINVAL; 2127815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 21281da177e4SLinus Torvalds if (err) 21291da177e4SLinus Torvalds return err; 21301da177e4SLinus Torvalds 21318823c079SEric W. Biederman err = -EINVAL; 21324ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 21338823c079SEric W. Biederman goto out; 21348823c079SEric W. Biederman 213584d17192SAl Viro mp = lock_mount(path); 213684d17192SAl Viro err = PTR_ERR(mp); 213784d17192SAl Viro if (IS_ERR(mp)) 21389676f0c6SRam Pai goto out; 21399676f0c6SRam Pai 214087129cc0SAl Viro old = real_mount(old_path.mnt); 214184d17192SAl Viro parent = real_mount(path->mnt); 214287129cc0SAl Viro 2143b12cea91SAl Viro err = -EINVAL; 2144fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2145b12cea91SAl Viro goto out2; 2146b12cea91SAl Viro 2147e149ed2bSAl Viro if (!check_mnt(parent)) 2148e149ed2bSAl Viro goto out2; 2149e149ed2bSAl Viro 2150e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2151b12cea91SAl Viro goto out2; 2152ccd48bc7SAl Viro 21535ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 21545ff9d8a6SEric W. Biederman goto out2; 21555ff9d8a6SEric W. Biederman 21561da177e4SLinus Torvalds if (recurse) 21574ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 21581da177e4SLinus Torvalds else 215987129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 21601da177e4SLinus Torvalds 2161be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2162be34d1a3SDavid Howells err = PTR_ERR(mnt); 2163e9c5d8a5SAndrey Vagin goto out2; 2164be34d1a3SDavid Howells } 2165ccd48bc7SAl Viro 21665ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 21675ff9d8a6SEric W. Biederman 216884d17192SAl Viro err = graft_tree(mnt, parent, mp); 21691da177e4SLinus Torvalds if (err) { 2170719ea2fbSAl Viro lock_mount_hash(); 2171e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2172719ea2fbSAl Viro unlock_mount_hash(); 21735b83d2c5SRam Pai } 2174b12cea91SAl Viro out2: 217584d17192SAl Viro unlock_mount(mp); 2176ccd48bc7SAl Viro out: 21772d92ab3cSAl Viro path_put(&old_path); 21781da177e4SLinus Torvalds return err; 21791da177e4SLinus Torvalds } 21801da177e4SLinus Torvalds 21812e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 21822e4b7fcdSDave Hansen { 21832e4b7fcdSDave Hansen int error = 0; 21842e4b7fcdSDave Hansen int readonly_request = 0; 21852e4b7fcdSDave Hansen 21862e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 21872e4b7fcdSDave Hansen readonly_request = 1; 21882e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 21892e4b7fcdSDave Hansen return 0; 21902e4b7fcdSDave Hansen 21912e4b7fcdSDave Hansen if (readonly_request) 219283adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 21932e4b7fcdSDave Hansen else 219483adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 21952e4b7fcdSDave Hansen return error; 21962e4b7fcdSDave Hansen } 21972e4b7fcdSDave Hansen 21981da177e4SLinus Torvalds /* 21991da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 22001da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 22011da177e4SLinus Torvalds * on it - tough luck. 22021da177e4SLinus Torvalds */ 22030a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 22041da177e4SLinus Torvalds void *data) 22051da177e4SLinus Torvalds { 22061da177e4SLinus Torvalds int err; 22072d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2208143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 22091da177e4SLinus Torvalds 2210143c8c91SAl Viro if (!check_mnt(mnt)) 22111da177e4SLinus Torvalds return -EINVAL; 22121da177e4SLinus Torvalds 22132d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 22141da177e4SLinus Torvalds return -EINVAL; 22151da177e4SLinus Torvalds 221607b64558SEric W. Biederman /* Don't allow changing of locked mnt flags. 221707b64558SEric W. Biederman * 221807b64558SEric W. Biederman * No locks need to be held here while testing the various 221907b64558SEric W. Biederman * MNT_LOCK flags because those flags can never be cleared 222007b64558SEric W. Biederman * once they are set. 222107b64558SEric W. Biederman */ 222207b64558SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_READONLY) && 222307b64558SEric W. Biederman !(mnt_flags & MNT_READONLY)) { 222407b64558SEric W. Biederman return -EPERM; 222507b64558SEric W. Biederman } 22269566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NODEV) && 22279566d674SEric W. Biederman !(mnt_flags & MNT_NODEV)) { 22289566d674SEric W. Biederman return -EPERM; 22299566d674SEric W. Biederman } 22309566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOSUID) && 22319566d674SEric W. Biederman !(mnt_flags & MNT_NOSUID)) { 22329566d674SEric W. Biederman return -EPERM; 22339566d674SEric W. Biederman } 22349566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_NOEXEC) && 22359566d674SEric W. Biederman !(mnt_flags & MNT_NOEXEC)) { 22369566d674SEric W. Biederman return -EPERM; 22379566d674SEric W. Biederman } 22389566d674SEric W. Biederman if ((mnt->mnt.mnt_flags & MNT_LOCK_ATIME) && 22399566d674SEric W. Biederman ((mnt->mnt.mnt_flags & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) { 22409566d674SEric W. Biederman return -EPERM; 22419566d674SEric W. Biederman } 22429566d674SEric W. Biederman 2243ff36fe2cSEric Paris err = security_sb_remount(sb, data); 2244ff36fe2cSEric Paris if (err) 2245ff36fe2cSEric Paris return err; 2246ff36fe2cSEric Paris 22471da177e4SLinus Torvalds down_write(&sb->s_umount); 22482e4b7fcdSDave Hansen if (flags & MS_BIND) 22492d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 225057eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 225157eccb83SAl Viro err = -EPERM; 22524aa98cf7SAl Viro else 22531da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 22547b43a79fSAl Viro if (!err) { 2255719ea2fbSAl Viro lock_mount_hash(); 2256a6138db8SEric W. Biederman mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 2257143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 2258143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2259719ea2fbSAl Viro unlock_mount_hash(); 22600e55a7ccSDan Williams } 22616339dab8SAl Viro up_write(&sb->s_umount); 22621da177e4SLinus Torvalds return err; 22631da177e4SLinus Torvalds } 22641da177e4SLinus Torvalds 2265cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 22669676f0c6SRam Pai { 2267315fc83eSAl Viro struct mount *p; 2268909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2269fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 22709676f0c6SRam Pai return 1; 22719676f0c6SRam Pai } 22729676f0c6SRam Pai return 0; 22739676f0c6SRam Pai } 22749676f0c6SRam Pai 2275808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 22761da177e4SLinus Torvalds { 22772d92ab3cSAl Viro struct path old_path, parent_path; 2278676da58dSAl Viro struct mount *p; 22790fb54e50SAl Viro struct mount *old; 228084d17192SAl Viro struct mountpoint *mp; 228157eccb83SAl Viro int err; 22821da177e4SLinus Torvalds if (!old_name || !*old_name) 22831da177e4SLinus Torvalds return -EINVAL; 22842d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 22851da177e4SLinus Torvalds if (err) 22861da177e4SLinus Torvalds return err; 22871da177e4SLinus Torvalds 228884d17192SAl Viro mp = lock_mount(path); 228984d17192SAl Viro err = PTR_ERR(mp); 229084d17192SAl Viro if (IS_ERR(mp)) 2291cc53ce53SDavid Howells goto out; 2292cc53ce53SDavid Howells 2293143c8c91SAl Viro old = real_mount(old_path.mnt); 2294fc7be130SAl Viro p = real_mount(path->mnt); 2295143c8c91SAl Viro 22961da177e4SLinus Torvalds err = -EINVAL; 2297fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 22981da177e4SLinus Torvalds goto out1; 22991da177e4SLinus Torvalds 23005ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 23015ff9d8a6SEric W. Biederman goto out1; 23025ff9d8a6SEric W. Biederman 23031da177e4SLinus Torvalds err = -EINVAL; 23042d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 230521444403SRam Pai goto out1; 23061da177e4SLinus Torvalds 2307676da58dSAl Viro if (!mnt_has_parent(old)) 230821444403SRam Pai goto out1; 23091da177e4SLinus Torvalds 2310e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2311e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 231221444403SRam Pai goto out1; 231321444403SRam Pai /* 231421444403SRam Pai * Don't move a mount residing in a shared parent. 231521444403SRam Pai */ 2316fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 231721444403SRam Pai goto out1; 23189676f0c6SRam Pai /* 23199676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 23209676f0c6SRam Pai * mount which is shared. 23219676f0c6SRam Pai */ 2322fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 23239676f0c6SRam Pai goto out1; 23241da177e4SLinus Torvalds err = -ELOOP; 2325fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2326676da58dSAl Viro if (p == old) 232721444403SRam Pai goto out1; 23281da177e4SLinus Torvalds 232984d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 23304ac91378SJan Blunck if (err) 233121444403SRam Pai goto out1; 23321da177e4SLinus Torvalds 23331da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 23341da177e4SLinus Torvalds * automatically */ 23356776db3dSAl Viro list_del_init(&old->mnt_expire); 23361da177e4SLinus Torvalds out1: 233784d17192SAl Viro unlock_mount(mp); 23381da177e4SLinus Torvalds out: 23391da177e4SLinus Torvalds if (!err) 23401a390689SAl Viro path_put(&parent_path); 23412d92ab3cSAl Viro path_put(&old_path); 23421da177e4SLinus Torvalds return err; 23431da177e4SLinus Torvalds } 23441da177e4SLinus Torvalds 23459d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 23469d412a43SAl Viro { 23479d412a43SAl Viro int err; 23489d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 23499d412a43SAl Viro if (subtype) { 23509d412a43SAl Viro subtype++; 23519d412a43SAl Viro err = -EINVAL; 23529d412a43SAl Viro if (!subtype[0]) 23539d412a43SAl Viro goto err; 23549d412a43SAl Viro } else 23559d412a43SAl Viro subtype = ""; 23569d412a43SAl Viro 23579d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 23589d412a43SAl Viro err = -ENOMEM; 23599d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 23609d412a43SAl Viro goto err; 23619d412a43SAl Viro return mnt; 23629d412a43SAl Viro 23639d412a43SAl Viro err: 23649d412a43SAl Viro mntput(mnt); 23659d412a43SAl Viro return ERR_PTR(err); 23669d412a43SAl Viro } 23679d412a43SAl Viro 23689d412a43SAl Viro /* 23699d412a43SAl Viro * add a mount into a namespace's mount tree 23709d412a43SAl Viro */ 237195bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 23729d412a43SAl Viro { 237384d17192SAl Viro struct mountpoint *mp; 237484d17192SAl Viro struct mount *parent; 23759d412a43SAl Viro int err; 23769d412a43SAl Viro 2377f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 23789d412a43SAl Viro 237984d17192SAl Viro mp = lock_mount(path); 238084d17192SAl Viro if (IS_ERR(mp)) 238184d17192SAl Viro return PTR_ERR(mp); 23829d412a43SAl Viro 238384d17192SAl Viro parent = real_mount(path->mnt); 23849d412a43SAl Viro err = -EINVAL; 238584d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2386156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2387156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 23889d412a43SAl Viro goto unlock; 2389156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 239084d17192SAl Viro if (!parent->mnt_ns) 2391156cacb1SAl Viro goto unlock; 2392156cacb1SAl Viro } 23939d412a43SAl Viro 23949d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 23959d412a43SAl Viro err = -EBUSY; 239695bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 23979d412a43SAl Viro path->mnt->mnt_root == path->dentry) 23989d412a43SAl Viro goto unlock; 23999d412a43SAl Viro 24009d412a43SAl Viro err = -EINVAL; 2401e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 24029d412a43SAl Viro goto unlock; 24039d412a43SAl Viro 240495bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 240584d17192SAl Viro err = graft_tree(newmnt, parent, mp); 24069d412a43SAl Viro 24079d412a43SAl Viro unlock: 240884d17192SAl Viro unlock_mount(mp); 24099d412a43SAl Viro return err; 24109d412a43SAl Viro } 2411b1e75df4SAl Viro 24128654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); 24131b852bceSEric W. Biederman 24141da177e4SLinus Torvalds /* 24151da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 24161da177e4SLinus Torvalds * namespace's tree 24171da177e4SLinus Torvalds */ 24180c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2419808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 24201da177e4SLinus Torvalds { 24210c55cfc4SEric W. Biederman struct file_system_type *type; 24221da177e4SLinus Torvalds struct vfsmount *mnt; 242315f9a3f3SAl Viro int err; 24241da177e4SLinus Torvalds 24250c55cfc4SEric W. Biederman if (!fstype) 24261da177e4SLinus Torvalds return -EINVAL; 24271da177e4SLinus Torvalds 24280c55cfc4SEric W. Biederman type = get_fs_type(fstype); 24290c55cfc4SEric W. Biederman if (!type) 24300c55cfc4SEric W. Biederman return -ENODEV; 24310c55cfc4SEric W. Biederman 24320c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 24330c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 24340c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 24350c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 24360c55cfc4SEric W. Biederman 24370c55cfc4SEric W. Biederman put_filesystem(type); 24381da177e4SLinus Torvalds if (IS_ERR(mnt)) 24391da177e4SLinus Torvalds return PTR_ERR(mnt); 24401da177e4SLinus Torvalds 24418654df4eSEric W. Biederman if (mount_too_revealing(mnt, &mnt_flags)) { 24428654df4eSEric W. Biederman mntput(mnt); 24438654df4eSEric W. Biederman return -EPERM; 24448654df4eSEric W. Biederman } 24458654df4eSEric W. Biederman 244695bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 244715f9a3f3SAl Viro if (err) 244815f9a3f3SAl Viro mntput(mnt); 244915f9a3f3SAl Viro return err; 24501da177e4SLinus Torvalds } 24511da177e4SLinus Torvalds 245219a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 245319a167afSAl Viro { 24546776db3dSAl Viro struct mount *mnt = real_mount(m); 245519a167afSAl Viro int err; 245619a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 245719a167afSAl Viro * expired before we get a chance to add it 245819a167afSAl Viro */ 24596776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 246019a167afSAl Viro 246119a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 246219a167afSAl Viro m->mnt_root == path->dentry) { 2463b1e75df4SAl Viro err = -ELOOP; 2464b1e75df4SAl Viro goto fail; 246519a167afSAl Viro } 246619a167afSAl Viro 246795bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2468b1e75df4SAl Viro if (!err) 2469b1e75df4SAl Viro return 0; 2470b1e75df4SAl Viro fail: 2471b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 24726776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 247397216be0SAl Viro namespace_lock(); 24746776db3dSAl Viro list_del_init(&mnt->mnt_expire); 247597216be0SAl Viro namespace_unlock(); 247619a167afSAl Viro } 2477b1e75df4SAl Viro mntput(m); 2478b1e75df4SAl Viro mntput(m); 247919a167afSAl Viro return err; 248019a167afSAl Viro } 248119a167afSAl Viro 2482ea5b778aSDavid Howells /** 2483ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2484ea5b778aSDavid Howells * @mnt: The mount to list. 2485ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2486ea5b778aSDavid Howells */ 2487ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2488ea5b778aSDavid Howells { 248997216be0SAl Viro namespace_lock(); 2490ea5b778aSDavid Howells 24916776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2492ea5b778aSDavid Howells 249397216be0SAl Viro namespace_unlock(); 2494ea5b778aSDavid Howells } 2495ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2496ea5b778aSDavid Howells 2497ea5b778aSDavid Howells /* 24981da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 24991da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 25001da177e4SLinus Torvalds * here 25011da177e4SLinus Torvalds */ 25021da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 25031da177e4SLinus Torvalds { 2504761d5c38SAl Viro struct mount *mnt, *next; 25051da177e4SLinus Torvalds LIST_HEAD(graveyard); 25061da177e4SLinus Torvalds 25071da177e4SLinus Torvalds if (list_empty(mounts)) 25081da177e4SLinus Torvalds return; 25091da177e4SLinus Torvalds 251097216be0SAl Viro namespace_lock(); 2511719ea2fbSAl Viro lock_mount_hash(); 25121da177e4SLinus Torvalds 25131da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 25141da177e4SLinus Torvalds * following criteria: 25151da177e4SLinus Torvalds * - only referenced by its parent vfsmount 25161da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 25171da177e4SLinus Torvalds * cleared by mntput()) 25181da177e4SLinus Torvalds */ 25196776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2520863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 25211ab59738SAl Viro propagate_mount_busy(mnt, 1)) 25221da177e4SLinus Torvalds continue; 25236776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 25241da177e4SLinus Torvalds } 2525bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 25266776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2527143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2528e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2529bcc5c7d2SAl Viro } 2530719ea2fbSAl Viro unlock_mount_hash(); 25313ab6abeeSAl Viro namespace_unlock(); 25321da177e4SLinus Torvalds } 25331da177e4SLinus Torvalds 25341da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 25351da177e4SLinus Torvalds 25361da177e4SLinus Torvalds /* 25375528f911STrond Myklebust * Ripoff of 'select_parent()' 25385528f911STrond Myklebust * 25395528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 25405528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 25415528f911STrond Myklebust */ 2542692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 25435528f911STrond Myklebust { 2544692afc31SAl Viro struct mount *this_parent = parent; 25455528f911STrond Myklebust struct list_head *next; 25465528f911STrond Myklebust int found = 0; 25475528f911STrond Myklebust 25485528f911STrond Myklebust repeat: 25496b41d536SAl Viro next = this_parent->mnt_mounts.next; 25505528f911STrond Myklebust resume: 25516b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 25525528f911STrond Myklebust struct list_head *tmp = next; 25536b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 25545528f911STrond Myklebust 25555528f911STrond Myklebust next = tmp->next; 2556692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 25575528f911STrond Myklebust continue; 25585528f911STrond Myklebust /* 25595528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 25605528f911STrond Myklebust */ 25616b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 25625528f911STrond Myklebust this_parent = mnt; 25635528f911STrond Myklebust goto repeat; 25645528f911STrond Myklebust } 25655528f911STrond Myklebust 25661ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 25676776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 25685528f911STrond Myklebust found++; 25695528f911STrond Myklebust } 25705528f911STrond Myklebust } 25715528f911STrond Myklebust /* 25725528f911STrond Myklebust * All done at this level ... ascend and resume the search 25735528f911STrond Myklebust */ 25745528f911STrond Myklebust if (this_parent != parent) { 25756b41d536SAl Viro next = this_parent->mnt_child.next; 25760714a533SAl Viro this_parent = this_parent->mnt_parent; 25775528f911STrond Myklebust goto resume; 25785528f911STrond Myklebust } 25795528f911STrond Myklebust return found; 25805528f911STrond Myklebust } 25815528f911STrond Myklebust 25825528f911STrond Myklebust /* 25835528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 25845528f911STrond Myklebust * submounts of a specific parent mountpoint 258599b7db7bSNick Piggin * 258648a066e7SAl Viro * mount_lock must be held for write 25875528f911STrond Myklebust */ 2588b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 25895528f911STrond Myklebust { 25905528f911STrond Myklebust LIST_HEAD(graveyard); 2591761d5c38SAl Viro struct mount *m; 25925528f911STrond Myklebust 25935528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2594c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2595bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2596761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 25976776db3dSAl Viro mnt_expire); 2598143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2599e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2600bcc5c7d2SAl Viro } 2601bcc5c7d2SAl Viro } 26025528f911STrond Myklebust } 26035528f911STrond Myklebust 26045528f911STrond Myklebust /* 26051da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 26061da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 26071da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 26081da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 26091da177e4SLinus Torvalds */ 2610b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2611b58fed8bSRam Pai unsigned long n) 26121da177e4SLinus Torvalds { 26131da177e4SLinus Torvalds char *t = to; 26141da177e4SLinus Torvalds const char __user *f = from; 26151da177e4SLinus Torvalds char c; 26161da177e4SLinus Torvalds 26171da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 26181da177e4SLinus Torvalds return n; 26191da177e4SLinus Torvalds 26201da177e4SLinus Torvalds while (n) { 26211da177e4SLinus Torvalds if (__get_user(c, f)) { 26221da177e4SLinus Torvalds memset(t, 0, n); 26231da177e4SLinus Torvalds break; 26241da177e4SLinus Torvalds } 26251da177e4SLinus Torvalds *t++ = c; 26261da177e4SLinus Torvalds f++; 26271da177e4SLinus Torvalds n--; 26281da177e4SLinus Torvalds } 26291da177e4SLinus Torvalds return n; 26301da177e4SLinus Torvalds } 26311da177e4SLinus Torvalds 2632b40ef869SAl Viro void *copy_mount_options(const void __user * data) 26331da177e4SLinus Torvalds { 26341da177e4SLinus Torvalds int i; 26351da177e4SLinus Torvalds unsigned long size; 2636b40ef869SAl Viro char *copy; 26371da177e4SLinus Torvalds 26381da177e4SLinus Torvalds if (!data) 2639b40ef869SAl Viro return NULL; 26401da177e4SLinus Torvalds 2641b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2642b40ef869SAl Viro if (!copy) 2643b40ef869SAl Viro return ERR_PTR(-ENOMEM); 26441da177e4SLinus Torvalds 26451da177e4SLinus Torvalds /* We only care that *some* data at the address the user 26461da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 26471da177e4SLinus Torvalds * the remainder of the page. 26481da177e4SLinus Torvalds */ 26491da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 26501da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 26511da177e4SLinus Torvalds if (size > PAGE_SIZE) 26521da177e4SLinus Torvalds size = PAGE_SIZE; 26531da177e4SLinus Torvalds 2654b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 26551da177e4SLinus Torvalds if (!i) { 2656b40ef869SAl Viro kfree(copy); 2657b40ef869SAl Viro return ERR_PTR(-EFAULT); 26581da177e4SLinus Torvalds } 26591da177e4SLinus Torvalds if (i != PAGE_SIZE) 2660b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2661b40ef869SAl Viro return copy; 26621da177e4SLinus Torvalds } 26631da177e4SLinus Torvalds 2664b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2665eca6f534SVegard Nossum { 2666b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2667eca6f534SVegard Nossum } 2668eca6f534SVegard Nossum 26691da177e4SLinus Torvalds /* 26701da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 26711da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 26721da177e4SLinus Torvalds * 26731da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 26741da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 26751da177e4SLinus Torvalds * information (or be NULL). 26761da177e4SLinus Torvalds * 26771da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 26781da177e4SLinus Torvalds * When the flags word was introduced its top half was required 26791da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 26801da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 26811da177e4SLinus Torvalds * and must be discarded. 26821da177e4SLinus Torvalds */ 26835e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2684808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 26851da177e4SLinus Torvalds { 26862d92ab3cSAl Viro struct path path; 26871da177e4SLinus Torvalds int retval = 0; 26881da177e4SLinus Torvalds int mnt_flags = 0; 26891da177e4SLinus Torvalds 26901da177e4SLinus Torvalds /* Discard magic */ 26911da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 26921da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 26931da177e4SLinus Torvalds 26941da177e4SLinus Torvalds /* Basic sanity checks */ 26951da177e4SLinus Torvalds if (data_page) 26961da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 26971da177e4SLinus Torvalds 2698a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 26995e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2700a27ab9f2STetsuo Handa if (retval) 2701a27ab9f2STetsuo Handa return retval; 2702a27ab9f2STetsuo Handa 2703a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2704a27ab9f2STetsuo Handa type_page, flags, data_page); 27050d5cadb8SAl Viro if (!retval && !may_mount()) 27060d5cadb8SAl Viro retval = -EPERM; 27079e8925b6SJeff Layton if (!retval && (flags & MS_MANDLOCK) && !may_mandlock()) 27089e8925b6SJeff Layton retval = -EPERM; 2709a27ab9f2STetsuo Handa if (retval) 2710a27ab9f2STetsuo Handa goto dput_out; 2711a27ab9f2STetsuo Handa 2712613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2713613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 27140a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 27150a1c01c9SMatthew Garrett 27161da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 27171da177e4SLinus Torvalds if (flags & MS_NOSUID) 27181da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 27191da177e4SLinus Torvalds if (flags & MS_NODEV) 27201da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 27211da177e4SLinus Torvalds if (flags & MS_NOEXEC) 27221da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2723fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2724fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2725fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2726fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2727d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2728d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 27292e4b7fcdSDave Hansen if (flags & MS_RDONLY) 27302e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2731fc33a7bbSChristoph Hellwig 2732ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2733ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2734ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2735ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2736ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2737ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2738ffbc6f0eSEric W. Biederman } 2739ffbc6f0eSEric W. Biederman 27407a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2741d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2742c568d683SMiklos Szeredi MS_STRICTATIME | MS_NOREMOTELOCK); 27431da177e4SLinus Torvalds 27441da177e4SLinus Torvalds if (flags & MS_REMOUNT) 27452d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 27461da177e4SLinus Torvalds data_page); 27471da177e4SLinus Torvalds else if (flags & MS_BIND) 27482d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 27499676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 27502d92ab3cSAl Viro retval = do_change_type(&path, flags); 27511da177e4SLinus Torvalds else if (flags & MS_MOVE) 27522d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 27531da177e4SLinus Torvalds else 27542d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 27551da177e4SLinus Torvalds dev_name, data_page); 27561da177e4SLinus Torvalds dput_out: 27572d92ab3cSAl Viro path_put(&path); 27581da177e4SLinus Torvalds return retval; 27591da177e4SLinus Torvalds } 27601da177e4SLinus Torvalds 2761537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 2762537f7ccbSEric W. Biederman { 2763537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 2764537f7ccbSEric W. Biederman } 2765537f7ccbSEric W. Biederman 2766537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 2767537f7ccbSEric W. Biederman { 2768537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 2769537f7ccbSEric W. Biederman } 2770537f7ccbSEric W. Biederman 2771771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2772771b1371SEric W. Biederman { 27736344c433SAl Viro ns_free_inum(&ns->ns); 2774537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 2775771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2776771b1371SEric W. Biederman kfree(ns); 2777771b1371SEric W. Biederman } 2778771b1371SEric W. Biederman 27798823c079SEric W. Biederman /* 27808823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 27818823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 27828823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 27838823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 27848823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 27858823c079SEric W. Biederman */ 27868823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 27878823c079SEric W. Biederman 2788771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2789cf8d2c11STrond Myklebust { 2790cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2791537f7ccbSEric W. Biederman struct ucounts *ucounts; 279298f842e6SEric W. Biederman int ret; 2793cf8d2c11STrond Myklebust 2794537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 2795537f7ccbSEric W. Biederman if (!ucounts) 2796df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 2797537f7ccbSEric W. Biederman 2798cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2799537f7ccbSEric W. Biederman if (!new_ns) { 2800537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 2801cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2802537f7ccbSEric W. Biederman } 28036344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 280498f842e6SEric W. Biederman if (ret) { 280598f842e6SEric W. Biederman kfree(new_ns); 2806537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 280798f842e6SEric W. Biederman return ERR_PTR(ret); 280898f842e6SEric W. Biederman } 280933c42940SAl Viro new_ns->ns.ops = &mntns_operations; 28108823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2811cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2812cf8d2c11STrond Myklebust new_ns->root = NULL; 2813cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2814cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2815cf8d2c11STrond Myklebust new_ns->event = 0; 2816771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2817537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 2818d2921684SEric W. Biederman new_ns->mounts = 0; 2819d2921684SEric W. Biederman new_ns->pending_mounts = 0; 2820cf8d2c11STrond Myklebust return new_ns; 2821cf8d2c11STrond Myklebust } 2822cf8d2c11STrond Myklebust 28230766f788SEmese Revfy __latent_entropy 28249559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 28259559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 28261da177e4SLinus Torvalds { 28276b3286edSKirill Korotaev struct mnt_namespace *new_ns; 28287f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2829315fc83eSAl Viro struct mount *p, *q; 28309559f689SAl Viro struct mount *old; 2831cb338d06SAl Viro struct mount *new; 28327a472ef4SEric W. Biederman int copy_flags; 28331da177e4SLinus Torvalds 28349559f689SAl Viro BUG_ON(!ns); 28359559f689SAl Viro 28369559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 28379559f689SAl Viro get_mnt_ns(ns); 28389559f689SAl Viro return ns; 28399559f689SAl Viro } 28409559f689SAl Viro 28419559f689SAl Viro old = ns->root; 28429559f689SAl Viro 2843771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2844cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2845cf8d2c11STrond Myklebust return new_ns; 28461da177e4SLinus Torvalds 284797216be0SAl Viro namespace_lock(); 28481da177e4SLinus Torvalds /* First pass: copy the tree topology */ 28494ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 28509559f689SAl Viro if (user_ns != ns->user_ns) 2851132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 28527a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2853be34d1a3SDavid Howells if (IS_ERR(new)) { 2854328e6d90SAl Viro namespace_unlock(); 2855771b1371SEric W. Biederman free_mnt_ns(new_ns); 2856be34d1a3SDavid Howells return ERR_CAST(new); 28571da177e4SLinus Torvalds } 2858be08d6d2SAl Viro new_ns->root = new; 28591a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 28601da177e4SLinus Torvalds 28611da177e4SLinus Torvalds /* 28621da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 28631da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 28641da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 28651da177e4SLinus Torvalds */ 2866909b0a88SAl Viro p = old; 2867cb338d06SAl Viro q = new; 28681da177e4SLinus Torvalds while (p) { 2869143c8c91SAl Viro q->mnt_ns = new_ns; 2870d2921684SEric W. Biederman new_ns->mounts++; 28719559f689SAl Viro if (new_fs) { 28729559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 28739559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2874315fc83eSAl Viro rootmnt = &p->mnt; 28751da177e4SLinus Torvalds } 28769559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 28779559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2878315fc83eSAl Viro pwdmnt = &p->mnt; 28791da177e4SLinus Torvalds } 28801da177e4SLinus Torvalds } 2881909b0a88SAl Viro p = next_mnt(p, old); 2882909b0a88SAl Viro q = next_mnt(q, new); 28834ce5d2b1SEric W. Biederman if (!q) 28844ce5d2b1SEric W. Biederman break; 28854ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 28864ce5d2b1SEric W. Biederman p = next_mnt(p, old); 28871da177e4SLinus Torvalds } 2888328e6d90SAl Viro namespace_unlock(); 28891da177e4SLinus Torvalds 28901da177e4SLinus Torvalds if (rootmnt) 2891f03c6599SAl Viro mntput(rootmnt); 28921da177e4SLinus Torvalds if (pwdmnt) 2893f03c6599SAl Viro mntput(pwdmnt); 28941da177e4SLinus Torvalds 2895741a2951SJANAK DESAI return new_ns; 2896741a2951SJANAK DESAI } 2897741a2951SJANAK DESAI 2898cf8d2c11STrond Myklebust /** 2899cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2900cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2901cf8d2c11STrond Myklebust */ 29021a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2903cf8d2c11STrond Myklebust { 2904771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2905cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 29061a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 29071a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2908be08d6d2SAl Viro new_ns->root = mnt; 2909d2921684SEric W. Biederman new_ns->mounts++; 2910b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2911c1334495SAl Viro } else { 29121a4eeaf2SAl Viro mntput(m); 2913cf8d2c11STrond Myklebust } 2914cf8d2c11STrond Myklebust return new_ns; 2915cf8d2c11STrond Myklebust } 2916cf8d2c11STrond Myklebust 2917ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2918ea441d11SAl Viro { 2919ea441d11SAl Viro struct mnt_namespace *ns; 2920d31da0f0SAl Viro struct super_block *s; 2921ea441d11SAl Viro struct path path; 2922ea441d11SAl Viro int err; 2923ea441d11SAl Viro 2924ea441d11SAl Viro ns = create_mnt_ns(mnt); 2925ea441d11SAl Viro if (IS_ERR(ns)) 2926ea441d11SAl Viro return ERR_CAST(ns); 2927ea441d11SAl Viro 2928ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2929ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2930ea441d11SAl Viro 2931ea441d11SAl Viro put_mnt_ns(ns); 2932ea441d11SAl Viro 2933ea441d11SAl Viro if (err) 2934ea441d11SAl Viro return ERR_PTR(err); 2935ea441d11SAl Viro 2936ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2937d31da0f0SAl Viro s = path.mnt->mnt_sb; 2938d31da0f0SAl Viro atomic_inc(&s->s_active); 2939ea441d11SAl Viro mntput(path.mnt); 2940ea441d11SAl Viro /* lock the sucker */ 2941d31da0f0SAl Viro down_write(&s->s_umount); 2942ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2943ea441d11SAl Viro return path.dentry; 2944ea441d11SAl Viro } 2945ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2946ea441d11SAl Viro 2947bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2948bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 29491da177e4SLinus Torvalds { 2950eca6f534SVegard Nossum int ret; 2951eca6f534SVegard Nossum char *kernel_type; 2952eca6f534SVegard Nossum char *kernel_dev; 2953b40ef869SAl Viro void *options; 29541da177e4SLinus Torvalds 2955b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 2956b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 2957b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 2958eca6f534SVegard Nossum goto out_type; 29591da177e4SLinus Torvalds 2960b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 2961b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 2962b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 2963eca6f534SVegard Nossum goto out_dev; 29641da177e4SLinus Torvalds 2965b40ef869SAl Viro options = copy_mount_options(data); 2966b40ef869SAl Viro ret = PTR_ERR(options); 2967b40ef869SAl Viro if (IS_ERR(options)) 2968eca6f534SVegard Nossum goto out_data; 29691da177e4SLinus Torvalds 2970b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 2971eca6f534SVegard Nossum 2972b40ef869SAl Viro kfree(options); 2973eca6f534SVegard Nossum out_data: 2974eca6f534SVegard Nossum kfree(kernel_dev); 2975eca6f534SVegard Nossum out_dev: 2976eca6f534SVegard Nossum kfree(kernel_type); 2977eca6f534SVegard Nossum out_type: 2978eca6f534SVegard Nossum return ret; 29791da177e4SLinus Torvalds } 29801da177e4SLinus Torvalds 29811da177e4SLinus Torvalds /* 2982afac7cbaSAl Viro * Return true if path is reachable from root 2983afac7cbaSAl Viro * 298448a066e7SAl Viro * namespace_sem or mount_lock is held 2985afac7cbaSAl Viro */ 2986643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2987afac7cbaSAl Viro const struct path *root) 2988afac7cbaSAl Viro { 2989643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2990a73324daSAl Viro dentry = mnt->mnt_mountpoint; 29910714a533SAl Viro mnt = mnt->mnt_parent; 2992afac7cbaSAl Viro } 2993643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2994afac7cbaSAl Viro } 2995afac7cbaSAl Viro 299625ab4c9bSYaowei Bai bool path_is_under(struct path *path1, struct path *path2) 2997afac7cbaSAl Viro { 299825ab4c9bSYaowei Bai bool res; 299948a066e7SAl Viro read_seqlock_excl(&mount_lock); 3000643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 300148a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3002afac7cbaSAl Viro return res; 3003afac7cbaSAl Viro } 3004afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3005afac7cbaSAl Viro 3006afac7cbaSAl Viro /* 30071da177e4SLinus Torvalds * pivot_root Semantics: 30081da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 30091da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 30101da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 30111da177e4SLinus Torvalds * 30121da177e4SLinus Torvalds * Restrictions: 30131da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 30141da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 30151da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 30161da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 30171da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 30181da177e4SLinus Torvalds * 30194a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 30204a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 30214a0d11faSNeil Brown * in this situation. 30224a0d11faSNeil Brown * 30231da177e4SLinus Torvalds * Notes: 30241da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 30251da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 30261da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 30271da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 30281da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 30291da177e4SLinus Torvalds * first. 30301da177e4SLinus Torvalds */ 30313480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 30323480b257SHeiko Carstens const char __user *, put_old) 30331da177e4SLinus Torvalds { 30342d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 303584d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 303684d17192SAl Viro struct mountpoint *old_mp, *root_mp; 30371da177e4SLinus Torvalds int error; 30381da177e4SLinus Torvalds 30399b40bc90SAl Viro if (!may_mount()) 30401da177e4SLinus Torvalds return -EPERM; 30411da177e4SLinus Torvalds 30422d8f3038SAl Viro error = user_path_dir(new_root, &new); 30431da177e4SLinus Torvalds if (error) 30441da177e4SLinus Torvalds goto out0; 30451da177e4SLinus Torvalds 30462d8f3038SAl Viro error = user_path_dir(put_old, &old); 30471da177e4SLinus Torvalds if (error) 30481da177e4SLinus Torvalds goto out1; 30491da177e4SLinus Torvalds 30502d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3051b12cea91SAl Viro if (error) 3052b12cea91SAl Viro goto out2; 30531da177e4SLinus Torvalds 3054f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 305584d17192SAl Viro old_mp = lock_mount(&old); 305684d17192SAl Viro error = PTR_ERR(old_mp); 305784d17192SAl Viro if (IS_ERR(old_mp)) 3058b12cea91SAl Viro goto out3; 3059b12cea91SAl Viro 30601da177e4SLinus Torvalds error = -EINVAL; 3061419148daSAl Viro new_mnt = real_mount(new.mnt); 3062419148daSAl Viro root_mnt = real_mount(root.mnt); 306384d17192SAl Viro old_mnt = real_mount(old.mnt); 306484d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3065fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3066fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3067b12cea91SAl Viro goto out4; 3068143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3069b12cea91SAl Viro goto out4; 30705ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 30715ff9d8a6SEric W. Biederman goto out4; 30721da177e4SLinus Torvalds error = -ENOENT; 3073f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3074b12cea91SAl Viro goto out4; 30751da177e4SLinus Torvalds error = -EBUSY; 307684d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3077b12cea91SAl Viro goto out4; /* loop, on the same file system */ 30781da177e4SLinus Torvalds error = -EINVAL; 30798c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3080b12cea91SAl Viro goto out4; /* not a mountpoint */ 3081676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3082b12cea91SAl Viro goto out4; /* not attached */ 308384d17192SAl Viro root_mp = root_mnt->mnt_mp; 30842d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3085b12cea91SAl Viro goto out4; /* not a mountpoint */ 3086676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3087b12cea91SAl Viro goto out4; /* not attached */ 30884ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 308984d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3090b12cea91SAl Viro goto out4; 30910d082601SEric W. Biederman /* make certain new is below the root */ 30920d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 30930d082601SEric W. Biederman goto out4; 309484d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3095719ea2fbSAl Viro lock_mount_hash(); 3096419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3097419148daSAl Viro detach_mnt(root_mnt, &root_parent); 30985ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 30995ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 31005ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 31015ff9d8a6SEric W. Biederman } 31024ac91378SJan Blunck /* mount old root on put_old */ 310384d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 31044ac91378SJan Blunck /* mount new_root on / */ 310584d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 31066b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 31074fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 31084fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 3109719ea2fbSAl Viro unlock_mount_hash(); 31102d8f3038SAl Viro chroot_fs_refs(&root, &new); 311184d17192SAl Viro put_mountpoint(root_mp); 31121da177e4SLinus Torvalds error = 0; 3113b12cea91SAl Viro out4: 311484d17192SAl Viro unlock_mount(old_mp); 3115b12cea91SAl Viro if (!error) { 31161a390689SAl Viro path_put(&root_parent); 31171a390689SAl Viro path_put(&parent_path); 3118b12cea91SAl Viro } 3119b12cea91SAl Viro out3: 31208c3ee42eSAl Viro path_put(&root); 3121b12cea91SAl Viro out2: 31222d8f3038SAl Viro path_put(&old); 31231da177e4SLinus Torvalds out1: 31242d8f3038SAl Viro path_put(&new); 31251da177e4SLinus Torvalds out0: 31261da177e4SLinus Torvalds return error; 31271da177e4SLinus Torvalds } 31281da177e4SLinus Torvalds 31291da177e4SLinus Torvalds static void __init init_mount_tree(void) 31301da177e4SLinus Torvalds { 31311da177e4SLinus Torvalds struct vfsmount *mnt; 31326b3286edSKirill Korotaev struct mnt_namespace *ns; 3133ac748a09SJan Blunck struct path root; 31340c55cfc4SEric W. Biederman struct file_system_type *type; 31351da177e4SLinus Torvalds 31360c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 31370c55cfc4SEric W. Biederman if (!type) 31380c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 31390c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 31400c55cfc4SEric W. Biederman put_filesystem(type); 31411da177e4SLinus Torvalds if (IS_ERR(mnt)) 31421da177e4SLinus Torvalds panic("Can't create rootfs"); 3143b3e19d92SNick Piggin 31443b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 31453b22edc5STrond Myklebust if (IS_ERR(ns)) 31461da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 31471da177e4SLinus Torvalds 31486b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 31496b3286edSKirill Korotaev get_mnt_ns(ns); 31501da177e4SLinus Torvalds 3151be08d6d2SAl Viro root.mnt = mnt; 3152be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3153da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3154ac748a09SJan Blunck 3155ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3156ac748a09SJan Blunck set_fs_root(current->fs, &root); 31571da177e4SLinus Torvalds } 31581da177e4SLinus Torvalds 315974bf17cfSDenis Cheng void __init mnt_init(void) 31601da177e4SLinus Torvalds { 316113f14b4dSEric Dumazet unsigned u; 316215a67dd8SRandy Dunlap int err; 31631da177e4SLinus Torvalds 31647d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 316520c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 31661da177e4SLinus Torvalds 31670818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 316838129a13SAl Viro sizeof(struct hlist_head), 31690818bf27SAl Viro mhash_entries, 19, 31700818bf27SAl Viro 0, 31710818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 31720818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 31730818bf27SAl Viro sizeof(struct hlist_head), 31740818bf27SAl Viro mphash_entries, 19, 31750818bf27SAl Viro 0, 31760818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 31771da177e4SLinus Torvalds 317884d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 31791da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 31801da177e4SLinus Torvalds 31810818bf27SAl Viro for (u = 0; u <= m_hash_mask; u++) 318238129a13SAl Viro INIT_HLIST_HEAD(&mount_hashtable[u]); 31830818bf27SAl Viro for (u = 0; u <= mp_hash_mask; u++) 31840818bf27SAl Viro INIT_HLIST_HEAD(&mountpoint_hashtable[u]); 31851da177e4SLinus Torvalds 31864b93dc9bSTejun Heo kernfs_init(); 31874b93dc9bSTejun Heo 318815a67dd8SRandy Dunlap err = sysfs_init(); 318915a67dd8SRandy Dunlap if (err) 319015a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 31918e24eea7SHarvey Harrison __func__, err); 319200d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 319300d26666SGreg Kroah-Hartman if (!fs_kobj) 31948e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 31951da177e4SLinus Torvalds init_rootfs(); 31961da177e4SLinus Torvalds init_mount_tree(); 31971da177e4SLinus Torvalds } 31981da177e4SLinus Torvalds 3199616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 32001da177e4SLinus Torvalds { 3201d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3202616511d0STrond Myklebust return; 32037b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3204771b1371SEric W. Biederman free_mnt_ns(ns); 32051da177e4SLinus Torvalds } 32069d412a43SAl Viro 32079d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 32089d412a43SAl Viro { 3209423e0ab0STim Chen struct vfsmount *mnt; 3210423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 3211423e0ab0STim Chen if (!IS_ERR(mnt)) { 3212423e0ab0STim Chen /* 3213423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3214423e0ab0STim Chen * we unmount before file sys is unregistered 3215423e0ab0STim Chen */ 3216f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3217423e0ab0STim Chen } 3218423e0ab0STim Chen return mnt; 32199d412a43SAl Viro } 32209d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3221423e0ab0STim Chen 3222423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3223423e0ab0STim Chen { 3224423e0ab0STim Chen /* release long term mount so mount point can be released */ 3225423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3226f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 322748a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3228423e0ab0STim Chen mntput(mnt); 3229423e0ab0STim Chen } 3230423e0ab0STim Chen } 3231423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 323202125a82SAl Viro 323302125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 323402125a82SAl Viro { 3235143c8c91SAl Viro return check_mnt(real_mount(mnt)); 323602125a82SAl Viro } 32378823c079SEric W. Biederman 32383151527eSEric W. Biederman bool current_chrooted(void) 32393151527eSEric W. Biederman { 32403151527eSEric W. Biederman /* Does the current process have a non-standard root */ 32413151527eSEric W. Biederman struct path ns_root; 32423151527eSEric W. Biederman struct path fs_root; 32433151527eSEric W. Biederman bool chrooted; 32443151527eSEric W. Biederman 32453151527eSEric W. Biederman /* Find the namespace root */ 32463151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 32473151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 32483151527eSEric W. Biederman path_get(&ns_root); 32493151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 32503151527eSEric W. Biederman ; 32513151527eSEric W. Biederman 32523151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 32533151527eSEric W. Biederman 32543151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 32553151527eSEric W. Biederman 32563151527eSEric W. Biederman path_put(&fs_root); 32573151527eSEric W. Biederman path_put(&ns_root); 32583151527eSEric W. Biederman 32593151527eSEric W. Biederman return chrooted; 32603151527eSEric W. Biederman } 32613151527eSEric W. Biederman 32628654df4eSEric W. Biederman static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, 32638654df4eSEric W. Biederman int *new_mnt_flags) 326487a8ebd6SEric W. Biederman { 32658c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 326687a8ebd6SEric W. Biederman struct mount *mnt; 3267e51db735SEric W. Biederman bool visible = false; 326887a8ebd6SEric W. Biederman 326944bb4385SAl Viro down_read(&namespace_sem); 327087a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3271e51db735SEric W. Biederman struct mount *child; 327277b1a97dSEric W. Biederman int mnt_flags; 327377b1a97dSEric W. Biederman 32748654df4eSEric W. Biederman if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) 3275e51db735SEric W. Biederman continue; 3276e51db735SEric W. Biederman 32777e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 32787e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 32797e96c1b0SEric W. Biederman */ 32807e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 32817e96c1b0SEric W. Biederman continue; 32827e96c1b0SEric W. Biederman 3283a1935c17SEric W. Biederman /* A local view of the mount flags */ 328477b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 328577b1a97dSEric W. Biederman 3286695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3287695e9df0SEric W. Biederman if (mnt->mnt.mnt_sb->s_flags & MS_RDONLY) 3288695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3289695e9df0SEric W. Biederman 32908c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 32918c6cf9ccSEric W. Biederman * than the proposed new mount. 32928c6cf9ccSEric W. Biederman */ 329377b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 32948c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 32958c6cf9ccSEric W. Biederman continue; 329677b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 329777b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 32988c6cf9ccSEric W. Biederman continue; 32998c6cf9ccSEric W. Biederman 3300ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3301ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3302ceeb0e5dSEric W. Biederman * empty directories. 3303e51db735SEric W. Biederman */ 3304e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3305e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3306ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3307d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3308ceeb0e5dSEric W. Biederman continue; 33097236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 33107236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3311e51db735SEric W. Biederman goto next; 331287a8ebd6SEric W. Biederman } 33138c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 331477b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 33158c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3316e51db735SEric W. Biederman visible = true; 3317e51db735SEric W. Biederman goto found; 3318e51db735SEric W. Biederman next: ; 331987a8ebd6SEric W. Biederman } 3320e51db735SEric W. Biederman found: 332144bb4385SAl Viro up_read(&namespace_sem); 3322e51db735SEric W. Biederman return visible; 332387a8ebd6SEric W. Biederman } 332487a8ebd6SEric W. Biederman 33258654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) 33268654df4eSEric W. Biederman { 3327a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 33288654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 33298654df4eSEric W. Biederman unsigned long s_iflags; 33308654df4eSEric W. Biederman 33318654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 33328654df4eSEric W. Biederman return false; 33338654df4eSEric W. Biederman 33348654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 33358654df4eSEric W. Biederman s_iflags = mnt->mnt_sb->s_iflags; 33368654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 33378654df4eSEric W. Biederman return false; 33388654df4eSEric W. Biederman 3339a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3340a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3341a1935c17SEric W. Biederman required_iflags); 3342a1935c17SEric W. Biederman return true; 3343a1935c17SEric W. Biederman } 3344a1935c17SEric W. Biederman 33458654df4eSEric W. Biederman return !mnt_already_visible(ns, mnt, new_mnt_flags); 33468654df4eSEric W. Biederman } 33478654df4eSEric W. Biederman 3348380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3349380cf5baSAndy Lutomirski { 3350380cf5baSAndy Lutomirski /* 3351380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3352380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3353380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3354380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3355380cf5baSAndy Lutomirski * in other namespaces. 3356380cf5baSAndy Lutomirski */ 3357380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3358380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3359380cf5baSAndy Lutomirski } 3360380cf5baSAndy Lutomirski 336164964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 33628823c079SEric W. Biederman { 336358be2825SAl Viro struct ns_common *ns = NULL; 33648823c079SEric W. Biederman struct nsproxy *nsproxy; 33658823c079SEric W. Biederman 3366728dba3aSEric W. Biederman task_lock(task); 3367728dba3aSEric W. Biederman nsproxy = task->nsproxy; 33688823c079SEric W. Biederman if (nsproxy) { 336958be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 337058be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 33718823c079SEric W. Biederman } 3372728dba3aSEric W. Biederman task_unlock(task); 33738823c079SEric W. Biederman 33748823c079SEric W. Biederman return ns; 33758823c079SEric W. Biederman } 33768823c079SEric W. Biederman 337764964528SAl Viro static void mntns_put(struct ns_common *ns) 33788823c079SEric W. Biederman { 337958be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 33808823c079SEric W. Biederman } 33818823c079SEric W. Biederman 338264964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 33838823c079SEric W. Biederman { 33848823c079SEric W. Biederman struct fs_struct *fs = current->fs; 338558be2825SAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns); 33868823c079SEric W. Biederman struct path root; 33878823c079SEric W. Biederman 33880c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3389c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3390c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3391ae11e0f1SZhao Hongjiang return -EPERM; 33928823c079SEric W. Biederman 33938823c079SEric W. Biederman if (fs->users != 1) 33948823c079SEric W. Biederman return -EINVAL; 33958823c079SEric W. Biederman 33968823c079SEric W. Biederman get_mnt_ns(mnt_ns); 33978823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 33988823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 33998823c079SEric W. Biederman 34008823c079SEric W. Biederman /* Find the root */ 34018823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 34028823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 34038823c079SEric W. Biederman path_get(&root); 34048823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 34058823c079SEric W. Biederman ; 34068823c079SEric W. Biederman 34078823c079SEric W. Biederman /* Update the pwd and root */ 34088823c079SEric W. Biederman set_fs_pwd(fs, &root); 34098823c079SEric W. Biederman set_fs_root(fs, &root); 34108823c079SEric W. Biederman 34118823c079SEric W. Biederman path_put(&root); 34128823c079SEric W. Biederman return 0; 34138823c079SEric W. Biederman } 34148823c079SEric W. Biederman 3415bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3416bcac25a5SAndrey Vagin { 3417bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3418bcac25a5SAndrey Vagin } 3419bcac25a5SAndrey Vagin 34208823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 34218823c079SEric W. Biederman .name = "mnt", 34228823c079SEric W. Biederman .type = CLONE_NEWNS, 34238823c079SEric W. Biederman .get = mntns_get, 34248823c079SEric W. Biederman .put = mntns_put, 34258823c079SEric W. Biederman .install = mntns_install, 3426bcac25a5SAndrey Vagin .owner = mntns_owner, 34278823c079SEric W. Biederman }; 3428