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> 2607b20889SRam Pai #include "pnode.h" 27948730b0SAdrian Bunk #include "internal.h" 281da177e4SLinus Torvalds 290818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 300818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 310818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 320818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 330818bf27SAl Viro 340818bf27SAl Viro static __initdata unsigned long mhash_entries; 350818bf27SAl Viro static int __init set_mhash_entries(char *str) 360818bf27SAl Viro { 370818bf27SAl Viro if (!str) 380818bf27SAl Viro return 0; 390818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 400818bf27SAl Viro return 1; 410818bf27SAl Viro } 420818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 430818bf27SAl Viro 440818bf27SAl Viro static __initdata unsigned long mphash_entries; 450818bf27SAl Viro static int __init set_mphash_entries(char *str) 460818bf27SAl Viro { 470818bf27SAl Viro if (!str) 480818bf27SAl Viro return 0; 490818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 500818bf27SAl Viro return 1; 510818bf27SAl Viro } 520818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 5313f14b4dSEric Dumazet 54c7999c36SAl Viro static u64 event; 5573cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 56719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 5799b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 58f21f6220SAl Viro static int mnt_id_start = 0; 59f21f6220SAl Viro static int mnt_group_start = 1; 605addc5ddSAl Viro 6138129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 620818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 63e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 6459aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 651da177e4SLinus Torvalds 66f87fd4c2SMiklos Szeredi /* /sys/fs */ 6700d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 6800d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 69f87fd4c2SMiklos Szeredi 7099b7db7bSNick Piggin /* 7199b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7299b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 7399b7db7bSNick Piggin * up the tree. 7499b7db7bSNick Piggin * 7599b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 7699b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 7799b7db7bSNick Piggin */ 7848a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 7999b7db7bSNick Piggin 8038129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 811da177e4SLinus Torvalds { 821da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 831da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 840818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 850818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 860818bf27SAl Viro } 870818bf27SAl Viro 880818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 890818bf27SAl Viro { 900818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 910818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 920818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 931da177e4SLinus Torvalds } 941da177e4SLinus Torvalds 9599b7db7bSNick Piggin /* 9699b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 9799b7db7bSNick Piggin * serialize with freeing. 9899b7db7bSNick Piggin */ 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 192b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1931da177e4SLinus Torvalds { 194c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 195c63181e6SAl Viro if (mnt) { 19673cd49ecSMiklos Szeredi int err; 19773cd49ecSMiklos Szeredi 198c63181e6SAl Viro err = mnt_alloc_id(mnt); 19988b38782SLi Zefan if (err) 20088b38782SLi Zefan goto out_free_cache; 20188b38782SLi Zefan 20288b38782SLi Zefan if (name) { 203c63181e6SAl Viro mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 204c63181e6SAl Viro if (!mnt->mnt_devname) 20588b38782SLi Zefan goto out_free_id; 20673cd49ecSMiklos Szeredi } 20773cd49ecSMiklos Szeredi 208b3e19d92SNick Piggin #ifdef CONFIG_SMP 209c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 210c63181e6SAl Viro if (!mnt->mnt_pcp) 211b3e19d92SNick Piggin goto out_free_devname; 212b3e19d92SNick Piggin 213c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 214b3e19d92SNick Piggin #else 215c63181e6SAl Viro mnt->mnt_count = 1; 216c63181e6SAl Viro mnt->mnt_writers = 0; 217b3e19d92SNick Piggin #endif 218b3e19d92SNick Piggin 21938129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 220c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 221c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 222c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 223c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 224c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 225c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 226c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2272504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 2282504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 2292504c5d6SAndreas Gruenbacher #endif 2301da177e4SLinus Torvalds } 231c63181e6SAl Viro return mnt; 23288b38782SLi Zefan 233d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 234d3ef3d73Snpiggin@suse.de out_free_devname: 235c63181e6SAl Viro kfree(mnt->mnt_devname); 236d3ef3d73Snpiggin@suse.de #endif 23788b38782SLi Zefan out_free_id: 238c63181e6SAl Viro mnt_free_id(mnt); 23988b38782SLi Zefan out_free_cache: 240c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 24188b38782SLi Zefan return NULL; 2421da177e4SLinus Torvalds } 2431da177e4SLinus Torvalds 2448366025eSDave Hansen /* 2458366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2468366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2478366025eSDave Hansen * We must keep track of when those operations start 2488366025eSDave Hansen * (for permission checks) and when they end, so that 2498366025eSDave Hansen * we can determine when writes are able to occur to 2508366025eSDave Hansen * a filesystem. 2518366025eSDave Hansen */ 2523d733633SDave Hansen /* 2533d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2543d733633SDave Hansen * @mnt: the mount to check for its write status 2553d733633SDave Hansen * 2563d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2573d733633SDave Hansen * It does not guarantee that the filesystem will stay 2583d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2593d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2603d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2613d733633SDave Hansen * r/w. 2623d733633SDave Hansen */ 2633d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2643d733633SDave Hansen { 2652e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2662e4b7fcdSDave Hansen return 1; 2672e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2682e4b7fcdSDave Hansen return 1; 2692e4b7fcdSDave Hansen return 0; 2703d733633SDave Hansen } 2713d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2723d733633SDave Hansen 27383adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2743d733633SDave Hansen { 275d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 27668e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 277d3ef3d73Snpiggin@suse.de #else 27868e8a9feSAl Viro mnt->mnt_writers++; 279d3ef3d73Snpiggin@suse.de #endif 2803d733633SDave Hansen } 2813d733633SDave Hansen 28283adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2833d733633SDave Hansen { 284d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 28568e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 286d3ef3d73Snpiggin@suse.de #else 28768e8a9feSAl Viro mnt->mnt_writers--; 288d3ef3d73Snpiggin@suse.de #endif 289d3ef3d73Snpiggin@suse.de } 290d3ef3d73Snpiggin@suse.de 29183adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 292d3ef3d73Snpiggin@suse.de { 293d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 294d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2953d733633SDave Hansen int cpu; 2963d733633SDave Hansen 2973d733633SDave Hansen for_each_possible_cpu(cpu) { 29868e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2993d733633SDave Hansen } 3003d733633SDave Hansen 301d3ef3d73Snpiggin@suse.de return count; 302d3ef3d73Snpiggin@suse.de #else 303d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 304d3ef3d73Snpiggin@suse.de #endif 3053d733633SDave Hansen } 3063d733633SDave Hansen 3074ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 3084ed5e82fSMiklos Szeredi { 3094ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 3104ed5e82fSMiklos Szeredi return 1; 3114ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 3124ed5e82fSMiklos Szeredi smp_rmb(); 3134ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 3144ed5e82fSMiklos Szeredi } 3154ed5e82fSMiklos Szeredi 3163d733633SDave Hansen /* 317eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 318eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 319eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 320eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3213d733633SDave Hansen */ 3228366025eSDave Hansen /** 323eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 32483adc753SAl Viro * @m: the mount on which to take a write 3258366025eSDave Hansen * 326eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 327eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 328eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 329eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 330eb04c282SJan Kara * called. This is effectively a refcount. 3318366025eSDave Hansen */ 332eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3338366025eSDave Hansen { 33483adc753SAl Viro struct mount *mnt = real_mount(m); 3353d733633SDave Hansen int ret = 0; 3363d733633SDave Hansen 337d3ef3d73Snpiggin@suse.de preempt_disable(); 338c6653a83SNick Piggin mnt_inc_writers(mnt); 339d3ef3d73Snpiggin@suse.de /* 340c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 341d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 342d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 343d3ef3d73Snpiggin@suse.de */ 344d3ef3d73Snpiggin@suse.de smp_mb(); 3451e75529eSMiao Xie while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 346d3ef3d73Snpiggin@suse.de cpu_relax(); 347d3ef3d73Snpiggin@suse.de /* 348d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 349d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 350d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 351d3ef3d73Snpiggin@suse.de */ 352d3ef3d73Snpiggin@suse.de smp_rmb(); 3534ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 354c6653a83SNick Piggin mnt_dec_writers(mnt); 3553d733633SDave Hansen ret = -EROFS; 3563d733633SDave Hansen } 357d3ef3d73Snpiggin@suse.de preempt_enable(); 358eb04c282SJan Kara 359eb04c282SJan Kara return ret; 360eb04c282SJan Kara } 361eb04c282SJan Kara 362eb04c282SJan Kara /** 363eb04c282SJan Kara * mnt_want_write - get write access to a mount 364eb04c282SJan Kara * @m: the mount on which to take a write 365eb04c282SJan Kara * 366eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 367eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 368eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 369eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 370eb04c282SJan Kara */ 371eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 372eb04c282SJan Kara { 373eb04c282SJan Kara int ret; 374eb04c282SJan Kara 375eb04c282SJan Kara sb_start_write(m->mnt_sb); 376eb04c282SJan Kara ret = __mnt_want_write(m); 377eb04c282SJan Kara if (ret) 378eb04c282SJan Kara sb_end_write(m->mnt_sb); 3793d733633SDave Hansen return ret; 3808366025eSDave Hansen } 3818366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3828366025eSDave Hansen 3838366025eSDave Hansen /** 38496029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 38596029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 38696029c4eSnpiggin@suse.de * 38796029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 38896029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 38996029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 39096029c4eSnpiggin@suse.de * on it. This allows some optimisation. 39196029c4eSnpiggin@suse.de * 39296029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 39396029c4eSnpiggin@suse.de * drop the reference. 39496029c4eSnpiggin@suse.de */ 39596029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 39696029c4eSnpiggin@suse.de { 39796029c4eSnpiggin@suse.de /* superblock may be r/o */ 39896029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 39996029c4eSnpiggin@suse.de return -EROFS; 40096029c4eSnpiggin@suse.de preempt_disable(); 40183adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 40296029c4eSnpiggin@suse.de preempt_enable(); 40396029c4eSnpiggin@suse.de return 0; 40496029c4eSnpiggin@suse.de } 40596029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 40696029c4eSnpiggin@suse.de 40796029c4eSnpiggin@suse.de /** 408eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 409eb04c282SJan Kara * @file: the file who's mount on which to take a write 410eb04c282SJan Kara * 411eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 412eb04c282SJan Kara * do some optimisations if the file is open for write already 413eb04c282SJan Kara */ 414eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 415eb04c282SJan Kara { 41683f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 417eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 418eb04c282SJan Kara else 419eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 420eb04c282SJan Kara } 421eb04c282SJan Kara 422eb04c282SJan Kara /** 42396029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 42496029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 42596029c4eSnpiggin@suse.de * 42696029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 42796029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 42896029c4eSnpiggin@suse.de */ 42996029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 43096029c4eSnpiggin@suse.de { 431eb04c282SJan Kara int ret; 432eb04c282SJan Kara 433eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 434eb04c282SJan Kara ret = __mnt_want_write_file(file); 435eb04c282SJan Kara if (ret) 436eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 437eb04c282SJan Kara return ret; 43896029c4eSnpiggin@suse.de } 43996029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 44096029c4eSnpiggin@suse.de 44196029c4eSnpiggin@suse.de /** 442eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4438366025eSDave Hansen * @mnt: the mount on which to give up write access 4448366025eSDave Hansen * 4458366025eSDave Hansen * Tells the low-level filesystem that we are done 4468366025eSDave Hansen * performing writes to it. Must be matched with 447eb04c282SJan Kara * __mnt_want_write() call above. 4488366025eSDave Hansen */ 449eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4508366025eSDave Hansen { 451d3ef3d73Snpiggin@suse.de preempt_disable(); 45283adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 453d3ef3d73Snpiggin@suse.de preempt_enable(); 4548366025eSDave Hansen } 455eb04c282SJan Kara 456eb04c282SJan Kara /** 457eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 458eb04c282SJan Kara * @mnt: the mount on which to give up write access 459eb04c282SJan Kara * 460eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 461eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 462eb04c282SJan Kara * mnt_want_write() call above. 463eb04c282SJan Kara */ 464eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 465eb04c282SJan Kara { 466eb04c282SJan Kara __mnt_drop_write(mnt); 467eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 468eb04c282SJan Kara } 4698366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4708366025eSDave Hansen 471eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 472eb04c282SJan Kara { 473eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 474eb04c282SJan Kara } 475eb04c282SJan Kara 4762a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4772a79f17eSAl Viro { 4782a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4792a79f17eSAl Viro } 4802a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4812a79f17eSAl Viro 48283adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4838366025eSDave Hansen { 4843d733633SDave Hansen int ret = 0; 4853d733633SDave Hansen 486719ea2fbSAl Viro lock_mount_hash(); 48783adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 488d3ef3d73Snpiggin@suse.de /* 489d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 490d3ef3d73Snpiggin@suse.de * should be visible before we do. 491d3ef3d73Snpiggin@suse.de */ 492d3ef3d73Snpiggin@suse.de smp_mb(); 493d3ef3d73Snpiggin@suse.de 494d3ef3d73Snpiggin@suse.de /* 495d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 496d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 497d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 498d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 499d3ef3d73Snpiggin@suse.de * 500d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 501d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 502d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 503d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 504d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 505d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 506d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 507d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 508d3ef3d73Snpiggin@suse.de * we're counting up here. 509d3ef3d73Snpiggin@suse.de */ 510c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 511d3ef3d73Snpiggin@suse.de ret = -EBUSY; 512d3ef3d73Snpiggin@suse.de else 51383adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 514d3ef3d73Snpiggin@suse.de /* 515d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 516d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 517d3ef3d73Snpiggin@suse.de */ 518d3ef3d73Snpiggin@suse.de smp_wmb(); 51983adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 520719ea2fbSAl Viro unlock_mount_hash(); 5213d733633SDave Hansen return ret; 5223d733633SDave Hansen } 5238366025eSDave Hansen 52483adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5252e4b7fcdSDave Hansen { 526719ea2fbSAl Viro lock_mount_hash(); 52783adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 528719ea2fbSAl Viro unlock_mount_hash(); 5292e4b7fcdSDave Hansen } 5302e4b7fcdSDave Hansen 5314ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5324ed5e82fSMiklos Szeredi { 5334ed5e82fSMiklos Szeredi struct mount *mnt; 5344ed5e82fSMiklos Szeredi int err = 0; 5354ed5e82fSMiklos Szeredi 5368e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5378e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5388e8b8796SMiklos Szeredi return -EBUSY; 5398e8b8796SMiklos Szeredi 540719ea2fbSAl Viro lock_mount_hash(); 5414ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5424ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5434ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5444ed5e82fSMiklos Szeredi smp_mb(); 5454ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5464ed5e82fSMiklos Szeredi err = -EBUSY; 5474ed5e82fSMiklos Szeredi break; 5484ed5e82fSMiklos Szeredi } 5494ed5e82fSMiklos Szeredi } 5504ed5e82fSMiklos Szeredi } 5518e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5528e8b8796SMiklos Szeredi err = -EBUSY; 5538e8b8796SMiklos Szeredi 5544ed5e82fSMiklos Szeredi if (!err) { 5554ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5564ed5e82fSMiklos Szeredi smp_wmb(); 5574ed5e82fSMiklos Szeredi } 5584ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5594ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5604ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5614ed5e82fSMiklos Szeredi } 562719ea2fbSAl Viro unlock_mount_hash(); 5634ed5e82fSMiklos Szeredi 5644ed5e82fSMiklos Szeredi return err; 5654ed5e82fSMiklos Szeredi } 5664ed5e82fSMiklos Szeredi 567b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5681da177e4SLinus Torvalds { 56952ba1621SAl Viro kfree(mnt->mnt_devname); 570d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 57168e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 572d3ef3d73Snpiggin@suse.de #endif 573b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds 5768ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5778ffcb32eSDavid Howells { 5788ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5798ffcb32eSDavid Howells } 5808ffcb32eSDavid Howells 58148a066e7SAl Viro /* call under rcu_read_lock */ 58248a066e7SAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 58348a066e7SAl Viro { 58448a066e7SAl Viro struct mount *mnt; 58548a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 58648a066e7SAl Viro return false; 58748a066e7SAl Viro if (bastard == NULL) 58848a066e7SAl Viro return true; 58948a066e7SAl Viro mnt = real_mount(bastard); 59048a066e7SAl Viro mnt_add_count(mnt, 1); 59148a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 59248a066e7SAl Viro return true; 59348a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 59448a066e7SAl Viro mnt_add_count(mnt, -1); 59548a066e7SAl Viro return false; 59648a066e7SAl Viro } 59748a066e7SAl Viro rcu_read_unlock(); 59848a066e7SAl Viro mntput(bastard); 59948a066e7SAl Viro rcu_read_lock(); 60048a066e7SAl Viro return false; 60148a066e7SAl Viro } 60248a066e7SAl Viro 6031da177e4SLinus Torvalds /* 604474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 60548a066e7SAl Viro * call under rcu_read_lock() 6061da177e4SLinus Torvalds */ 607474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6081da177e4SLinus Torvalds { 60938129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 610474279dcSAl Viro struct mount *p; 6111da177e4SLinus Torvalds 61238129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 613474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 614474279dcSAl Viro return p; 615474279dcSAl Viro return NULL; 6161da177e4SLinus Torvalds } 617474279dcSAl Viro 618474279dcSAl Viro /* 619474279dcSAl Viro * find the last mount at @dentry on vfsmount @mnt. 62048a066e7SAl Viro * mount_lock must be held. 621474279dcSAl Viro */ 622474279dcSAl Viro struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) 623474279dcSAl Viro { 62438129a13SAl Viro struct mount *p, *res; 62538129a13SAl Viro res = p = __lookup_mnt(mnt, dentry); 62638129a13SAl Viro if (!p) 62738129a13SAl Viro goto out; 62838129a13SAl Viro hlist_for_each_entry_continue(p, mnt_hash) { 6291d6a32acSAl Viro if (&p->mnt_parent->mnt != mnt || p->mnt_mountpoint != dentry) 6301d6a32acSAl Viro break; 6311d6a32acSAl Viro res = p; 6321d6a32acSAl Viro } 63338129a13SAl Viro out: 6341d6a32acSAl Viro return res; 6351da177e4SLinus Torvalds } 6361da177e4SLinus Torvalds 637a05964f3SRam Pai /* 638f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 639f015f126SDavid Howells * 640f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 641f015f126SDavid Howells * following mounts: 642f015f126SDavid Howells * 643f015f126SDavid Howells * mount /dev/sda1 /mnt 644f015f126SDavid Howells * mount /dev/sda2 /mnt 645f015f126SDavid Howells * mount /dev/sda3 /mnt 646f015f126SDavid Howells * 647f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 648f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 649f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 650f015f126SDavid Howells * 651f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 652a05964f3SRam Pai */ 6531c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 654a05964f3SRam Pai { 655c7105365SAl Viro struct mount *child_mnt; 65648a066e7SAl Viro struct vfsmount *m; 65748a066e7SAl Viro unsigned seq; 65899b7db7bSNick Piggin 65948a066e7SAl Viro rcu_read_lock(); 66048a066e7SAl Viro do { 66148a066e7SAl Viro seq = read_seqbegin(&mount_lock); 662474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 66348a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 66448a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 66548a066e7SAl Viro rcu_read_unlock(); 66648a066e7SAl Viro return m; 667a05964f3SRam Pai } 668a05964f3SRam Pai 66984d17192SAl Viro static struct mountpoint *new_mountpoint(struct dentry *dentry) 67084d17192SAl Viro { 6710818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 67284d17192SAl Viro struct mountpoint *mp; 673eed81007SMiklos Szeredi int ret; 67484d17192SAl Viro 6750818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 67684d17192SAl Viro if (mp->m_dentry == dentry) { 67784d17192SAl Viro /* might be worth a WARN_ON() */ 67884d17192SAl Viro if (d_unlinked(dentry)) 67984d17192SAl Viro return ERR_PTR(-ENOENT); 68084d17192SAl Viro mp->m_count++; 68184d17192SAl Viro return mp; 68284d17192SAl Viro } 68384d17192SAl Viro } 68484d17192SAl Viro 68584d17192SAl Viro mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 68684d17192SAl Viro if (!mp) 68784d17192SAl Viro return ERR_PTR(-ENOMEM); 68884d17192SAl Viro 689eed81007SMiklos Szeredi ret = d_set_mounted(dentry); 690eed81007SMiklos Szeredi if (ret) { 69184d17192SAl Viro kfree(mp); 692eed81007SMiklos Szeredi return ERR_PTR(ret); 69384d17192SAl Viro } 694eed81007SMiklos Szeredi 69584d17192SAl Viro mp->m_dentry = dentry; 69684d17192SAl Viro mp->m_count = 1; 6970818bf27SAl Viro hlist_add_head(&mp->m_hash, chain); 69884d17192SAl Viro return mp; 69984d17192SAl Viro } 70084d17192SAl Viro 70184d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 70284d17192SAl Viro { 70384d17192SAl Viro if (!--mp->m_count) { 70484d17192SAl Viro struct dentry *dentry = mp->m_dentry; 70584d17192SAl Viro spin_lock(&dentry->d_lock); 70684d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 70784d17192SAl Viro spin_unlock(&dentry->d_lock); 7080818bf27SAl Viro hlist_del(&mp->m_hash); 70984d17192SAl Viro kfree(mp); 71084d17192SAl Viro } 71184d17192SAl Viro } 71284d17192SAl Viro 713143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7141da177e4SLinus Torvalds { 7156b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 71899b7db7bSNick Piggin /* 71999b7db7bSNick Piggin * vfsmount lock must be held for write 72099b7db7bSNick Piggin */ 7216b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7225addc5ddSAl Viro { 7235addc5ddSAl Viro if (ns) { 7245addc5ddSAl Viro ns->event = ++event; 7255addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7265addc5ddSAl Viro } 7275addc5ddSAl Viro } 7285addc5ddSAl Viro 72999b7db7bSNick Piggin /* 73099b7db7bSNick Piggin * vfsmount lock must be held for write 73199b7db7bSNick Piggin */ 7326b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 7335addc5ddSAl Viro { 7345addc5ddSAl Viro if (ns && ns->event != event) { 7355addc5ddSAl Viro ns->event = event; 7365addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7375addc5ddSAl Viro } 7385addc5ddSAl Viro } 7395addc5ddSAl Viro 74099b7db7bSNick Piggin /* 74199b7db7bSNick Piggin * vfsmount lock must be held for write 74299b7db7bSNick Piggin */ 743419148daSAl Viro static void detach_mnt(struct mount *mnt, struct path *old_path) 7441da177e4SLinus Torvalds { 745a73324daSAl Viro old_path->dentry = mnt->mnt_mountpoint; 7460714a533SAl Viro old_path->mnt = &mnt->mnt_parent->mnt; 7470714a533SAl Viro mnt->mnt_parent = mnt; 748a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 7496b41d536SAl Viro list_del_init(&mnt->mnt_child); 75038129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 75184d17192SAl Viro put_mountpoint(mnt->mnt_mp); 75284d17192SAl Viro mnt->mnt_mp = NULL; 7531da177e4SLinus Torvalds } 7541da177e4SLinus Torvalds 75599b7db7bSNick Piggin /* 75699b7db7bSNick Piggin * vfsmount lock must be held for write 75799b7db7bSNick Piggin */ 75884d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 75984d17192SAl Viro struct mountpoint *mp, 76044d964d6SAl Viro struct mount *child_mnt) 761b90fa9aeSRam Pai { 76284d17192SAl Viro mp->m_count++; 7633a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 76484d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 7653a2393d7SAl Viro child_mnt->mnt_parent = mnt; 76684d17192SAl Viro child_mnt->mnt_mp = mp; 767b90fa9aeSRam Pai } 768b90fa9aeSRam Pai 76999b7db7bSNick Piggin /* 77099b7db7bSNick Piggin * vfsmount lock must be held for write 77199b7db7bSNick Piggin */ 77284d17192SAl Viro static void attach_mnt(struct mount *mnt, 77384d17192SAl Viro struct mount *parent, 77484d17192SAl Viro struct mountpoint *mp) 7751da177e4SLinus Torvalds { 77684d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 77738129a13SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, m_hash(&parent->mnt, mp->m_dentry)); 77884d17192SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 779b90fa9aeSRam Pai } 780b90fa9aeSRam Pai 78112a5b529SAl Viro static void attach_shadowed(struct mount *mnt, 78212a5b529SAl Viro struct mount *parent, 78312a5b529SAl Viro struct mount *shadows) 78412a5b529SAl Viro { 78512a5b529SAl Viro if (shadows) { 78612a5b529SAl Viro hlist_add_after_rcu(&shadows->mnt_hash, &mnt->mnt_hash); 78712a5b529SAl Viro list_add(&mnt->mnt_child, &shadows->mnt_child); 78812a5b529SAl Viro } else { 78912a5b529SAl Viro hlist_add_head_rcu(&mnt->mnt_hash, 79012a5b529SAl Viro m_hash(&parent->mnt, mnt->mnt_mountpoint)); 79112a5b529SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 79212a5b529SAl Viro } 79312a5b529SAl Viro } 79412a5b529SAl Viro 795b90fa9aeSRam Pai /* 79699b7db7bSNick Piggin * vfsmount lock must be held for write 797b90fa9aeSRam Pai */ 7981d6a32acSAl Viro static void commit_tree(struct mount *mnt, struct mount *shadows) 799b90fa9aeSRam Pai { 8000714a533SAl Viro struct mount *parent = mnt->mnt_parent; 80183adc753SAl Viro struct mount *m; 802b90fa9aeSRam Pai LIST_HEAD(head); 803143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 804b90fa9aeSRam Pai 8050714a533SAl Viro BUG_ON(parent == mnt); 806b90fa9aeSRam Pai 8071a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 808f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 809143c8c91SAl Viro m->mnt_ns = n; 810f03c6599SAl Viro 811b90fa9aeSRam Pai list_splice(&head, n->list.prev); 812b90fa9aeSRam Pai 81312a5b529SAl Viro attach_shadowed(mnt, parent, shadows); 8146b3286edSKirill Korotaev touch_mnt_namespace(n); 8151da177e4SLinus Torvalds } 8161da177e4SLinus Torvalds 817909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 8181da177e4SLinus Torvalds { 8196b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 8206b41d536SAl Viro if (next == &p->mnt_mounts) { 8211da177e4SLinus Torvalds while (1) { 822909b0a88SAl Viro if (p == root) 8231da177e4SLinus Torvalds return NULL; 8246b41d536SAl Viro next = p->mnt_child.next; 8256b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 8261da177e4SLinus Torvalds break; 8270714a533SAl Viro p = p->mnt_parent; 8281da177e4SLinus Torvalds } 8291da177e4SLinus Torvalds } 8306b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 8311da177e4SLinus Torvalds } 8321da177e4SLinus Torvalds 833315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 8349676f0c6SRam Pai { 8356b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 8366b41d536SAl Viro while (prev != &p->mnt_mounts) { 8376b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 8386b41d536SAl Viro prev = p->mnt_mounts.prev; 8399676f0c6SRam Pai } 8409676f0c6SRam Pai return p; 8419676f0c6SRam Pai } 8429676f0c6SRam Pai 8439d412a43SAl Viro struct vfsmount * 8449d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 8459d412a43SAl Viro { 846b105e270SAl Viro struct mount *mnt; 8479d412a43SAl Viro struct dentry *root; 8489d412a43SAl Viro 8499d412a43SAl Viro if (!type) 8509d412a43SAl Viro return ERR_PTR(-ENODEV); 8519d412a43SAl Viro 8529d412a43SAl Viro mnt = alloc_vfsmnt(name); 8539d412a43SAl Viro if (!mnt) 8549d412a43SAl Viro return ERR_PTR(-ENOMEM); 8559d412a43SAl Viro 8569d412a43SAl Viro if (flags & MS_KERNMOUNT) 857b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 8589d412a43SAl Viro 8599d412a43SAl Viro root = mount_fs(type, flags, name, data); 8609d412a43SAl Viro if (IS_ERR(root)) { 8618ffcb32eSDavid Howells mnt_free_id(mnt); 8629d412a43SAl Viro free_vfsmnt(mnt); 8639d412a43SAl Viro return ERR_CAST(root); 8649d412a43SAl Viro } 8659d412a43SAl Viro 866b105e270SAl Viro mnt->mnt.mnt_root = root; 867b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 868a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8690714a533SAl Viro mnt->mnt_parent = mnt; 870719ea2fbSAl Viro lock_mount_hash(); 87139f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 872719ea2fbSAl Viro unlock_mount_hash(); 873b105e270SAl Viro return &mnt->mnt; 8749d412a43SAl Viro } 8759d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 8769d412a43SAl Viro 87787129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 87836341f64SRam Pai int flag) 8791da177e4SLinus Torvalds { 88087129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 881be34d1a3SDavid Howells struct mount *mnt; 882be34d1a3SDavid Howells int err; 8831da177e4SLinus Torvalds 884be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 885be34d1a3SDavid Howells if (!mnt) 886be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 887be34d1a3SDavid Howells 8887a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 88915169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 890719f5d7fSMiklos Szeredi else 89115169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 892719f5d7fSMiklos Szeredi 89315169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 894be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 895719f5d7fSMiklos Szeredi if (err) 896719f5d7fSMiklos Szeredi goto out_free; 897719f5d7fSMiklos Szeredi } 898719f5d7fSMiklos Szeredi 899f2ebb3a9SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~(MNT_WRITE_HOLD|MNT_MARKED); 900132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 901132c94e3SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY)) 902132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 903132c94e3SEric W. Biederman 9045ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 9055ff9d8a6SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) 9065ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 9075ff9d8a6SEric W. Biederman 9081da177e4SLinus Torvalds atomic_inc(&sb->s_active); 909b105e270SAl Viro mnt->mnt.mnt_sb = sb; 910b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 911a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9120714a533SAl Viro mnt->mnt_parent = mnt; 913719ea2fbSAl Viro lock_mount_hash(); 91439f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 915719ea2fbSAl Viro unlock_mount_hash(); 916b90fa9aeSRam Pai 9177a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 9187a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 9196776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 92032301920SAl Viro mnt->mnt_master = old; 921fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 9228aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 923fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 9246776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 925d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 9266776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 927d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 9285afe0022SRam Pai } 929b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 9300f0afb1dSAl Viro set_mnt_shared(mnt); 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 9331da177e4SLinus Torvalds * as the original if that was on one */ 93436341f64SRam Pai if (flag & CL_EXPIRE) { 9356776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 9366776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 9371da177e4SLinus Torvalds } 938be34d1a3SDavid Howells 939cb338d06SAl Viro return mnt; 940719f5d7fSMiklos Szeredi 941719f5d7fSMiklos Szeredi out_free: 9428ffcb32eSDavid Howells mnt_free_id(mnt); 943719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 944be34d1a3SDavid Howells return ERR_PTR(err); 9451da177e4SLinus Torvalds } 9461da177e4SLinus Torvalds 947900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 9487b7b1aceSAl Viro { 94948a066e7SAl Viro rcu_read_lock(); 950aa9c0e07SAl Viro mnt_add_count(mnt, -1); 95148a066e7SAl Viro if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ 95248a066e7SAl Viro rcu_read_unlock(); 95399b7db7bSNick Piggin return; 954b3e19d92SNick Piggin } 955719ea2fbSAl Viro lock_mount_hash(); 956b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 95748a066e7SAl Viro rcu_read_unlock(); 958719ea2fbSAl Viro unlock_mount_hash(); 95999b7db7bSNick Piggin return; 96099b7db7bSNick Piggin } 96148a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 96248a066e7SAl Viro rcu_read_unlock(); 96348a066e7SAl Viro unlock_mount_hash(); 96448a066e7SAl Viro return; 96548a066e7SAl Viro } 96648a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 96748a066e7SAl Viro rcu_read_unlock(); 968962830dfSAndi Kleen 96939f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 970719ea2fbSAl Viro unlock_mount_hash(); 971649a795aSAl Viro 972649a795aSAl Viro /* 973649a795aSAl Viro * This probably indicates that somebody messed 974649a795aSAl Viro * up a mnt_want/drop_write() pair. If this 975649a795aSAl Viro * happens, the filesystem was probably unable 976649a795aSAl Viro * to make r/w->r/o transitions. 977649a795aSAl Viro */ 978649a795aSAl Viro /* 979649a795aSAl Viro * The locking used to deal with mnt_count decrement provides barriers, 980649a795aSAl Viro * so mnt_get_writers() below is safe. 981649a795aSAl Viro */ 982649a795aSAl Viro WARN_ON(mnt_get_writers(mnt)); 9833064c356SAl Viro if (unlikely(mnt->mnt_pins.first)) 9843064c356SAl Viro mnt_pin_kill(mnt); 985649a795aSAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 986649a795aSAl Viro dput(mnt->mnt.mnt_root); 987649a795aSAl Viro deactivate_super(mnt->mnt.mnt_sb); 98848a066e7SAl Viro mnt_free_id(mnt); 9898ffcb32eSDavid Howells call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 990b3e19d92SNick Piggin } 991b3e19d92SNick Piggin 992b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 993b3e19d92SNick Piggin { 994b3e19d92SNick Piggin if (mnt) { 995863d684fSAl Viro struct mount *m = real_mount(mnt); 996b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 997863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 998863d684fSAl Viro m->mnt_expiry_mark = 0; 999863d684fSAl Viro mntput_no_expire(m); 1000b3e19d92SNick Piggin } 1001b3e19d92SNick Piggin } 1002b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1003b3e19d92SNick Piggin 1004b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1005b3e19d92SNick Piggin { 1006b3e19d92SNick Piggin if (mnt) 100783adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1008b3e19d92SNick Piggin return mnt; 1009b3e19d92SNick Piggin } 1010b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1011b3e19d92SNick Piggin 10123064c356SAl Viro struct vfsmount *mnt_clone_internal(struct path *path) 10137b7b1aceSAl Viro { 10143064c356SAl Viro struct mount *p; 10153064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 10163064c356SAl Viro if (IS_ERR(p)) 10173064c356SAl Viro return ERR_CAST(p); 10183064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 10193064c356SAl Viro return &p->mnt; 10207b7b1aceSAl Viro } 10211da177e4SLinus Torvalds 1022b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 1023b3b304a2SMiklos Szeredi { 1024b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 1025b3b304a2SMiklos Szeredi } 1026b3b304a2SMiklos Szeredi 1027b3b304a2SMiklos Szeredi /* 1028b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 1029b3b304a2SMiklos Szeredi * implement more complex mount option showing. 1030b3b304a2SMiklos Szeredi * 1031b3b304a2SMiklos Szeredi * See also save_mount_options(). 1032b3b304a2SMiklos Szeredi */ 103334c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 1034b3b304a2SMiklos Szeredi { 10352a32cebdSAl Viro const char *options; 10362a32cebdSAl Viro 10372a32cebdSAl Viro rcu_read_lock(); 103834c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 1039b3b304a2SMiklos Szeredi 1040b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 1041b3b304a2SMiklos Szeredi seq_putc(m, ','); 1042b3b304a2SMiklos Szeredi mangle(m, options); 1043b3b304a2SMiklos Szeredi } 10442a32cebdSAl Viro rcu_read_unlock(); 1045b3b304a2SMiklos Szeredi 1046b3b304a2SMiklos Szeredi return 0; 1047b3b304a2SMiklos Szeredi } 1048b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 1049b3b304a2SMiklos Szeredi 1050b3b304a2SMiklos Szeredi /* 1051b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 1052b3b304a2SMiklos Szeredi * called from the fill_super() callback. 1053b3b304a2SMiklos Szeredi * 1054b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 1055b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1056b3b304a2SMiklos Szeredi * remount fails. 1057b3b304a2SMiklos Szeredi * 1058b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1059b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1060b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1061b3b304a2SMiklos Szeredi * any more. 1062b3b304a2SMiklos Szeredi */ 1063b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1064b3b304a2SMiklos Szeredi { 10652a32cebdSAl Viro BUG_ON(sb->s_options); 10662a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1067b3b304a2SMiklos Szeredi } 1068b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1069b3b304a2SMiklos Szeredi 10702a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 10712a32cebdSAl Viro { 10722a32cebdSAl Viro char *old = sb->s_options; 10732a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 10742a32cebdSAl Viro if (old) { 10752a32cebdSAl Viro synchronize_rcu(); 10762a32cebdSAl Viro kfree(old); 10772a32cebdSAl Viro } 10782a32cebdSAl Viro } 10792a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 10802a32cebdSAl Viro 1081a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 10820226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 10831da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 10841da177e4SLinus Torvalds { 10856ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 10861da177e4SLinus Torvalds 1087390c6843SRam Pai down_read(&namespace_sem); 1088c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1089c7999c36SAl Viro void *v = p->cached_mount; 1090c7999c36SAl Viro if (*pos == p->cached_index) 1091c7999c36SAl Viro return v; 1092c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1093c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1094c7999c36SAl Viro return p->cached_mount = v; 1095c7999c36SAl Viro } 1096c7999c36SAl Viro } 1097c7999c36SAl Viro 1098c7999c36SAl Viro p->cached_event = p->ns->event; 1099c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1100c7999c36SAl Viro p->cached_index = *pos; 1101c7999c36SAl Viro return p->cached_mount; 11021da177e4SLinus Torvalds } 11031da177e4SLinus Torvalds 11041da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 11051da177e4SLinus Torvalds { 11066ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 1107b0765fb8SPavel Emelianov 1108c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1109c7999c36SAl Viro p->cached_index = *pos; 1110c7999c36SAl Viro return p->cached_mount; 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds 11131da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 11141da177e4SLinus Torvalds { 1115390c6843SRam Pai up_read(&namespace_sem); 11161da177e4SLinus Torvalds } 11171da177e4SLinus Torvalds 11180226f492SAl Viro static int m_show(struct seq_file *m, void *v) 11199f5596afSAl Viro { 11206ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 11211a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 11220226f492SAl Viro return p->show(m, &r->mnt); 11231da177e4SLinus Torvalds } 11241da177e4SLinus Torvalds 1125a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 11261da177e4SLinus Torvalds .start = m_start, 11271da177e4SLinus Torvalds .next = m_next, 11281da177e4SLinus Torvalds .stop = m_stop, 11290226f492SAl Viro .show = m_show, 1130b4629fe2SChuck Lever }; 1131a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1132b4629fe2SChuck Lever 11331da177e4SLinus Torvalds /** 11341da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 11351da177e4SLinus Torvalds * @mnt: root of mount tree 11361da177e4SLinus Torvalds * 11371da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 11381da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 11391da177e4SLinus Torvalds * busy. 11401da177e4SLinus Torvalds */ 1141909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 11421da177e4SLinus Torvalds { 1143909b0a88SAl Viro struct mount *mnt = real_mount(m); 114436341f64SRam Pai int actual_refs = 0; 114536341f64SRam Pai int minimum_refs = 0; 1146315fc83eSAl Viro struct mount *p; 1147909b0a88SAl Viro BUG_ON(!m); 11481da177e4SLinus Torvalds 1149b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1150719ea2fbSAl Viro lock_mount_hash(); 1151909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 115283adc753SAl Viro actual_refs += mnt_get_count(p); 11531da177e4SLinus Torvalds minimum_refs += 2; 11541da177e4SLinus Torvalds } 1155719ea2fbSAl Viro unlock_mount_hash(); 11561da177e4SLinus Torvalds 11571da177e4SLinus Torvalds if (actual_refs > minimum_refs) 11581da177e4SLinus Torvalds return 0; 1159e3474a8eSIan Kent 1160e3474a8eSIan Kent return 1; 11611da177e4SLinus Torvalds } 11621da177e4SLinus Torvalds 11631da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 11641da177e4SLinus Torvalds 11651da177e4SLinus Torvalds /** 11661da177e4SLinus Torvalds * may_umount - check if a mount point is busy 11671da177e4SLinus Torvalds * @mnt: root of mount 11681da177e4SLinus Torvalds * 11691da177e4SLinus Torvalds * This is called to check if a mount point has any 11701da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 11711da177e4SLinus Torvalds * mount has sub mounts this will return busy 11721da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 11731da177e4SLinus Torvalds * 11741da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 11751da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 11761da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 11771da177e4SLinus Torvalds */ 11781da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 11791da177e4SLinus Torvalds { 1180e3474a8eSIan Kent int ret = 1; 11818ad08d8aSAl Viro down_read(&namespace_sem); 1182719ea2fbSAl Viro lock_mount_hash(); 11831ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1184e3474a8eSIan Kent ret = 0; 1185719ea2fbSAl Viro unlock_mount_hash(); 11868ad08d8aSAl Viro up_read(&namespace_sem); 1187a05964f3SRam Pai return ret; 11881da177e4SLinus Torvalds } 11891da177e4SLinus Torvalds 11901da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 11911da177e4SLinus Torvalds 119238129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1193e3197d83SAl Viro 119497216be0SAl Viro static void namespace_unlock(void) 11951da177e4SLinus Torvalds { 1196d5e50f74SAl Viro struct mount *mnt; 119738129a13SAl Viro struct hlist_head head = unmounted; 119897216be0SAl Viro 119938129a13SAl Viro if (likely(hlist_empty(&head))) { 120097216be0SAl Viro up_write(&namespace_sem); 120197216be0SAl Viro return; 120297216be0SAl Viro } 120397216be0SAl Viro 120438129a13SAl Viro head.first->pprev = &head.first; 120538129a13SAl Viro INIT_HLIST_HEAD(&unmounted); 120638129a13SAl Viro 120797216be0SAl Viro up_write(&namespace_sem); 120897216be0SAl Viro 120948a066e7SAl Viro synchronize_rcu(); 121048a066e7SAl Viro 121138129a13SAl Viro while (!hlist_empty(&head)) { 121238129a13SAl Viro mnt = hlist_entry(head.first, struct mount, mnt_hash); 121338129a13SAl Viro hlist_del_init(&mnt->mnt_hash); 1214aba809cfSAl Viro if (mnt->mnt_ex_mountpoint.mnt) 1215aba809cfSAl Viro path_put(&mnt->mnt_ex_mountpoint); 1216d5e50f74SAl Viro mntput(&mnt->mnt); 121770fbcdf4SRam Pai } 121870fbcdf4SRam Pai } 121970fbcdf4SRam Pai 122097216be0SAl Viro static inline void namespace_lock(void) 1221e3197d83SAl Viro { 122297216be0SAl Viro down_write(&namespace_sem); 1223e3197d83SAl Viro } 1224e3197d83SAl Viro 122599b7db7bSNick Piggin /* 122648a066e7SAl Viro * mount_lock must be held 122799b7db7bSNick Piggin * namespace_sem must be held for write 122848a066e7SAl Viro * how = 0 => just this tree, don't propagate 122948a066e7SAl Viro * how = 1 => propagate; we know that nobody else has reference to any victims 123048a066e7SAl Viro * how = 2 => lazy umount 123199b7db7bSNick Piggin */ 123248a066e7SAl Viro void umount_tree(struct mount *mnt, int how) 123370fbcdf4SRam Pai { 123438129a13SAl Viro HLIST_HEAD(tmp_list); 1235315fc83eSAl Viro struct mount *p; 123638129a13SAl Viro struct mount *last = NULL; 123770fbcdf4SRam Pai 123838129a13SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 123938129a13SAl Viro hlist_del_init_rcu(&p->mnt_hash); 124038129a13SAl Viro hlist_add_head(&p->mnt_hash, &tmp_list); 124138129a13SAl Viro } 124270fbcdf4SRam Pai 124348a066e7SAl Viro if (how) 12447b8a53fdSAl Viro propagate_umount(&tmp_list); 1245a05964f3SRam Pai 124638129a13SAl Viro hlist_for_each_entry(p, &tmp_list, mnt_hash) { 12476776db3dSAl Viro list_del_init(&p->mnt_expire); 12481a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1249143c8c91SAl Viro __touch_mnt_namespace(p->mnt_ns); 125063d37a84SAl Viro p->mnt_ns = NULL; 125148a066e7SAl Viro if (how < 2) 125248a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 12536b41d536SAl Viro list_del_init(&p->mnt_child); 1254676da58dSAl Viro if (mnt_has_parent(p)) { 125584d17192SAl Viro put_mountpoint(p->mnt_mp); 1256aba809cfSAl Viro /* move the reference to mountpoint into ->mnt_ex_mountpoint */ 1257aba809cfSAl Viro p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; 1258aba809cfSAl Viro p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; 1259aba809cfSAl Viro p->mnt_mountpoint = p->mnt.mnt_root; 1260aba809cfSAl Viro p->mnt_parent = p; 126184d17192SAl Viro p->mnt_mp = NULL; 12627c4b93d8SAl Viro } 12630f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 126438129a13SAl Viro last = p; 12651da177e4SLinus Torvalds } 126638129a13SAl Viro if (last) { 126738129a13SAl Viro last->mnt_hash.next = unmounted.first; 126838129a13SAl Viro unmounted.first = tmp_list.first; 126938129a13SAl Viro unmounted.first->pprev = &unmounted.first; 127038129a13SAl Viro } 12711da177e4SLinus Torvalds } 12721da177e4SLinus Torvalds 1273b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1274c35038beSAl Viro 12751ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 12761da177e4SLinus Torvalds { 12771ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 12781da177e4SLinus Torvalds int retval; 12791da177e4SLinus Torvalds 12801ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 12811da177e4SLinus Torvalds if (retval) 12821da177e4SLinus Torvalds return retval; 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds /* 12851da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 12861da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 12871da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 12881da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 12891da177e4SLinus Torvalds */ 12901da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 12911ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 12921da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 12931da177e4SLinus Torvalds return -EINVAL; 12941da177e4SLinus Torvalds 1295b3e19d92SNick Piggin /* 1296b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1297b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1298b3e19d92SNick Piggin */ 1299719ea2fbSAl Viro lock_mount_hash(); 130083adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1301719ea2fbSAl Viro unlock_mount_hash(); 13021da177e4SLinus Torvalds return -EBUSY; 1303b3e19d92SNick Piggin } 1304719ea2fbSAl Viro unlock_mount_hash(); 13051da177e4SLinus Torvalds 1306863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 13071da177e4SLinus Torvalds return -EAGAIN; 13081da177e4SLinus Torvalds } 13091da177e4SLinus Torvalds 13101da177e4SLinus Torvalds /* 13111da177e4SLinus Torvalds * If we may have to abort operations to get out of this 13121da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 13131da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 13141da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 13151da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 13161da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 13171da177e4SLinus Torvalds * about for the moment. 13181da177e4SLinus Torvalds */ 13191da177e4SLinus Torvalds 132042faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 132142faad99SAl Viro sb->s_op->umount_begin(sb); 132242faad99SAl Viro } 13231da177e4SLinus Torvalds 13241da177e4SLinus Torvalds /* 13251da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 13261da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 13271da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 13281da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 13291da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 13301da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 13311da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 13321da177e4SLinus Torvalds */ 13331ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 13341da177e4SLinus Torvalds /* 13351da177e4SLinus Torvalds * Special case for "unmounting" root ... 13361da177e4SLinus Torvalds * we just try to remount it readonly. 13371da177e4SLinus Torvalds */ 13381da177e4SLinus Torvalds down_write(&sb->s_umount); 13394aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 13401da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 13411da177e4SLinus Torvalds up_write(&sb->s_umount); 13421da177e4SLinus Torvalds return retval; 13431da177e4SLinus Torvalds } 13441da177e4SLinus Torvalds 134597216be0SAl Viro namespace_lock(); 1346719ea2fbSAl Viro lock_mount_hash(); 13475addc5ddSAl Viro event++; 13481da177e4SLinus Torvalds 134948a066e7SAl Viro if (flags & MNT_DETACH) { 135048a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 135148a066e7SAl Viro umount_tree(mnt, 2); 135248a066e7SAl Viro retval = 0; 135348a066e7SAl Viro } else { 1354b54b9be7SAl Viro shrink_submounts(mnt); 13551da177e4SLinus Torvalds retval = -EBUSY; 135648a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 13571a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1358328e6d90SAl Viro umount_tree(mnt, 1); 13591da177e4SLinus Torvalds retval = 0; 13601da177e4SLinus Torvalds } 136148a066e7SAl Viro } 1362719ea2fbSAl Viro unlock_mount_hash(); 1363e3197d83SAl Viro namespace_unlock(); 13641da177e4SLinus Torvalds return retval; 13651da177e4SLinus Torvalds } 13661da177e4SLinus Torvalds 13671da177e4SLinus Torvalds /* 13689b40bc90SAl Viro * Is the caller allowed to modify his namespace? 13699b40bc90SAl Viro */ 13709b40bc90SAl Viro static inline bool may_mount(void) 13719b40bc90SAl Viro { 13729b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 13739b40bc90SAl Viro } 13749b40bc90SAl Viro 13759b40bc90SAl Viro /* 13761da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 13771da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 13781da177e4SLinus Torvalds * 13791da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 13801da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 13811da177e4SLinus Torvalds */ 13821da177e4SLinus Torvalds 1383bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 13841da177e4SLinus Torvalds { 13852d8f3038SAl Viro struct path path; 1386900148dcSAl Viro struct mount *mnt; 13871da177e4SLinus Torvalds int retval; 1388db1f05bbSMiklos Szeredi int lookup_flags = 0; 13891da177e4SLinus Torvalds 1390db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1391db1f05bbSMiklos Szeredi return -EINVAL; 1392db1f05bbSMiklos Szeredi 13939b40bc90SAl Viro if (!may_mount()) 13949b40bc90SAl Viro return -EPERM; 13959b40bc90SAl Viro 1396db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1397db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1398db1f05bbSMiklos Szeredi 1399197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 14001da177e4SLinus Torvalds if (retval) 14011da177e4SLinus Torvalds goto out; 1402900148dcSAl Viro mnt = real_mount(path.mnt); 14031da177e4SLinus Torvalds retval = -EINVAL; 14042d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 14051da177e4SLinus Torvalds goto dput_and_out; 1406143c8c91SAl Viro if (!check_mnt(mnt)) 14071da177e4SLinus Torvalds goto dput_and_out; 14085ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 14095ff9d8a6SEric W. Biederman goto dput_and_out; 14101da177e4SLinus Torvalds 1411900148dcSAl Viro retval = do_umount(mnt, flags); 14121da177e4SLinus Torvalds dput_and_out: 1413429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 14142d8f3038SAl Viro dput(path.dentry); 1415900148dcSAl Viro mntput_no_expire(mnt); 14161da177e4SLinus Torvalds out: 14171da177e4SLinus Torvalds return retval; 14181da177e4SLinus Torvalds } 14191da177e4SLinus Torvalds 14201da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 14211da177e4SLinus Torvalds 14221da177e4SLinus Torvalds /* 14231da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 14241da177e4SLinus Torvalds */ 1425bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 14261da177e4SLinus Torvalds { 14271da177e4SLinus Torvalds return sys_umount(name, 0); 14281da177e4SLinus Torvalds } 14291da177e4SLinus Torvalds 14301da177e4SLinus Torvalds #endif 14311da177e4SLinus Torvalds 14324ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 14338823c079SEric W. Biederman { 14344ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 14354ce5d2b1SEric W. Biederman struct inode *inode = dentry->d_inode; 14360bb80f24SDavid Howells struct proc_ns *ei; 14378823c079SEric W. Biederman 14388823c079SEric W. Biederman if (!proc_ns_inode(inode)) 14398823c079SEric W. Biederman return false; 14408823c079SEric W. Biederman 14410bb80f24SDavid Howells ei = get_proc_ns(inode); 14428823c079SEric W. Biederman if (ei->ns_ops != &mntns_operations) 14438823c079SEric W. Biederman return false; 14448823c079SEric W. Biederman 14454ce5d2b1SEric W. Biederman return true; 14464ce5d2b1SEric W. Biederman } 14474ce5d2b1SEric W. Biederman 14484ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 14494ce5d2b1SEric W. Biederman { 14504ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 14514ce5d2b1SEric W. Biederman * mount namespace loop? 14524ce5d2b1SEric W. Biederman */ 14534ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 14544ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 14554ce5d2b1SEric W. Biederman return false; 14564ce5d2b1SEric W. Biederman 14574ce5d2b1SEric W. Biederman mnt_ns = get_proc_ns(dentry->d_inode)->ns; 14588823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 14598823c079SEric W. Biederman } 14608823c079SEric W. Biederman 146187129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 146236341f64SRam Pai int flag) 14631da177e4SLinus Torvalds { 146484d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 14651da177e4SLinus Torvalds 14664ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 14674ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 14684ce5d2b1SEric W. Biederman 14694ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1470be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 14719676f0c6SRam Pai 147236341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1473be34d1a3SDavid Howells if (IS_ERR(q)) 1474be34d1a3SDavid Howells return q; 1475be34d1a3SDavid Howells 14765ff9d8a6SEric W. Biederman q->mnt.mnt_flags &= ~MNT_LOCKED; 1477a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds p = mnt; 14806b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1481315fc83eSAl Viro struct mount *s; 14827ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 14831da177e4SLinus Torvalds continue; 14841da177e4SLinus Torvalds 1485909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 148612a5b529SAl Viro struct mount *t = NULL; 14874ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 14884ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 14894ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 14904ce5d2b1SEric W. Biederman continue; 14914ce5d2b1SEric W. Biederman } 14924ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 14934ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 14949676f0c6SRam Pai s = skip_mnt_tree(s); 14959676f0c6SRam Pai continue; 14969676f0c6SRam Pai } 14970714a533SAl Viro while (p != s->mnt_parent) { 14980714a533SAl Viro p = p->mnt_parent; 14990714a533SAl Viro q = q->mnt_parent; 15001da177e4SLinus Torvalds } 150187129cc0SAl Viro p = s; 150284d17192SAl Viro parent = q; 150387129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1504be34d1a3SDavid Howells if (IS_ERR(q)) 1505be34d1a3SDavid Howells goto out; 1506719ea2fbSAl Viro lock_mount_hash(); 15071a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 150812a5b529SAl Viro mnt_set_mountpoint(parent, p->mnt_mp, q); 150912a5b529SAl Viro if (!list_empty(&parent->mnt_mounts)) { 151012a5b529SAl Viro t = list_last_entry(&parent->mnt_mounts, 151112a5b529SAl Viro struct mount, mnt_child); 151212a5b529SAl Viro if (t->mnt_mp != p->mnt_mp) 151312a5b529SAl Viro t = NULL; 151412a5b529SAl Viro } 151512a5b529SAl Viro attach_shadowed(q, parent, t); 1516719ea2fbSAl Viro unlock_mount_hash(); 15171da177e4SLinus Torvalds } 15181da177e4SLinus Torvalds } 15191da177e4SLinus Torvalds return res; 1520be34d1a3SDavid Howells out: 15211da177e4SLinus Torvalds if (res) { 1522719ea2fbSAl Viro lock_mount_hash(); 1523328e6d90SAl Viro umount_tree(res, 0); 1524719ea2fbSAl Viro unlock_mount_hash(); 15251da177e4SLinus Torvalds } 1526be34d1a3SDavid Howells return q; 15271da177e4SLinus Torvalds } 15281da177e4SLinus Torvalds 1529be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1530be34d1a3SDavid Howells 1531589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 15328aec0809SAl Viro { 1533cb338d06SAl Viro struct mount *tree; 153497216be0SAl Viro namespace_lock(); 153587129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 153687129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1537328e6d90SAl Viro namespace_unlock(); 1538be34d1a3SDavid Howells if (IS_ERR(tree)) 153952e220d3SDan Carpenter return ERR_CAST(tree); 1540be34d1a3SDavid Howells return &tree->mnt; 15418aec0809SAl Viro } 15428aec0809SAl Viro 15438aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 15448aec0809SAl Viro { 154597216be0SAl Viro namespace_lock(); 1546719ea2fbSAl Viro lock_mount_hash(); 1547328e6d90SAl Viro umount_tree(real_mount(mnt), 0); 1548719ea2fbSAl Viro unlock_mount_hash(); 15493ab6abeeSAl Viro namespace_unlock(); 15508aec0809SAl Viro } 15518aec0809SAl Viro 15521f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 15531f707137SAl Viro struct vfsmount *root) 15541f707137SAl Viro { 15551a4eeaf2SAl Viro struct mount *mnt; 15561f707137SAl Viro int res = f(root, arg); 15571f707137SAl Viro if (res) 15581f707137SAl Viro return res; 15591a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 15601a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 15611f707137SAl Viro if (res) 15621f707137SAl Viro return res; 15631f707137SAl Viro } 15641f707137SAl Viro return 0; 15651f707137SAl Viro } 15661f707137SAl Viro 15674b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1568719f5d7fSMiklos Szeredi { 1569315fc83eSAl Viro struct mount *p; 1570719f5d7fSMiklos Szeredi 1571909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1572fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 15734b8b21f4SAl Viro mnt_release_group_id(p); 1574719f5d7fSMiklos Szeredi } 1575719f5d7fSMiklos Szeredi } 1576719f5d7fSMiklos Szeredi 15774b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1578719f5d7fSMiklos Szeredi { 1579315fc83eSAl Viro struct mount *p; 1580719f5d7fSMiklos Szeredi 1581909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1582fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 15834b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1584719f5d7fSMiklos Szeredi if (err) { 15854b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1586719f5d7fSMiklos Szeredi return err; 1587719f5d7fSMiklos Szeredi } 1588719f5d7fSMiklos Szeredi } 1589719f5d7fSMiklos Szeredi } 1590719f5d7fSMiklos Szeredi 1591719f5d7fSMiklos Szeredi return 0; 1592719f5d7fSMiklos Szeredi } 1593719f5d7fSMiklos Szeredi 1594b90fa9aeSRam Pai /* 1595b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1596b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 159721444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 159821444403SRam Pai * store the parent mount and mountpoint dentry. 159921444403SRam Pai * (done when source_mnt is moved) 1600b90fa9aeSRam Pai * 1601b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1602b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 16039676f0c6SRam Pai * --------------------------------------------------------------------------- 1604b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 16059676f0c6SRam Pai * |************************************************************************** 16069676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 16079676f0c6SRam Pai * | dest | | | | | 16089676f0c6SRam Pai * | | | | | | | 16099676f0c6SRam Pai * | v | | | | | 16109676f0c6SRam Pai * |************************************************************************** 16119676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 16125afe0022SRam Pai * | | | | | | 16139676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 16149676f0c6SRam Pai * *************************************************************************** 1615b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1616b90fa9aeSRam Pai * destination mount. 1617b90fa9aeSRam Pai * 1618b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1619b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1620b90fa9aeSRam Pai * the peer group of the source mount. 1621b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1622b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1623b90fa9aeSRam Pai * mount. 16245afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 16255afe0022SRam Pai * of the destination mount and the cloned mount is made slave 16265afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 16275afe0022SRam Pai * is marked as 'shared and slave'. 16285afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 16295afe0022SRam Pai * source mount. 16305afe0022SRam Pai * 16319676f0c6SRam Pai * --------------------------------------------------------------------------- 163221444403SRam Pai * | MOVE MOUNT OPERATION | 16339676f0c6SRam Pai * |************************************************************************** 16349676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 16359676f0c6SRam Pai * | dest | | | | | 16369676f0c6SRam Pai * | | | | | | | 16379676f0c6SRam Pai * | v | | | | | 16389676f0c6SRam Pai * |************************************************************************** 16399676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 16405afe0022SRam Pai * | | | | | | 16419676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 16429676f0c6SRam Pai * *************************************************************************** 16435afe0022SRam Pai * 16445afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 16455afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 164621444403SRam Pai * (+*) the mount is moved to the destination. 16475afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 16485afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 16495afe0022SRam Pai * the mount is marked as 'shared and slave'. 16505afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1651b90fa9aeSRam Pai * 1652b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1653b90fa9aeSRam Pai * applied to each mount in the tree. 1654b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1655b90fa9aeSRam Pai * in allocations. 1656b90fa9aeSRam Pai */ 16570fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 165884d17192SAl Viro struct mount *dest_mnt, 165984d17192SAl Viro struct mountpoint *dest_mp, 166084d17192SAl Viro struct path *parent_path) 1661b90fa9aeSRam Pai { 166238129a13SAl Viro HLIST_HEAD(tree_list); 1663315fc83eSAl Viro struct mount *child, *p; 166438129a13SAl Viro struct hlist_node *n; 1665719f5d7fSMiklos Szeredi int err; 1666b90fa9aeSRam Pai 1667fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 16680fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1669719f5d7fSMiklos Szeredi if (err) 1670719f5d7fSMiklos Szeredi goto out; 167184d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1672f2ebb3a9SAl Viro lock_mount_hash(); 1673719f5d7fSMiklos Szeredi if (err) 1674719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1675909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 16760f0afb1dSAl Viro set_mnt_shared(p); 16770b1b901bSAl Viro } else { 16780b1b901bSAl Viro lock_mount_hash(); 1679b90fa9aeSRam Pai } 16801a390689SAl Viro if (parent_path) { 16810fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 168284d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1683143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 168421444403SRam Pai } else { 168584d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 16861d6a32acSAl Viro commit_tree(source_mnt, NULL); 168721444403SRam Pai } 1688b90fa9aeSRam Pai 168938129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 16901d6a32acSAl Viro struct mount *q; 169138129a13SAl Viro hlist_del_init(&child->mnt_hash); 16921d6a32acSAl Viro q = __lookup_mnt_last(&child->mnt_parent->mnt, 16931d6a32acSAl Viro child->mnt_mountpoint); 16941d6a32acSAl Viro commit_tree(child, q); 1695b90fa9aeSRam Pai } 1696719ea2fbSAl Viro unlock_mount_hash(); 169799b7db7bSNick Piggin 1698b90fa9aeSRam Pai return 0; 1699719f5d7fSMiklos Szeredi 1700719f5d7fSMiklos Szeredi out_cleanup_ids: 1701f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 1702f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 1703f2ebb3a9SAl Viro umount_tree(child, 0); 1704f2ebb3a9SAl Viro } 1705f2ebb3a9SAl Viro unlock_mount_hash(); 17060fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1707719f5d7fSMiklos Szeredi out: 1708719f5d7fSMiklos Szeredi return err; 1709b90fa9aeSRam Pai } 1710b90fa9aeSRam Pai 171184d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 1712b12cea91SAl Viro { 1713b12cea91SAl Viro struct vfsmount *mnt; 171484d17192SAl Viro struct dentry *dentry = path->dentry; 1715b12cea91SAl Viro retry: 171684d17192SAl Viro mutex_lock(&dentry->d_inode->i_mutex); 171784d17192SAl Viro if (unlikely(cant_mount(dentry))) { 171884d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 171984d17192SAl Viro return ERR_PTR(-ENOENT); 1720b12cea91SAl Viro } 172197216be0SAl Viro namespace_lock(); 1722b12cea91SAl Viro mnt = lookup_mnt(path); 172384d17192SAl Viro if (likely(!mnt)) { 172484d17192SAl Viro struct mountpoint *mp = new_mountpoint(dentry); 172584d17192SAl Viro if (IS_ERR(mp)) { 172697216be0SAl Viro namespace_unlock(); 172784d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 172884d17192SAl Viro return mp; 172984d17192SAl Viro } 173084d17192SAl Viro return mp; 173184d17192SAl Viro } 173297216be0SAl Viro namespace_unlock(); 1733b12cea91SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 1734b12cea91SAl Viro path_put(path); 1735b12cea91SAl Viro path->mnt = mnt; 173684d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 1737b12cea91SAl Viro goto retry; 1738b12cea91SAl Viro } 1739b12cea91SAl Viro 174084d17192SAl Viro static void unlock_mount(struct mountpoint *where) 1741b12cea91SAl Viro { 174284d17192SAl Viro struct dentry *dentry = where->m_dentry; 174384d17192SAl Viro put_mountpoint(where); 1744328e6d90SAl Viro namespace_unlock(); 174584d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 1746b12cea91SAl Viro } 1747b12cea91SAl Viro 174884d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 17491da177e4SLinus Torvalds { 175095bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 17511da177e4SLinus Torvalds return -EINVAL; 17521da177e4SLinus Torvalds 175384d17192SAl Viro if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != 175495bc5f25SAl Viro S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) 17551da177e4SLinus Torvalds return -ENOTDIR; 17561da177e4SLinus Torvalds 175784d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 17581da177e4SLinus Torvalds } 17591da177e4SLinus Torvalds 17601da177e4SLinus Torvalds /* 17617a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 17627a2e8a8fSValerie Aurora */ 17637a2e8a8fSValerie Aurora 17647a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 17657a2e8a8fSValerie Aurora { 17667c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 17677a2e8a8fSValerie Aurora 17687a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 17697a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 17707a2e8a8fSValerie Aurora return 0; 17717a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 17727a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 17737a2e8a8fSValerie Aurora return 0; 17747a2e8a8fSValerie Aurora return type; 17757a2e8a8fSValerie Aurora } 17767a2e8a8fSValerie Aurora 17777a2e8a8fSValerie Aurora /* 177807b20889SRam Pai * recursively change the type of the mountpoint. 177907b20889SRam Pai */ 17800a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 178107b20889SRam Pai { 1782315fc83eSAl Viro struct mount *m; 17834b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 178407b20889SRam Pai int recurse = flag & MS_REC; 17857a2e8a8fSValerie Aurora int type; 1786719f5d7fSMiklos Szeredi int err = 0; 178707b20889SRam Pai 17882d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 178907b20889SRam Pai return -EINVAL; 179007b20889SRam Pai 17917a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 17927a2e8a8fSValerie Aurora if (!type) 17937a2e8a8fSValerie Aurora return -EINVAL; 17947a2e8a8fSValerie Aurora 179597216be0SAl Viro namespace_lock(); 1796719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1797719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1798719f5d7fSMiklos Szeredi if (err) 1799719f5d7fSMiklos Szeredi goto out_unlock; 1800719f5d7fSMiklos Szeredi } 1801719f5d7fSMiklos Szeredi 1802719ea2fbSAl Viro lock_mount_hash(); 1803909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 18040f0afb1dSAl Viro change_mnt_propagation(m, type); 1805719ea2fbSAl Viro unlock_mount_hash(); 1806719f5d7fSMiklos Szeredi 1807719f5d7fSMiklos Szeredi out_unlock: 180897216be0SAl Viro namespace_unlock(); 1809719f5d7fSMiklos Szeredi return err; 181007b20889SRam Pai } 181107b20889SRam Pai 18125ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 18135ff9d8a6SEric W. Biederman { 18145ff9d8a6SEric W. Biederman struct mount *child; 18155ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 18165ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 18175ff9d8a6SEric W. Biederman continue; 18185ff9d8a6SEric W. Biederman 18195ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 18205ff9d8a6SEric W. Biederman return true; 18215ff9d8a6SEric W. Biederman } 18225ff9d8a6SEric W. Biederman return false; 18235ff9d8a6SEric W. Biederman } 18245ff9d8a6SEric W. Biederman 182507b20889SRam Pai /* 18261da177e4SLinus Torvalds * do loopback mount. 18271da177e4SLinus Torvalds */ 1828808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 18292dafe1c4SEric Sandeen int recurse) 18301da177e4SLinus Torvalds { 18312d92ab3cSAl Viro struct path old_path; 183284d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 183384d17192SAl Viro struct mountpoint *mp; 183457eccb83SAl Viro int err; 18351da177e4SLinus Torvalds if (!old_name || !*old_name) 18361da177e4SLinus Torvalds return -EINVAL; 1837815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 18381da177e4SLinus Torvalds if (err) 18391da177e4SLinus Torvalds return err; 18401da177e4SLinus Torvalds 18418823c079SEric W. Biederman err = -EINVAL; 18424ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 18438823c079SEric W. Biederman goto out; 18448823c079SEric W. Biederman 184584d17192SAl Viro mp = lock_mount(path); 184684d17192SAl Viro err = PTR_ERR(mp); 184784d17192SAl Viro if (IS_ERR(mp)) 18489676f0c6SRam Pai goto out; 18499676f0c6SRam Pai 185087129cc0SAl Viro old = real_mount(old_path.mnt); 185184d17192SAl Viro parent = real_mount(path->mnt); 185287129cc0SAl Viro 1853b12cea91SAl Viro err = -EINVAL; 1854fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 1855b12cea91SAl Viro goto out2; 1856b12cea91SAl Viro 185784d17192SAl Viro if (!check_mnt(parent) || !check_mnt(old)) 1858b12cea91SAl Viro goto out2; 1859ccd48bc7SAl Viro 18605ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 18615ff9d8a6SEric W. Biederman goto out2; 18625ff9d8a6SEric W. Biederman 18631da177e4SLinus Torvalds if (recurse) 18644ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 18651da177e4SLinus Torvalds else 186687129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 18671da177e4SLinus Torvalds 1868be34d1a3SDavid Howells if (IS_ERR(mnt)) { 1869be34d1a3SDavid Howells err = PTR_ERR(mnt); 1870e9c5d8a5SAndrey Vagin goto out2; 1871be34d1a3SDavid Howells } 1872ccd48bc7SAl Viro 18735ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 18745ff9d8a6SEric W. Biederman 187584d17192SAl Viro err = graft_tree(mnt, parent, mp); 18761da177e4SLinus Torvalds if (err) { 1877719ea2fbSAl Viro lock_mount_hash(); 1878328e6d90SAl Viro umount_tree(mnt, 0); 1879719ea2fbSAl Viro unlock_mount_hash(); 18805b83d2c5SRam Pai } 1881b12cea91SAl Viro out2: 188284d17192SAl Viro unlock_mount(mp); 1883ccd48bc7SAl Viro out: 18842d92ab3cSAl Viro path_put(&old_path); 18851da177e4SLinus Torvalds return err; 18861da177e4SLinus Torvalds } 18871da177e4SLinus Torvalds 18882e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 18892e4b7fcdSDave Hansen { 18902e4b7fcdSDave Hansen int error = 0; 18912e4b7fcdSDave Hansen int readonly_request = 0; 18922e4b7fcdSDave Hansen 18932e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 18942e4b7fcdSDave Hansen readonly_request = 1; 18952e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 18962e4b7fcdSDave Hansen return 0; 18972e4b7fcdSDave Hansen 189890563b19SEric W. Biederman if (mnt->mnt_flags & MNT_LOCK_READONLY) 189990563b19SEric W. Biederman return -EPERM; 190090563b19SEric W. Biederman 19012e4b7fcdSDave Hansen if (readonly_request) 190283adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 19032e4b7fcdSDave Hansen else 190483adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 19052e4b7fcdSDave Hansen return error; 19062e4b7fcdSDave Hansen } 19072e4b7fcdSDave Hansen 19081da177e4SLinus Torvalds /* 19091da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 19101da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 19111da177e4SLinus Torvalds * on it - tough luck. 19121da177e4SLinus Torvalds */ 19130a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 19141da177e4SLinus Torvalds void *data) 19151da177e4SLinus Torvalds { 19161da177e4SLinus Torvalds int err; 19172d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 1918143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 19191da177e4SLinus Torvalds 1920143c8c91SAl Viro if (!check_mnt(mnt)) 19211da177e4SLinus Torvalds return -EINVAL; 19221da177e4SLinus Torvalds 19232d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 19241da177e4SLinus Torvalds return -EINVAL; 19251da177e4SLinus Torvalds 1926ff36fe2cSEric Paris err = security_sb_remount(sb, data); 1927ff36fe2cSEric Paris if (err) 1928ff36fe2cSEric Paris return err; 1929ff36fe2cSEric Paris 19301da177e4SLinus Torvalds down_write(&sb->s_umount); 19312e4b7fcdSDave Hansen if (flags & MS_BIND) 19322d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 193357eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 193457eccb83SAl Viro err = -EPERM; 19354aa98cf7SAl Viro else 19361da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 19377b43a79fSAl Viro if (!err) { 1938719ea2fbSAl Viro lock_mount_hash(); 1939143c8c91SAl Viro mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; 1940143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 1941143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1942719ea2fbSAl Viro unlock_mount_hash(); 19430e55a7ccSDan Williams } 19446339dab8SAl Viro up_write(&sb->s_umount); 19451da177e4SLinus Torvalds return err; 19461da177e4SLinus Torvalds } 19471da177e4SLinus Torvalds 1948cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 19499676f0c6SRam Pai { 1950315fc83eSAl Viro struct mount *p; 1951909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 1952fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 19539676f0c6SRam Pai return 1; 19549676f0c6SRam Pai } 19559676f0c6SRam Pai return 0; 19569676f0c6SRam Pai } 19579676f0c6SRam Pai 1958808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 19591da177e4SLinus Torvalds { 19602d92ab3cSAl Viro struct path old_path, parent_path; 1961676da58dSAl Viro struct mount *p; 19620fb54e50SAl Viro struct mount *old; 196384d17192SAl Viro struct mountpoint *mp; 196457eccb83SAl Viro int err; 19651da177e4SLinus Torvalds if (!old_name || !*old_name) 19661da177e4SLinus Torvalds return -EINVAL; 19672d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 19681da177e4SLinus Torvalds if (err) 19691da177e4SLinus Torvalds return err; 19701da177e4SLinus Torvalds 197184d17192SAl Viro mp = lock_mount(path); 197284d17192SAl Viro err = PTR_ERR(mp); 197384d17192SAl Viro if (IS_ERR(mp)) 1974cc53ce53SDavid Howells goto out; 1975cc53ce53SDavid Howells 1976143c8c91SAl Viro old = real_mount(old_path.mnt); 1977fc7be130SAl Viro p = real_mount(path->mnt); 1978143c8c91SAl Viro 19791da177e4SLinus Torvalds err = -EINVAL; 1980fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 19811da177e4SLinus Torvalds goto out1; 19821da177e4SLinus Torvalds 19835ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 19845ff9d8a6SEric W. Biederman goto out1; 19855ff9d8a6SEric W. Biederman 19861da177e4SLinus Torvalds err = -EINVAL; 19872d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 198821444403SRam Pai goto out1; 19891da177e4SLinus Torvalds 1990676da58dSAl Viro if (!mnt_has_parent(old)) 199121444403SRam Pai goto out1; 19921da177e4SLinus Torvalds 19932d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 19942d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 199521444403SRam Pai goto out1; 199621444403SRam Pai /* 199721444403SRam Pai * Don't move a mount residing in a shared parent. 199821444403SRam Pai */ 1999fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 200021444403SRam Pai goto out1; 20019676f0c6SRam Pai /* 20029676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 20039676f0c6SRam Pai * mount which is shared. 20049676f0c6SRam Pai */ 2005fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 20069676f0c6SRam Pai goto out1; 20071da177e4SLinus Torvalds err = -ELOOP; 2008fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2009676da58dSAl Viro if (p == old) 201021444403SRam Pai goto out1; 20111da177e4SLinus Torvalds 201284d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 20134ac91378SJan Blunck if (err) 201421444403SRam Pai goto out1; 20151da177e4SLinus Torvalds 20161da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 20171da177e4SLinus Torvalds * automatically */ 20186776db3dSAl Viro list_del_init(&old->mnt_expire); 20191da177e4SLinus Torvalds out1: 202084d17192SAl Viro unlock_mount(mp); 20211da177e4SLinus Torvalds out: 20221da177e4SLinus Torvalds if (!err) 20231a390689SAl Viro path_put(&parent_path); 20242d92ab3cSAl Viro path_put(&old_path); 20251da177e4SLinus Torvalds return err; 20261da177e4SLinus Torvalds } 20271da177e4SLinus Torvalds 20289d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 20299d412a43SAl Viro { 20309d412a43SAl Viro int err; 20319d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 20329d412a43SAl Viro if (subtype) { 20339d412a43SAl Viro subtype++; 20349d412a43SAl Viro err = -EINVAL; 20359d412a43SAl Viro if (!subtype[0]) 20369d412a43SAl Viro goto err; 20379d412a43SAl Viro } else 20389d412a43SAl Viro subtype = ""; 20399d412a43SAl Viro 20409d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 20419d412a43SAl Viro err = -ENOMEM; 20429d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 20439d412a43SAl Viro goto err; 20449d412a43SAl Viro return mnt; 20459d412a43SAl Viro 20469d412a43SAl Viro err: 20479d412a43SAl Viro mntput(mnt); 20489d412a43SAl Viro return ERR_PTR(err); 20499d412a43SAl Viro } 20509d412a43SAl Viro 20519d412a43SAl Viro /* 20529d412a43SAl Viro * add a mount into a namespace's mount tree 20539d412a43SAl Viro */ 205495bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 20559d412a43SAl Viro { 205684d17192SAl Viro struct mountpoint *mp; 205784d17192SAl Viro struct mount *parent; 20589d412a43SAl Viro int err; 20599d412a43SAl Viro 2060f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 20619d412a43SAl Viro 206284d17192SAl Viro mp = lock_mount(path); 206384d17192SAl Viro if (IS_ERR(mp)) 206484d17192SAl Viro return PTR_ERR(mp); 20659d412a43SAl Viro 206684d17192SAl Viro parent = real_mount(path->mnt); 20679d412a43SAl Viro err = -EINVAL; 206884d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2069156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2070156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 20719d412a43SAl Viro goto unlock; 2072156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 207384d17192SAl Viro if (!parent->mnt_ns) 2074156cacb1SAl Viro goto unlock; 2075156cacb1SAl Viro } 20769d412a43SAl Viro 20779d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 20789d412a43SAl Viro err = -EBUSY; 207995bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 20809d412a43SAl Viro path->mnt->mnt_root == path->dentry) 20819d412a43SAl Viro goto unlock; 20829d412a43SAl Viro 20839d412a43SAl Viro err = -EINVAL; 208495bc5f25SAl Viro if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) 20859d412a43SAl Viro goto unlock; 20869d412a43SAl Viro 208795bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 208884d17192SAl Viro err = graft_tree(newmnt, parent, mp); 20899d412a43SAl Viro 20909d412a43SAl Viro unlock: 209184d17192SAl Viro unlock_mount(mp); 20929d412a43SAl Viro return err; 20939d412a43SAl Viro } 2094b1e75df4SAl Viro 20951da177e4SLinus Torvalds /* 20961da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 20971da177e4SLinus Torvalds * namespace's tree 20981da177e4SLinus Torvalds */ 20990c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2100808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 21011da177e4SLinus Torvalds { 21020c55cfc4SEric W. Biederman struct file_system_type *type; 21039b40bc90SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 21041da177e4SLinus Torvalds struct vfsmount *mnt; 210515f9a3f3SAl Viro int err; 21061da177e4SLinus Torvalds 21070c55cfc4SEric W. Biederman if (!fstype) 21081da177e4SLinus Torvalds return -EINVAL; 21091da177e4SLinus Torvalds 21100c55cfc4SEric W. Biederman type = get_fs_type(fstype); 21110c55cfc4SEric W. Biederman if (!type) 21120c55cfc4SEric W. Biederman return -ENODEV; 21130c55cfc4SEric W. Biederman 21140c55cfc4SEric W. Biederman if (user_ns != &init_user_ns) { 21150c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_MOUNT)) { 21160c55cfc4SEric W. Biederman put_filesystem(type); 21170c55cfc4SEric W. Biederman return -EPERM; 21180c55cfc4SEric W. Biederman } 21190c55cfc4SEric W. Biederman /* Only in special cases allow devices from mounts 21200c55cfc4SEric W. Biederman * created outside the initial user namespace. 21210c55cfc4SEric W. Biederman */ 21220c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { 21230c55cfc4SEric W. Biederman flags |= MS_NODEV; 21240c55cfc4SEric W. Biederman mnt_flags |= MNT_NODEV; 21250c55cfc4SEric W. Biederman } 21260c55cfc4SEric W. Biederman } 21270c55cfc4SEric W. Biederman 21280c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 21290c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 21300c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 21310c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 21320c55cfc4SEric W. Biederman 21330c55cfc4SEric W. Biederman put_filesystem(type); 21341da177e4SLinus Torvalds if (IS_ERR(mnt)) 21351da177e4SLinus Torvalds return PTR_ERR(mnt); 21361da177e4SLinus Torvalds 213795bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 213815f9a3f3SAl Viro if (err) 213915f9a3f3SAl Viro mntput(mnt); 214015f9a3f3SAl Viro return err; 21411da177e4SLinus Torvalds } 21421da177e4SLinus Torvalds 214319a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 214419a167afSAl Viro { 21456776db3dSAl Viro struct mount *mnt = real_mount(m); 214619a167afSAl Viro int err; 214719a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 214819a167afSAl Viro * expired before we get a chance to add it 214919a167afSAl Viro */ 21506776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 215119a167afSAl Viro 215219a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 215319a167afSAl Viro m->mnt_root == path->dentry) { 2154b1e75df4SAl Viro err = -ELOOP; 2155b1e75df4SAl Viro goto fail; 215619a167afSAl Viro } 215719a167afSAl Viro 215895bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2159b1e75df4SAl Viro if (!err) 2160b1e75df4SAl Viro return 0; 2161b1e75df4SAl Viro fail: 2162b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 21636776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 216497216be0SAl Viro namespace_lock(); 21656776db3dSAl Viro list_del_init(&mnt->mnt_expire); 216697216be0SAl Viro namespace_unlock(); 216719a167afSAl Viro } 2168b1e75df4SAl Viro mntput(m); 2169b1e75df4SAl Viro mntput(m); 217019a167afSAl Viro return err; 217119a167afSAl Viro } 217219a167afSAl Viro 2173ea5b778aSDavid Howells /** 2174ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2175ea5b778aSDavid Howells * @mnt: The mount to list. 2176ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2177ea5b778aSDavid Howells */ 2178ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2179ea5b778aSDavid Howells { 218097216be0SAl Viro namespace_lock(); 2181ea5b778aSDavid Howells 21826776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2183ea5b778aSDavid Howells 218497216be0SAl Viro namespace_unlock(); 2185ea5b778aSDavid Howells } 2186ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2187ea5b778aSDavid Howells 2188ea5b778aSDavid Howells /* 21891da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 21901da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 21911da177e4SLinus Torvalds * here 21921da177e4SLinus Torvalds */ 21931da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 21941da177e4SLinus Torvalds { 2195761d5c38SAl Viro struct mount *mnt, *next; 21961da177e4SLinus Torvalds LIST_HEAD(graveyard); 21971da177e4SLinus Torvalds 21981da177e4SLinus Torvalds if (list_empty(mounts)) 21991da177e4SLinus Torvalds return; 22001da177e4SLinus Torvalds 220197216be0SAl Viro namespace_lock(); 2202719ea2fbSAl Viro lock_mount_hash(); 22031da177e4SLinus Torvalds 22041da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 22051da177e4SLinus Torvalds * following criteria: 22061da177e4SLinus Torvalds * - only referenced by its parent vfsmount 22071da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 22081da177e4SLinus Torvalds * cleared by mntput()) 22091da177e4SLinus Torvalds */ 22106776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2211863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 22121ab59738SAl Viro propagate_mount_busy(mnt, 1)) 22131da177e4SLinus Torvalds continue; 22146776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 22151da177e4SLinus Torvalds } 2216bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 22176776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2218143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2219328e6d90SAl Viro umount_tree(mnt, 1); 2220bcc5c7d2SAl Viro } 2221719ea2fbSAl Viro unlock_mount_hash(); 22223ab6abeeSAl Viro namespace_unlock(); 22231da177e4SLinus Torvalds } 22241da177e4SLinus Torvalds 22251da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 22261da177e4SLinus Torvalds 22271da177e4SLinus Torvalds /* 22285528f911STrond Myklebust * Ripoff of 'select_parent()' 22295528f911STrond Myklebust * 22305528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 22315528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 22325528f911STrond Myklebust */ 2233692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 22345528f911STrond Myklebust { 2235692afc31SAl Viro struct mount *this_parent = parent; 22365528f911STrond Myklebust struct list_head *next; 22375528f911STrond Myklebust int found = 0; 22385528f911STrond Myklebust 22395528f911STrond Myklebust repeat: 22406b41d536SAl Viro next = this_parent->mnt_mounts.next; 22415528f911STrond Myklebust resume: 22426b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 22435528f911STrond Myklebust struct list_head *tmp = next; 22446b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 22455528f911STrond Myklebust 22465528f911STrond Myklebust next = tmp->next; 2247692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 22485528f911STrond Myklebust continue; 22495528f911STrond Myklebust /* 22505528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 22515528f911STrond Myklebust */ 22526b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 22535528f911STrond Myklebust this_parent = mnt; 22545528f911STrond Myklebust goto repeat; 22555528f911STrond Myklebust } 22565528f911STrond Myklebust 22571ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 22586776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 22595528f911STrond Myklebust found++; 22605528f911STrond Myklebust } 22615528f911STrond Myklebust } 22625528f911STrond Myklebust /* 22635528f911STrond Myklebust * All done at this level ... ascend and resume the search 22645528f911STrond Myklebust */ 22655528f911STrond Myklebust if (this_parent != parent) { 22666b41d536SAl Viro next = this_parent->mnt_child.next; 22670714a533SAl Viro this_parent = this_parent->mnt_parent; 22685528f911STrond Myklebust goto resume; 22695528f911STrond Myklebust } 22705528f911STrond Myklebust return found; 22715528f911STrond Myklebust } 22725528f911STrond Myklebust 22735528f911STrond Myklebust /* 22745528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 22755528f911STrond Myklebust * submounts of a specific parent mountpoint 227699b7db7bSNick Piggin * 227748a066e7SAl Viro * mount_lock must be held for write 22785528f911STrond Myklebust */ 2279b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 22805528f911STrond Myklebust { 22815528f911STrond Myklebust LIST_HEAD(graveyard); 2282761d5c38SAl Viro struct mount *m; 22835528f911STrond Myklebust 22845528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2285c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2286bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2287761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 22886776db3dSAl Viro mnt_expire); 2289143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2290328e6d90SAl Viro umount_tree(m, 1); 2291bcc5c7d2SAl Viro } 2292bcc5c7d2SAl Viro } 22935528f911STrond Myklebust } 22945528f911STrond Myklebust 22955528f911STrond Myklebust /* 22961da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 22971da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 22981da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 22991da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 23001da177e4SLinus Torvalds */ 2301b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2302b58fed8bSRam Pai unsigned long n) 23031da177e4SLinus Torvalds { 23041da177e4SLinus Torvalds char *t = to; 23051da177e4SLinus Torvalds const char __user *f = from; 23061da177e4SLinus Torvalds char c; 23071da177e4SLinus Torvalds 23081da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 23091da177e4SLinus Torvalds return n; 23101da177e4SLinus Torvalds 23111da177e4SLinus Torvalds while (n) { 23121da177e4SLinus Torvalds if (__get_user(c, f)) { 23131da177e4SLinus Torvalds memset(t, 0, n); 23141da177e4SLinus Torvalds break; 23151da177e4SLinus Torvalds } 23161da177e4SLinus Torvalds *t++ = c; 23171da177e4SLinus Torvalds f++; 23181da177e4SLinus Torvalds n--; 23191da177e4SLinus Torvalds } 23201da177e4SLinus Torvalds return n; 23211da177e4SLinus Torvalds } 23221da177e4SLinus Torvalds 23231da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 23241da177e4SLinus Torvalds { 23251da177e4SLinus Torvalds int i; 23261da177e4SLinus Torvalds unsigned long page; 23271da177e4SLinus Torvalds unsigned long size; 23281da177e4SLinus Torvalds 23291da177e4SLinus Torvalds *where = 0; 23301da177e4SLinus Torvalds if (!data) 23311da177e4SLinus Torvalds return 0; 23321da177e4SLinus Torvalds 23331da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 23341da177e4SLinus Torvalds return -ENOMEM; 23351da177e4SLinus Torvalds 23361da177e4SLinus Torvalds /* We only care that *some* data at the address the user 23371da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 23381da177e4SLinus Torvalds * the remainder of the page. 23391da177e4SLinus Torvalds */ 23401da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 23411da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 23421da177e4SLinus Torvalds if (size > PAGE_SIZE) 23431da177e4SLinus Torvalds size = PAGE_SIZE; 23441da177e4SLinus Torvalds 23451da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 23461da177e4SLinus Torvalds if (!i) { 23471da177e4SLinus Torvalds free_page(page); 23481da177e4SLinus Torvalds return -EFAULT; 23491da177e4SLinus Torvalds } 23501da177e4SLinus Torvalds if (i != PAGE_SIZE) 23511da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 23521da177e4SLinus Torvalds *where = page; 23531da177e4SLinus Torvalds return 0; 23541da177e4SLinus Torvalds } 23551da177e4SLinus Torvalds 2356eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 2357eca6f534SVegard Nossum { 2358eca6f534SVegard Nossum char *tmp; 2359eca6f534SVegard Nossum 2360eca6f534SVegard Nossum if (!data) { 2361eca6f534SVegard Nossum *where = NULL; 2362eca6f534SVegard Nossum return 0; 2363eca6f534SVegard Nossum } 2364eca6f534SVegard Nossum 2365eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 2366eca6f534SVegard Nossum if (IS_ERR(tmp)) 2367eca6f534SVegard Nossum return PTR_ERR(tmp); 2368eca6f534SVegard Nossum 2369eca6f534SVegard Nossum *where = tmp; 2370eca6f534SVegard Nossum return 0; 2371eca6f534SVegard Nossum } 2372eca6f534SVegard Nossum 23731da177e4SLinus Torvalds /* 23741da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 23751da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 23761da177e4SLinus Torvalds * 23771da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 23781da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 23791da177e4SLinus Torvalds * information (or be NULL). 23801da177e4SLinus Torvalds * 23811da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 23821da177e4SLinus Torvalds * When the flags word was introduced its top half was required 23831da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 23841da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 23851da177e4SLinus Torvalds * and must be discarded. 23861da177e4SLinus Torvalds */ 2387808d4e3cSAl Viro long do_mount(const char *dev_name, const char *dir_name, 2388808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 23891da177e4SLinus Torvalds { 23902d92ab3cSAl Viro struct path path; 23911da177e4SLinus Torvalds int retval = 0; 23921da177e4SLinus Torvalds int mnt_flags = 0; 23931da177e4SLinus Torvalds 23941da177e4SLinus Torvalds /* Discard magic */ 23951da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 23961da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 23971da177e4SLinus Torvalds 23981da177e4SLinus Torvalds /* Basic sanity checks */ 23991da177e4SLinus Torvalds 24001da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 24011da177e4SLinus Torvalds return -EINVAL; 24021da177e4SLinus Torvalds 24031da177e4SLinus Torvalds if (data_page) 24041da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 24051da177e4SLinus Torvalds 2406a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 2407a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2408a27ab9f2STetsuo Handa if (retval) 2409a27ab9f2STetsuo Handa return retval; 2410a27ab9f2STetsuo Handa 2411a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2412a27ab9f2STetsuo Handa type_page, flags, data_page); 24130d5cadb8SAl Viro if (!retval && !may_mount()) 24140d5cadb8SAl Viro retval = -EPERM; 2415a27ab9f2STetsuo Handa if (retval) 2416a27ab9f2STetsuo Handa goto dput_out; 2417a27ab9f2STetsuo Handa 2418613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2419613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 24200a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 24210a1c01c9SMatthew Garrett 24221da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 24231da177e4SLinus Torvalds if (flags & MS_NOSUID) 24241da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 24251da177e4SLinus Torvalds if (flags & MS_NODEV) 24261da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 24271da177e4SLinus Torvalds if (flags & MS_NOEXEC) 24281da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2429fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2430fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2431fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2432fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2433d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2434d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 24352e4b7fcdSDave Hansen if (flags & MS_RDONLY) 24362e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2437fc33a7bbSChristoph Hellwig 24387a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2439d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2440d0adde57SMatthew Garrett MS_STRICTATIME); 24411da177e4SLinus Torvalds 24421da177e4SLinus Torvalds if (flags & MS_REMOUNT) 24432d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 24441da177e4SLinus Torvalds data_page); 24451da177e4SLinus Torvalds else if (flags & MS_BIND) 24462d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 24479676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 24482d92ab3cSAl Viro retval = do_change_type(&path, flags); 24491da177e4SLinus Torvalds else if (flags & MS_MOVE) 24502d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 24511da177e4SLinus Torvalds else 24522d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 24531da177e4SLinus Torvalds dev_name, data_page); 24541da177e4SLinus Torvalds dput_out: 24552d92ab3cSAl Viro path_put(&path); 24561da177e4SLinus Torvalds return retval; 24571da177e4SLinus Torvalds } 24581da177e4SLinus Torvalds 2459771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2460771b1371SEric W. Biederman { 246198f842e6SEric W. Biederman proc_free_inum(ns->proc_inum); 2462771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2463771b1371SEric W. Biederman kfree(ns); 2464771b1371SEric W. Biederman } 2465771b1371SEric W. Biederman 24668823c079SEric W. Biederman /* 24678823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 24688823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 24698823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 24708823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 24718823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 24728823c079SEric W. Biederman */ 24738823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 24748823c079SEric W. Biederman 2475771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2476cf8d2c11STrond Myklebust { 2477cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 247898f842e6SEric W. Biederman int ret; 2479cf8d2c11STrond Myklebust 2480cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2481cf8d2c11STrond Myklebust if (!new_ns) 2482cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 248398f842e6SEric W. Biederman ret = proc_alloc_inum(&new_ns->proc_inum); 248498f842e6SEric W. Biederman if (ret) { 248598f842e6SEric W. Biederman kfree(new_ns); 248698f842e6SEric W. Biederman return ERR_PTR(ret); 248798f842e6SEric W. Biederman } 24888823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2489cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2490cf8d2c11STrond Myklebust new_ns->root = NULL; 2491cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2492cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2493cf8d2c11STrond Myklebust new_ns->event = 0; 2494771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2495cf8d2c11STrond Myklebust return new_ns; 2496cf8d2c11STrond Myklebust } 2497cf8d2c11STrond Myklebust 24989559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 24999559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 25001da177e4SLinus Torvalds { 25016b3286edSKirill Korotaev struct mnt_namespace *new_ns; 25027f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2503315fc83eSAl Viro struct mount *p, *q; 25049559f689SAl Viro struct mount *old; 2505cb338d06SAl Viro struct mount *new; 25067a472ef4SEric W. Biederman int copy_flags; 25071da177e4SLinus Torvalds 25089559f689SAl Viro BUG_ON(!ns); 25099559f689SAl Viro 25109559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 25119559f689SAl Viro get_mnt_ns(ns); 25129559f689SAl Viro return ns; 25139559f689SAl Viro } 25149559f689SAl Viro 25159559f689SAl Viro old = ns->root; 25169559f689SAl Viro 2517771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2518cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2519cf8d2c11STrond Myklebust return new_ns; 25201da177e4SLinus Torvalds 252197216be0SAl Viro namespace_lock(); 25221da177e4SLinus Torvalds /* First pass: copy the tree topology */ 25234ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 25249559f689SAl Viro if (user_ns != ns->user_ns) 2525132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 25267a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2527be34d1a3SDavid Howells if (IS_ERR(new)) { 2528328e6d90SAl Viro namespace_unlock(); 2529771b1371SEric W. Biederman free_mnt_ns(new_ns); 2530be34d1a3SDavid Howells return ERR_CAST(new); 25311da177e4SLinus Torvalds } 2532be08d6d2SAl Viro new_ns->root = new; 25331a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 25341da177e4SLinus Torvalds 25351da177e4SLinus Torvalds /* 25361da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 25371da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 25381da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 25391da177e4SLinus Torvalds */ 2540909b0a88SAl Viro p = old; 2541cb338d06SAl Viro q = new; 25421da177e4SLinus Torvalds while (p) { 2543143c8c91SAl Viro q->mnt_ns = new_ns; 25449559f689SAl Viro if (new_fs) { 25459559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 25469559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2547315fc83eSAl Viro rootmnt = &p->mnt; 25481da177e4SLinus Torvalds } 25499559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 25509559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2551315fc83eSAl Viro pwdmnt = &p->mnt; 25521da177e4SLinus Torvalds } 25531da177e4SLinus Torvalds } 2554909b0a88SAl Viro p = next_mnt(p, old); 2555909b0a88SAl Viro q = next_mnt(q, new); 25564ce5d2b1SEric W. Biederman if (!q) 25574ce5d2b1SEric W. Biederman break; 25584ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 25594ce5d2b1SEric W. Biederman p = next_mnt(p, old); 25601da177e4SLinus Torvalds } 2561328e6d90SAl Viro namespace_unlock(); 25621da177e4SLinus Torvalds 25631da177e4SLinus Torvalds if (rootmnt) 2564f03c6599SAl Viro mntput(rootmnt); 25651da177e4SLinus Torvalds if (pwdmnt) 2566f03c6599SAl Viro mntput(pwdmnt); 25671da177e4SLinus Torvalds 2568741a2951SJANAK DESAI return new_ns; 2569741a2951SJANAK DESAI } 2570741a2951SJANAK DESAI 2571cf8d2c11STrond Myklebust /** 2572cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2573cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2574cf8d2c11STrond Myklebust */ 25751a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2576cf8d2c11STrond Myklebust { 2577771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2578cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 25791a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 25801a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2581be08d6d2SAl Viro new_ns->root = mnt; 2582b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2583c1334495SAl Viro } else { 25841a4eeaf2SAl Viro mntput(m); 2585cf8d2c11STrond Myklebust } 2586cf8d2c11STrond Myklebust return new_ns; 2587cf8d2c11STrond Myklebust } 2588cf8d2c11STrond Myklebust 2589ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2590ea441d11SAl Viro { 2591ea441d11SAl Viro struct mnt_namespace *ns; 2592d31da0f0SAl Viro struct super_block *s; 2593ea441d11SAl Viro struct path path; 2594ea441d11SAl Viro int err; 2595ea441d11SAl Viro 2596ea441d11SAl Viro ns = create_mnt_ns(mnt); 2597ea441d11SAl Viro if (IS_ERR(ns)) 2598ea441d11SAl Viro return ERR_CAST(ns); 2599ea441d11SAl Viro 2600ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2601ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2602ea441d11SAl Viro 2603ea441d11SAl Viro put_mnt_ns(ns); 2604ea441d11SAl Viro 2605ea441d11SAl Viro if (err) 2606ea441d11SAl Viro return ERR_PTR(err); 2607ea441d11SAl Viro 2608ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2609d31da0f0SAl Viro s = path.mnt->mnt_sb; 2610d31da0f0SAl Viro atomic_inc(&s->s_active); 2611ea441d11SAl Viro mntput(path.mnt); 2612ea441d11SAl Viro /* lock the sucker */ 2613d31da0f0SAl Viro down_write(&s->s_umount); 2614ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2615ea441d11SAl Viro return path.dentry; 2616ea441d11SAl Viro } 2617ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2618ea441d11SAl Viro 2619bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2620bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 26211da177e4SLinus Torvalds { 2622eca6f534SVegard Nossum int ret; 2623eca6f534SVegard Nossum char *kernel_type; 262491a27b2aSJeff Layton struct filename *kernel_dir; 2625eca6f534SVegard Nossum char *kernel_dev; 26261da177e4SLinus Torvalds unsigned long data_page; 26271da177e4SLinus Torvalds 2628eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2629eca6f534SVegard Nossum if (ret < 0) 2630eca6f534SVegard Nossum goto out_type; 26311da177e4SLinus Torvalds 2632eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2633eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2634eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2635eca6f534SVegard Nossum goto out_dir; 2636eca6f534SVegard Nossum } 26371da177e4SLinus Torvalds 2638eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2639eca6f534SVegard Nossum if (ret < 0) 2640eca6f534SVegard Nossum goto out_dev; 26411da177e4SLinus Torvalds 2642eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2643eca6f534SVegard Nossum if (ret < 0) 2644eca6f534SVegard Nossum goto out_data; 26451da177e4SLinus Torvalds 264691a27b2aSJeff Layton ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, 2647eca6f534SVegard Nossum (void *) data_page); 2648eca6f534SVegard Nossum 26491da177e4SLinus Torvalds free_page(data_page); 2650eca6f534SVegard Nossum out_data: 2651eca6f534SVegard Nossum kfree(kernel_dev); 2652eca6f534SVegard Nossum out_dev: 2653eca6f534SVegard Nossum putname(kernel_dir); 2654eca6f534SVegard Nossum out_dir: 2655eca6f534SVegard Nossum kfree(kernel_type); 2656eca6f534SVegard Nossum out_type: 2657eca6f534SVegard Nossum return ret; 26581da177e4SLinus Torvalds } 26591da177e4SLinus Torvalds 26601da177e4SLinus Torvalds /* 2661afac7cbaSAl Viro * Return true if path is reachable from root 2662afac7cbaSAl Viro * 266348a066e7SAl Viro * namespace_sem or mount_lock is held 2664afac7cbaSAl Viro */ 2665643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2666afac7cbaSAl Viro const struct path *root) 2667afac7cbaSAl Viro { 2668643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2669a73324daSAl Viro dentry = mnt->mnt_mountpoint; 26700714a533SAl Viro mnt = mnt->mnt_parent; 2671afac7cbaSAl Viro } 2672643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2673afac7cbaSAl Viro } 2674afac7cbaSAl Viro 2675afac7cbaSAl Viro int path_is_under(struct path *path1, struct path *path2) 2676afac7cbaSAl Viro { 2677afac7cbaSAl Viro int res; 267848a066e7SAl Viro read_seqlock_excl(&mount_lock); 2679643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 268048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 2681afac7cbaSAl Viro return res; 2682afac7cbaSAl Viro } 2683afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 2684afac7cbaSAl Viro 2685afac7cbaSAl Viro /* 26861da177e4SLinus Torvalds * pivot_root Semantics: 26871da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 26881da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 26891da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 26901da177e4SLinus Torvalds * 26911da177e4SLinus Torvalds * Restrictions: 26921da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 26931da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 26941da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 26951da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 26961da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 26971da177e4SLinus Torvalds * 26984a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 26994a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 27004a0d11faSNeil Brown * in this situation. 27014a0d11faSNeil Brown * 27021da177e4SLinus Torvalds * Notes: 27031da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 27041da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 27051da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 27061da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 27071da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 27081da177e4SLinus Torvalds * first. 27091da177e4SLinus Torvalds */ 27103480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 27113480b257SHeiko Carstens const char __user *, put_old) 27121da177e4SLinus Torvalds { 27132d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 271484d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 271584d17192SAl Viro struct mountpoint *old_mp, *root_mp; 27161da177e4SLinus Torvalds int error; 27171da177e4SLinus Torvalds 27189b40bc90SAl Viro if (!may_mount()) 27191da177e4SLinus Torvalds return -EPERM; 27201da177e4SLinus Torvalds 27212d8f3038SAl Viro error = user_path_dir(new_root, &new); 27221da177e4SLinus Torvalds if (error) 27231da177e4SLinus Torvalds goto out0; 27241da177e4SLinus Torvalds 27252d8f3038SAl Viro error = user_path_dir(put_old, &old); 27261da177e4SLinus Torvalds if (error) 27271da177e4SLinus Torvalds goto out1; 27281da177e4SLinus Torvalds 27292d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 2730b12cea91SAl Viro if (error) 2731b12cea91SAl Viro goto out2; 27321da177e4SLinus Torvalds 2733f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 273484d17192SAl Viro old_mp = lock_mount(&old); 273584d17192SAl Viro error = PTR_ERR(old_mp); 273684d17192SAl Viro if (IS_ERR(old_mp)) 2737b12cea91SAl Viro goto out3; 2738b12cea91SAl Viro 27391da177e4SLinus Torvalds error = -EINVAL; 2740419148daSAl Viro new_mnt = real_mount(new.mnt); 2741419148daSAl Viro root_mnt = real_mount(root.mnt); 274284d17192SAl Viro old_mnt = real_mount(old.mnt); 274384d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 2744fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 2745fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 2746b12cea91SAl Viro goto out4; 2747143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 2748b12cea91SAl Viro goto out4; 27495ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 27505ff9d8a6SEric W. Biederman goto out4; 27511da177e4SLinus Torvalds error = -ENOENT; 2752f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 2753b12cea91SAl Viro goto out4; 27541da177e4SLinus Torvalds error = -EBUSY; 275584d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 2756b12cea91SAl Viro goto out4; /* loop, on the same file system */ 27571da177e4SLinus Torvalds error = -EINVAL; 27588c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 2759b12cea91SAl Viro goto out4; /* not a mountpoint */ 2760676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 2761b12cea91SAl Viro goto out4; /* not attached */ 276284d17192SAl Viro root_mp = root_mnt->mnt_mp; 27632d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 2764b12cea91SAl Viro goto out4; /* not a mountpoint */ 2765676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 2766b12cea91SAl Viro goto out4; /* not attached */ 27674ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 276884d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 2769b12cea91SAl Viro goto out4; 277084d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 2771719ea2fbSAl Viro lock_mount_hash(); 2772419148daSAl Viro detach_mnt(new_mnt, &parent_path); 2773419148daSAl Viro detach_mnt(root_mnt, &root_parent); 27745ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 27755ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 27765ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 27775ff9d8a6SEric W. Biederman } 27784ac91378SJan Blunck /* mount old root on put_old */ 277984d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 27804ac91378SJan Blunck /* mount new_root on / */ 278184d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 27826b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 2783719ea2fbSAl Viro unlock_mount_hash(); 27842d8f3038SAl Viro chroot_fs_refs(&root, &new); 278584d17192SAl Viro put_mountpoint(root_mp); 27861da177e4SLinus Torvalds error = 0; 2787b12cea91SAl Viro out4: 278884d17192SAl Viro unlock_mount(old_mp); 2789b12cea91SAl Viro if (!error) { 27901a390689SAl Viro path_put(&root_parent); 27911a390689SAl Viro path_put(&parent_path); 2792b12cea91SAl Viro } 2793b12cea91SAl Viro out3: 27948c3ee42eSAl Viro path_put(&root); 2795b12cea91SAl Viro out2: 27962d8f3038SAl Viro path_put(&old); 27971da177e4SLinus Torvalds out1: 27982d8f3038SAl Viro path_put(&new); 27991da177e4SLinus Torvalds out0: 28001da177e4SLinus Torvalds return error; 28011da177e4SLinus Torvalds } 28021da177e4SLinus Torvalds 28031da177e4SLinus Torvalds static void __init init_mount_tree(void) 28041da177e4SLinus Torvalds { 28051da177e4SLinus Torvalds struct vfsmount *mnt; 28066b3286edSKirill Korotaev struct mnt_namespace *ns; 2807ac748a09SJan Blunck struct path root; 28080c55cfc4SEric W. Biederman struct file_system_type *type; 28091da177e4SLinus Torvalds 28100c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 28110c55cfc4SEric W. Biederman if (!type) 28120c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 28130c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 28140c55cfc4SEric W. Biederman put_filesystem(type); 28151da177e4SLinus Torvalds if (IS_ERR(mnt)) 28161da177e4SLinus Torvalds panic("Can't create rootfs"); 2817b3e19d92SNick Piggin 28183b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 28193b22edc5STrond Myklebust if (IS_ERR(ns)) 28201da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 28211da177e4SLinus Torvalds 28226b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 28236b3286edSKirill Korotaev get_mnt_ns(ns); 28241da177e4SLinus Torvalds 2825be08d6d2SAl Viro root.mnt = mnt; 2826be08d6d2SAl Viro root.dentry = mnt->mnt_root; 2827ac748a09SJan Blunck 2828ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2829ac748a09SJan Blunck set_fs_root(current->fs, &root); 28301da177e4SLinus Torvalds } 28311da177e4SLinus Torvalds 283274bf17cfSDenis Cheng void __init mnt_init(void) 28331da177e4SLinus Torvalds { 283413f14b4dSEric Dumazet unsigned u; 283515a67dd8SRandy Dunlap int err; 28361da177e4SLinus Torvalds 28377d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 283820c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 28391da177e4SLinus Torvalds 28400818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 284138129a13SAl Viro sizeof(struct hlist_head), 28420818bf27SAl Viro mhash_entries, 19, 28430818bf27SAl Viro 0, 28440818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 28450818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 28460818bf27SAl Viro sizeof(struct hlist_head), 28470818bf27SAl Viro mphash_entries, 19, 28480818bf27SAl Viro 0, 28490818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 28501da177e4SLinus Torvalds 285184d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 28521da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 28531da177e4SLinus Torvalds 28540818bf27SAl Viro for (u = 0; u <= m_hash_mask; u++) 285538129a13SAl Viro INIT_HLIST_HEAD(&mount_hashtable[u]); 28560818bf27SAl Viro for (u = 0; u <= mp_hash_mask; u++) 28570818bf27SAl Viro INIT_HLIST_HEAD(&mountpoint_hashtable[u]); 28581da177e4SLinus Torvalds 28594b93dc9bSTejun Heo kernfs_init(); 28604b93dc9bSTejun Heo 286115a67dd8SRandy Dunlap err = sysfs_init(); 286215a67dd8SRandy Dunlap if (err) 286315a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 28648e24eea7SHarvey Harrison __func__, err); 286500d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 286600d26666SGreg Kroah-Hartman if (!fs_kobj) 28678e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 28681da177e4SLinus Torvalds init_rootfs(); 28691da177e4SLinus Torvalds init_mount_tree(); 28701da177e4SLinus Torvalds } 28711da177e4SLinus Torvalds 2872616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 28731da177e4SLinus Torvalds { 2874d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 2875616511d0STrond Myklebust return; 28767b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 2877771b1371SEric W. Biederman free_mnt_ns(ns); 28781da177e4SLinus Torvalds } 28799d412a43SAl Viro 28809d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 28819d412a43SAl Viro { 2882423e0ab0STim Chen struct vfsmount *mnt; 2883423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 2884423e0ab0STim Chen if (!IS_ERR(mnt)) { 2885423e0ab0STim Chen /* 2886423e0ab0STim Chen * it is a longterm mount, don't release mnt until 2887423e0ab0STim Chen * we unmount before file sys is unregistered 2888423e0ab0STim Chen */ 2889f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 2890423e0ab0STim Chen } 2891423e0ab0STim Chen return mnt; 28929d412a43SAl Viro } 28939d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 2894423e0ab0STim Chen 2895423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 2896423e0ab0STim Chen { 2897423e0ab0STim Chen /* release long term mount so mount point can be released */ 2898423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 2899f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 290048a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 2901423e0ab0STim Chen mntput(mnt); 2902423e0ab0STim Chen } 2903423e0ab0STim Chen } 2904423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 290502125a82SAl Viro 290602125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 290702125a82SAl Viro { 2908143c8c91SAl Viro return check_mnt(real_mount(mnt)); 290902125a82SAl Viro } 29108823c079SEric W. Biederman 29113151527eSEric W. Biederman bool current_chrooted(void) 29123151527eSEric W. Biederman { 29133151527eSEric W. Biederman /* Does the current process have a non-standard root */ 29143151527eSEric W. Biederman struct path ns_root; 29153151527eSEric W. Biederman struct path fs_root; 29163151527eSEric W. Biederman bool chrooted; 29173151527eSEric W. Biederman 29183151527eSEric W. Biederman /* Find the namespace root */ 29193151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 29203151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 29213151527eSEric W. Biederman path_get(&ns_root); 29223151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 29233151527eSEric W. Biederman ; 29243151527eSEric W. Biederman 29253151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 29263151527eSEric W. Biederman 29273151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 29283151527eSEric W. Biederman 29293151527eSEric W. Biederman path_put(&fs_root); 29303151527eSEric W. Biederman path_put(&ns_root); 29313151527eSEric W. Biederman 29323151527eSEric W. Biederman return chrooted; 29333151527eSEric W. Biederman } 29343151527eSEric W. Biederman 2935e51db735SEric W. Biederman bool fs_fully_visible(struct file_system_type *type) 293687a8ebd6SEric W. Biederman { 293787a8ebd6SEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 293887a8ebd6SEric W. Biederman struct mount *mnt; 2939e51db735SEric W. Biederman bool visible = false; 294087a8ebd6SEric W. Biederman 2941e51db735SEric W. Biederman if (unlikely(!ns)) 2942e51db735SEric W. Biederman return false; 2943e51db735SEric W. Biederman 294444bb4385SAl Viro down_read(&namespace_sem); 294587a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 2946e51db735SEric W. Biederman struct mount *child; 2947e51db735SEric W. Biederman if (mnt->mnt.mnt_sb->s_type != type) 2948e51db735SEric W. Biederman continue; 2949e51db735SEric W. Biederman 2950e51db735SEric W. Biederman /* This mount is not fully visible if there are any child mounts 2951e51db735SEric W. Biederman * that cover anything except for empty directories. 2952e51db735SEric W. Biederman */ 2953e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 2954e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 2955e51db735SEric W. Biederman if (!S_ISDIR(inode->i_mode)) 2956e51db735SEric W. Biederman goto next; 295741301ae7SEric W. Biederman if (inode->i_nlink > 2) 2958e51db735SEric W. Biederman goto next; 295987a8ebd6SEric W. Biederman } 2960e51db735SEric W. Biederman visible = true; 2961e51db735SEric W. Biederman goto found; 2962e51db735SEric W. Biederman next: ; 296387a8ebd6SEric W. Biederman } 2964e51db735SEric W. Biederman found: 296544bb4385SAl Viro up_read(&namespace_sem); 2966e51db735SEric W. Biederman return visible; 296787a8ebd6SEric W. Biederman } 296887a8ebd6SEric W. Biederman 29698823c079SEric W. Biederman static void *mntns_get(struct task_struct *task) 29708823c079SEric W. Biederman { 29718823c079SEric W. Biederman struct mnt_namespace *ns = NULL; 29728823c079SEric W. Biederman struct nsproxy *nsproxy; 29738823c079SEric W. Biederman 29748823c079SEric W. Biederman rcu_read_lock(); 29758823c079SEric W. Biederman nsproxy = task_nsproxy(task); 29768823c079SEric W. Biederman if (nsproxy) { 29778823c079SEric W. Biederman ns = nsproxy->mnt_ns; 29788823c079SEric W. Biederman get_mnt_ns(ns); 29798823c079SEric W. Biederman } 29808823c079SEric W. Biederman rcu_read_unlock(); 29818823c079SEric W. Biederman 29828823c079SEric W. Biederman return ns; 29838823c079SEric W. Biederman } 29848823c079SEric W. Biederman 29858823c079SEric W. Biederman static void mntns_put(void *ns) 29868823c079SEric W. Biederman { 29878823c079SEric W. Biederman put_mnt_ns(ns); 29888823c079SEric W. Biederman } 29898823c079SEric W. Biederman 29908823c079SEric W. Biederman static int mntns_install(struct nsproxy *nsproxy, void *ns) 29918823c079SEric W. Biederman { 29928823c079SEric W. Biederman struct fs_struct *fs = current->fs; 29938823c079SEric W. Biederman struct mnt_namespace *mnt_ns = ns; 29948823c079SEric W. Biederman struct path root; 29958823c079SEric W. Biederman 29960c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 2997c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 2998c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 2999ae11e0f1SZhao Hongjiang return -EPERM; 30008823c079SEric W. Biederman 30018823c079SEric W. Biederman if (fs->users != 1) 30028823c079SEric W. Biederman return -EINVAL; 30038823c079SEric W. Biederman 30048823c079SEric W. Biederman get_mnt_ns(mnt_ns); 30058823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 30068823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 30078823c079SEric W. Biederman 30088823c079SEric W. Biederman /* Find the root */ 30098823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 30108823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 30118823c079SEric W. Biederman path_get(&root); 30128823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 30138823c079SEric W. Biederman ; 30148823c079SEric W. Biederman 30158823c079SEric W. Biederman /* Update the pwd and root */ 30168823c079SEric W. Biederman set_fs_pwd(fs, &root); 30178823c079SEric W. Biederman set_fs_root(fs, &root); 30188823c079SEric W. Biederman 30198823c079SEric W. Biederman path_put(&root); 30208823c079SEric W. Biederman return 0; 30218823c079SEric W. Biederman } 30228823c079SEric W. Biederman 302398f842e6SEric W. Biederman static unsigned int mntns_inum(void *ns) 302498f842e6SEric W. Biederman { 302598f842e6SEric W. Biederman struct mnt_namespace *mnt_ns = ns; 302698f842e6SEric W. Biederman return mnt_ns->proc_inum; 302798f842e6SEric W. Biederman } 302898f842e6SEric W. Biederman 30298823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 30308823c079SEric W. Biederman .name = "mnt", 30318823c079SEric W. Biederman .type = CLONE_NEWNS, 30328823c079SEric W. Biederman .get = mntns_get, 30338823c079SEric W. Biederman .put = mntns_put, 30348823c079SEric W. Biederman .install = mntns_install, 303598f842e6SEric W. Biederman .inum = mntns_inum, 30368823c079SEric W. Biederman }; 3037