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> 185b825c3aSIngo Molnar #include <linux/cred.h> 1973cd49ecSMiklos Szeredi #include <linux/idr.h> 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23d10577a8SAl Viro #include <linux/uaccess.h> 240bb80f24SDavid Howells #include <linux/proc_ns.h> 2520b4fb48SLinus Torvalds #include <linux/magic.h> 2657c8a661SMike Rapoport #include <linux/memblock.h> 279ea459e1SAl Viro #include <linux/task_work.h> 289164bb4aSIngo Molnar #include <linux/sched/task.h> 29e262e32dSDavid Howells #include <uapi/linux/mount.h> 309bc61ab1SDavid Howells #include <linux/fs_context.h> 319164bb4aSIngo Molnar 3207b20889SRam Pai #include "pnode.h" 33948730b0SAdrian Bunk #include "internal.h" 341da177e4SLinus Torvalds 35d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 36d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 37d2921684SEric W. Biederman 380818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 390818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 400818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 410818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 420818bf27SAl Viro 430818bf27SAl Viro static __initdata unsigned long mhash_entries; 440818bf27SAl Viro static int __init set_mhash_entries(char *str) 450818bf27SAl Viro { 460818bf27SAl Viro if (!str) 470818bf27SAl Viro return 0; 480818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 490818bf27SAl Viro return 1; 500818bf27SAl Viro } 510818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 520818bf27SAl Viro 530818bf27SAl Viro static __initdata unsigned long mphash_entries; 540818bf27SAl Viro static int __init set_mphash_entries(char *str) 550818bf27SAl Viro { 560818bf27SAl Viro if (!str) 570818bf27SAl Viro return 0; 580818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 590818bf27SAl Viro return 1; 600818bf27SAl Viro } 610818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6213f14b4dSEric Dumazet 63c7999c36SAl Viro static u64 event; 6473cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 65719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 665addc5ddSAl Viro 6738129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 680818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 69e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 7059aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 711da177e4SLinus Torvalds 72f87fd4c2SMiklos Szeredi /* /sys/fs */ 7300d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7400d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 75f87fd4c2SMiklos Szeredi 7699b7db7bSNick Piggin /* 7799b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7899b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 7999b7db7bSNick Piggin * up the tree. 8099b7db7bSNick Piggin * 8199b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8299b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8399b7db7bSNick Piggin */ 8448a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8599b7db7bSNick Piggin 8638129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 871da177e4SLinus Torvalds { 881da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 891da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 900818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 910818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 920818bf27SAl Viro } 930818bf27SAl Viro 940818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 950818bf27SAl Viro { 960818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 970818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 980818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 991da177e4SLinus Torvalds } 1001da177e4SLinus Torvalds 101b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10273cd49ecSMiklos Szeredi { 103169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 10473cd49ecSMiklos Szeredi 105169b480eSMatthew Wilcox if (res < 0) 10673cd49ecSMiklos Szeredi return res; 107169b480eSMatthew Wilcox mnt->mnt_id = res; 108169b480eSMatthew Wilcox return 0; 10973cd49ecSMiklos Szeredi } 11073cd49ecSMiklos Szeredi 111b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11273cd49ecSMiklos Szeredi { 113169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 11473cd49ecSMiklos Szeredi } 11573cd49ecSMiklos Szeredi 116719f5d7fSMiklos Szeredi /* 117719f5d7fSMiklos Szeredi * Allocate a new peer group ID 118719f5d7fSMiklos Szeredi */ 1194b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 120719f5d7fSMiklos Szeredi { 121169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 122f21f6220SAl Viro 123169b480eSMatthew Wilcox if (res < 0) 124f21f6220SAl Viro return res; 125169b480eSMatthew Wilcox mnt->mnt_group_id = res; 126169b480eSMatthew Wilcox return 0; 127719f5d7fSMiklos Szeredi } 128719f5d7fSMiklos Szeredi 129719f5d7fSMiklos Szeredi /* 130719f5d7fSMiklos Szeredi * Release a peer group ID 131719f5d7fSMiklos Szeredi */ 1324b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 133719f5d7fSMiklos Szeredi { 134169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 13515169fe7SAl Viro mnt->mnt_group_id = 0; 136719f5d7fSMiklos Szeredi } 137719f5d7fSMiklos Szeredi 138b3e19d92SNick Piggin /* 139b3e19d92SNick Piggin * vfsmount lock must be held for read 140b3e19d92SNick Piggin */ 14183adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 142b3e19d92SNick Piggin { 143b3e19d92SNick Piggin #ifdef CONFIG_SMP 14468e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 145b3e19d92SNick Piggin #else 146b3e19d92SNick Piggin preempt_disable(); 14768e8a9feSAl Viro mnt->mnt_count += n; 148b3e19d92SNick Piggin preempt_enable(); 149b3e19d92SNick Piggin #endif 150b3e19d92SNick Piggin } 151b3e19d92SNick Piggin 152b3e19d92SNick Piggin /* 153b3e19d92SNick Piggin * vfsmount lock must be held for write 154b3e19d92SNick Piggin */ 15583adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 156b3e19d92SNick Piggin { 157b3e19d92SNick Piggin #ifdef CONFIG_SMP 158f03c6599SAl Viro unsigned int count = 0; 159b3e19d92SNick Piggin int cpu; 160b3e19d92SNick Piggin 161b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 16268e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 163b3e19d92SNick Piggin } 164b3e19d92SNick Piggin 165b3e19d92SNick Piggin return count; 166b3e19d92SNick Piggin #else 16768e8a9feSAl Viro return mnt->mnt_count; 168b3e19d92SNick Piggin #endif 169b3e19d92SNick Piggin } 170b3e19d92SNick Piggin 17187b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 17287b95ce0SAl Viro { 17387b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 17487b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 17587b95ce0SAl Viro pin_remove(p); 17687b95ce0SAl Viro mntput(&m->mnt); 17787b95ce0SAl Viro } 17887b95ce0SAl Viro 179b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1801da177e4SLinus Torvalds { 181c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 182c63181e6SAl Viro if (mnt) { 18373cd49ecSMiklos Szeredi int err; 18473cd49ecSMiklos Szeredi 185c63181e6SAl Viro err = mnt_alloc_id(mnt); 18688b38782SLi Zefan if (err) 18788b38782SLi Zefan goto out_free_cache; 18888b38782SLi Zefan 18988b38782SLi Zefan if (name) { 190fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 191c63181e6SAl Viro if (!mnt->mnt_devname) 19288b38782SLi Zefan goto out_free_id; 19373cd49ecSMiklos Szeredi } 19473cd49ecSMiklos Szeredi 195b3e19d92SNick Piggin #ifdef CONFIG_SMP 196c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 197c63181e6SAl Viro if (!mnt->mnt_pcp) 198b3e19d92SNick Piggin goto out_free_devname; 199b3e19d92SNick Piggin 200c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 201b3e19d92SNick Piggin #else 202c63181e6SAl Viro mnt->mnt_count = 1; 203c63181e6SAl Viro mnt->mnt_writers = 0; 204b3e19d92SNick Piggin #endif 205b3e19d92SNick Piggin 20638129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 207c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 208c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 209c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 210c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 211c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 212c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 213c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2140a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 21599b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 21687b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2171da177e4SLinus Torvalds } 218c63181e6SAl Viro return mnt; 21988b38782SLi Zefan 220d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 221d3ef3d73Snpiggin@suse.de out_free_devname: 222fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 223d3ef3d73Snpiggin@suse.de #endif 22488b38782SLi Zefan out_free_id: 225c63181e6SAl Viro mnt_free_id(mnt); 22688b38782SLi Zefan out_free_cache: 227c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 22888b38782SLi Zefan return NULL; 2291da177e4SLinus Torvalds } 2301da177e4SLinus Torvalds 2318366025eSDave Hansen /* 2328366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2338366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2348366025eSDave Hansen * We must keep track of when those operations start 2358366025eSDave Hansen * (for permission checks) and when they end, so that 2368366025eSDave Hansen * we can determine when writes are able to occur to 2378366025eSDave Hansen * a filesystem. 2388366025eSDave Hansen */ 2393d733633SDave Hansen /* 2403d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2413d733633SDave Hansen * @mnt: the mount to check for its write status 2423d733633SDave Hansen * 2433d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2443d733633SDave Hansen * It does not guarantee that the filesystem will stay 2453d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2463d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2473d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2483d733633SDave Hansen * r/w. 2493d733633SDave Hansen */ 25043f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 2513d733633SDave Hansen { 25243f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 2533d733633SDave Hansen } 2543d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2553d733633SDave Hansen 25683adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2573d733633SDave Hansen { 258d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 25968e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 260d3ef3d73Snpiggin@suse.de #else 26168e8a9feSAl Viro mnt->mnt_writers++; 262d3ef3d73Snpiggin@suse.de #endif 2633d733633SDave Hansen } 2643d733633SDave Hansen 26583adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2663d733633SDave Hansen { 267d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 26868e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 269d3ef3d73Snpiggin@suse.de #else 27068e8a9feSAl Viro mnt->mnt_writers--; 271d3ef3d73Snpiggin@suse.de #endif 272d3ef3d73Snpiggin@suse.de } 273d3ef3d73Snpiggin@suse.de 27483adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 275d3ef3d73Snpiggin@suse.de { 276d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 277d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2783d733633SDave Hansen int cpu; 2793d733633SDave Hansen 2803d733633SDave Hansen for_each_possible_cpu(cpu) { 28168e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2823d733633SDave Hansen } 2833d733633SDave Hansen 284d3ef3d73Snpiggin@suse.de return count; 285d3ef3d73Snpiggin@suse.de #else 286d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 287d3ef3d73Snpiggin@suse.de #endif 2883d733633SDave Hansen } 2893d733633SDave Hansen 2904ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2914ed5e82fSMiklos Szeredi { 2924ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2934ed5e82fSMiklos Szeredi return 1; 2944ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2954ed5e82fSMiklos Szeredi smp_rmb(); 2964ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2974ed5e82fSMiklos Szeredi } 2984ed5e82fSMiklos Szeredi 2993d733633SDave Hansen /* 300eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 301eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 302eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 303eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3043d733633SDave Hansen */ 3058366025eSDave Hansen /** 306eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 30783adc753SAl Viro * @m: the mount on which to take a write 3088366025eSDave Hansen * 309eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 310eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 311eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 312eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 313eb04c282SJan Kara * called. This is effectively a refcount. 3148366025eSDave Hansen */ 315eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3168366025eSDave Hansen { 31783adc753SAl Viro struct mount *mnt = real_mount(m); 3183d733633SDave Hansen int ret = 0; 3193d733633SDave Hansen 320d3ef3d73Snpiggin@suse.de preempt_disable(); 321c6653a83SNick Piggin mnt_inc_writers(mnt); 322d3ef3d73Snpiggin@suse.de /* 323c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 324d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 325d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 326d3ef3d73Snpiggin@suse.de */ 327d3ef3d73Snpiggin@suse.de smp_mb(); 3286aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 329d3ef3d73Snpiggin@suse.de cpu_relax(); 330d3ef3d73Snpiggin@suse.de /* 331d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 332d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 333d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 334d3ef3d73Snpiggin@suse.de */ 335d3ef3d73Snpiggin@suse.de smp_rmb(); 3364ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 337c6653a83SNick Piggin mnt_dec_writers(mnt); 3383d733633SDave Hansen ret = -EROFS; 3393d733633SDave Hansen } 340d3ef3d73Snpiggin@suse.de preempt_enable(); 341eb04c282SJan Kara 342eb04c282SJan Kara return ret; 343eb04c282SJan Kara } 344eb04c282SJan Kara 345eb04c282SJan Kara /** 346eb04c282SJan Kara * mnt_want_write - get write access to a mount 347eb04c282SJan Kara * @m: the mount on which to take a write 348eb04c282SJan Kara * 349eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 350eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 351eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 352eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 353eb04c282SJan Kara */ 354eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 355eb04c282SJan Kara { 356eb04c282SJan Kara int ret; 357eb04c282SJan Kara 358eb04c282SJan Kara sb_start_write(m->mnt_sb); 359eb04c282SJan Kara ret = __mnt_want_write(m); 360eb04c282SJan Kara if (ret) 361eb04c282SJan Kara sb_end_write(m->mnt_sb); 3623d733633SDave Hansen return ret; 3638366025eSDave Hansen } 3648366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3658366025eSDave Hansen 3668366025eSDave Hansen /** 36796029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 36896029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 36996029c4eSnpiggin@suse.de * 37096029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 37196029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 37296029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 37396029c4eSnpiggin@suse.de * on it. This allows some optimisation. 37496029c4eSnpiggin@suse.de * 37596029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 37696029c4eSnpiggin@suse.de * drop the reference. 37796029c4eSnpiggin@suse.de */ 37896029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 37996029c4eSnpiggin@suse.de { 38096029c4eSnpiggin@suse.de /* superblock may be r/o */ 38196029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 38296029c4eSnpiggin@suse.de return -EROFS; 38396029c4eSnpiggin@suse.de preempt_disable(); 38483adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 38596029c4eSnpiggin@suse.de preempt_enable(); 38696029c4eSnpiggin@suse.de return 0; 38796029c4eSnpiggin@suse.de } 38896029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 38996029c4eSnpiggin@suse.de 39096029c4eSnpiggin@suse.de /** 391eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 392eb04c282SJan Kara * @file: the file who's mount on which to take a write 393eb04c282SJan Kara * 394eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 395eb04c282SJan Kara * do some optimisations if the file is open for write already 396eb04c282SJan Kara */ 397eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 398eb04c282SJan Kara { 39983f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 400eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 401eb04c282SJan Kara else 402eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 403eb04c282SJan Kara } 404eb04c282SJan Kara 405eb04c282SJan Kara /** 4067c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4077c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4087c6893e3SMiklos Szeredi * 4097c6893e3SMiklos Szeredi * This is like mnt_want_write, but it takes a file and can 4107c6893e3SMiklos Szeredi * do some optimisations if the file is open for write already 4117c6893e3SMiklos Szeredi */ 4127c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4137c6893e3SMiklos Szeredi { 4147c6893e3SMiklos Szeredi int ret; 4157c6893e3SMiklos Szeredi 4167c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4177c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4187c6893e3SMiklos Szeredi if (ret) 4197c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4207c6893e3SMiklos Szeredi return ret; 4217c6893e3SMiklos Szeredi } 42296029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 42396029c4eSnpiggin@suse.de 42496029c4eSnpiggin@suse.de /** 425eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4268366025eSDave Hansen * @mnt: the mount on which to give up write access 4278366025eSDave Hansen * 4288366025eSDave Hansen * Tells the low-level filesystem that we are done 4298366025eSDave Hansen * performing writes to it. Must be matched with 430eb04c282SJan Kara * __mnt_want_write() call above. 4318366025eSDave Hansen */ 432eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4338366025eSDave Hansen { 434d3ef3d73Snpiggin@suse.de preempt_disable(); 43583adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 436d3ef3d73Snpiggin@suse.de preempt_enable(); 4378366025eSDave Hansen } 438eb04c282SJan Kara 439eb04c282SJan Kara /** 440eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 441eb04c282SJan Kara * @mnt: the mount on which to give up write access 442eb04c282SJan Kara * 443eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 444eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 445eb04c282SJan Kara * mnt_want_write() call above. 446eb04c282SJan Kara */ 447eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 448eb04c282SJan Kara { 449eb04c282SJan Kara __mnt_drop_write(mnt); 450eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 451eb04c282SJan Kara } 4528366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4538366025eSDave Hansen 454eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 455eb04c282SJan Kara { 456eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 457eb04c282SJan Kara } 458eb04c282SJan Kara 4597c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4607c6893e3SMiklos Szeredi { 461a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4627c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4637c6893e3SMiklos Szeredi } 4642a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4652a79f17eSAl Viro 46683adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4678366025eSDave Hansen { 4683d733633SDave Hansen int ret = 0; 4693d733633SDave Hansen 470719ea2fbSAl Viro lock_mount_hash(); 47183adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 472d3ef3d73Snpiggin@suse.de /* 473d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 474d3ef3d73Snpiggin@suse.de * should be visible before we do. 475d3ef3d73Snpiggin@suse.de */ 476d3ef3d73Snpiggin@suse.de smp_mb(); 477d3ef3d73Snpiggin@suse.de 478d3ef3d73Snpiggin@suse.de /* 479d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 480d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 481d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 482d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 483d3ef3d73Snpiggin@suse.de * 484d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 485d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 486d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 487d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 488d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 489d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 490d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 491d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 492d3ef3d73Snpiggin@suse.de * we're counting up here. 493d3ef3d73Snpiggin@suse.de */ 494c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 495d3ef3d73Snpiggin@suse.de ret = -EBUSY; 496d3ef3d73Snpiggin@suse.de else 49783adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 498d3ef3d73Snpiggin@suse.de /* 499d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 500d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 501d3ef3d73Snpiggin@suse.de */ 502d3ef3d73Snpiggin@suse.de smp_wmb(); 50383adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 504719ea2fbSAl Viro unlock_mount_hash(); 5053d733633SDave Hansen return ret; 5063d733633SDave Hansen } 5078366025eSDave Hansen 50843f5e655SDavid Howells static int __mnt_unmake_readonly(struct mount *mnt) 5092e4b7fcdSDave Hansen { 510719ea2fbSAl Viro lock_mount_hash(); 51183adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 512719ea2fbSAl Viro unlock_mount_hash(); 51343f5e655SDavid Howells return 0; 5142e4b7fcdSDave Hansen } 5152e4b7fcdSDave Hansen 5164ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5174ed5e82fSMiklos Szeredi { 5184ed5e82fSMiklos Szeredi struct mount *mnt; 5194ed5e82fSMiklos Szeredi int err = 0; 5204ed5e82fSMiklos Szeredi 5218e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5228e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5238e8b8796SMiklos Szeredi return -EBUSY; 5248e8b8796SMiklos Szeredi 525719ea2fbSAl Viro lock_mount_hash(); 5264ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5274ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5284ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5294ed5e82fSMiklos Szeredi smp_mb(); 5304ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5314ed5e82fSMiklos Szeredi err = -EBUSY; 5324ed5e82fSMiklos Szeredi break; 5334ed5e82fSMiklos Szeredi } 5344ed5e82fSMiklos Szeredi } 5354ed5e82fSMiklos Szeredi } 5368e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5378e8b8796SMiklos Szeredi err = -EBUSY; 5388e8b8796SMiklos Szeredi 5394ed5e82fSMiklos Szeredi if (!err) { 5404ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5414ed5e82fSMiklos Szeredi smp_wmb(); 5424ed5e82fSMiklos Szeredi } 5434ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5444ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5454ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5464ed5e82fSMiklos Szeredi } 547719ea2fbSAl Viro unlock_mount_hash(); 5484ed5e82fSMiklos Szeredi 5494ed5e82fSMiklos Szeredi return err; 5504ed5e82fSMiklos Szeredi } 5514ed5e82fSMiklos Szeredi 552b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5531da177e4SLinus Torvalds { 554fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 555d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 55668e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 557d3ef3d73Snpiggin@suse.de #endif 558b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5591da177e4SLinus Torvalds } 5601da177e4SLinus Torvalds 5618ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5628ffcb32eSDavid Howells { 5638ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5648ffcb32eSDavid Howells } 5658ffcb32eSDavid Howells 56648a066e7SAl Viro /* call under rcu_read_lock */ 567294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 56848a066e7SAl Viro { 56948a066e7SAl Viro struct mount *mnt; 57048a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 571294d71ffSAl Viro return 1; 57248a066e7SAl Viro if (bastard == NULL) 573294d71ffSAl Viro return 0; 57448a066e7SAl Viro mnt = real_mount(bastard); 57548a066e7SAl Viro mnt_add_count(mnt, 1); 576119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 57748a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 578294d71ffSAl Viro return 0; 57948a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 58048a066e7SAl Viro mnt_add_count(mnt, -1); 581294d71ffSAl Viro return 1; 58248a066e7SAl Viro } 583119e1ef8SAl Viro lock_mount_hash(); 584119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 585119e1ef8SAl Viro mnt_add_count(mnt, -1); 586119e1ef8SAl Viro unlock_mount_hash(); 587119e1ef8SAl Viro return 1; 588119e1ef8SAl Viro } 589119e1ef8SAl Viro unlock_mount_hash(); 590119e1ef8SAl Viro /* caller will mntput() */ 591294d71ffSAl Viro return -1; 592294d71ffSAl Viro } 593294d71ffSAl Viro 594294d71ffSAl Viro /* call under rcu_read_lock */ 595294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 596294d71ffSAl Viro { 597294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 598294d71ffSAl Viro if (likely(!res)) 599294d71ffSAl Viro return true; 600294d71ffSAl Viro if (unlikely(res < 0)) { 60148a066e7SAl Viro rcu_read_unlock(); 60248a066e7SAl Viro mntput(bastard); 60348a066e7SAl Viro rcu_read_lock(); 604294d71ffSAl Viro } 60548a066e7SAl Viro return false; 60648a066e7SAl Viro } 60748a066e7SAl Viro 6081da177e4SLinus Torvalds /* 609474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 61048a066e7SAl Viro * call under rcu_read_lock() 6111da177e4SLinus Torvalds */ 612474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6131da177e4SLinus Torvalds { 61438129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 615474279dcSAl Viro struct mount *p; 6161da177e4SLinus Torvalds 61738129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 618474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 619474279dcSAl Viro return p; 620474279dcSAl Viro return NULL; 6211da177e4SLinus Torvalds } 622474279dcSAl Viro 623474279dcSAl Viro /* 624f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 625f015f126SDavid Howells * 626f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 627f015f126SDavid Howells * following mounts: 628f015f126SDavid Howells * 629f015f126SDavid Howells * mount /dev/sda1 /mnt 630f015f126SDavid Howells * mount /dev/sda2 /mnt 631f015f126SDavid Howells * mount /dev/sda3 /mnt 632f015f126SDavid Howells * 633f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 634f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 635f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 636f015f126SDavid Howells * 637f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 638a05964f3SRam Pai */ 639ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 640a05964f3SRam Pai { 641c7105365SAl Viro struct mount *child_mnt; 64248a066e7SAl Viro struct vfsmount *m; 64348a066e7SAl Viro unsigned seq; 64499b7db7bSNick Piggin 64548a066e7SAl Viro rcu_read_lock(); 64648a066e7SAl Viro do { 64748a066e7SAl Viro seq = read_seqbegin(&mount_lock); 648474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 64948a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 65048a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 65148a066e7SAl Viro rcu_read_unlock(); 65248a066e7SAl Viro return m; 653a05964f3SRam Pai } 654a05964f3SRam Pai 6557af1364fSEric W. Biederman /* 6567af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6577af1364fSEric W. Biederman * current mount namespace. 6587af1364fSEric W. Biederman * 6597af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6607af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6617af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6627af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6637af1364fSEric W. Biederman * is a mountpoint. 6647af1364fSEric W. Biederman * 6657af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6667af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6677af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6687af1364fSEric W. Biederman * parent mount. 6697af1364fSEric W. Biederman */ 6707af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6717af1364fSEric W. Biederman { 6727af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 6737af1364fSEric W. Biederman struct mount *mnt; 6747af1364fSEric W. Biederman bool is_covered = false; 6757af1364fSEric W. Biederman 6767af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 6777af1364fSEric W. Biederman goto out; 6787af1364fSEric W. Biederman 6797af1364fSEric W. Biederman down_read(&namespace_sem); 6807af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 6817af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 6827af1364fSEric W. Biederman if (is_covered) 6837af1364fSEric W. Biederman break; 6847af1364fSEric W. Biederman } 6857af1364fSEric W. Biederman up_read(&namespace_sem); 6867af1364fSEric W. Biederman out: 6877af1364fSEric W. Biederman return is_covered; 6887af1364fSEric W. Biederman } 6897af1364fSEric W. Biederman 690e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 69184d17192SAl Viro { 6920818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 69384d17192SAl Viro struct mountpoint *mp; 69484d17192SAl Viro 6950818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 69684d17192SAl Viro if (mp->m_dentry == dentry) { 69784d17192SAl Viro mp->m_count++; 69884d17192SAl Viro return mp; 69984d17192SAl Viro } 70084d17192SAl Viro } 701e2dfa935SEric W. Biederman return NULL; 702e2dfa935SEric W. Biederman } 703e2dfa935SEric W. Biederman 7043895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 705e2dfa935SEric W. Biederman { 7063895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 707e2dfa935SEric W. Biederman int ret; 70884d17192SAl Viro 7093895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7101e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7111e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7121e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7133895dbf8SEric W. Biederman mountpoint: 7143895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7153895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7163895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7173895dbf8SEric W. Biederman if (mp) 7183895dbf8SEric W. Biederman goto done; 71984d17192SAl Viro } 720eed81007SMiklos Szeredi 7213895dbf8SEric W. Biederman if (!new) 7223895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7233895dbf8SEric W. Biederman if (!new) 7243895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7253895dbf8SEric W. Biederman 7263895dbf8SEric W. Biederman 7273895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7283895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7293895dbf8SEric W. Biederman 7303895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7313895dbf8SEric W. Biederman if (ret == -EBUSY) 7323895dbf8SEric W. Biederman goto mountpoint; 7333895dbf8SEric W. Biederman 7343895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7353895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7363895dbf8SEric W. Biederman if (ret) 7373895dbf8SEric W. Biederman goto done; 7383895dbf8SEric W. Biederman 7393895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7403895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7413895dbf8SEric W. Biederman new->m_dentry = dentry; 7423895dbf8SEric W. Biederman new->m_count = 1; 7433895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7443895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7453895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7463895dbf8SEric W. Biederman 7473895dbf8SEric W. Biederman mp = new; 7483895dbf8SEric W. Biederman new = NULL; 7493895dbf8SEric W. Biederman done: 7503895dbf8SEric W. Biederman kfree(new); 75184d17192SAl Viro return mp; 75284d17192SAl Viro } 75384d17192SAl Viro 75484d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 75584d17192SAl Viro { 75684d17192SAl Viro if (!--mp->m_count) { 75784d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7580a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 75984d17192SAl Viro spin_lock(&dentry->d_lock); 76084d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 76184d17192SAl Viro spin_unlock(&dentry->d_lock); 7620818bf27SAl Viro hlist_del(&mp->m_hash); 76384d17192SAl Viro kfree(mp); 76484d17192SAl Viro } 76584d17192SAl Viro } 76684d17192SAl Viro 767143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7681da177e4SLinus Torvalds { 7696b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7701da177e4SLinus Torvalds } 7711da177e4SLinus Torvalds 77299b7db7bSNick Piggin /* 77399b7db7bSNick Piggin * vfsmount lock must be held for write 77499b7db7bSNick Piggin */ 7756b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7765addc5ddSAl Viro { 7775addc5ddSAl Viro if (ns) { 7785addc5ddSAl Viro ns->event = ++event; 7795addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7805addc5ddSAl Viro } 7815addc5ddSAl Viro } 7825addc5ddSAl Viro 78399b7db7bSNick Piggin /* 78499b7db7bSNick Piggin * vfsmount lock must be held for write 78599b7db7bSNick Piggin */ 7866b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 7875addc5ddSAl Viro { 7885addc5ddSAl Viro if (ns && ns->event != event) { 7895addc5ddSAl Viro ns->event = event; 7905addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7915addc5ddSAl Viro } 7925addc5ddSAl Viro } 7935addc5ddSAl Viro 79499b7db7bSNick Piggin /* 79599b7db7bSNick Piggin * vfsmount lock must be held for write 79699b7db7bSNick Piggin */ 7977bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 7981da177e4SLinus Torvalds { 7990714a533SAl Viro mnt->mnt_parent = mnt; 800a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8016b41d536SAl Viro list_del_init(&mnt->mnt_child); 80238129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8030a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 80484d17192SAl Viro put_mountpoint(mnt->mnt_mp); 80584d17192SAl Viro mnt->mnt_mp = NULL; 8061da177e4SLinus Torvalds } 8071da177e4SLinus Torvalds 80899b7db7bSNick Piggin /* 80999b7db7bSNick Piggin * vfsmount lock must be held for write 81099b7db7bSNick Piggin */ 8117bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8127bdb11deSEric W. Biederman { 8137bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8147bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8157bdb11deSEric W. Biederman unhash_mnt(mnt); 8167bdb11deSEric W. Biederman } 8177bdb11deSEric W. Biederman 8187bdb11deSEric W. Biederman /* 8197bdb11deSEric W. Biederman * vfsmount lock must be held for write 8207bdb11deSEric W. Biederman */ 8216a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8226a46c573SEric W. Biederman { 8236a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8246a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8256a46c573SEric W. Biederman unhash_mnt(mnt); 8266a46c573SEric W. Biederman } 8276a46c573SEric W. Biederman 8286a46c573SEric W. Biederman /* 8296a46c573SEric W. Biederman * vfsmount lock must be held for write 8306a46c573SEric W. Biederman */ 83184d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 83284d17192SAl Viro struct mountpoint *mp, 83344d964d6SAl Viro struct mount *child_mnt) 834b90fa9aeSRam Pai { 83584d17192SAl Viro mp->m_count++; 8363a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 83784d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8383a2393d7SAl Viro child_mnt->mnt_parent = mnt; 83984d17192SAl Viro child_mnt->mnt_mp = mp; 8400a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 841b90fa9aeSRam Pai } 842b90fa9aeSRam Pai 8431064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8441064f874SEric W. Biederman { 8451064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8461064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8471064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8481064f874SEric W. Biederman } 8491064f874SEric W. Biederman 85099b7db7bSNick Piggin /* 85199b7db7bSNick Piggin * vfsmount lock must be held for write 85299b7db7bSNick Piggin */ 85384d17192SAl Viro static void attach_mnt(struct mount *mnt, 85484d17192SAl Viro struct mount *parent, 85584d17192SAl Viro struct mountpoint *mp) 8561da177e4SLinus Torvalds { 85784d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8581064f874SEric W. Biederman __attach_mnt(mnt, parent); 859b90fa9aeSRam Pai } 860b90fa9aeSRam Pai 8611064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 86212a5b529SAl Viro { 8631064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8641064f874SEric W. Biederman struct dentry *old_mountpoint = mnt->mnt_mountpoint; 8651064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8661064f874SEric W. Biederman 8671064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8681064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8691064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8701064f874SEric W. Biederman 8711064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 8721064f874SEric W. Biederman 8731064f874SEric W. Biederman put_mountpoint(old_mp); 8741064f874SEric W. Biederman 8751064f874SEric W. Biederman /* 8761064f874SEric W. Biederman * Safely avoid even the suggestion this code might sleep or 8771064f874SEric W. Biederman * lock the mount hash by taking advantage of the knowledge that 8781064f874SEric W. Biederman * mnt_change_mountpoint will not release the final reference 8791064f874SEric W. Biederman * to a mountpoint. 8801064f874SEric W. Biederman * 8811064f874SEric W. Biederman * During mounting, the mount passed in as the parent mount will 8821064f874SEric W. Biederman * continue to use the old mountpoint and during unmounting, the 8831064f874SEric W. Biederman * old mountpoint will continue to exist until namespace_unlock, 8841064f874SEric W. Biederman * which happens well after mnt_change_mountpoint. 8851064f874SEric W. Biederman */ 8861064f874SEric W. Biederman spin_lock(&old_mountpoint->d_lock); 8871064f874SEric W. Biederman old_mountpoint->d_lockref.count--; 8881064f874SEric W. Biederman spin_unlock(&old_mountpoint->d_lock); 8891064f874SEric W. Biederman 8901064f874SEric W. Biederman mnt_add_count(old_parent, -1); 89112a5b529SAl Viro } 89212a5b529SAl Viro 893b90fa9aeSRam Pai /* 89499b7db7bSNick Piggin * vfsmount lock must be held for write 895b90fa9aeSRam Pai */ 8961064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 897b90fa9aeSRam Pai { 8980714a533SAl Viro struct mount *parent = mnt->mnt_parent; 89983adc753SAl Viro struct mount *m; 900b90fa9aeSRam Pai LIST_HEAD(head); 901143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 902b90fa9aeSRam Pai 9030714a533SAl Viro BUG_ON(parent == mnt); 904b90fa9aeSRam Pai 9051a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 906f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 907143c8c91SAl Viro m->mnt_ns = n; 908f03c6599SAl Viro 909b90fa9aeSRam Pai list_splice(&head, n->list.prev); 910b90fa9aeSRam Pai 911d2921684SEric W. Biederman n->mounts += n->pending_mounts; 912d2921684SEric W. Biederman n->pending_mounts = 0; 913d2921684SEric W. Biederman 9141064f874SEric W. Biederman __attach_mnt(mnt, parent); 9156b3286edSKirill Korotaev touch_mnt_namespace(n); 9161da177e4SLinus Torvalds } 9171da177e4SLinus Torvalds 918909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9191da177e4SLinus Torvalds { 9206b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9216b41d536SAl Viro if (next == &p->mnt_mounts) { 9221da177e4SLinus Torvalds while (1) { 923909b0a88SAl Viro if (p == root) 9241da177e4SLinus Torvalds return NULL; 9256b41d536SAl Viro next = p->mnt_child.next; 9266b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9271da177e4SLinus Torvalds break; 9280714a533SAl Viro p = p->mnt_parent; 9291da177e4SLinus Torvalds } 9301da177e4SLinus Torvalds } 9316b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9321da177e4SLinus Torvalds } 9331da177e4SLinus Torvalds 934315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9359676f0c6SRam Pai { 9366b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9376b41d536SAl Viro while (prev != &p->mnt_mounts) { 9386b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9396b41d536SAl Viro prev = p->mnt_mounts.prev; 9409676f0c6SRam Pai } 9419676f0c6SRam Pai return p; 9429676f0c6SRam Pai } 9439676f0c6SRam Pai 9448f291889SAl Viro /** 9458f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 9468f291889SAl Viro * @fc: The configuration context with the superblock attached 9478f291889SAl Viro * 9488f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 9498f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 9508f291889SAl Viro * 9518f291889SAl Viro * Note that this does not attach the mount to anything. 9528f291889SAl Viro */ 9538f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 9548f291889SAl Viro { 9558f291889SAl Viro struct mount *mnt; 9568f291889SAl Viro 9578f291889SAl Viro if (!fc->root) 9588f291889SAl Viro return ERR_PTR(-EINVAL); 9598f291889SAl Viro 9608f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 9618f291889SAl Viro if (!mnt) 9628f291889SAl Viro return ERR_PTR(-ENOMEM); 9638f291889SAl Viro 9648f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 9658f291889SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9668f291889SAl Viro 9678f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 9688f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 9698f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 9708f291889SAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9718f291889SAl Viro mnt->mnt_parent = mnt; 9728f291889SAl Viro 9738f291889SAl Viro lock_mount_hash(); 9748f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 9758f291889SAl Viro unlock_mount_hash(); 9768f291889SAl Viro return &mnt->mnt; 9778f291889SAl Viro } 9788f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 9798f291889SAl Viro 9808f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 9818f291889SAl Viro { 9828f291889SAl Viro int err = vfs_get_tree(fc); 9838f291889SAl Viro if (!err) { 9848f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 9858f291889SAl Viro return vfs_create_mount(fc); 9868f291889SAl Viro } 9878f291889SAl Viro return ERR_PTR(err); 9888f291889SAl Viro } 9898f291889SAl Viro EXPORT_SYMBOL(fc_mount); 9908f291889SAl Viro 9919bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 9929bc61ab1SDavid Howells int flags, const char *name, 9939bc61ab1SDavid Howells void *data) 9949d412a43SAl Viro { 9959bc61ab1SDavid Howells struct fs_context *fc; 9968f291889SAl Viro struct vfsmount *mnt; 9979bc61ab1SDavid Howells int ret = 0; 9989d412a43SAl Viro 9999d412a43SAl Viro if (!type) 10009d412a43SAl Viro return ERR_PTR(-ENODEV); 10019d412a43SAl Viro 10029bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 10039bc61ab1SDavid Howells if (IS_ERR(fc)) 10049bc61ab1SDavid Howells return ERR_CAST(fc); 10059bc61ab1SDavid Howells 10069bc61ab1SDavid Howells if (name) { 10079bc61ab1SDavid Howells fc->source = kstrdup(name, GFP_KERNEL); 10089bc61ab1SDavid Howells if (!fc->source) 10099bc61ab1SDavid Howells ret = -ENOMEM; 10109bc61ab1SDavid Howells } 10119bc61ab1SDavid Howells if (!ret) 10129bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 10139bc61ab1SDavid Howells if (!ret) 10148f291889SAl Viro mnt = fc_mount(fc); 10158f291889SAl Viro else 10168f291889SAl Viro mnt = ERR_PTR(ret); 10179d412a43SAl Viro 10189bc61ab1SDavid Howells put_fs_context(fc); 10198f291889SAl Viro return mnt; 10209d412a43SAl Viro } 10219d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10229d412a43SAl Viro 102393faccbbSEric W. Biederman struct vfsmount * 102493faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 102593faccbbSEric W. Biederman const char *name, void *data) 102693faccbbSEric W. Biederman { 102793faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 102893faccbbSEric W. Biederman * through from the parent mount to the submount don't support 102993faccbbSEric W. Biederman * unprivileged mounts with submounts. 103093faccbbSEric W. Biederman */ 103193faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 103293faccbbSEric W. Biederman return ERR_PTR(-EPERM); 103393faccbbSEric W. Biederman 1034e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 103593faccbbSEric W. Biederman } 103693faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 103793faccbbSEric W. Biederman 103887129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 103936341f64SRam Pai int flag) 10401da177e4SLinus Torvalds { 104187129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1042be34d1a3SDavid Howells struct mount *mnt; 1043be34d1a3SDavid Howells int err; 10441da177e4SLinus Torvalds 1045be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1046be34d1a3SDavid Howells if (!mnt) 1047be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1048be34d1a3SDavid Howells 10497a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 105015169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1051719f5d7fSMiklos Szeredi else 105215169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1053719f5d7fSMiklos Szeredi 105415169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1055be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1056719f5d7fSMiklos Szeredi if (err) 1057719f5d7fSMiklos Szeredi goto out_free; 1058719f5d7fSMiklos Szeredi } 1059719f5d7fSMiklos Szeredi 106016a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 106116a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 10625ff9d8a6SEric W. Biederman 10631da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1064b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1065b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1066a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10670714a533SAl Viro mnt->mnt_parent = mnt; 1068719ea2fbSAl Viro lock_mount_hash(); 106939f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1070719ea2fbSAl Viro unlock_mount_hash(); 1071b90fa9aeSRam Pai 10727a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10737a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10746776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 107532301920SAl Viro mnt->mnt_master = old; 1076fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10778aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1078fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10796776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1080d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10816776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1082d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10835235d448SAl Viro } else { 10845235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10855afe0022SRam Pai } 1086b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10870f0afb1dSAl Viro set_mnt_shared(mnt); 10881da177e4SLinus Torvalds 10891da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10901da177e4SLinus Torvalds * as the original if that was on one */ 109136341f64SRam Pai if (flag & CL_EXPIRE) { 10926776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10936776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10941da177e4SLinus Torvalds } 1095be34d1a3SDavid Howells 1096cb338d06SAl Viro return mnt; 1097719f5d7fSMiklos Szeredi 1098719f5d7fSMiklos Szeredi out_free: 10998ffcb32eSDavid Howells mnt_free_id(mnt); 1100719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1101be34d1a3SDavid Howells return ERR_PTR(err); 11021da177e4SLinus Torvalds } 11031da177e4SLinus Torvalds 11049ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11059ea459e1SAl Viro { 11069ea459e1SAl Viro /* 11079ea459e1SAl Viro * This probably indicates that somebody messed 11089ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 11099ea459e1SAl Viro * happens, the filesystem was probably unable 11109ea459e1SAl Viro * to make r/w->r/o transitions. 11119ea459e1SAl Viro */ 11129ea459e1SAl Viro /* 11139ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11149ea459e1SAl Viro * so mnt_get_writers() below is safe. 11159ea459e1SAl Viro */ 11169ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11179ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11189ea459e1SAl Viro mnt_pin_kill(mnt); 11199ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11209ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11219ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11229ea459e1SAl Viro mnt_free_id(mnt); 11239ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11249ea459e1SAl Viro } 11259ea459e1SAl Viro 11269ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11279ea459e1SAl Viro { 11289ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11299ea459e1SAl Viro } 11309ea459e1SAl Viro 11319ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11329ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11339ea459e1SAl Viro { 11349ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 113529785735SByungchul Park struct mount *m, *t; 11369ea459e1SAl Viro 113729785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 113829785735SByungchul Park cleanup_mnt(m); 11399ea459e1SAl Viro } 11409ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11419ea459e1SAl Viro 1142900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11437b7b1aceSAl Viro { 114448a066e7SAl Viro rcu_read_lock(); 11459ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11469ea0a46cSAl Viro /* 11479ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11489ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11499ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11509ea0a46cSAl Viro * be dropped until after an RCU delay done after 11519ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11529ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11539ea0a46cSAl Viro * we are dropping is not the final one. 11549ea0a46cSAl Viro */ 1155aa9c0e07SAl Viro mnt_add_count(mnt, -1); 115648a066e7SAl Viro rcu_read_unlock(); 115799b7db7bSNick Piggin return; 1158b3e19d92SNick Piggin } 1159719ea2fbSAl Viro lock_mount_hash(); 1160119e1ef8SAl Viro /* 1161119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1162119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1163119e1ef8SAl Viro */ 1164119e1ef8SAl Viro smp_mb(); 11659ea0a46cSAl Viro mnt_add_count(mnt, -1); 1166b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 116748a066e7SAl Viro rcu_read_unlock(); 1168719ea2fbSAl Viro unlock_mount_hash(); 116999b7db7bSNick Piggin return; 117099b7db7bSNick Piggin } 117148a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 117248a066e7SAl Viro rcu_read_unlock(); 117348a066e7SAl Viro unlock_mount_hash(); 117448a066e7SAl Viro return; 117548a066e7SAl Viro } 117648a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 117748a066e7SAl Viro rcu_read_unlock(); 1178962830dfSAndi Kleen 117939f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1180ce07d891SEric W. Biederman 1181ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1182ce07d891SEric W. Biederman struct mount *p, *tmp; 1183ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1184ce07d891SEric W. Biederman umount_mnt(p); 1185ce07d891SEric W. Biederman } 1186ce07d891SEric W. Biederman } 1187719ea2fbSAl Viro unlock_mount_hash(); 1188649a795aSAl Viro 11899ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11909ea459e1SAl Viro struct task_struct *task = current; 11919ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11929ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11939ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11949ea459e1SAl Viro return; 11959ea459e1SAl Viro } 11969ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11979ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11989ea459e1SAl Viro return; 11999ea459e1SAl Viro } 12009ea459e1SAl Viro cleanup_mnt(mnt); 1201b3e19d92SNick Piggin } 1202b3e19d92SNick Piggin 1203b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1204b3e19d92SNick Piggin { 1205b3e19d92SNick Piggin if (mnt) { 1206863d684fSAl Viro struct mount *m = real_mount(mnt); 1207b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1208863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1209863d684fSAl Viro m->mnt_expiry_mark = 0; 1210863d684fSAl Viro mntput_no_expire(m); 1211b3e19d92SNick Piggin } 1212b3e19d92SNick Piggin } 1213b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1214b3e19d92SNick Piggin 1215b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1216b3e19d92SNick Piggin { 1217b3e19d92SNick Piggin if (mnt) 121883adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1219b3e19d92SNick Piggin return mnt; 1220b3e19d92SNick Piggin } 1221b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1222b3e19d92SNick Piggin 1223c6609c0aSIan Kent /* path_is_mountpoint() - Check if path is a mount in the current 1224c6609c0aSIan Kent * namespace. 1225c6609c0aSIan Kent * 1226c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1227c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1228c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1229c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1230c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1231c6609c0aSIan Kent * alone. 1232c6609c0aSIan Kent */ 1233c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1234c6609c0aSIan Kent { 1235c6609c0aSIan Kent unsigned seq; 1236c6609c0aSIan Kent bool res; 1237c6609c0aSIan Kent 1238c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1239c6609c0aSIan Kent return false; 1240c6609c0aSIan Kent 1241c6609c0aSIan Kent rcu_read_lock(); 1242c6609c0aSIan Kent do { 1243c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1244c6609c0aSIan Kent res = __path_is_mountpoint(path); 1245c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1246c6609c0aSIan Kent rcu_read_unlock(); 1247c6609c0aSIan Kent 1248c6609c0aSIan Kent return res; 1249c6609c0aSIan Kent } 1250c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1251c6609c0aSIan Kent 1252ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12537b7b1aceSAl Viro { 12543064c356SAl Viro struct mount *p; 12553064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12563064c356SAl Viro if (IS_ERR(p)) 12573064c356SAl Viro return ERR_CAST(p); 12583064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12593064c356SAl Viro return &p->mnt; 12607b7b1aceSAl Viro } 12611da177e4SLinus Torvalds 1262a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12630226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12641da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12651da177e4SLinus Torvalds { 1266ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12671da177e4SLinus Torvalds 1268390c6843SRam Pai down_read(&namespace_sem); 1269c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1270c7999c36SAl Viro void *v = p->cached_mount; 1271c7999c36SAl Viro if (*pos == p->cached_index) 1272c7999c36SAl Viro return v; 1273c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1274c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1275c7999c36SAl Viro return p->cached_mount = v; 1276c7999c36SAl Viro } 1277c7999c36SAl Viro } 1278c7999c36SAl Viro 1279c7999c36SAl Viro p->cached_event = p->ns->event; 1280c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1281c7999c36SAl Viro p->cached_index = *pos; 1282c7999c36SAl Viro return p->cached_mount; 12831da177e4SLinus Torvalds } 12841da177e4SLinus Torvalds 12851da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12861da177e4SLinus Torvalds { 1287ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1288b0765fb8SPavel Emelianov 1289c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1290c7999c36SAl Viro p->cached_index = *pos; 1291c7999c36SAl Viro return p->cached_mount; 12921da177e4SLinus Torvalds } 12931da177e4SLinus Torvalds 12941da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12951da177e4SLinus Torvalds { 1296390c6843SRam Pai up_read(&namespace_sem); 12971da177e4SLinus Torvalds } 12981da177e4SLinus Torvalds 12990226f492SAl Viro static int m_show(struct seq_file *m, void *v) 13009f5596afSAl Viro { 1301ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13021a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 13030226f492SAl Viro return p->show(m, &r->mnt); 13041da177e4SLinus Torvalds } 13051da177e4SLinus Torvalds 1306a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13071da177e4SLinus Torvalds .start = m_start, 13081da177e4SLinus Torvalds .next = m_next, 13091da177e4SLinus Torvalds .stop = m_stop, 13100226f492SAl Viro .show = m_show, 1311b4629fe2SChuck Lever }; 1312a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1313b4629fe2SChuck Lever 13141da177e4SLinus Torvalds /** 13151da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 13161da177e4SLinus Torvalds * @mnt: root of mount tree 13171da177e4SLinus Torvalds * 13181da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 13191da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 13201da177e4SLinus Torvalds * busy. 13211da177e4SLinus Torvalds */ 1322909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13231da177e4SLinus Torvalds { 1324909b0a88SAl Viro struct mount *mnt = real_mount(m); 132536341f64SRam Pai int actual_refs = 0; 132636341f64SRam Pai int minimum_refs = 0; 1327315fc83eSAl Viro struct mount *p; 1328909b0a88SAl Viro BUG_ON(!m); 13291da177e4SLinus Torvalds 1330b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1331719ea2fbSAl Viro lock_mount_hash(); 1332909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 133383adc753SAl Viro actual_refs += mnt_get_count(p); 13341da177e4SLinus Torvalds minimum_refs += 2; 13351da177e4SLinus Torvalds } 1336719ea2fbSAl Viro unlock_mount_hash(); 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13391da177e4SLinus Torvalds return 0; 1340e3474a8eSIan Kent 1341e3474a8eSIan Kent return 1; 13421da177e4SLinus Torvalds } 13431da177e4SLinus Torvalds 13441da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds /** 13471da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13481da177e4SLinus Torvalds * @mnt: root of mount 13491da177e4SLinus Torvalds * 13501da177e4SLinus Torvalds * This is called to check if a mount point has any 13511da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13521da177e4SLinus Torvalds * mount has sub mounts this will return busy 13531da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13541da177e4SLinus Torvalds * 13551da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13561da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13571da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13581da177e4SLinus Torvalds */ 13591da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13601da177e4SLinus Torvalds { 1361e3474a8eSIan Kent int ret = 1; 13628ad08d8aSAl Viro down_read(&namespace_sem); 1363719ea2fbSAl Viro lock_mount_hash(); 13641ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1365e3474a8eSIan Kent ret = 0; 1366719ea2fbSAl Viro unlock_mount_hash(); 13678ad08d8aSAl Viro up_read(&namespace_sem); 1368a05964f3SRam Pai return ret; 13691da177e4SLinus Torvalds } 13701da177e4SLinus Torvalds 13711da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13721da177e4SLinus Torvalds 137338129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1374e3197d83SAl Viro 137597216be0SAl Viro static void namespace_unlock(void) 13761da177e4SLinus Torvalds { 1377a3b3c562SEric W. Biederman struct hlist_head head; 137897216be0SAl Viro 1379a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1380a3b3c562SEric W. Biederman 138197216be0SAl Viro up_write(&namespace_sem); 1382a3b3c562SEric W. Biederman 1383a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 138497216be0SAl Viro return; 138597216be0SAl Viro 138622cb7405SNeilBrown synchronize_rcu_expedited(); 138748a066e7SAl Viro 138887b95ce0SAl Viro group_pin_kill(&head); 138970fbcdf4SRam Pai } 139070fbcdf4SRam Pai 139197216be0SAl Viro static inline void namespace_lock(void) 1392e3197d83SAl Viro { 139397216be0SAl Viro down_write(&namespace_sem); 1394e3197d83SAl Viro } 1395e3197d83SAl Viro 1396e819f152SEric W. Biederman enum umount_tree_flags { 1397e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1398e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1399e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1400e819f152SEric W. Biederman }; 1401f2d0a123SEric W. Biederman 1402f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1403f2d0a123SEric W. Biederman { 1404f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1405f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1406f2d0a123SEric W. Biederman return true; 1407f2d0a123SEric W. Biederman 1408f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1409f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1410f2d0a123SEric W. Biederman return true; 1411f2d0a123SEric W. Biederman 1412f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1413f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1414f2d0a123SEric W. Biederman * connected to mounted mounts. 1415f2d0a123SEric W. Biederman */ 1416f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1417f2d0a123SEric W. Biederman return true; 1418f2d0a123SEric W. Biederman 1419f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1420f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1421f2d0a123SEric W. Biederman return false; 1422f2d0a123SEric W. Biederman 1423f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1424f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1425f2d0a123SEric W. Biederman return false; 1426f2d0a123SEric W. Biederman 1427f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1428f2d0a123SEric W. Biederman return true; 1429f2d0a123SEric W. Biederman } 1430f2d0a123SEric W. Biederman 143199b7db7bSNick Piggin /* 143248a066e7SAl Viro * mount_lock must be held 143399b7db7bSNick Piggin * namespace_sem must be held for write 143499b7db7bSNick Piggin */ 1435e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 143670fbcdf4SRam Pai { 1437c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1438315fc83eSAl Viro struct mount *p; 143970fbcdf4SRam Pai 14405d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14415d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14425d88457eSEric W. Biederman 1443c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1444590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1445590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1446c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1447590ce4bcSEric W. Biederman } 1448c003b26fSEric W. Biederman 1449411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1450c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1451c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 145238129a13SAl Viro } 145370fbcdf4SRam Pai 1454c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1455e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14567b8a53fdSAl Viro propagate_umount(&tmp_list); 1457a05964f3SRam Pai 1458c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1459d2921684SEric W. Biederman struct mnt_namespace *ns; 1460ce07d891SEric W. Biederman bool disconnect; 1461c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 14626776db3dSAl Viro list_del_init(&p->mnt_expire); 14631a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1464d2921684SEric W. Biederman ns = p->mnt_ns; 1465d2921684SEric W. Biederman if (ns) { 1466d2921684SEric W. Biederman ns->mounts--; 1467d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1468d2921684SEric W. Biederman } 146963d37a84SAl Viro p->mnt_ns = NULL; 1470e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 147148a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 147287b95ce0SAl Viro 1473f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1474ce07d891SEric W. Biederman 1475ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1476ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1477676da58dSAl Viro if (mnt_has_parent(p)) { 147881b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1479ce07d891SEric W. Biederman if (!disconnect) { 1480ce07d891SEric W. Biederman /* Don't forget about p */ 1481ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1482ce07d891SEric W. Biederman } else { 14836a46c573SEric W. Biederman umount_mnt(p); 14847c4b93d8SAl Viro } 1485ce07d891SEric W. Biederman } 14860f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 148738129a13SAl Viro } 14881da177e4SLinus Torvalds } 14891da177e4SLinus Torvalds 1490b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1491c35038beSAl Viro 14921ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 14931da177e4SLinus Torvalds { 14941ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 14951da177e4SLinus Torvalds int retval; 14961da177e4SLinus Torvalds 14971ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 14981da177e4SLinus Torvalds if (retval) 14991da177e4SLinus Torvalds return retval; 15001da177e4SLinus Torvalds 15011da177e4SLinus Torvalds /* 15021da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 15031da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 15041da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 15051da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 15061da177e4SLinus Torvalds */ 15071da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 15081ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 15091da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 15101da177e4SLinus Torvalds return -EINVAL; 15111da177e4SLinus Torvalds 1512b3e19d92SNick Piggin /* 1513b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1514b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1515b3e19d92SNick Piggin */ 1516719ea2fbSAl Viro lock_mount_hash(); 151783adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1518719ea2fbSAl Viro unlock_mount_hash(); 15191da177e4SLinus Torvalds return -EBUSY; 1520b3e19d92SNick Piggin } 1521719ea2fbSAl Viro unlock_mount_hash(); 15221da177e4SLinus Torvalds 1523863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 15241da177e4SLinus Torvalds return -EAGAIN; 15251da177e4SLinus Torvalds } 15261da177e4SLinus Torvalds 15271da177e4SLinus Torvalds /* 15281da177e4SLinus Torvalds * If we may have to abort operations to get out of this 15291da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 15301da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 15311da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 15321da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 15331da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 15341da177e4SLinus Torvalds * about for the moment. 15351da177e4SLinus Torvalds */ 15361da177e4SLinus Torvalds 153742faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 153842faad99SAl Viro sb->s_op->umount_begin(sb); 153942faad99SAl Viro } 15401da177e4SLinus Torvalds 15411da177e4SLinus Torvalds /* 15421da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15431da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15441da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15451da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15461da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15471da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15481da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15491da177e4SLinus Torvalds */ 15501ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15511da177e4SLinus Torvalds /* 15521da177e4SLinus Torvalds * Special case for "unmounting" root ... 15531da177e4SLinus Torvalds * we just try to remount it readonly. 15541da177e4SLinus Torvalds */ 1555bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1556a1480dccSAndy Lutomirski return -EPERM; 15571da177e4SLinus Torvalds down_write(&sb->s_umount); 1558bc98a42cSDavid Howells if (!sb_rdonly(sb)) 1559e462ec50SDavid Howells retval = do_remount_sb(sb, SB_RDONLY, NULL, 0); 15601da177e4SLinus Torvalds up_write(&sb->s_umount); 15611da177e4SLinus Torvalds return retval; 15621da177e4SLinus Torvalds } 15631da177e4SLinus Torvalds 156497216be0SAl Viro namespace_lock(); 1565719ea2fbSAl Viro lock_mount_hash(); 15661da177e4SLinus Torvalds 156725d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 156825d202edSEric W. Biederman retval = -EINVAL; 156925d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 157025d202edSEric W. Biederman goto out; 157125d202edSEric W. Biederman 157225d202edSEric W. Biederman event++; 157348a066e7SAl Viro if (flags & MNT_DETACH) { 157448a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1575e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 157648a066e7SAl Viro retval = 0; 157748a066e7SAl Viro } else { 1578b54b9be7SAl Viro shrink_submounts(mnt); 15791da177e4SLinus Torvalds retval = -EBUSY; 158048a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 15811a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1582e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 15831da177e4SLinus Torvalds retval = 0; 15841da177e4SLinus Torvalds } 158548a066e7SAl Viro } 158625d202edSEric W. Biederman out: 1587719ea2fbSAl Viro unlock_mount_hash(); 1588e3197d83SAl Viro namespace_unlock(); 15891da177e4SLinus Torvalds return retval; 15901da177e4SLinus Torvalds } 15911da177e4SLinus Torvalds 15921da177e4SLinus Torvalds /* 159380b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 159480b5dce8SEric W. Biederman * 159580b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 159680b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 159780b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 159880b5dce8SEric W. Biederman * leaking them. 159980b5dce8SEric W. Biederman * 160080b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 160180b5dce8SEric W. Biederman */ 160280b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 160380b5dce8SEric W. Biederman { 160480b5dce8SEric W. Biederman struct mountpoint *mp; 160580b5dce8SEric W. Biederman struct mount *mnt; 160680b5dce8SEric W. Biederman 160780b5dce8SEric W. Biederman namespace_lock(); 16083895dbf8SEric W. Biederman lock_mount_hash(); 160980b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1610f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 161180b5dce8SEric W. Biederman goto out_unlock; 161280b5dce8SEric W. Biederman 1613e06b933eSAndrey Ulanov event++; 161480b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 161580b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1616ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1617fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1618fe78fcc8SEric W. Biederman umount_mnt(mnt); 1619ce07d891SEric W. Biederman } 1620e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 162180b5dce8SEric W. Biederman } 162280b5dce8SEric W. Biederman put_mountpoint(mp); 162380b5dce8SEric W. Biederman out_unlock: 16243895dbf8SEric W. Biederman unlock_mount_hash(); 162580b5dce8SEric W. Biederman namespace_unlock(); 162680b5dce8SEric W. Biederman } 162780b5dce8SEric W. Biederman 162880b5dce8SEric W. Biederman /* 16299b40bc90SAl Viro * Is the caller allowed to modify his namespace? 16309b40bc90SAl Viro */ 16319b40bc90SAl Viro static inline bool may_mount(void) 16329b40bc90SAl Viro { 16339b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 16349b40bc90SAl Viro } 16359b40bc90SAl Viro 16369e8925b6SJeff Layton static inline bool may_mandlock(void) 16379e8925b6SJeff Layton { 16389e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 16399e8925b6SJeff Layton return false; 16409e8925b6SJeff Layton #endif 164195ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 16429e8925b6SJeff Layton } 16439e8925b6SJeff Layton 16449b40bc90SAl Viro /* 16451da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16461da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16471da177e4SLinus Torvalds * 16481da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16491da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16501da177e4SLinus Torvalds */ 16511da177e4SLinus Torvalds 16523a18ef5cSDominik Brodowski int ksys_umount(char __user *name, int flags) 16531da177e4SLinus Torvalds { 16542d8f3038SAl Viro struct path path; 1655900148dcSAl Viro struct mount *mnt; 16561da177e4SLinus Torvalds int retval; 1657db1f05bbSMiklos Szeredi int lookup_flags = 0; 16581da177e4SLinus Torvalds 1659db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1660db1f05bbSMiklos Szeredi return -EINVAL; 1661db1f05bbSMiklos Szeredi 16629b40bc90SAl Viro if (!may_mount()) 16639b40bc90SAl Viro return -EPERM; 16649b40bc90SAl Viro 1665db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1666db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1667db1f05bbSMiklos Szeredi 1668197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 16691da177e4SLinus Torvalds if (retval) 16701da177e4SLinus Torvalds goto out; 1671900148dcSAl Viro mnt = real_mount(path.mnt); 16721da177e4SLinus Torvalds retval = -EINVAL; 16732d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 16741da177e4SLinus Torvalds goto dput_and_out; 1675143c8c91SAl Viro if (!check_mnt(mnt)) 16761da177e4SLinus Torvalds goto dput_and_out; 167725d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 16785ff9d8a6SEric W. Biederman goto dput_and_out; 1679b2f5d4dcSEric W. Biederman retval = -EPERM; 1680b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1681b2f5d4dcSEric W. Biederman goto dput_and_out; 16821da177e4SLinus Torvalds 1683900148dcSAl Viro retval = do_umount(mnt, flags); 16841da177e4SLinus Torvalds dput_and_out: 1685429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 16862d8f3038SAl Viro dput(path.dentry); 1687900148dcSAl Viro mntput_no_expire(mnt); 16881da177e4SLinus Torvalds out: 16891da177e4SLinus Torvalds return retval; 16901da177e4SLinus Torvalds } 16911da177e4SLinus Torvalds 16923a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 16933a18ef5cSDominik Brodowski { 16943a18ef5cSDominik Brodowski return ksys_umount(name, flags); 16953a18ef5cSDominik Brodowski } 16963a18ef5cSDominik Brodowski 16971da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 16981da177e4SLinus Torvalds 16991da177e4SLinus Torvalds /* 17001da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17011da177e4SLinus Torvalds */ 1702bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17031da177e4SLinus Torvalds { 17043a18ef5cSDominik Brodowski return ksys_umount(name, 0); 17051da177e4SLinus Torvalds } 17061da177e4SLinus Torvalds 17071da177e4SLinus Torvalds #endif 17081da177e4SLinus Torvalds 17094ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 17108823c079SEric W. Biederman { 17114ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1712e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1713e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 17144ce5d2b1SEric W. Biederman } 17154ce5d2b1SEric W. Biederman 171658be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 171758be2825SAl Viro { 171858be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 171958be2825SAl Viro } 172058be2825SAl Viro 17214ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 17224ce5d2b1SEric W. Biederman { 17234ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 17244ce5d2b1SEric W. Biederman * mount namespace loop? 17254ce5d2b1SEric W. Biederman */ 17264ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 17274ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 17284ce5d2b1SEric W. Biederman return false; 17294ce5d2b1SEric W. Biederman 1730f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 17318823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 17328823c079SEric W. Biederman } 17338823c079SEric W. Biederman 173487129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 173536341f64SRam Pai int flag) 17361da177e4SLinus Torvalds { 173784d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 17381da177e4SLinus Torvalds 17394ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 17404ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 17414ce5d2b1SEric W. Biederman 17424ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1743be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 17449676f0c6SRam Pai 174536341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1746be34d1a3SDavid Howells if (IS_ERR(q)) 1747be34d1a3SDavid Howells return q; 1748be34d1a3SDavid Howells 1749a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17501da177e4SLinus Torvalds 17511da177e4SLinus Torvalds p = mnt; 17526b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1753315fc83eSAl Viro struct mount *s; 17547ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17551da177e4SLinus Torvalds continue; 17561da177e4SLinus Torvalds 1757909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 17584ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17594ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1760df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1761df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1762df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1763df7342b2SEric W. Biederman goto out; 1764df7342b2SEric W. Biederman } else { 17654ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17664ce5d2b1SEric W. Biederman continue; 17674ce5d2b1SEric W. Biederman } 1768df7342b2SEric W. Biederman } 17694ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17704ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17719676f0c6SRam Pai s = skip_mnt_tree(s); 17729676f0c6SRam Pai continue; 17739676f0c6SRam Pai } 17740714a533SAl Viro while (p != s->mnt_parent) { 17750714a533SAl Viro p = p->mnt_parent; 17760714a533SAl Viro q = q->mnt_parent; 17771da177e4SLinus Torvalds } 177887129cc0SAl Viro p = s; 177984d17192SAl Viro parent = q; 178087129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1781be34d1a3SDavid Howells if (IS_ERR(q)) 1782be34d1a3SDavid Howells goto out; 1783719ea2fbSAl Viro lock_mount_hash(); 17841a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 17851064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1786719ea2fbSAl Viro unlock_mount_hash(); 17871da177e4SLinus Torvalds } 17881da177e4SLinus Torvalds } 17891da177e4SLinus Torvalds return res; 1790be34d1a3SDavid Howells out: 17911da177e4SLinus Torvalds if (res) { 1792719ea2fbSAl Viro lock_mount_hash(); 1793e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1794719ea2fbSAl Viro unlock_mount_hash(); 17951da177e4SLinus Torvalds } 1796be34d1a3SDavid Howells return q; 17971da177e4SLinus Torvalds } 17981da177e4SLinus Torvalds 1799be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1800be34d1a3SDavid Howells 1801ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 18028aec0809SAl Viro { 1803cb338d06SAl Viro struct mount *tree; 180497216be0SAl Viro namespace_lock(); 1805cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1806cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1807cd4a4017SEric W. Biederman else 180887129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 180987129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1810328e6d90SAl Viro namespace_unlock(); 1811be34d1a3SDavid Howells if (IS_ERR(tree)) 181252e220d3SDan Carpenter return ERR_CAST(tree); 1813be34d1a3SDavid Howells return &tree->mnt; 18148aec0809SAl Viro } 18158aec0809SAl Viro 18168aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 18178aec0809SAl Viro { 181897216be0SAl Viro namespace_lock(); 1819719ea2fbSAl Viro lock_mount_hash(); 18209c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1821719ea2fbSAl Viro unlock_mount_hash(); 18223ab6abeeSAl Viro namespace_unlock(); 18238aec0809SAl Viro } 18248aec0809SAl Viro 1825c771d683SMiklos Szeredi /** 1826c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1827c771d683SMiklos Szeredi * 1828c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1829c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1830c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1831c771d683SMiklos Szeredi * 1832c771d683SMiklos Szeredi * Release with mntput(). 1833c771d683SMiklos Szeredi */ 1834ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1835c771d683SMiklos Szeredi { 1836c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1837c771d683SMiklos Szeredi struct mount *new_mnt; 1838c771d683SMiklos Szeredi 1839c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1840c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1841c771d683SMiklos Szeredi 1842c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1843c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1844c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1845c771d683SMiklos Szeredi 1846c771d683SMiklos Szeredi return &new_mnt->mnt; 1847c771d683SMiklos Szeredi } 1848c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1849c771d683SMiklos Szeredi 18501f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18511f707137SAl Viro struct vfsmount *root) 18521f707137SAl Viro { 18531a4eeaf2SAl Viro struct mount *mnt; 18541f707137SAl Viro int res = f(root, arg); 18551f707137SAl Viro if (res) 18561f707137SAl Viro return res; 18571a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18581a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18591f707137SAl Viro if (res) 18601f707137SAl Viro return res; 18611f707137SAl Viro } 18621f707137SAl Viro return 0; 18631f707137SAl Viro } 18641f707137SAl Viro 18653bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 18663bd045ccSAl Viro { 18673bd045ccSAl Viro struct mount *p; 18683bd045ccSAl Viro 18693bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 18703bd045ccSAl Viro int flags = p->mnt.mnt_flags; 18713bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 18723bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 18733bd045ccSAl Viro 18743bd045ccSAl Viro if (flags & MNT_READONLY) 18753bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 18763bd045ccSAl Viro 18773bd045ccSAl Viro if (flags & MNT_NODEV) 18783bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 18793bd045ccSAl Viro 18803bd045ccSAl Viro if (flags & MNT_NOSUID) 18813bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 18823bd045ccSAl Viro 18833bd045ccSAl Viro if (flags & MNT_NOEXEC) 18843bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 18853bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 18863bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 18873bd045ccSAl Viro flags |= MNT_LOCKED; 18883bd045ccSAl Viro p->mnt.mnt_flags = flags; 18893bd045ccSAl Viro } 18903bd045ccSAl Viro } 18913bd045ccSAl Viro 18924b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1893719f5d7fSMiklos Szeredi { 1894315fc83eSAl Viro struct mount *p; 1895719f5d7fSMiklos Szeredi 1896909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1897fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 18984b8b21f4SAl Viro mnt_release_group_id(p); 1899719f5d7fSMiklos Szeredi } 1900719f5d7fSMiklos Szeredi } 1901719f5d7fSMiklos Szeredi 19024b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1903719f5d7fSMiklos Szeredi { 1904315fc83eSAl Viro struct mount *p; 1905719f5d7fSMiklos Szeredi 1906909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1907fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 19084b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1909719f5d7fSMiklos Szeredi if (err) { 19104b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1911719f5d7fSMiklos Szeredi return err; 1912719f5d7fSMiklos Szeredi } 1913719f5d7fSMiklos Szeredi } 1914719f5d7fSMiklos Szeredi } 1915719f5d7fSMiklos Szeredi 1916719f5d7fSMiklos Szeredi return 0; 1917719f5d7fSMiklos Szeredi } 1918719f5d7fSMiklos Szeredi 1919d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1920d2921684SEric W. Biederman { 1921d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1922d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1923d2921684SEric W. Biederman struct mount *p; 1924d2921684SEric W. Biederman 1925d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1926d2921684SEric W. Biederman mounts++; 1927d2921684SEric W. Biederman 1928d2921684SEric W. Biederman old = ns->mounts; 1929d2921684SEric W. Biederman pending = ns->pending_mounts; 1930d2921684SEric W. Biederman sum = old + pending; 1931d2921684SEric W. Biederman if ((old > sum) || 1932d2921684SEric W. Biederman (pending > sum) || 1933d2921684SEric W. Biederman (max < sum) || 1934d2921684SEric W. Biederman (mounts > (max - sum))) 1935d2921684SEric W. Biederman return -ENOSPC; 1936d2921684SEric W. Biederman 1937d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1938d2921684SEric W. Biederman return 0; 1939d2921684SEric W. Biederman } 1940d2921684SEric W. Biederman 1941b90fa9aeSRam Pai /* 1942b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1943b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 194421444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 194521444403SRam Pai * store the parent mount and mountpoint dentry. 194621444403SRam Pai * (done when source_mnt is moved) 1947b90fa9aeSRam Pai * 1948b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1949b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 19509676f0c6SRam Pai * --------------------------------------------------------------------------- 1951b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 19529676f0c6SRam Pai * |************************************************************************** 19539676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19549676f0c6SRam Pai * | dest | | | | | 19559676f0c6SRam Pai * | | | | | | | 19569676f0c6SRam Pai * | v | | | | | 19579676f0c6SRam Pai * |************************************************************************** 19589676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 19595afe0022SRam Pai * | | | | | | 19609676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 19619676f0c6SRam Pai * *************************************************************************** 1962b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1963b90fa9aeSRam Pai * destination mount. 1964b90fa9aeSRam Pai * 1965b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1966b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1967b90fa9aeSRam Pai * the peer group of the source mount. 1968b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1969b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1970b90fa9aeSRam Pai * mount. 19715afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 19725afe0022SRam Pai * of the destination mount and the cloned mount is made slave 19735afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 19745afe0022SRam Pai * is marked as 'shared and slave'. 19755afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 19765afe0022SRam Pai * source mount. 19775afe0022SRam Pai * 19789676f0c6SRam Pai * --------------------------------------------------------------------------- 197921444403SRam Pai * | MOVE MOUNT OPERATION | 19809676f0c6SRam Pai * |************************************************************************** 19819676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19829676f0c6SRam Pai * | dest | | | | | 19839676f0c6SRam Pai * | | | | | | | 19849676f0c6SRam Pai * | v | | | | | 19859676f0c6SRam Pai * |************************************************************************** 19869676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 19875afe0022SRam Pai * | | | | | | 19889676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 19899676f0c6SRam Pai * *************************************************************************** 19905afe0022SRam Pai * 19915afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 19925afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 199321444403SRam Pai * (+*) the mount is moved to the destination. 19945afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 19955afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 19965afe0022SRam Pai * the mount is marked as 'shared and slave'. 19975afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1998b90fa9aeSRam Pai * 1999b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2000b90fa9aeSRam Pai * applied to each mount in the tree. 2001b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2002b90fa9aeSRam Pai * in allocations. 2003b90fa9aeSRam Pai */ 20040fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 200584d17192SAl Viro struct mount *dest_mnt, 200684d17192SAl Viro struct mountpoint *dest_mp, 200784d17192SAl Viro struct path *parent_path) 2008b90fa9aeSRam Pai { 20093bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 201038129a13SAl Viro HLIST_HEAD(tree_list); 2011d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 20121064f874SEric W. Biederman struct mountpoint *smp; 2013315fc83eSAl Viro struct mount *child, *p; 201438129a13SAl Viro struct hlist_node *n; 2015719f5d7fSMiklos Szeredi int err; 2016b90fa9aeSRam Pai 20171064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 20181064f874SEric W. Biederman * to be tucked under other mounts. 20191064f874SEric W. Biederman */ 20201064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 20211064f874SEric W. Biederman if (IS_ERR(smp)) 20221064f874SEric W. Biederman return PTR_ERR(smp); 20231064f874SEric W. Biederman 2024d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 2025d2921684SEric W. Biederman if (!parent_path) { 2026d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2027d2921684SEric W. Biederman if (err) 2028d2921684SEric W. Biederman goto out; 2029d2921684SEric W. Biederman } 2030d2921684SEric W. Biederman 2031fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 20320fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2033719f5d7fSMiklos Szeredi if (err) 2034719f5d7fSMiklos Szeredi goto out; 203584d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2036f2ebb3a9SAl Viro lock_mount_hash(); 2037719f5d7fSMiklos Szeredi if (err) 2038719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2039909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 20400f0afb1dSAl Viro set_mnt_shared(p); 20410b1b901bSAl Viro } else { 20420b1b901bSAl Viro lock_mount_hash(); 2043b90fa9aeSRam Pai } 20441a390689SAl Viro if (parent_path) { 20450fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 204684d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2047143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 204821444403SRam Pai } else { 204984d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 20501064f874SEric W. Biederman commit_tree(source_mnt); 205121444403SRam Pai } 2052b90fa9aeSRam Pai 205338129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 20541d6a32acSAl Viro struct mount *q; 205538129a13SAl Viro hlist_del_init(&child->mnt_hash); 20561064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 20571d6a32acSAl Viro child->mnt_mountpoint); 20581064f874SEric W. Biederman if (q) 20591064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 20603bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 20613bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 20623bd045ccSAl Viro lock_mnt_tree(child); 20631064f874SEric W. Biederman commit_tree(child); 2064b90fa9aeSRam Pai } 20651064f874SEric W. Biederman put_mountpoint(smp); 2066719ea2fbSAl Viro unlock_mount_hash(); 206799b7db7bSNick Piggin 2068b90fa9aeSRam Pai return 0; 2069719f5d7fSMiklos Szeredi 2070719f5d7fSMiklos Szeredi out_cleanup_ids: 2071f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2072f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2073d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2074e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2075f2ebb3a9SAl Viro } 2076f2ebb3a9SAl Viro unlock_mount_hash(); 20770fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2078719f5d7fSMiklos Szeredi out: 2079d2921684SEric W. Biederman ns->pending_mounts = 0; 20801064f874SEric W. Biederman 20811064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 20821064f874SEric W. Biederman put_mountpoint(smp); 20831064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 20841064f874SEric W. Biederman 2085719f5d7fSMiklos Szeredi return err; 2086b90fa9aeSRam Pai } 2087b90fa9aeSRam Pai 208884d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2089b12cea91SAl Viro { 2090b12cea91SAl Viro struct vfsmount *mnt; 209184d17192SAl Viro struct dentry *dentry = path->dentry; 2092b12cea91SAl Viro retry: 20935955102cSAl Viro inode_lock(dentry->d_inode); 209484d17192SAl Viro if (unlikely(cant_mount(dentry))) { 20955955102cSAl Viro inode_unlock(dentry->d_inode); 209684d17192SAl Viro return ERR_PTR(-ENOENT); 2097b12cea91SAl Viro } 209897216be0SAl Viro namespace_lock(); 2099b12cea91SAl Viro mnt = lookup_mnt(path); 210084d17192SAl Viro if (likely(!mnt)) { 21013895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 210284d17192SAl Viro if (IS_ERR(mp)) { 210397216be0SAl Viro namespace_unlock(); 21045955102cSAl Viro inode_unlock(dentry->d_inode); 210584d17192SAl Viro return mp; 210684d17192SAl Viro } 210784d17192SAl Viro return mp; 210884d17192SAl Viro } 210997216be0SAl Viro namespace_unlock(); 21105955102cSAl Viro inode_unlock(path->dentry->d_inode); 2111b12cea91SAl Viro path_put(path); 2112b12cea91SAl Viro path->mnt = mnt; 211384d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2114b12cea91SAl Viro goto retry; 2115b12cea91SAl Viro } 2116b12cea91SAl Viro 211784d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2118b12cea91SAl Viro { 211984d17192SAl Viro struct dentry *dentry = where->m_dentry; 21203895dbf8SEric W. Biederman 21213895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 212284d17192SAl Viro put_mountpoint(where); 21233895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 21243895dbf8SEric W. Biederman 2125328e6d90SAl Viro namespace_unlock(); 21265955102cSAl Viro inode_unlock(dentry->d_inode); 2127b12cea91SAl Viro } 2128b12cea91SAl Viro 212984d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 21301da177e4SLinus Torvalds { 2131e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 21321da177e4SLinus Torvalds return -EINVAL; 21331da177e4SLinus Torvalds 2134e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2135e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 21361da177e4SLinus Torvalds return -ENOTDIR; 21371da177e4SLinus Torvalds 213884d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 21391da177e4SLinus Torvalds } 21401da177e4SLinus Torvalds 21411da177e4SLinus Torvalds /* 21427a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 21437a2e8a8fSValerie Aurora */ 21447a2e8a8fSValerie Aurora 2145e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 21467a2e8a8fSValerie Aurora { 2147e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 21487a2e8a8fSValerie Aurora 21497a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 21507a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 21517a2e8a8fSValerie Aurora return 0; 21527a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 21537a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 21547a2e8a8fSValerie Aurora return 0; 21557a2e8a8fSValerie Aurora return type; 21567a2e8a8fSValerie Aurora } 21577a2e8a8fSValerie Aurora 21587a2e8a8fSValerie Aurora /* 215907b20889SRam Pai * recursively change the type of the mountpoint. 216007b20889SRam Pai */ 2161e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 216207b20889SRam Pai { 2163315fc83eSAl Viro struct mount *m; 21644b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2165e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 21667a2e8a8fSValerie Aurora int type; 2167719f5d7fSMiklos Szeredi int err = 0; 216807b20889SRam Pai 21692d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 217007b20889SRam Pai return -EINVAL; 217107b20889SRam Pai 2172e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 21737a2e8a8fSValerie Aurora if (!type) 21747a2e8a8fSValerie Aurora return -EINVAL; 21757a2e8a8fSValerie Aurora 217697216be0SAl Viro namespace_lock(); 2177719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2178719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2179719f5d7fSMiklos Szeredi if (err) 2180719f5d7fSMiklos Szeredi goto out_unlock; 2181719f5d7fSMiklos Szeredi } 2182719f5d7fSMiklos Szeredi 2183719ea2fbSAl Viro lock_mount_hash(); 2184909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 21850f0afb1dSAl Viro change_mnt_propagation(m, type); 2186719ea2fbSAl Viro unlock_mount_hash(); 2187719f5d7fSMiklos Szeredi 2188719f5d7fSMiklos Szeredi out_unlock: 218997216be0SAl Viro namespace_unlock(); 2190719f5d7fSMiklos Szeredi return err; 219107b20889SRam Pai } 219207b20889SRam Pai 21935ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 21945ff9d8a6SEric W. Biederman { 21955ff9d8a6SEric W. Biederman struct mount *child; 21965ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 21975ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 21985ff9d8a6SEric W. Biederman continue; 21995ff9d8a6SEric W. Biederman 22005ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 22015ff9d8a6SEric W. Biederman return true; 22025ff9d8a6SEric W. Biederman } 22035ff9d8a6SEric W. Biederman return false; 22045ff9d8a6SEric W. Biederman } 22055ff9d8a6SEric W. Biederman 220607b20889SRam Pai /* 22071da177e4SLinus Torvalds * do loopback mount. 22081da177e4SLinus Torvalds */ 2209808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 22102dafe1c4SEric Sandeen int recurse) 22111da177e4SLinus Torvalds { 22122d92ab3cSAl Viro struct path old_path; 221384d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 221484d17192SAl Viro struct mountpoint *mp; 221557eccb83SAl Viro int err; 22161da177e4SLinus Torvalds if (!old_name || !*old_name) 22171da177e4SLinus Torvalds return -EINVAL; 2218815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 22191da177e4SLinus Torvalds if (err) 22201da177e4SLinus Torvalds return err; 22211da177e4SLinus Torvalds 22228823c079SEric W. Biederman err = -EINVAL; 22234ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 22248823c079SEric W. Biederman goto out; 22258823c079SEric W. Biederman 222684d17192SAl Viro mp = lock_mount(path); 222784d17192SAl Viro err = PTR_ERR(mp); 222884d17192SAl Viro if (IS_ERR(mp)) 22299676f0c6SRam Pai goto out; 22309676f0c6SRam Pai 223187129cc0SAl Viro old = real_mount(old_path.mnt); 223284d17192SAl Viro parent = real_mount(path->mnt); 223387129cc0SAl Viro 2234b12cea91SAl Viro err = -EINVAL; 2235fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 2236b12cea91SAl Viro goto out2; 2237b12cea91SAl Viro 2238e149ed2bSAl Viro if (!check_mnt(parent)) 2239e149ed2bSAl Viro goto out2; 2240e149ed2bSAl Viro 2241e149ed2bSAl Viro if (!check_mnt(old) && old_path.dentry->d_op != &ns_dentry_operations) 2242b12cea91SAl Viro goto out2; 2243ccd48bc7SAl Viro 22445ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 22455ff9d8a6SEric W. Biederman goto out2; 22465ff9d8a6SEric W. Biederman 22471da177e4SLinus Torvalds if (recurse) 22484ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 22491da177e4SLinus Torvalds else 225087129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 22511da177e4SLinus Torvalds 2252be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2253be34d1a3SDavid Howells err = PTR_ERR(mnt); 2254e9c5d8a5SAndrey Vagin goto out2; 2255be34d1a3SDavid Howells } 2256ccd48bc7SAl Viro 22575ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 22585ff9d8a6SEric W. Biederman 225984d17192SAl Viro err = graft_tree(mnt, parent, mp); 22601da177e4SLinus Torvalds if (err) { 2261719ea2fbSAl Viro lock_mount_hash(); 2262e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2263719ea2fbSAl Viro unlock_mount_hash(); 22645b83d2c5SRam Pai } 2265b12cea91SAl Viro out2: 226684d17192SAl Viro unlock_mount(mp); 2267ccd48bc7SAl Viro out: 22682d92ab3cSAl Viro path_put(&old_path); 22691da177e4SLinus Torvalds return err; 22701da177e4SLinus Torvalds } 22711da177e4SLinus Torvalds 227243f5e655SDavid Howells /* 227343f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 227443f5e655SDavid Howells * 227543f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 227643f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 227743f5e655SDavid Howells */ 227843f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 22792e4b7fcdSDave Hansen { 228043f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 22812e4b7fcdSDave Hansen 228243f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 228343f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 228443f5e655SDavid Howells return false; 228543f5e655SDavid Howells 228643f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 228743f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 228843f5e655SDavid Howells return false; 228943f5e655SDavid Howells 229043f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 229143f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 229243f5e655SDavid Howells return false; 229343f5e655SDavid Howells 229443f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 229543f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 229643f5e655SDavid Howells return false; 229743f5e655SDavid Howells 229843f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 229943f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 230043f5e655SDavid Howells return false; 230143f5e655SDavid Howells 230243f5e655SDavid Howells return true; 230343f5e655SDavid Howells } 230443f5e655SDavid Howells 230543f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 230643f5e655SDavid Howells { 230743f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 230843f5e655SDavid Howells 230943f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 23102e4b7fcdSDave Hansen return 0; 23112e4b7fcdSDave Hansen 23122e4b7fcdSDave Hansen if (readonly_request) 231343f5e655SDavid Howells return mnt_make_readonly(mnt); 231443f5e655SDavid Howells 231543f5e655SDavid Howells return __mnt_unmake_readonly(mnt); 231643f5e655SDavid Howells } 231743f5e655SDavid Howells 231843f5e655SDavid Howells /* 231943f5e655SDavid Howells * Update the user-settable attributes on a mount. The caller must hold 232043f5e655SDavid Howells * sb->s_umount for writing. 232143f5e655SDavid Howells */ 232243f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 232343f5e655SDavid Howells { 232443f5e655SDavid Howells lock_mount_hash(); 232543f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 232643f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 232743f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 232843f5e655SDavid Howells unlock_mount_hash(); 232943f5e655SDavid Howells } 233043f5e655SDavid Howells 233143f5e655SDavid Howells /* 233243f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 233343f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 233443f5e655SDavid Howells * to mount(2). 233543f5e655SDavid Howells */ 233643f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 233743f5e655SDavid Howells { 233843f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 233943f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 234043f5e655SDavid Howells int ret; 234143f5e655SDavid Howells 234243f5e655SDavid Howells if (!check_mnt(mnt)) 234343f5e655SDavid Howells return -EINVAL; 234443f5e655SDavid Howells 234543f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 234643f5e655SDavid Howells return -EINVAL; 234743f5e655SDavid Howells 234843f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 234943f5e655SDavid Howells return -EPERM; 235043f5e655SDavid Howells 235143f5e655SDavid Howells down_write(&sb->s_umount); 235243f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 235343f5e655SDavid Howells if (ret == 0) 235443f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 235543f5e655SDavid Howells up_write(&sb->s_umount); 235643f5e655SDavid Howells return ret; 23572e4b7fcdSDave Hansen } 23582e4b7fcdSDave Hansen 23591da177e4SLinus Torvalds /* 23601da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 23611da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 23621da177e4SLinus Torvalds * on it - tough luck. 23631da177e4SLinus Torvalds */ 2364e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2365e462ec50SDavid Howells int mnt_flags, void *data) 23661da177e4SLinus Torvalds { 23671da177e4SLinus Torvalds int err; 23682d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2369143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 2370204cc0ccSAl Viro void *sec_opts = NULL; 23711da177e4SLinus Torvalds 2372143c8c91SAl Viro if (!check_mnt(mnt)) 23731da177e4SLinus Torvalds return -EINVAL; 23741da177e4SLinus Torvalds 23752d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 23761da177e4SLinus Torvalds return -EINVAL; 23771da177e4SLinus Torvalds 237843f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 237907b64558SEric W. Biederman return -EPERM; 23809566d674SEric W. Biederman 2381c039bc3cSAl Viro if (data && !(sb->s_type->fs_flags & FS_BINARY_MOUNTDATA)) { 2382204cc0ccSAl Viro err = security_sb_eat_lsm_opts(data, &sec_opts); 2383c039bc3cSAl Viro if (err) 2384c039bc3cSAl Viro return err; 2385c039bc3cSAl Viro } 2386204cc0ccSAl Viro err = security_sb_remount(sb, sec_opts); 2387204cc0ccSAl Viro security_free_mnt_opts(&sec_opts); 2388ff36fe2cSEric Paris if (err) 2389ff36fe2cSEric Paris return err; 2390ff36fe2cSEric Paris 23911da177e4SLinus Torvalds down_write(&sb->s_umount); 239257eccb83SAl Viro err = -EPERM; 239343f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 2394e462ec50SDavid Howells err = do_remount_sb(sb, sb_flags, data, 0); 239543f5e655SDavid Howells if (!err) 239643f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 23970e55a7ccSDan Williams } 23986339dab8SAl Viro up_write(&sb->s_umount); 23991da177e4SLinus Torvalds return err; 24001da177e4SLinus Torvalds } 24011da177e4SLinus Torvalds 2402cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 24039676f0c6SRam Pai { 2404315fc83eSAl Viro struct mount *p; 2405909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2406fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 24079676f0c6SRam Pai return 1; 24089676f0c6SRam Pai } 24099676f0c6SRam Pai return 0; 24109676f0c6SRam Pai } 24119676f0c6SRam Pai 2412808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 24131da177e4SLinus Torvalds { 24142d92ab3cSAl Viro struct path old_path, parent_path; 2415676da58dSAl Viro struct mount *p; 24160fb54e50SAl Viro struct mount *old; 241784d17192SAl Viro struct mountpoint *mp; 241857eccb83SAl Viro int err; 24191da177e4SLinus Torvalds if (!old_name || !*old_name) 24201da177e4SLinus Torvalds return -EINVAL; 24212d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 24221da177e4SLinus Torvalds if (err) 24231da177e4SLinus Torvalds return err; 24241da177e4SLinus Torvalds 242584d17192SAl Viro mp = lock_mount(path); 242684d17192SAl Viro err = PTR_ERR(mp); 242784d17192SAl Viro if (IS_ERR(mp)) 2428cc53ce53SDavid Howells goto out; 2429cc53ce53SDavid Howells 2430143c8c91SAl Viro old = real_mount(old_path.mnt); 2431fc7be130SAl Viro p = real_mount(path->mnt); 2432143c8c91SAl Viro 24331da177e4SLinus Torvalds err = -EINVAL; 2434fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 24351da177e4SLinus Torvalds goto out1; 24361da177e4SLinus Torvalds 24375ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 24385ff9d8a6SEric W. Biederman goto out1; 24395ff9d8a6SEric W. Biederman 24401da177e4SLinus Torvalds err = -EINVAL; 24412d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 244221444403SRam Pai goto out1; 24431da177e4SLinus Torvalds 2444676da58dSAl Viro if (!mnt_has_parent(old)) 244521444403SRam Pai goto out1; 24461da177e4SLinus Torvalds 2447e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2448e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 244921444403SRam Pai goto out1; 245021444403SRam Pai /* 245121444403SRam Pai * Don't move a mount residing in a shared parent. 245221444403SRam Pai */ 2453fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 245421444403SRam Pai goto out1; 24559676f0c6SRam Pai /* 24569676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 24579676f0c6SRam Pai * mount which is shared. 24589676f0c6SRam Pai */ 2459fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 24609676f0c6SRam Pai goto out1; 24611da177e4SLinus Torvalds err = -ELOOP; 2462fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2463676da58dSAl Viro if (p == old) 246421444403SRam Pai goto out1; 24651da177e4SLinus Torvalds 246684d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 24674ac91378SJan Blunck if (err) 246821444403SRam Pai goto out1; 24691da177e4SLinus Torvalds 24701da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 24711da177e4SLinus Torvalds * automatically */ 24726776db3dSAl Viro list_del_init(&old->mnt_expire); 24731da177e4SLinus Torvalds out1: 247484d17192SAl Viro unlock_mount(mp); 24751da177e4SLinus Torvalds out: 24761da177e4SLinus Torvalds if (!err) 24771a390689SAl Viro path_put(&parent_path); 24782d92ab3cSAl Viro path_put(&old_path); 24791da177e4SLinus Torvalds return err; 24801da177e4SLinus Torvalds } 24811da177e4SLinus Torvalds 24829d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 24839d412a43SAl Viro { 24849d412a43SAl Viro int err; 24859d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 24869d412a43SAl Viro if (subtype) { 24879d412a43SAl Viro subtype++; 24889d412a43SAl Viro err = -EINVAL; 24899d412a43SAl Viro if (!subtype[0]) 24909d412a43SAl Viro goto err; 24919d412a43SAl Viro } else 24929d412a43SAl Viro subtype = ""; 24939d412a43SAl Viro 24949d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 24959d412a43SAl Viro err = -ENOMEM; 24969d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 24979d412a43SAl Viro goto err; 24989d412a43SAl Viro return mnt; 24999d412a43SAl Viro 25009d412a43SAl Viro err: 25019d412a43SAl Viro mntput(mnt); 25029d412a43SAl Viro return ERR_PTR(err); 25039d412a43SAl Viro } 25049d412a43SAl Viro 25059d412a43SAl Viro /* 25069d412a43SAl Viro * add a mount into a namespace's mount tree 25079d412a43SAl Viro */ 250895bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 25099d412a43SAl Viro { 251084d17192SAl Viro struct mountpoint *mp; 251184d17192SAl Viro struct mount *parent; 25129d412a43SAl Viro int err; 25139d412a43SAl Viro 2514f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 25159d412a43SAl Viro 251684d17192SAl Viro mp = lock_mount(path); 251784d17192SAl Viro if (IS_ERR(mp)) 251884d17192SAl Viro return PTR_ERR(mp); 25199d412a43SAl Viro 252084d17192SAl Viro parent = real_mount(path->mnt); 25219d412a43SAl Viro err = -EINVAL; 252284d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2523156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2524156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 25259d412a43SAl Viro goto unlock; 2526156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 252784d17192SAl Viro if (!parent->mnt_ns) 2528156cacb1SAl Viro goto unlock; 2529156cacb1SAl Viro } 25309d412a43SAl Viro 25319d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 25329d412a43SAl Viro err = -EBUSY; 253395bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 25349d412a43SAl Viro path->mnt->mnt_root == path->dentry) 25359d412a43SAl Viro goto unlock; 25369d412a43SAl Viro 25379d412a43SAl Viro err = -EINVAL; 2538e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 25399d412a43SAl Viro goto unlock; 25409d412a43SAl Viro 254195bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 254284d17192SAl Viro err = graft_tree(newmnt, parent, mp); 25439d412a43SAl Viro 25449d412a43SAl Viro unlock: 254584d17192SAl Viro unlock_mount(mp); 25469d412a43SAl Viro return err; 25479d412a43SAl Viro } 2548b1e75df4SAl Viro 25498654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags); 25501b852bceSEric W. Biederman 25511da177e4SLinus Torvalds /* 25521da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 25531da177e4SLinus Torvalds * namespace's tree 25541da177e4SLinus Torvalds */ 2555e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2556808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 25571da177e4SLinus Torvalds { 25580c55cfc4SEric W. Biederman struct file_system_type *type; 25591da177e4SLinus Torvalds struct vfsmount *mnt; 256015f9a3f3SAl Viro int err; 25611da177e4SLinus Torvalds 25620c55cfc4SEric W. Biederman if (!fstype) 25631da177e4SLinus Torvalds return -EINVAL; 25641da177e4SLinus Torvalds 25650c55cfc4SEric W. Biederman type = get_fs_type(fstype); 25660c55cfc4SEric W. Biederman if (!type) 25670c55cfc4SEric W. Biederman return -ENODEV; 25680c55cfc4SEric W. Biederman 2569e462ec50SDavid Howells mnt = vfs_kern_mount(type, sb_flags, name, data); 25700c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 25710c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 25720c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 25730c55cfc4SEric W. Biederman 25740c55cfc4SEric W. Biederman put_filesystem(type); 25751da177e4SLinus Torvalds if (IS_ERR(mnt)) 25761da177e4SLinus Torvalds return PTR_ERR(mnt); 25771da177e4SLinus Torvalds 25788654df4eSEric W. Biederman if (mount_too_revealing(mnt, &mnt_flags)) { 25798654df4eSEric W. Biederman mntput(mnt); 25808654df4eSEric W. Biederman return -EPERM; 25818654df4eSEric W. Biederman } 25828654df4eSEric W. Biederman 258395bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 258415f9a3f3SAl Viro if (err) 258515f9a3f3SAl Viro mntput(mnt); 258615f9a3f3SAl Viro return err; 25871da177e4SLinus Torvalds } 25881da177e4SLinus Torvalds 258919a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 259019a167afSAl Viro { 25916776db3dSAl Viro struct mount *mnt = real_mount(m); 259219a167afSAl Viro int err; 259319a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 259419a167afSAl Viro * expired before we get a chance to add it 259519a167afSAl Viro */ 25966776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 259719a167afSAl Viro 259819a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 259919a167afSAl Viro m->mnt_root == path->dentry) { 2600b1e75df4SAl Viro err = -ELOOP; 2601b1e75df4SAl Viro goto fail; 260219a167afSAl Viro } 260319a167afSAl Viro 260495bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2605b1e75df4SAl Viro if (!err) 2606b1e75df4SAl Viro return 0; 2607b1e75df4SAl Viro fail: 2608b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 26096776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 261097216be0SAl Viro namespace_lock(); 26116776db3dSAl Viro list_del_init(&mnt->mnt_expire); 261297216be0SAl Viro namespace_unlock(); 261319a167afSAl Viro } 2614b1e75df4SAl Viro mntput(m); 2615b1e75df4SAl Viro mntput(m); 261619a167afSAl Viro return err; 261719a167afSAl Viro } 261819a167afSAl Viro 2619ea5b778aSDavid Howells /** 2620ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2621ea5b778aSDavid Howells * @mnt: The mount to list. 2622ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2623ea5b778aSDavid Howells */ 2624ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2625ea5b778aSDavid Howells { 262697216be0SAl Viro namespace_lock(); 2627ea5b778aSDavid Howells 26286776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2629ea5b778aSDavid Howells 263097216be0SAl Viro namespace_unlock(); 2631ea5b778aSDavid Howells } 2632ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2633ea5b778aSDavid Howells 2634ea5b778aSDavid Howells /* 26351da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 26361da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 26371da177e4SLinus Torvalds * here 26381da177e4SLinus Torvalds */ 26391da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 26401da177e4SLinus Torvalds { 2641761d5c38SAl Viro struct mount *mnt, *next; 26421da177e4SLinus Torvalds LIST_HEAD(graveyard); 26431da177e4SLinus Torvalds 26441da177e4SLinus Torvalds if (list_empty(mounts)) 26451da177e4SLinus Torvalds return; 26461da177e4SLinus Torvalds 264797216be0SAl Viro namespace_lock(); 2648719ea2fbSAl Viro lock_mount_hash(); 26491da177e4SLinus Torvalds 26501da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 26511da177e4SLinus Torvalds * following criteria: 26521da177e4SLinus Torvalds * - only referenced by its parent vfsmount 26531da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 26541da177e4SLinus Torvalds * cleared by mntput()) 26551da177e4SLinus Torvalds */ 26566776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2657863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 26581ab59738SAl Viro propagate_mount_busy(mnt, 1)) 26591da177e4SLinus Torvalds continue; 26606776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 26611da177e4SLinus Torvalds } 2662bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 26636776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2664143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2665e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2666bcc5c7d2SAl Viro } 2667719ea2fbSAl Viro unlock_mount_hash(); 26683ab6abeeSAl Viro namespace_unlock(); 26691da177e4SLinus Torvalds } 26701da177e4SLinus Torvalds 26711da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 26721da177e4SLinus Torvalds 26731da177e4SLinus Torvalds /* 26745528f911STrond Myklebust * Ripoff of 'select_parent()' 26755528f911STrond Myklebust * 26765528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 26775528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 26785528f911STrond Myklebust */ 2679692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 26805528f911STrond Myklebust { 2681692afc31SAl Viro struct mount *this_parent = parent; 26825528f911STrond Myklebust struct list_head *next; 26835528f911STrond Myklebust int found = 0; 26845528f911STrond Myklebust 26855528f911STrond Myklebust repeat: 26866b41d536SAl Viro next = this_parent->mnt_mounts.next; 26875528f911STrond Myklebust resume: 26886b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 26895528f911STrond Myklebust struct list_head *tmp = next; 26906b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 26915528f911STrond Myklebust 26925528f911STrond Myklebust next = tmp->next; 2693692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 26945528f911STrond Myklebust continue; 26955528f911STrond Myklebust /* 26965528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 26975528f911STrond Myklebust */ 26986b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 26995528f911STrond Myklebust this_parent = mnt; 27005528f911STrond Myklebust goto repeat; 27015528f911STrond Myklebust } 27025528f911STrond Myklebust 27031ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 27046776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 27055528f911STrond Myklebust found++; 27065528f911STrond Myklebust } 27075528f911STrond Myklebust } 27085528f911STrond Myklebust /* 27095528f911STrond Myklebust * All done at this level ... ascend and resume the search 27105528f911STrond Myklebust */ 27115528f911STrond Myklebust if (this_parent != parent) { 27126b41d536SAl Viro next = this_parent->mnt_child.next; 27130714a533SAl Viro this_parent = this_parent->mnt_parent; 27145528f911STrond Myklebust goto resume; 27155528f911STrond Myklebust } 27165528f911STrond Myklebust return found; 27175528f911STrond Myklebust } 27185528f911STrond Myklebust 27195528f911STrond Myklebust /* 27205528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 27215528f911STrond Myklebust * submounts of a specific parent mountpoint 272299b7db7bSNick Piggin * 272348a066e7SAl Viro * mount_lock must be held for write 27245528f911STrond Myklebust */ 2725b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 27265528f911STrond Myklebust { 27275528f911STrond Myklebust LIST_HEAD(graveyard); 2728761d5c38SAl Viro struct mount *m; 27295528f911STrond Myklebust 27305528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2731c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2732bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2733761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 27346776db3dSAl Viro mnt_expire); 2735143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2736e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2737bcc5c7d2SAl Viro } 2738bcc5c7d2SAl Viro } 27395528f911STrond Myklebust } 27405528f911STrond Myklebust 27415528f911STrond Myklebust /* 27421da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 27431da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 27441da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 27451da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 27461da177e4SLinus Torvalds */ 2747b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2748b58fed8bSRam Pai unsigned long n) 27491da177e4SLinus Torvalds { 27501da177e4SLinus Torvalds char *t = to; 27511da177e4SLinus Torvalds const char __user *f = from; 27521da177e4SLinus Torvalds char c; 27531da177e4SLinus Torvalds 275496d4f267SLinus Torvalds if (!access_ok(from, n)) 27551da177e4SLinus Torvalds return n; 27561da177e4SLinus Torvalds 27579da3f2b7SJann Horn current->kernel_uaccess_faults_ok++; 27581da177e4SLinus Torvalds while (n) { 27591da177e4SLinus Torvalds if (__get_user(c, f)) { 27601da177e4SLinus Torvalds memset(t, 0, n); 27611da177e4SLinus Torvalds break; 27621da177e4SLinus Torvalds } 27631da177e4SLinus Torvalds *t++ = c; 27641da177e4SLinus Torvalds f++; 27651da177e4SLinus Torvalds n--; 27661da177e4SLinus Torvalds } 27679da3f2b7SJann Horn current->kernel_uaccess_faults_ok--; 27681da177e4SLinus Torvalds return n; 27691da177e4SLinus Torvalds } 27701da177e4SLinus Torvalds 2771b40ef869SAl Viro void *copy_mount_options(const void __user * data) 27721da177e4SLinus Torvalds { 27731da177e4SLinus Torvalds int i; 27741da177e4SLinus Torvalds unsigned long size; 2775b40ef869SAl Viro char *copy; 27761da177e4SLinus Torvalds 27771da177e4SLinus Torvalds if (!data) 2778b40ef869SAl Viro return NULL; 27791da177e4SLinus Torvalds 2780b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2781b40ef869SAl Viro if (!copy) 2782b40ef869SAl Viro return ERR_PTR(-ENOMEM); 27831da177e4SLinus Torvalds 27841da177e4SLinus Torvalds /* We only care that *some* data at the address the user 27851da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 27861da177e4SLinus Torvalds * the remainder of the page. 27871da177e4SLinus Torvalds */ 27881da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 27891da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 27901da177e4SLinus Torvalds if (size > PAGE_SIZE) 27911da177e4SLinus Torvalds size = PAGE_SIZE; 27921da177e4SLinus Torvalds 2793b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 27941da177e4SLinus Torvalds if (!i) { 2795b40ef869SAl Viro kfree(copy); 2796b40ef869SAl Viro return ERR_PTR(-EFAULT); 27971da177e4SLinus Torvalds } 27981da177e4SLinus Torvalds if (i != PAGE_SIZE) 2799b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2800b40ef869SAl Viro return copy; 28011da177e4SLinus Torvalds } 28021da177e4SLinus Torvalds 2803b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2804eca6f534SVegard Nossum { 2805b8850d1fSTim Gardner return data ? strndup_user(data, PAGE_SIZE) : NULL; 2806eca6f534SVegard Nossum } 2807eca6f534SVegard Nossum 28081da177e4SLinus Torvalds /* 28091da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 28101da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 28111da177e4SLinus Torvalds * 28121da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 28131da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 28141da177e4SLinus Torvalds * information (or be NULL). 28151da177e4SLinus Torvalds * 28161da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 28171da177e4SLinus Torvalds * When the flags word was introduced its top half was required 28181da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 28191da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 28201da177e4SLinus Torvalds * and must be discarded. 28211da177e4SLinus Torvalds */ 28225e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2823808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 28241da177e4SLinus Torvalds { 28252d92ab3cSAl Viro struct path path; 2826e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 28271da177e4SLinus Torvalds int retval = 0; 28281da177e4SLinus Torvalds 28291da177e4SLinus Torvalds /* Discard magic */ 28301da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 28311da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 28321da177e4SLinus Torvalds 28331da177e4SLinus Torvalds /* Basic sanity checks */ 28341da177e4SLinus Torvalds if (data_page) 28351da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 28361da177e4SLinus Torvalds 2837e462ec50SDavid Howells if (flags & MS_NOUSER) 2838e462ec50SDavid Howells return -EINVAL; 2839e462ec50SDavid Howells 2840a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 28415e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2842a27ab9f2STetsuo Handa if (retval) 2843a27ab9f2STetsuo Handa return retval; 2844a27ab9f2STetsuo Handa 2845a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2846a27ab9f2STetsuo Handa type_page, flags, data_page); 28470d5cadb8SAl Viro if (!retval && !may_mount()) 28480d5cadb8SAl Viro retval = -EPERM; 2849e462ec50SDavid Howells if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) 28509e8925b6SJeff Layton retval = -EPERM; 2851a27ab9f2STetsuo Handa if (retval) 2852a27ab9f2STetsuo Handa goto dput_out; 2853a27ab9f2STetsuo Handa 2854613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2855613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 28560a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 28570a1c01c9SMatthew Garrett 28581da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 28591da177e4SLinus Torvalds if (flags & MS_NOSUID) 28601da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 28611da177e4SLinus Torvalds if (flags & MS_NODEV) 28621da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 28631da177e4SLinus Torvalds if (flags & MS_NOEXEC) 28641da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2865fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2866fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2867fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2868fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2869d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2870d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 2871a9e5b732SDavid Howells if (flags & MS_RDONLY) 28722e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2873fc33a7bbSChristoph Hellwig 2874ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 2875ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 2876ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 2877ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 2878ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 2879ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 2880ffbc6f0eSEric W. Biederman } 2881ffbc6f0eSEric W. Biederman 2882e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 2883e462ec50SDavid Howells SB_SYNCHRONOUS | 2884e462ec50SDavid Howells SB_MANDLOCK | 2885e462ec50SDavid Howells SB_DIRSYNC | 2886e462ec50SDavid Howells SB_SILENT | 2887917086ffSMimi Zohar SB_POSIXACL | 2888d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 2889917086ffSMimi Zohar SB_I_VERSION); 28901da177e4SLinus Torvalds 289143f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 289243f5e655SDavid Howells retval = do_reconfigure_mnt(&path, mnt_flags); 289343f5e655SDavid Howells else if (flags & MS_REMOUNT) 2894e462ec50SDavid Howells retval = do_remount(&path, flags, sb_flags, mnt_flags, 28951da177e4SLinus Torvalds data_page); 28961da177e4SLinus Torvalds else if (flags & MS_BIND) 28972d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 28989676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 28992d92ab3cSAl Viro retval = do_change_type(&path, flags); 29001da177e4SLinus Torvalds else if (flags & MS_MOVE) 29012d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 29021da177e4SLinus Torvalds else 2903e462ec50SDavid Howells retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, 29041da177e4SLinus Torvalds dev_name, data_page); 29051da177e4SLinus Torvalds dput_out: 29062d92ab3cSAl Viro path_put(&path); 29071da177e4SLinus Torvalds return retval; 29081da177e4SLinus Torvalds } 29091da177e4SLinus Torvalds 2910537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 2911537f7ccbSEric W. Biederman { 2912537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 2913537f7ccbSEric W. Biederman } 2914537f7ccbSEric W. Biederman 2915537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 2916537f7ccbSEric W. Biederman { 2917537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 2918537f7ccbSEric W. Biederman } 2919537f7ccbSEric W. Biederman 2920771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2921771b1371SEric W. Biederman { 292274e83122SAl Viro if (!is_anon_ns(ns)) 29236344c433SAl Viro ns_free_inum(&ns->ns); 2924537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 2925771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2926771b1371SEric W. Biederman kfree(ns); 2927771b1371SEric W. Biederman } 2928771b1371SEric W. Biederman 29298823c079SEric W. Biederman /* 29308823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 29318823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 29328823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 29338823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 29348823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 29358823c079SEric W. Biederman */ 29368823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 29378823c079SEric W. Biederman 293874e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 2939cf8d2c11STrond Myklebust { 2940cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2941537f7ccbSEric W. Biederman struct ucounts *ucounts; 294298f842e6SEric W. Biederman int ret; 2943cf8d2c11STrond Myklebust 2944537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 2945537f7ccbSEric W. Biederman if (!ucounts) 2946df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 2947537f7ccbSEric W. Biederman 294874e83122SAl Viro new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2949537f7ccbSEric W. Biederman if (!new_ns) { 2950537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 2951cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2952537f7ccbSEric W. Biederman } 295374e83122SAl Viro if (!anon) { 29546344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 295598f842e6SEric W. Biederman if (ret) { 295698f842e6SEric W. Biederman kfree(new_ns); 2957537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 295898f842e6SEric W. Biederman return ERR_PTR(ret); 295998f842e6SEric W. Biederman } 296074e83122SAl Viro } 296133c42940SAl Viro new_ns->ns.ops = &mntns_operations; 296274e83122SAl Viro if (!anon) 29638823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2964cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2965cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2966cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2967771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2968537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 2969cf8d2c11STrond Myklebust return new_ns; 2970cf8d2c11STrond Myklebust } 2971cf8d2c11STrond Myklebust 29720766f788SEmese Revfy __latent_entropy 29739559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 29749559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 29751da177e4SLinus Torvalds { 29766b3286edSKirill Korotaev struct mnt_namespace *new_ns; 29777f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2978315fc83eSAl Viro struct mount *p, *q; 29799559f689SAl Viro struct mount *old; 2980cb338d06SAl Viro struct mount *new; 29817a472ef4SEric W. Biederman int copy_flags; 29821da177e4SLinus Torvalds 29839559f689SAl Viro BUG_ON(!ns); 29849559f689SAl Viro 29859559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 29869559f689SAl Viro get_mnt_ns(ns); 29879559f689SAl Viro return ns; 29889559f689SAl Viro } 29899559f689SAl Viro 29909559f689SAl Viro old = ns->root; 29919559f689SAl Viro 299274e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 2993cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2994cf8d2c11STrond Myklebust return new_ns; 29951da177e4SLinus Torvalds 299697216be0SAl Viro namespace_lock(); 29971da177e4SLinus Torvalds /* First pass: copy the tree topology */ 29984ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 29999559f689SAl Viro if (user_ns != ns->user_ns) 30003bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 30017a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3002be34d1a3SDavid Howells if (IS_ERR(new)) { 3003328e6d90SAl Viro namespace_unlock(); 3004771b1371SEric W. Biederman free_mnt_ns(new_ns); 3005be34d1a3SDavid Howells return ERR_CAST(new); 30061da177e4SLinus Torvalds } 30073bd045ccSAl Viro if (user_ns != ns->user_ns) { 30083bd045ccSAl Viro lock_mount_hash(); 30093bd045ccSAl Viro lock_mnt_tree(new); 30103bd045ccSAl Viro unlock_mount_hash(); 30113bd045ccSAl Viro } 3012be08d6d2SAl Viro new_ns->root = new; 30131a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 30141da177e4SLinus Torvalds 30151da177e4SLinus Torvalds /* 30161da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 30171da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 30181da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 30191da177e4SLinus Torvalds */ 3020909b0a88SAl Viro p = old; 3021cb338d06SAl Viro q = new; 30221da177e4SLinus Torvalds while (p) { 3023143c8c91SAl Viro q->mnt_ns = new_ns; 3024d2921684SEric W. Biederman new_ns->mounts++; 30259559f689SAl Viro if (new_fs) { 30269559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 30279559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3028315fc83eSAl Viro rootmnt = &p->mnt; 30291da177e4SLinus Torvalds } 30309559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 30319559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3032315fc83eSAl Viro pwdmnt = &p->mnt; 30331da177e4SLinus Torvalds } 30341da177e4SLinus Torvalds } 3035909b0a88SAl Viro p = next_mnt(p, old); 3036909b0a88SAl Viro q = next_mnt(q, new); 30374ce5d2b1SEric W. Biederman if (!q) 30384ce5d2b1SEric W. Biederman break; 30394ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 30404ce5d2b1SEric W. Biederman p = next_mnt(p, old); 30411da177e4SLinus Torvalds } 3042328e6d90SAl Viro namespace_unlock(); 30431da177e4SLinus Torvalds 30441da177e4SLinus Torvalds if (rootmnt) 3045f03c6599SAl Viro mntput(rootmnt); 30461da177e4SLinus Torvalds if (pwdmnt) 3047f03c6599SAl Viro mntput(pwdmnt); 30481da177e4SLinus Torvalds 3049741a2951SJANAK DESAI return new_ns; 3050741a2951SJANAK DESAI } 3051741a2951SJANAK DESAI 305274e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3053cf8d2c11STrond Myklebust { 30541a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3055ea441d11SAl Viro struct mnt_namespace *ns; 3056d31da0f0SAl Viro struct super_block *s; 3057ea441d11SAl Viro struct path path; 3058ea441d11SAl Viro int err; 3059ea441d11SAl Viro 306074e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 306174e83122SAl Viro if (IS_ERR(ns)) { 306274e83122SAl Viro mntput(m); 3063ea441d11SAl Viro return ERR_CAST(ns); 306474e83122SAl Viro } 306574e83122SAl Viro mnt->mnt_ns = ns; 306674e83122SAl Viro ns->root = mnt; 306774e83122SAl Viro ns->mounts++; 306874e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3069ea441d11SAl Viro 307074e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3071ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3072ea441d11SAl Viro 3073ea441d11SAl Viro put_mnt_ns(ns); 3074ea441d11SAl Viro 3075ea441d11SAl Viro if (err) 3076ea441d11SAl Viro return ERR_PTR(err); 3077ea441d11SAl Viro 3078ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3079d31da0f0SAl Viro s = path.mnt->mnt_sb; 3080d31da0f0SAl Viro atomic_inc(&s->s_active); 3081ea441d11SAl Viro mntput(path.mnt); 3082ea441d11SAl Viro /* lock the sucker */ 3083d31da0f0SAl Viro down_write(&s->s_umount); 3084ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3085ea441d11SAl Viro return path.dentry; 3086ea441d11SAl Viro } 3087ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3088ea441d11SAl Viro 3089312db1aaSDominik Brodowski int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, 3090312db1aaSDominik Brodowski unsigned long flags, void __user *data) 30911da177e4SLinus Torvalds { 3092eca6f534SVegard Nossum int ret; 3093eca6f534SVegard Nossum char *kernel_type; 3094eca6f534SVegard Nossum char *kernel_dev; 3095b40ef869SAl Viro void *options; 30961da177e4SLinus Torvalds 3097b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3098b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3099b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3100eca6f534SVegard Nossum goto out_type; 31011da177e4SLinus Torvalds 3102b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3103b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3104b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3105eca6f534SVegard Nossum goto out_dev; 31061da177e4SLinus Torvalds 3107b40ef869SAl Viro options = copy_mount_options(data); 3108b40ef869SAl Viro ret = PTR_ERR(options); 3109b40ef869SAl Viro if (IS_ERR(options)) 3110eca6f534SVegard Nossum goto out_data; 31111da177e4SLinus Torvalds 3112b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3113eca6f534SVegard Nossum 3114b40ef869SAl Viro kfree(options); 3115eca6f534SVegard Nossum out_data: 3116eca6f534SVegard Nossum kfree(kernel_dev); 3117eca6f534SVegard Nossum out_dev: 3118eca6f534SVegard Nossum kfree(kernel_type); 3119eca6f534SVegard Nossum out_type: 3120eca6f534SVegard Nossum return ret; 31211da177e4SLinus Torvalds } 31221da177e4SLinus Torvalds 3123312db1aaSDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3124312db1aaSDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 3125312db1aaSDominik Brodowski { 3126312db1aaSDominik Brodowski return ksys_mount(dev_name, dir_name, type, flags, data); 3127312db1aaSDominik Brodowski } 3128312db1aaSDominik Brodowski 31291da177e4SLinus Torvalds /* 3130afac7cbaSAl Viro * Return true if path is reachable from root 3131afac7cbaSAl Viro * 313248a066e7SAl Viro * namespace_sem or mount_lock is held 3133afac7cbaSAl Viro */ 3134643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3135afac7cbaSAl Viro const struct path *root) 3136afac7cbaSAl Viro { 3137643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3138a73324daSAl Viro dentry = mnt->mnt_mountpoint; 31390714a533SAl Viro mnt = mnt->mnt_parent; 3140afac7cbaSAl Viro } 3141643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3142afac7cbaSAl Viro } 3143afac7cbaSAl Viro 3144640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3145afac7cbaSAl Viro { 314625ab4c9bSYaowei Bai bool res; 314748a066e7SAl Viro read_seqlock_excl(&mount_lock); 3148643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 314948a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3150afac7cbaSAl Viro return res; 3151afac7cbaSAl Viro } 3152afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3153afac7cbaSAl Viro 3154afac7cbaSAl Viro /* 31551da177e4SLinus Torvalds * pivot_root Semantics: 31561da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 31571da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 31581da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 31591da177e4SLinus Torvalds * 31601da177e4SLinus Torvalds * Restrictions: 31611da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 31621da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 31631da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 31641da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 31651da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 31661da177e4SLinus Torvalds * 31674a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 31684a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 31694a0d11faSNeil Brown * in this situation. 31704a0d11faSNeil Brown * 31711da177e4SLinus Torvalds * Notes: 31721da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 31731da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 31741da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 31751da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 31761da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 31771da177e4SLinus Torvalds * first. 31781da177e4SLinus Torvalds */ 31793480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 31803480b257SHeiko Carstens const char __user *, put_old) 31811da177e4SLinus Torvalds { 31822d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 318384d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 318484d17192SAl Viro struct mountpoint *old_mp, *root_mp; 31851da177e4SLinus Torvalds int error; 31861da177e4SLinus Torvalds 31879b40bc90SAl Viro if (!may_mount()) 31881da177e4SLinus Torvalds return -EPERM; 31891da177e4SLinus Torvalds 31902d8f3038SAl Viro error = user_path_dir(new_root, &new); 31911da177e4SLinus Torvalds if (error) 31921da177e4SLinus Torvalds goto out0; 31931da177e4SLinus Torvalds 31942d8f3038SAl Viro error = user_path_dir(put_old, &old); 31951da177e4SLinus Torvalds if (error) 31961da177e4SLinus Torvalds goto out1; 31971da177e4SLinus Torvalds 31982d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3199b12cea91SAl Viro if (error) 3200b12cea91SAl Viro goto out2; 32011da177e4SLinus Torvalds 3202f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 320384d17192SAl Viro old_mp = lock_mount(&old); 320484d17192SAl Viro error = PTR_ERR(old_mp); 320584d17192SAl Viro if (IS_ERR(old_mp)) 3206b12cea91SAl Viro goto out3; 3207b12cea91SAl Viro 32081da177e4SLinus Torvalds error = -EINVAL; 3209419148daSAl Viro new_mnt = real_mount(new.mnt); 3210419148daSAl Viro root_mnt = real_mount(root.mnt); 321184d17192SAl Viro old_mnt = real_mount(old.mnt); 321284d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3213fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3214fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3215b12cea91SAl Viro goto out4; 3216143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3217b12cea91SAl Viro goto out4; 32185ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 32195ff9d8a6SEric W. Biederman goto out4; 32201da177e4SLinus Torvalds error = -ENOENT; 3221f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3222b12cea91SAl Viro goto out4; 32231da177e4SLinus Torvalds error = -EBUSY; 322484d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3225b12cea91SAl Viro goto out4; /* loop, on the same file system */ 32261da177e4SLinus Torvalds error = -EINVAL; 32278c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3228b12cea91SAl Viro goto out4; /* not a mountpoint */ 3229676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3230b12cea91SAl Viro goto out4; /* not attached */ 323184d17192SAl Viro root_mp = root_mnt->mnt_mp; 32322d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3233b12cea91SAl Viro goto out4; /* not a mountpoint */ 3234676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3235b12cea91SAl Viro goto out4; /* not attached */ 32364ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 323784d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3238b12cea91SAl Viro goto out4; 32390d082601SEric W. Biederman /* make certain new is below the root */ 32400d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 32410d082601SEric W. Biederman goto out4; 324284d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3243719ea2fbSAl Viro lock_mount_hash(); 3244419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3245419148daSAl Viro detach_mnt(root_mnt, &root_parent); 32465ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 32475ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 32485ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 32495ff9d8a6SEric W. Biederman } 32504ac91378SJan Blunck /* mount old root on put_old */ 325184d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 32524ac91378SJan Blunck /* mount new_root on / */ 325384d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 32546b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 32554fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 32564fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 32573895dbf8SEric W. Biederman put_mountpoint(root_mp); 3258719ea2fbSAl Viro unlock_mount_hash(); 32592d8f3038SAl Viro chroot_fs_refs(&root, &new); 32601da177e4SLinus Torvalds error = 0; 3261b12cea91SAl Viro out4: 326284d17192SAl Viro unlock_mount(old_mp); 3263b12cea91SAl Viro if (!error) { 32641a390689SAl Viro path_put(&root_parent); 32651a390689SAl Viro path_put(&parent_path); 3266b12cea91SAl Viro } 3267b12cea91SAl Viro out3: 32688c3ee42eSAl Viro path_put(&root); 3269b12cea91SAl Viro out2: 32702d8f3038SAl Viro path_put(&old); 32711da177e4SLinus Torvalds out1: 32722d8f3038SAl Viro path_put(&new); 32731da177e4SLinus Torvalds out0: 32741da177e4SLinus Torvalds return error; 32751da177e4SLinus Torvalds } 32761da177e4SLinus Torvalds 32771da177e4SLinus Torvalds static void __init init_mount_tree(void) 32781da177e4SLinus Torvalds { 32791da177e4SLinus Torvalds struct vfsmount *mnt; 328074e83122SAl Viro struct mount *m; 32816b3286edSKirill Korotaev struct mnt_namespace *ns; 3282ac748a09SJan Blunck struct path root; 32830c55cfc4SEric W. Biederman struct file_system_type *type; 32841da177e4SLinus Torvalds 32850c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 32860c55cfc4SEric W. Biederman if (!type) 32870c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 32880c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 32890c55cfc4SEric W. Biederman put_filesystem(type); 32901da177e4SLinus Torvalds if (IS_ERR(mnt)) 32911da177e4SLinus Torvalds panic("Can't create rootfs"); 3292b3e19d92SNick Piggin 329374e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 32943b22edc5STrond Myklebust if (IS_ERR(ns)) 32951da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 329674e83122SAl Viro m = real_mount(mnt); 329774e83122SAl Viro m->mnt_ns = ns; 329874e83122SAl Viro ns->root = m; 329974e83122SAl Viro ns->mounts = 1; 330074e83122SAl Viro list_add(&m->mnt_list, &ns->list); 33016b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 33026b3286edSKirill Korotaev get_mnt_ns(ns); 33031da177e4SLinus Torvalds 3304be08d6d2SAl Viro root.mnt = mnt; 3305be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3306da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3307ac748a09SJan Blunck 3308ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3309ac748a09SJan Blunck set_fs_root(current->fs, &root); 33101da177e4SLinus Torvalds } 33111da177e4SLinus Torvalds 331274bf17cfSDenis Cheng void __init mnt_init(void) 33131da177e4SLinus Torvalds { 331415a67dd8SRandy Dunlap int err; 33151da177e4SLinus Torvalds 33167d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 331720c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 33181da177e4SLinus Torvalds 33190818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 332038129a13SAl Viro sizeof(struct hlist_head), 33210818bf27SAl Viro mhash_entries, 19, 33223d375d78SPavel Tatashin HASH_ZERO, 33230818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 33240818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 33250818bf27SAl Viro sizeof(struct hlist_head), 33260818bf27SAl Viro mphash_entries, 19, 33273d375d78SPavel Tatashin HASH_ZERO, 33280818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 33291da177e4SLinus Torvalds 333084d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 33311da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 33321da177e4SLinus Torvalds 33334b93dc9bSTejun Heo kernfs_init(); 33344b93dc9bSTejun Heo 333515a67dd8SRandy Dunlap err = sysfs_init(); 333615a67dd8SRandy Dunlap if (err) 333715a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 33388e24eea7SHarvey Harrison __func__, err); 333900d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 334000d26666SGreg Kroah-Hartman if (!fs_kobj) 33418e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 33421da177e4SLinus Torvalds init_rootfs(); 33431da177e4SLinus Torvalds init_mount_tree(); 33441da177e4SLinus Torvalds } 33451da177e4SLinus Torvalds 3346616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 33471da177e4SLinus Torvalds { 3348d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3349616511d0STrond Myklebust return; 33507b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3351771b1371SEric W. Biederman free_mnt_ns(ns); 33521da177e4SLinus Torvalds } 33539d412a43SAl Viro 33549d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 33559d412a43SAl Viro { 3356423e0ab0STim Chen struct vfsmount *mnt; 3357e462ec50SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, data); 3358423e0ab0STim Chen if (!IS_ERR(mnt)) { 3359423e0ab0STim Chen /* 3360423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3361423e0ab0STim Chen * we unmount before file sys is unregistered 3362423e0ab0STim Chen */ 3363f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3364423e0ab0STim Chen } 3365423e0ab0STim Chen return mnt; 33669d412a43SAl Viro } 33679d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 3368423e0ab0STim Chen 3369423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3370423e0ab0STim Chen { 3371423e0ab0STim Chen /* release long term mount so mount point can be released */ 3372423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3373f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 337448a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3375423e0ab0STim Chen mntput(mnt); 3376423e0ab0STim Chen } 3377423e0ab0STim Chen } 3378423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 337902125a82SAl Viro 338002125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 338102125a82SAl Viro { 3382143c8c91SAl Viro return check_mnt(real_mount(mnt)); 338302125a82SAl Viro } 33848823c079SEric W. Biederman 33853151527eSEric W. Biederman bool current_chrooted(void) 33863151527eSEric W. Biederman { 33873151527eSEric W. Biederman /* Does the current process have a non-standard root */ 33883151527eSEric W. Biederman struct path ns_root; 33893151527eSEric W. Biederman struct path fs_root; 33903151527eSEric W. Biederman bool chrooted; 33913151527eSEric W. Biederman 33923151527eSEric W. Biederman /* Find the namespace root */ 33933151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 33943151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 33953151527eSEric W. Biederman path_get(&ns_root); 33963151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 33973151527eSEric W. Biederman ; 33983151527eSEric W. Biederman 33993151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 34003151527eSEric W. Biederman 34013151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 34023151527eSEric W. Biederman 34033151527eSEric W. Biederman path_put(&fs_root); 34043151527eSEric W. Biederman path_put(&ns_root); 34053151527eSEric W. Biederman 34063151527eSEric W. Biederman return chrooted; 34073151527eSEric W. Biederman } 34083151527eSEric W. Biederman 34098654df4eSEric W. Biederman static bool mnt_already_visible(struct mnt_namespace *ns, struct vfsmount *new, 34108654df4eSEric W. Biederman int *new_mnt_flags) 341187a8ebd6SEric W. Biederman { 34128c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 341387a8ebd6SEric W. Biederman struct mount *mnt; 3414e51db735SEric W. Biederman bool visible = false; 341587a8ebd6SEric W. Biederman 341644bb4385SAl Viro down_read(&namespace_sem); 341787a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3418e51db735SEric W. Biederman struct mount *child; 341977b1a97dSEric W. Biederman int mnt_flags; 342077b1a97dSEric W. Biederman 34218654df4eSEric W. Biederman if (mnt->mnt.mnt_sb->s_type != new->mnt_sb->s_type) 3422e51db735SEric W. Biederman continue; 3423e51db735SEric W. Biederman 34247e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 34257e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 34267e96c1b0SEric W. Biederman */ 34277e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 34287e96c1b0SEric W. Biederman continue; 34297e96c1b0SEric W. Biederman 3430a1935c17SEric W. Biederman /* A local view of the mount flags */ 343177b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 343277b1a97dSEric W. Biederman 3433695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3434bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 3435695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3436695e9df0SEric W. Biederman 34378c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 34388c6cf9ccSEric W. Biederman * than the proposed new mount. 34398c6cf9ccSEric W. Biederman */ 344077b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 34418c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 34428c6cf9ccSEric W. Biederman continue; 344377b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 344477b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 34458c6cf9ccSEric W. Biederman continue; 34468c6cf9ccSEric W. Biederman 3447ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3448ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3449ceeb0e5dSEric W. Biederman * empty directories. 3450e51db735SEric W. Biederman */ 3451e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3452e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3453ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3454d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3455ceeb0e5dSEric W. Biederman continue; 34567236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 34577236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3458e51db735SEric W. Biederman goto next; 345987a8ebd6SEric W. Biederman } 34608c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 346177b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 34628c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3463e51db735SEric W. Biederman visible = true; 3464e51db735SEric W. Biederman goto found; 3465e51db735SEric W. Biederman next: ; 346687a8ebd6SEric W. Biederman } 3467e51db735SEric W. Biederman found: 346844bb4385SAl Viro up_read(&namespace_sem); 3469e51db735SEric W. Biederman return visible; 347087a8ebd6SEric W. Biederman } 347187a8ebd6SEric W. Biederman 34728654df4eSEric W. Biederman static bool mount_too_revealing(struct vfsmount *mnt, int *new_mnt_flags) 34738654df4eSEric W. Biederman { 3474a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 34758654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 34768654df4eSEric W. Biederman unsigned long s_iflags; 34778654df4eSEric W. Biederman 34788654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 34798654df4eSEric W. Biederman return false; 34808654df4eSEric W. Biederman 34818654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 34828654df4eSEric W. Biederman s_iflags = mnt->mnt_sb->s_iflags; 34838654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 34848654df4eSEric W. Biederman return false; 34858654df4eSEric W. Biederman 3486a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3487a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3488a1935c17SEric W. Biederman required_iflags); 3489a1935c17SEric W. Biederman return true; 3490a1935c17SEric W. Biederman } 3491a1935c17SEric W. Biederman 34928654df4eSEric W. Biederman return !mnt_already_visible(ns, mnt, new_mnt_flags); 34938654df4eSEric W. Biederman } 34948654df4eSEric W. Biederman 3495380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3496380cf5baSAndy Lutomirski { 3497380cf5baSAndy Lutomirski /* 3498380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3499380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3500380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3501380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3502380cf5baSAndy Lutomirski * in other namespaces. 3503380cf5baSAndy Lutomirski */ 3504380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3505380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3506380cf5baSAndy Lutomirski } 3507380cf5baSAndy Lutomirski 350864964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 35098823c079SEric W. Biederman { 351058be2825SAl Viro struct ns_common *ns = NULL; 35118823c079SEric W. Biederman struct nsproxy *nsproxy; 35128823c079SEric W. Biederman 3513728dba3aSEric W. Biederman task_lock(task); 3514728dba3aSEric W. Biederman nsproxy = task->nsproxy; 35158823c079SEric W. Biederman if (nsproxy) { 351658be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 351758be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 35188823c079SEric W. Biederman } 3519728dba3aSEric W. Biederman task_unlock(task); 35208823c079SEric W. Biederman 35218823c079SEric W. Biederman return ns; 35228823c079SEric W. Biederman } 35238823c079SEric W. Biederman 352464964528SAl Viro static void mntns_put(struct ns_common *ns) 35258823c079SEric W. Biederman { 352658be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 35278823c079SEric W. Biederman } 35288823c079SEric W. Biederman 352964964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 35308823c079SEric W. Biederman { 35318823c079SEric W. Biederman struct fs_struct *fs = current->fs; 35324f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 35338823c079SEric W. Biederman struct path root; 35344f757f3cSAl Viro int err; 35358823c079SEric W. Biederman 35360c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3537c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3538c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3539ae11e0f1SZhao Hongjiang return -EPERM; 35408823c079SEric W. Biederman 354174e83122SAl Viro if (is_anon_ns(mnt_ns)) 354274e83122SAl Viro return -EINVAL; 354374e83122SAl Viro 35448823c079SEric W. Biederman if (fs->users != 1) 35458823c079SEric W. Biederman return -EINVAL; 35468823c079SEric W. Biederman 35478823c079SEric W. Biederman get_mnt_ns(mnt_ns); 35484f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 35498823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 35508823c079SEric W. Biederman 35518823c079SEric W. Biederman /* Find the root */ 35524f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 35534f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 35544f757f3cSAl Viro if (err) { 35554f757f3cSAl Viro /* revert to old namespace */ 35564f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 35574f757f3cSAl Viro put_mnt_ns(mnt_ns); 35584f757f3cSAl Viro return err; 35594f757f3cSAl Viro } 35608823c079SEric W. Biederman 35614068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 35624068367cSAndrei Vagin 35638823c079SEric W. Biederman /* Update the pwd and root */ 35648823c079SEric W. Biederman set_fs_pwd(fs, &root); 35658823c079SEric W. Biederman set_fs_root(fs, &root); 35668823c079SEric W. Biederman 35678823c079SEric W. Biederman path_put(&root); 35688823c079SEric W. Biederman return 0; 35698823c079SEric W. Biederman } 35708823c079SEric W. Biederman 3571bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3572bcac25a5SAndrey Vagin { 3573bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3574bcac25a5SAndrey Vagin } 3575bcac25a5SAndrey Vagin 35768823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 35778823c079SEric W. Biederman .name = "mnt", 35788823c079SEric W. Biederman .type = CLONE_NEWNS, 35798823c079SEric W. Biederman .get = mntns_get, 35808823c079SEric W. Biederman .put = mntns_put, 35818823c079SEric W. Biederman .install = mntns_install, 3582bcac25a5SAndrey Vagin .owner = mntns_owner, 35838823c079SEric W. Biederman }; 3584