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 */ 23a07b2000SAl Viro #include <linux/file.h> 24d10577a8SAl Viro #include <linux/uaccess.h> 250bb80f24SDavid Howells #include <linux/proc_ns.h> 2620b4fb48SLinus Torvalds #include <linux/magic.h> 2757c8a661SMike Rapoport #include <linux/memblock.h> 289ea459e1SAl Viro #include <linux/task_work.h> 299164bb4aSIngo Molnar #include <linux/sched/task.h> 30e262e32dSDavid Howells #include <uapi/linux/mount.h> 319bc61ab1SDavid Howells #include <linux/fs_context.h> 329164bb4aSIngo Molnar 3307b20889SRam Pai #include "pnode.h" 34948730b0SAdrian Bunk #include "internal.h" 351da177e4SLinus Torvalds 36d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 37d2921684SEric W. Biederman unsigned int sysctl_mount_max __read_mostly = 100000; 38d2921684SEric W. Biederman 390818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 400818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 410818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 420818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 430818bf27SAl Viro 440818bf27SAl Viro static __initdata unsigned long mhash_entries; 450818bf27SAl Viro static int __init set_mhash_entries(char *str) 460818bf27SAl Viro { 470818bf27SAl Viro if (!str) 480818bf27SAl Viro return 0; 490818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 500818bf27SAl Viro return 1; 510818bf27SAl Viro } 520818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 530818bf27SAl Viro 540818bf27SAl Viro static __initdata unsigned long mphash_entries; 550818bf27SAl Viro static int __init set_mphash_entries(char *str) 560818bf27SAl Viro { 570818bf27SAl Viro if (!str) 580818bf27SAl Viro return 0; 590818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 600818bf27SAl Viro return 1; 610818bf27SAl Viro } 620818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6313f14b4dSEric Dumazet 64c7999c36SAl Viro static u64 event; 6573cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 66719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 675addc5ddSAl Viro 6838129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 690818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 70e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 7159aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 721da177e4SLinus Torvalds 73f87fd4c2SMiklos Szeredi /* /sys/fs */ 7400d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 7500d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 76f87fd4c2SMiklos Szeredi 7799b7db7bSNick Piggin /* 7899b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 7999b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 8099b7db7bSNick Piggin * up the tree. 8199b7db7bSNick Piggin * 8299b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 8399b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 8499b7db7bSNick Piggin */ 8548a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 8699b7db7bSNick Piggin 8738129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 881da177e4SLinus Torvalds { 891da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 901da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 910818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 920818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 930818bf27SAl Viro } 940818bf27SAl Viro 950818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 960818bf27SAl Viro { 970818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 980818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 990818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 1001da177e4SLinus Torvalds } 1011da177e4SLinus Torvalds 102b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 10373cd49ecSMiklos Szeredi { 104169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 10573cd49ecSMiklos Szeredi 106169b480eSMatthew Wilcox if (res < 0) 10773cd49ecSMiklos Szeredi return res; 108169b480eSMatthew Wilcox mnt->mnt_id = res; 109169b480eSMatthew Wilcox return 0; 11073cd49ecSMiklos Szeredi } 11173cd49ecSMiklos Szeredi 112b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 11373cd49ecSMiklos Szeredi { 114169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 11573cd49ecSMiklos Szeredi } 11673cd49ecSMiklos Szeredi 117719f5d7fSMiklos Szeredi /* 118719f5d7fSMiklos Szeredi * Allocate a new peer group ID 119719f5d7fSMiklos Szeredi */ 1204b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 121719f5d7fSMiklos Szeredi { 122169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 123f21f6220SAl Viro 124169b480eSMatthew Wilcox if (res < 0) 125f21f6220SAl Viro return res; 126169b480eSMatthew Wilcox mnt->mnt_group_id = res; 127169b480eSMatthew Wilcox return 0; 128719f5d7fSMiklos Szeredi } 129719f5d7fSMiklos Szeredi 130719f5d7fSMiklos Szeredi /* 131719f5d7fSMiklos Szeredi * Release a peer group ID 132719f5d7fSMiklos Szeredi */ 1334b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 134719f5d7fSMiklos Szeredi { 135169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 13615169fe7SAl Viro mnt->mnt_group_id = 0; 137719f5d7fSMiklos Szeredi } 138719f5d7fSMiklos Szeredi 139b3e19d92SNick Piggin /* 140b3e19d92SNick Piggin * vfsmount lock must be held for read 141b3e19d92SNick Piggin */ 14283adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 143b3e19d92SNick Piggin { 144b3e19d92SNick Piggin #ifdef CONFIG_SMP 14568e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 146b3e19d92SNick Piggin #else 147b3e19d92SNick Piggin preempt_disable(); 14868e8a9feSAl Viro mnt->mnt_count += n; 149b3e19d92SNick Piggin preempt_enable(); 150b3e19d92SNick Piggin #endif 151b3e19d92SNick Piggin } 152b3e19d92SNick Piggin 153b3e19d92SNick Piggin /* 154b3e19d92SNick Piggin * vfsmount lock must be held for write 155b3e19d92SNick Piggin */ 15683adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 157b3e19d92SNick Piggin { 158b3e19d92SNick Piggin #ifdef CONFIG_SMP 159f03c6599SAl Viro unsigned int count = 0; 160b3e19d92SNick Piggin int cpu; 161b3e19d92SNick Piggin 162b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 16368e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 164b3e19d92SNick Piggin } 165b3e19d92SNick Piggin 166b3e19d92SNick Piggin return count; 167b3e19d92SNick Piggin #else 16868e8a9feSAl Viro return mnt->mnt_count; 169b3e19d92SNick Piggin #endif 170b3e19d92SNick Piggin } 171b3e19d92SNick Piggin 17287b95ce0SAl Viro static void drop_mountpoint(struct fs_pin *p) 17387b95ce0SAl Viro { 17487b95ce0SAl Viro struct mount *m = container_of(p, struct mount, mnt_umount); 17587b95ce0SAl Viro dput(m->mnt_ex_mountpoint); 17687b95ce0SAl Viro pin_remove(p); 17787b95ce0SAl Viro mntput(&m->mnt); 17887b95ce0SAl Viro } 17987b95ce0SAl Viro 180b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1811da177e4SLinus Torvalds { 182c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 183c63181e6SAl Viro if (mnt) { 18473cd49ecSMiklos Szeredi int err; 18573cd49ecSMiklos Szeredi 186c63181e6SAl Viro err = mnt_alloc_id(mnt); 18788b38782SLi Zefan if (err) 18888b38782SLi Zefan goto out_free_cache; 18988b38782SLi Zefan 19088b38782SLi Zefan if (name) { 191fcc139aeSAndrzej Hajda mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL); 192c63181e6SAl Viro if (!mnt->mnt_devname) 19388b38782SLi Zefan goto out_free_id; 19473cd49ecSMiklos Szeredi } 19573cd49ecSMiklos Szeredi 196b3e19d92SNick Piggin #ifdef CONFIG_SMP 197c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 198c63181e6SAl Viro if (!mnt->mnt_pcp) 199b3e19d92SNick Piggin goto out_free_devname; 200b3e19d92SNick Piggin 201c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 202b3e19d92SNick Piggin #else 203c63181e6SAl Viro mnt->mnt_count = 1; 204c63181e6SAl Viro mnt->mnt_writers = 0; 205b3e19d92SNick Piggin #endif 206b3e19d92SNick Piggin 20738129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 208c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 209c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 210c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 211c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 212c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 213c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 214c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2150a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 21699b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 21787b95ce0SAl Viro init_fs_pin(&mnt->mnt_umount, drop_mountpoint); 2181da177e4SLinus Torvalds } 219c63181e6SAl Viro return mnt; 22088b38782SLi Zefan 221d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 222d3ef3d73Snpiggin@suse.de out_free_devname: 223fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 224d3ef3d73Snpiggin@suse.de #endif 22588b38782SLi Zefan out_free_id: 226c63181e6SAl Viro mnt_free_id(mnt); 22788b38782SLi Zefan out_free_cache: 228c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 22988b38782SLi Zefan return NULL; 2301da177e4SLinus Torvalds } 2311da177e4SLinus Torvalds 2328366025eSDave Hansen /* 2338366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2348366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2358366025eSDave Hansen * We must keep track of when those operations start 2368366025eSDave Hansen * (for permission checks) and when they end, so that 2378366025eSDave Hansen * we can determine when writes are able to occur to 2388366025eSDave Hansen * a filesystem. 2398366025eSDave Hansen */ 2403d733633SDave Hansen /* 2413d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2423d733633SDave Hansen * @mnt: the mount to check for its write status 2433d733633SDave Hansen * 2443d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2453d733633SDave Hansen * It does not guarantee that the filesystem will stay 2463d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2473d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2483d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2493d733633SDave Hansen * r/w. 2503d733633SDave Hansen */ 25143f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 2523d733633SDave Hansen { 25343f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 2543d733633SDave Hansen } 2553d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2563d733633SDave Hansen 25783adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2583d733633SDave Hansen { 259d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 26068e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 261d3ef3d73Snpiggin@suse.de #else 26268e8a9feSAl Viro mnt->mnt_writers++; 263d3ef3d73Snpiggin@suse.de #endif 2643d733633SDave Hansen } 2653d733633SDave Hansen 26683adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2673d733633SDave Hansen { 268d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 26968e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 270d3ef3d73Snpiggin@suse.de #else 27168e8a9feSAl Viro mnt->mnt_writers--; 272d3ef3d73Snpiggin@suse.de #endif 273d3ef3d73Snpiggin@suse.de } 274d3ef3d73Snpiggin@suse.de 27583adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 276d3ef3d73Snpiggin@suse.de { 277d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 278d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2793d733633SDave Hansen int cpu; 2803d733633SDave Hansen 2813d733633SDave Hansen for_each_possible_cpu(cpu) { 28268e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2833d733633SDave Hansen } 2843d733633SDave Hansen 285d3ef3d73Snpiggin@suse.de return count; 286d3ef3d73Snpiggin@suse.de #else 287d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 288d3ef3d73Snpiggin@suse.de #endif 2893d733633SDave Hansen } 2903d733633SDave Hansen 2914ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2924ed5e82fSMiklos Szeredi { 2934ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2944ed5e82fSMiklos Szeredi return 1; 2954ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2964ed5e82fSMiklos Szeredi smp_rmb(); 2974ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2984ed5e82fSMiklos Szeredi } 2994ed5e82fSMiklos Szeredi 3003d733633SDave Hansen /* 301eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 302eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 303eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 304eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 3053d733633SDave Hansen */ 3068366025eSDave Hansen /** 307eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 30883adc753SAl Viro * @m: the mount on which to take a write 3098366025eSDave Hansen * 310eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 311eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 312eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 313eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 314eb04c282SJan Kara * called. This is effectively a refcount. 3158366025eSDave Hansen */ 316eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3178366025eSDave Hansen { 31883adc753SAl Viro struct mount *mnt = real_mount(m); 3193d733633SDave Hansen int ret = 0; 3203d733633SDave Hansen 321d3ef3d73Snpiggin@suse.de preempt_disable(); 322c6653a83SNick Piggin mnt_inc_writers(mnt); 323d3ef3d73Snpiggin@suse.de /* 324c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 325d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 326d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 327d3ef3d73Snpiggin@suse.de */ 328d3ef3d73Snpiggin@suse.de smp_mb(); 3296aa7de05SMark Rutland while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 330d3ef3d73Snpiggin@suse.de cpu_relax(); 331d3ef3d73Snpiggin@suse.de /* 332d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 333d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 334d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 335d3ef3d73Snpiggin@suse.de */ 336d3ef3d73Snpiggin@suse.de smp_rmb(); 3374ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 338c6653a83SNick Piggin mnt_dec_writers(mnt); 3393d733633SDave Hansen ret = -EROFS; 3403d733633SDave Hansen } 341d3ef3d73Snpiggin@suse.de preempt_enable(); 342eb04c282SJan Kara 343eb04c282SJan Kara return ret; 344eb04c282SJan Kara } 345eb04c282SJan Kara 346eb04c282SJan Kara /** 347eb04c282SJan Kara * mnt_want_write - get write access to a mount 348eb04c282SJan Kara * @m: the mount on which to take a write 349eb04c282SJan Kara * 350eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 351eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 352eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 353eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 354eb04c282SJan Kara */ 355eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 356eb04c282SJan Kara { 357eb04c282SJan Kara int ret; 358eb04c282SJan Kara 359eb04c282SJan Kara sb_start_write(m->mnt_sb); 360eb04c282SJan Kara ret = __mnt_want_write(m); 361eb04c282SJan Kara if (ret) 362eb04c282SJan Kara sb_end_write(m->mnt_sb); 3633d733633SDave Hansen return ret; 3648366025eSDave Hansen } 3658366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3668366025eSDave Hansen 3678366025eSDave Hansen /** 36896029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 36996029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 37096029c4eSnpiggin@suse.de * 37196029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 37296029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 37396029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 37496029c4eSnpiggin@suse.de * on it. This allows some optimisation. 37596029c4eSnpiggin@suse.de * 37696029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 37796029c4eSnpiggin@suse.de * drop the reference. 37896029c4eSnpiggin@suse.de */ 37996029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 38096029c4eSnpiggin@suse.de { 38196029c4eSnpiggin@suse.de /* superblock may be r/o */ 38296029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 38396029c4eSnpiggin@suse.de return -EROFS; 38496029c4eSnpiggin@suse.de preempt_disable(); 38583adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 38696029c4eSnpiggin@suse.de preempt_enable(); 38796029c4eSnpiggin@suse.de return 0; 38896029c4eSnpiggin@suse.de } 38996029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 39096029c4eSnpiggin@suse.de 39196029c4eSnpiggin@suse.de /** 392eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 393eb04c282SJan Kara * @file: the file who's mount on which to take a write 394eb04c282SJan Kara * 395eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 396eb04c282SJan Kara * do some optimisations if the file is open for write already 397eb04c282SJan Kara */ 398eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 399eb04c282SJan Kara { 40083f936c7SAl Viro if (!(file->f_mode & FMODE_WRITER)) 401eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 402eb04c282SJan Kara else 403eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 404eb04c282SJan Kara } 405eb04c282SJan Kara 406eb04c282SJan Kara /** 4077c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 4087c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 4097c6893e3SMiklos Szeredi * 4107c6893e3SMiklos Szeredi * This is like mnt_want_write, but it takes a file and can 4117c6893e3SMiklos Szeredi * do some optimisations if the file is open for write already 4127c6893e3SMiklos Szeredi */ 4137c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 4147c6893e3SMiklos Szeredi { 4157c6893e3SMiklos Szeredi int ret; 4167c6893e3SMiklos Szeredi 4177c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 4187c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 4197c6893e3SMiklos Szeredi if (ret) 4207c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4217c6893e3SMiklos Szeredi return ret; 4227c6893e3SMiklos Szeredi } 42396029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 42496029c4eSnpiggin@suse.de 42596029c4eSnpiggin@suse.de /** 426eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4278366025eSDave Hansen * @mnt: the mount on which to give up write access 4288366025eSDave Hansen * 4298366025eSDave Hansen * Tells the low-level filesystem that we are done 4308366025eSDave Hansen * performing writes to it. Must be matched with 431eb04c282SJan Kara * __mnt_want_write() call above. 4328366025eSDave Hansen */ 433eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4348366025eSDave Hansen { 435d3ef3d73Snpiggin@suse.de preempt_disable(); 43683adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 437d3ef3d73Snpiggin@suse.de preempt_enable(); 4388366025eSDave Hansen } 439eb04c282SJan Kara 440eb04c282SJan Kara /** 441eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 442eb04c282SJan Kara * @mnt: the mount on which to give up write access 443eb04c282SJan Kara * 444eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 445eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 446eb04c282SJan Kara * mnt_want_write() call above. 447eb04c282SJan Kara */ 448eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 449eb04c282SJan Kara { 450eb04c282SJan Kara __mnt_drop_write(mnt); 451eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 452eb04c282SJan Kara } 4538366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4548366025eSDave Hansen 455eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 456eb04c282SJan Kara { 457eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 458eb04c282SJan Kara } 459eb04c282SJan Kara 4607c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 4617c6893e3SMiklos Szeredi { 462a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 4637c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 4647c6893e3SMiklos Szeredi } 4652a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4662a79f17eSAl Viro 46783adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4688366025eSDave Hansen { 4693d733633SDave Hansen int ret = 0; 4703d733633SDave Hansen 471719ea2fbSAl Viro lock_mount_hash(); 47283adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 473d3ef3d73Snpiggin@suse.de /* 474d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 475d3ef3d73Snpiggin@suse.de * should be visible before we do. 476d3ef3d73Snpiggin@suse.de */ 477d3ef3d73Snpiggin@suse.de smp_mb(); 478d3ef3d73Snpiggin@suse.de 479d3ef3d73Snpiggin@suse.de /* 480d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 481d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 482d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 483d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 484d3ef3d73Snpiggin@suse.de * 485d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 486d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 487d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 488d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 489d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 490d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 491d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 492d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 493d3ef3d73Snpiggin@suse.de * we're counting up here. 494d3ef3d73Snpiggin@suse.de */ 495c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 496d3ef3d73Snpiggin@suse.de ret = -EBUSY; 497d3ef3d73Snpiggin@suse.de else 49883adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 499d3ef3d73Snpiggin@suse.de /* 500d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 501d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 502d3ef3d73Snpiggin@suse.de */ 503d3ef3d73Snpiggin@suse.de smp_wmb(); 50483adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 505719ea2fbSAl Viro unlock_mount_hash(); 5063d733633SDave Hansen return ret; 5073d733633SDave Hansen } 5088366025eSDave Hansen 50943f5e655SDavid Howells static int __mnt_unmake_readonly(struct mount *mnt) 5102e4b7fcdSDave Hansen { 511719ea2fbSAl Viro lock_mount_hash(); 51283adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 513719ea2fbSAl Viro unlock_mount_hash(); 51443f5e655SDavid Howells return 0; 5152e4b7fcdSDave Hansen } 5162e4b7fcdSDave Hansen 5174ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5184ed5e82fSMiklos Szeredi { 5194ed5e82fSMiklos Szeredi struct mount *mnt; 5204ed5e82fSMiklos Szeredi int err = 0; 5214ed5e82fSMiklos Szeredi 5228e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5238e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5248e8b8796SMiklos Szeredi return -EBUSY; 5258e8b8796SMiklos Szeredi 526719ea2fbSAl Viro lock_mount_hash(); 5274ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5284ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5294ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5304ed5e82fSMiklos Szeredi smp_mb(); 5314ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5324ed5e82fSMiklos Szeredi err = -EBUSY; 5334ed5e82fSMiklos Szeredi break; 5344ed5e82fSMiklos Szeredi } 5354ed5e82fSMiklos Szeredi } 5364ed5e82fSMiklos Szeredi } 5378e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5388e8b8796SMiklos Szeredi err = -EBUSY; 5398e8b8796SMiklos Szeredi 5404ed5e82fSMiklos Szeredi if (!err) { 5414ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5424ed5e82fSMiklos Szeredi smp_wmb(); 5434ed5e82fSMiklos Szeredi } 5444ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5454ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5464ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5474ed5e82fSMiklos Szeredi } 548719ea2fbSAl Viro unlock_mount_hash(); 5494ed5e82fSMiklos Szeredi 5504ed5e82fSMiklos Szeredi return err; 5514ed5e82fSMiklos Szeredi } 5524ed5e82fSMiklos Szeredi 553b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5541da177e4SLinus Torvalds { 555fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 556d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 55768e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 558d3ef3d73Snpiggin@suse.de #endif 559b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5601da177e4SLinus Torvalds } 5611da177e4SLinus Torvalds 5628ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 5638ffcb32eSDavid Howells { 5648ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 5658ffcb32eSDavid Howells } 5668ffcb32eSDavid Howells 56748a066e7SAl Viro /* call under rcu_read_lock */ 568294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 56948a066e7SAl Viro { 57048a066e7SAl Viro struct mount *mnt; 57148a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 572294d71ffSAl Viro return 1; 57348a066e7SAl Viro if (bastard == NULL) 574294d71ffSAl Viro return 0; 57548a066e7SAl Viro mnt = real_mount(bastard); 57648a066e7SAl Viro mnt_add_count(mnt, 1); 577119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 57848a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 579294d71ffSAl Viro return 0; 58048a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 58148a066e7SAl Viro mnt_add_count(mnt, -1); 582294d71ffSAl Viro return 1; 58348a066e7SAl Viro } 584119e1ef8SAl Viro lock_mount_hash(); 585119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 586119e1ef8SAl Viro mnt_add_count(mnt, -1); 587119e1ef8SAl Viro unlock_mount_hash(); 588119e1ef8SAl Viro return 1; 589119e1ef8SAl Viro } 590119e1ef8SAl Viro unlock_mount_hash(); 591119e1ef8SAl Viro /* caller will mntput() */ 592294d71ffSAl Viro return -1; 593294d71ffSAl Viro } 594294d71ffSAl Viro 595294d71ffSAl Viro /* call under rcu_read_lock */ 596294d71ffSAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 597294d71ffSAl Viro { 598294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 599294d71ffSAl Viro if (likely(!res)) 600294d71ffSAl Viro return true; 601294d71ffSAl Viro if (unlikely(res < 0)) { 60248a066e7SAl Viro rcu_read_unlock(); 60348a066e7SAl Viro mntput(bastard); 60448a066e7SAl Viro rcu_read_lock(); 605294d71ffSAl Viro } 60648a066e7SAl Viro return false; 60748a066e7SAl Viro } 60848a066e7SAl Viro 6091da177e4SLinus Torvalds /* 610474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 61148a066e7SAl Viro * call under rcu_read_lock() 6121da177e4SLinus Torvalds */ 613474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 6141da177e4SLinus Torvalds { 61538129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 616474279dcSAl Viro struct mount *p; 6171da177e4SLinus Torvalds 61838129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 619474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 620474279dcSAl Viro return p; 621474279dcSAl Viro return NULL; 6221da177e4SLinus Torvalds } 623474279dcSAl Viro 624474279dcSAl Viro /* 625f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 626f015f126SDavid Howells * 627f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 628f015f126SDavid Howells * following mounts: 629f015f126SDavid Howells * 630f015f126SDavid Howells * mount /dev/sda1 /mnt 631f015f126SDavid Howells * mount /dev/sda2 /mnt 632f015f126SDavid Howells * mount /dev/sda3 /mnt 633f015f126SDavid Howells * 634f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 635f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 636f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 637f015f126SDavid Howells * 638f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 639a05964f3SRam Pai */ 640ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 641a05964f3SRam Pai { 642c7105365SAl Viro struct mount *child_mnt; 64348a066e7SAl Viro struct vfsmount *m; 64448a066e7SAl Viro unsigned seq; 64599b7db7bSNick Piggin 64648a066e7SAl Viro rcu_read_lock(); 64748a066e7SAl Viro do { 64848a066e7SAl Viro seq = read_seqbegin(&mount_lock); 649474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 65048a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 65148a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 65248a066e7SAl Viro rcu_read_unlock(); 65348a066e7SAl Viro return m; 654a05964f3SRam Pai } 655a05964f3SRam Pai 6567af1364fSEric W. Biederman /* 6577af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 6587af1364fSEric W. Biederman * current mount namespace. 6597af1364fSEric W. Biederman * 6607af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 6617af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 6627af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 6637af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 6647af1364fSEric W. Biederman * is a mountpoint. 6657af1364fSEric W. Biederman * 6667af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 6677af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 6687af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 6697af1364fSEric W. Biederman * parent mount. 6707af1364fSEric W. Biederman */ 6717af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 6727af1364fSEric W. Biederman { 6737af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 6747af1364fSEric W. Biederman struct mount *mnt; 6757af1364fSEric W. Biederman bool is_covered = false; 6767af1364fSEric W. Biederman 6777af1364fSEric W. Biederman if (!d_mountpoint(dentry)) 6787af1364fSEric W. Biederman goto out; 6797af1364fSEric W. Biederman 6807af1364fSEric W. Biederman down_read(&namespace_sem); 6817af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 6827af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 6837af1364fSEric W. Biederman if (is_covered) 6847af1364fSEric W. Biederman break; 6857af1364fSEric W. Biederman } 6867af1364fSEric W. Biederman up_read(&namespace_sem); 6877af1364fSEric W. Biederman out: 6887af1364fSEric W. Biederman return is_covered; 6897af1364fSEric W. Biederman } 6907af1364fSEric W. Biederman 691e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 69284d17192SAl Viro { 6930818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 69484d17192SAl Viro struct mountpoint *mp; 69584d17192SAl Viro 6960818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 69784d17192SAl Viro if (mp->m_dentry == dentry) { 69884d17192SAl Viro mp->m_count++; 69984d17192SAl Viro return mp; 70084d17192SAl Viro } 70184d17192SAl Viro } 702e2dfa935SEric W. Biederman return NULL; 703e2dfa935SEric W. Biederman } 704e2dfa935SEric W. Biederman 7053895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 706e2dfa935SEric W. Biederman { 7073895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 708e2dfa935SEric W. Biederman int ret; 70984d17192SAl Viro 7103895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 7111e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 7121e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 7131e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 7143895dbf8SEric W. Biederman mountpoint: 7153895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7163895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 7173895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7183895dbf8SEric W. Biederman if (mp) 7193895dbf8SEric W. Biederman goto done; 72084d17192SAl Viro } 721eed81007SMiklos Szeredi 7223895dbf8SEric W. Biederman if (!new) 7233895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 7243895dbf8SEric W. Biederman if (!new) 7253895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 7263895dbf8SEric W. Biederman 7273895dbf8SEric W. Biederman 7283895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 7293895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 7303895dbf8SEric W. Biederman 7313895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 7323895dbf8SEric W. Biederman if (ret == -EBUSY) 7333895dbf8SEric W. Biederman goto mountpoint; 7343895dbf8SEric W. Biederman 7353895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 7363895dbf8SEric W. Biederman mp = ERR_PTR(ret); 7373895dbf8SEric W. Biederman if (ret) 7383895dbf8SEric W. Biederman goto done; 7393895dbf8SEric W. Biederman 7403895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 7413895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 7423895dbf8SEric W. Biederman new->m_dentry = dentry; 7433895dbf8SEric W. Biederman new->m_count = 1; 7443895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 7453895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 7463895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 7473895dbf8SEric W. Biederman 7483895dbf8SEric W. Biederman mp = new; 7493895dbf8SEric W. Biederman new = NULL; 7503895dbf8SEric W. Biederman done: 7513895dbf8SEric W. Biederman kfree(new); 75284d17192SAl Viro return mp; 75384d17192SAl Viro } 75484d17192SAl Viro 75584d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 75684d17192SAl Viro { 75784d17192SAl Viro if (!--mp->m_count) { 75884d17192SAl Viro struct dentry *dentry = mp->m_dentry; 7590a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 76084d17192SAl Viro spin_lock(&dentry->d_lock); 76184d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 76284d17192SAl Viro spin_unlock(&dentry->d_lock); 7630818bf27SAl Viro hlist_del(&mp->m_hash); 76484d17192SAl Viro kfree(mp); 76584d17192SAl Viro } 76684d17192SAl Viro } 76784d17192SAl Viro 768143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 7691da177e4SLinus Torvalds { 7706b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds 77399b7db7bSNick Piggin /* 77499b7db7bSNick Piggin * vfsmount lock must be held for write 77599b7db7bSNick Piggin */ 7766b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 7775addc5ddSAl Viro { 7785addc5ddSAl Viro if (ns) { 7795addc5ddSAl Viro ns->event = ++event; 7805addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7815addc5ddSAl Viro } 7825addc5ddSAl Viro } 7835addc5ddSAl Viro 78499b7db7bSNick Piggin /* 78599b7db7bSNick Piggin * vfsmount lock must be held for write 78699b7db7bSNick Piggin */ 7876b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 7885addc5ddSAl Viro { 7895addc5ddSAl Viro if (ns && ns->event != event) { 7905addc5ddSAl Viro ns->event = event; 7915addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7925addc5ddSAl Viro } 7935addc5ddSAl Viro } 7945addc5ddSAl Viro 79599b7db7bSNick Piggin /* 79699b7db7bSNick Piggin * vfsmount lock must be held for write 79799b7db7bSNick Piggin */ 7987bdb11deSEric W. Biederman static void unhash_mnt(struct mount *mnt) 7991da177e4SLinus Torvalds { 8000714a533SAl Viro mnt->mnt_parent = mnt; 801a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8026b41d536SAl Viro list_del_init(&mnt->mnt_child); 80338129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 8040a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 80584d17192SAl Viro put_mountpoint(mnt->mnt_mp); 80684d17192SAl Viro mnt->mnt_mp = NULL; 8071da177e4SLinus Torvalds } 8081da177e4SLinus Torvalds 80999b7db7bSNick Piggin /* 81099b7db7bSNick Piggin * vfsmount lock must be held for write 81199b7db7bSNick Piggin */ 8127bdb11deSEric W. Biederman static void detach_mnt(struct mount *mnt, struct path *old_path) 8137bdb11deSEric W. Biederman { 8147bdb11deSEric W. Biederman old_path->dentry = mnt->mnt_mountpoint; 8157bdb11deSEric W. Biederman old_path->mnt = &mnt->mnt_parent->mnt; 8167bdb11deSEric W. Biederman unhash_mnt(mnt); 8177bdb11deSEric W. Biederman } 8187bdb11deSEric W. Biederman 8197bdb11deSEric W. Biederman /* 8207bdb11deSEric W. Biederman * vfsmount lock must be held for write 8217bdb11deSEric W. Biederman */ 8226a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 8236a46c573SEric W. Biederman { 8246a46c573SEric W. Biederman /* old mountpoint will be dropped when we can do that */ 8256a46c573SEric W. Biederman mnt->mnt_ex_mountpoint = mnt->mnt_mountpoint; 8266a46c573SEric W. Biederman unhash_mnt(mnt); 8276a46c573SEric W. Biederman } 8286a46c573SEric W. Biederman 8296a46c573SEric W. Biederman /* 8306a46c573SEric W. Biederman * vfsmount lock must be held for write 8316a46c573SEric W. Biederman */ 83284d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 83384d17192SAl Viro struct mountpoint *mp, 83444d964d6SAl Viro struct mount *child_mnt) 835b90fa9aeSRam Pai { 83684d17192SAl Viro mp->m_count++; 8373a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 83884d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 8393a2393d7SAl Viro child_mnt->mnt_parent = mnt; 84084d17192SAl Viro child_mnt->mnt_mp = mp; 8410a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 842b90fa9aeSRam Pai } 843b90fa9aeSRam Pai 8441064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 8451064f874SEric W. Biederman { 8461064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 8471064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 8481064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 8491064f874SEric W. Biederman } 8501064f874SEric W. Biederman 85199b7db7bSNick Piggin /* 85299b7db7bSNick Piggin * vfsmount lock must be held for write 85399b7db7bSNick Piggin */ 85484d17192SAl Viro static void attach_mnt(struct mount *mnt, 85584d17192SAl Viro struct mount *parent, 85684d17192SAl Viro struct mountpoint *mp) 8571da177e4SLinus Torvalds { 85884d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 8591064f874SEric W. Biederman __attach_mnt(mnt, parent); 860b90fa9aeSRam Pai } 861b90fa9aeSRam Pai 8621064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 86312a5b529SAl Viro { 8641064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 8651064f874SEric W. Biederman struct dentry *old_mountpoint = mnt->mnt_mountpoint; 8661064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 8671064f874SEric W. Biederman 8681064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 8691064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 8701064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 8711064f874SEric W. Biederman 8721064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 8731064f874SEric W. Biederman 8741064f874SEric W. Biederman put_mountpoint(old_mp); 8751064f874SEric W. Biederman 8761064f874SEric W. Biederman /* 8771064f874SEric W. Biederman * Safely avoid even the suggestion this code might sleep or 8781064f874SEric W. Biederman * lock the mount hash by taking advantage of the knowledge that 8791064f874SEric W. Biederman * mnt_change_mountpoint will not release the final reference 8801064f874SEric W. Biederman * to a mountpoint. 8811064f874SEric W. Biederman * 8821064f874SEric W. Biederman * During mounting, the mount passed in as the parent mount will 8831064f874SEric W. Biederman * continue to use the old mountpoint and during unmounting, the 8841064f874SEric W. Biederman * old mountpoint will continue to exist until namespace_unlock, 8851064f874SEric W. Biederman * which happens well after mnt_change_mountpoint. 8861064f874SEric W. Biederman */ 8871064f874SEric W. Biederman spin_lock(&old_mountpoint->d_lock); 8881064f874SEric W. Biederman old_mountpoint->d_lockref.count--; 8891064f874SEric W. Biederman spin_unlock(&old_mountpoint->d_lock); 8901064f874SEric W. Biederman 8911064f874SEric W. Biederman mnt_add_count(old_parent, -1); 89212a5b529SAl Viro } 89312a5b529SAl Viro 894b90fa9aeSRam Pai /* 89599b7db7bSNick Piggin * vfsmount lock must be held for write 896b90fa9aeSRam Pai */ 8971064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 898b90fa9aeSRam Pai { 8990714a533SAl Viro struct mount *parent = mnt->mnt_parent; 90083adc753SAl Viro struct mount *m; 901b90fa9aeSRam Pai LIST_HEAD(head); 902143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 903b90fa9aeSRam Pai 9040714a533SAl Viro BUG_ON(parent == mnt); 905b90fa9aeSRam Pai 9061a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 907f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 908143c8c91SAl Viro m->mnt_ns = n; 909f03c6599SAl Viro 910b90fa9aeSRam Pai list_splice(&head, n->list.prev); 911b90fa9aeSRam Pai 912d2921684SEric W. Biederman n->mounts += n->pending_mounts; 913d2921684SEric W. Biederman n->pending_mounts = 0; 914d2921684SEric W. Biederman 9151064f874SEric W. Biederman __attach_mnt(mnt, parent); 9166b3286edSKirill Korotaev touch_mnt_namespace(n); 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds 919909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 9201da177e4SLinus Torvalds { 9216b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 9226b41d536SAl Viro if (next == &p->mnt_mounts) { 9231da177e4SLinus Torvalds while (1) { 924909b0a88SAl Viro if (p == root) 9251da177e4SLinus Torvalds return NULL; 9266b41d536SAl Viro next = p->mnt_child.next; 9276b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 9281da177e4SLinus Torvalds break; 9290714a533SAl Viro p = p->mnt_parent; 9301da177e4SLinus Torvalds } 9311da177e4SLinus Torvalds } 9326b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 9331da177e4SLinus Torvalds } 9341da177e4SLinus Torvalds 935315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 9369676f0c6SRam Pai { 9376b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 9386b41d536SAl Viro while (prev != &p->mnt_mounts) { 9396b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 9406b41d536SAl Viro prev = p->mnt_mounts.prev; 9419676f0c6SRam Pai } 9429676f0c6SRam Pai return p; 9439676f0c6SRam Pai } 9449676f0c6SRam Pai 9458f291889SAl Viro /** 9468f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 9478f291889SAl Viro * @fc: The configuration context with the superblock attached 9488f291889SAl Viro * 9498f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 9508f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 9518f291889SAl Viro * 9528f291889SAl Viro * Note that this does not attach the mount to anything. 9538f291889SAl Viro */ 9548f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 9559d412a43SAl Viro { 956b105e270SAl Viro struct mount *mnt; 9579d412a43SAl Viro 9588f291889SAl Viro if (!fc->root) 9598f291889SAl Viro return ERR_PTR(-EINVAL); 9609d412a43SAl Viro 9618f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 9629d412a43SAl Viro if (!mnt) 9639d412a43SAl Viro return ERR_PTR(-ENOMEM); 9649d412a43SAl Viro 9658f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 966b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 9679d412a43SAl Viro 9688f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 9698f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 9708f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 971a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9720714a533SAl Viro mnt->mnt_parent = mnt; 9738f291889SAl Viro 974719ea2fbSAl Viro lock_mount_hash(); 9758f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 976719ea2fbSAl Viro unlock_mount_hash(); 977b105e270SAl Viro return &mnt->mnt; 9789d412a43SAl Viro } 9798f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 9808f291889SAl Viro 9818f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 9828f291889SAl Viro { 9838f291889SAl Viro int err = vfs_get_tree(fc); 9848f291889SAl Viro if (!err) { 9858f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 9868f291889SAl Viro return vfs_create_mount(fc); 9878f291889SAl Viro } 9888f291889SAl Viro return ERR_PTR(err); 9898f291889SAl Viro } 9908f291889SAl Viro EXPORT_SYMBOL(fc_mount); 9918f291889SAl Viro 9929bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 9939bc61ab1SDavid Howells int flags, const char *name, 9949bc61ab1SDavid Howells void *data) 9951da177e4SLinus Torvalds { 9969bc61ab1SDavid Howells struct fs_context *fc; 9978f291889SAl Viro struct vfsmount *mnt; 9989bc61ab1SDavid Howells int ret = 0; 9999d412a43SAl Viro 10009d412a43SAl Viro if (!type) 10013e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 10029d412a43SAl Viro 10039bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 10049bc61ab1SDavid Howells if (IS_ERR(fc)) 10059bc61ab1SDavid Howells return ERR_CAST(fc); 10069bc61ab1SDavid Howells 10073e1aeb00SDavid Howells if (name) 10083e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 10093e1aeb00SDavid Howells name, strlen(name)); 10109bc61ab1SDavid Howells if (!ret) 10119bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 10129bc61ab1SDavid Howells if (!ret) 10138f291889SAl Viro mnt = fc_mount(fc); 10148f291889SAl Viro else 10158f291889SAl Viro mnt = ERR_PTR(ret); 10169d412a43SAl Viro 10179bc61ab1SDavid Howells put_fs_context(fc); 10188f291889SAl Viro return mnt; 10199d412a43SAl Viro } 10209d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 10219d412a43SAl Viro 102293faccbbSEric W. Biederman struct vfsmount * 102393faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 102493faccbbSEric W. Biederman const char *name, void *data) 102593faccbbSEric W. Biederman { 102693faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 102793faccbbSEric W. Biederman * through from the parent mount to the submount don't support 102893faccbbSEric W. Biederman * unprivileged mounts with submounts. 102993faccbbSEric W. Biederman */ 103093faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 103193faccbbSEric W. Biederman return ERR_PTR(-EPERM); 103293faccbbSEric W. Biederman 1033e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 103493faccbbSEric W. Biederman } 103593faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 103693faccbbSEric W. Biederman 103787129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 103836341f64SRam Pai int flag) 10391da177e4SLinus Torvalds { 104087129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1041be34d1a3SDavid Howells struct mount *mnt; 1042be34d1a3SDavid Howells int err; 10431da177e4SLinus Torvalds 1044be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1045be34d1a3SDavid Howells if (!mnt) 1046be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1047be34d1a3SDavid Howells 10487a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 104915169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1050719f5d7fSMiklos Szeredi else 105115169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1052719f5d7fSMiklos Szeredi 105315169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1054be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1055719f5d7fSMiklos Szeredi if (err) 1056719f5d7fSMiklos Szeredi goto out_free; 1057719f5d7fSMiklos Szeredi } 1058719f5d7fSMiklos Szeredi 105916a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 106016a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 10615ff9d8a6SEric W. Biederman 10621da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1063b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1064b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1065a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 10660714a533SAl Viro mnt->mnt_parent = mnt; 1067719ea2fbSAl Viro lock_mount_hash(); 106839f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1069719ea2fbSAl Viro unlock_mount_hash(); 1070b90fa9aeSRam Pai 10717a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 10727a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 10736776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 107432301920SAl Viro mnt->mnt_master = old; 1075fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 10768aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1077fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 10786776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1079d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 10806776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1081d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 10825235d448SAl Viro } else { 10835235d448SAl Viro CLEAR_MNT_SHARED(mnt); 10845afe0022SRam Pai } 1085b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 10860f0afb1dSAl Viro set_mnt_shared(mnt); 10871da177e4SLinus Torvalds 10881da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 10891da177e4SLinus Torvalds * as the original if that was on one */ 109036341f64SRam Pai if (flag & CL_EXPIRE) { 10916776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 10926776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 10931da177e4SLinus Torvalds } 1094be34d1a3SDavid Howells 1095cb338d06SAl Viro return mnt; 1096719f5d7fSMiklos Szeredi 1097719f5d7fSMiklos Szeredi out_free: 10988ffcb32eSDavid Howells mnt_free_id(mnt); 1099719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1100be34d1a3SDavid Howells return ERR_PTR(err); 11011da177e4SLinus Torvalds } 11021da177e4SLinus Torvalds 11039ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 11049ea459e1SAl Viro { 11059ea459e1SAl Viro /* 11069ea459e1SAl Viro * This probably indicates that somebody messed 11079ea459e1SAl Viro * up a mnt_want/drop_write() pair. If this 11089ea459e1SAl Viro * happens, the filesystem was probably unable 11099ea459e1SAl Viro * to make r/w->r/o transitions. 11109ea459e1SAl Viro */ 11119ea459e1SAl Viro /* 11129ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 11139ea459e1SAl Viro * so mnt_get_writers() below is safe. 11149ea459e1SAl Viro */ 11159ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 11169ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 11179ea459e1SAl Viro mnt_pin_kill(mnt); 11189ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 11199ea459e1SAl Viro dput(mnt->mnt.mnt_root); 11209ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 11219ea459e1SAl Viro mnt_free_id(mnt); 11229ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 11239ea459e1SAl Viro } 11249ea459e1SAl Viro 11259ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 11269ea459e1SAl Viro { 11279ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 11289ea459e1SAl Viro } 11299ea459e1SAl Viro 11309ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 11319ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 11329ea459e1SAl Viro { 11339ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 113429785735SByungchul Park struct mount *m, *t; 11359ea459e1SAl Viro 113629785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 113729785735SByungchul Park cleanup_mnt(m); 11389ea459e1SAl Viro } 11399ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 11409ea459e1SAl Viro 1141900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 11427b7b1aceSAl Viro { 114348a066e7SAl Viro rcu_read_lock(); 11449ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 11459ea0a46cSAl Viro /* 11469ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 11479ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 11489ea0a46cSAl Viro * non-NULL, then there's a reference that won't 11499ea0a46cSAl Viro * be dropped until after an RCU delay done after 11509ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 11519ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 11529ea0a46cSAl Viro * we are dropping is not the final one. 11539ea0a46cSAl Viro */ 1154aa9c0e07SAl Viro mnt_add_count(mnt, -1); 115548a066e7SAl Viro rcu_read_unlock(); 115699b7db7bSNick Piggin return; 1157b3e19d92SNick Piggin } 1158719ea2fbSAl Viro lock_mount_hash(); 1159119e1ef8SAl Viro /* 1160119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1161119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1162119e1ef8SAl Viro */ 1163119e1ef8SAl Viro smp_mb(); 11649ea0a46cSAl Viro mnt_add_count(mnt, -1); 1165b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 116648a066e7SAl Viro rcu_read_unlock(); 1167719ea2fbSAl Viro unlock_mount_hash(); 116899b7db7bSNick Piggin return; 116999b7db7bSNick Piggin } 117048a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 117148a066e7SAl Viro rcu_read_unlock(); 117248a066e7SAl Viro unlock_mount_hash(); 117348a066e7SAl Viro return; 117448a066e7SAl Viro } 117548a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 117648a066e7SAl Viro rcu_read_unlock(); 1177962830dfSAndi Kleen 117839f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1179ce07d891SEric W. Biederman 1180ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1181ce07d891SEric W. Biederman struct mount *p, *tmp; 1182ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 1183ce07d891SEric W. Biederman umount_mnt(p); 1184ce07d891SEric W. Biederman } 1185ce07d891SEric W. Biederman } 1186719ea2fbSAl Viro unlock_mount_hash(); 1187649a795aSAl Viro 11889ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 11899ea459e1SAl Viro struct task_struct *task = current; 11909ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 11919ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 11929ea459e1SAl Viro if (!task_work_add(task, &mnt->mnt_rcu, true)) 11939ea459e1SAl Viro return; 11949ea459e1SAl Viro } 11959ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 11969ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 11979ea459e1SAl Viro return; 11989ea459e1SAl Viro } 11999ea459e1SAl Viro cleanup_mnt(mnt); 1200b3e19d92SNick Piggin } 1201b3e19d92SNick Piggin 1202b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1203b3e19d92SNick Piggin { 1204b3e19d92SNick Piggin if (mnt) { 1205863d684fSAl Viro struct mount *m = real_mount(mnt); 1206b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1207863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1208863d684fSAl Viro m->mnt_expiry_mark = 0; 1209863d684fSAl Viro mntput_no_expire(m); 1210b3e19d92SNick Piggin } 1211b3e19d92SNick Piggin } 1212b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1213b3e19d92SNick Piggin 1214b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1215b3e19d92SNick Piggin { 1216b3e19d92SNick Piggin if (mnt) 121783adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1218b3e19d92SNick Piggin return mnt; 1219b3e19d92SNick Piggin } 1220b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1221b3e19d92SNick Piggin 1222c6609c0aSIan Kent /* path_is_mountpoint() - Check if path is a mount in the current 1223c6609c0aSIan Kent * namespace. 1224c6609c0aSIan Kent * 1225c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1226c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1227c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1228c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1229c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1230c6609c0aSIan Kent * alone. 1231c6609c0aSIan Kent */ 1232c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1233c6609c0aSIan Kent { 1234c6609c0aSIan Kent unsigned seq; 1235c6609c0aSIan Kent bool res; 1236c6609c0aSIan Kent 1237c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1238c6609c0aSIan Kent return false; 1239c6609c0aSIan Kent 1240c6609c0aSIan Kent rcu_read_lock(); 1241c6609c0aSIan Kent do { 1242c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1243c6609c0aSIan Kent res = __path_is_mountpoint(path); 1244c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1245c6609c0aSIan Kent rcu_read_unlock(); 1246c6609c0aSIan Kent 1247c6609c0aSIan Kent return res; 1248c6609c0aSIan Kent } 1249c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1250c6609c0aSIan Kent 1251ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 12527b7b1aceSAl Viro { 12533064c356SAl Viro struct mount *p; 12543064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 12553064c356SAl Viro if (IS_ERR(p)) 12563064c356SAl Viro return ERR_CAST(p); 12573064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 12583064c356SAl Viro return &p->mnt; 12597b7b1aceSAl Viro } 12601da177e4SLinus Torvalds 1261a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 12620226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 12631da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 12641da177e4SLinus Torvalds { 1265ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 12661da177e4SLinus Torvalds 1267390c6843SRam Pai down_read(&namespace_sem); 1268c7999c36SAl Viro if (p->cached_event == p->ns->event) { 1269c7999c36SAl Viro void *v = p->cached_mount; 1270c7999c36SAl Viro if (*pos == p->cached_index) 1271c7999c36SAl Viro return v; 1272c7999c36SAl Viro if (*pos == p->cached_index + 1) { 1273c7999c36SAl Viro v = seq_list_next(v, &p->ns->list, &p->cached_index); 1274c7999c36SAl Viro return p->cached_mount = v; 1275c7999c36SAl Viro } 1276c7999c36SAl Viro } 1277c7999c36SAl Viro 1278c7999c36SAl Viro p->cached_event = p->ns->event; 1279c7999c36SAl Viro p->cached_mount = seq_list_start(&p->ns->list, *pos); 1280c7999c36SAl Viro p->cached_index = *pos; 1281c7999c36SAl Viro return p->cached_mount; 12821da177e4SLinus Torvalds } 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 12851da177e4SLinus Torvalds { 1286ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 1287b0765fb8SPavel Emelianov 1288c7999c36SAl Viro p->cached_mount = seq_list_next(v, &p->ns->list, pos); 1289c7999c36SAl Viro p->cached_index = *pos; 1290c7999c36SAl Viro return p->cached_mount; 12911da177e4SLinus Torvalds } 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 12941da177e4SLinus Torvalds { 1295390c6843SRam Pai up_read(&namespace_sem); 12961da177e4SLinus Torvalds } 12971da177e4SLinus Torvalds 12980226f492SAl Viro static int m_show(struct seq_file *m, void *v) 12999f5596afSAl Viro { 1300ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 13011a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 13020226f492SAl Viro return p->show(m, &r->mnt); 13031da177e4SLinus Torvalds } 13041da177e4SLinus Torvalds 1305a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 13061da177e4SLinus Torvalds .start = m_start, 13071da177e4SLinus Torvalds .next = m_next, 13081da177e4SLinus Torvalds .stop = m_stop, 13090226f492SAl Viro .show = m_show, 1310b4629fe2SChuck Lever }; 1311a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1312b4629fe2SChuck Lever 13131da177e4SLinus Torvalds /** 13141da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 13151da177e4SLinus Torvalds * @mnt: root of mount tree 13161da177e4SLinus Torvalds * 13171da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 13181da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 13191da177e4SLinus Torvalds * busy. 13201da177e4SLinus Torvalds */ 1321909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 13221da177e4SLinus Torvalds { 1323909b0a88SAl Viro struct mount *mnt = real_mount(m); 132436341f64SRam Pai int actual_refs = 0; 132536341f64SRam Pai int minimum_refs = 0; 1326315fc83eSAl Viro struct mount *p; 1327909b0a88SAl Viro BUG_ON(!m); 13281da177e4SLinus Torvalds 1329b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1330719ea2fbSAl Viro lock_mount_hash(); 1331909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 133283adc753SAl Viro actual_refs += mnt_get_count(p); 13331da177e4SLinus Torvalds minimum_refs += 2; 13341da177e4SLinus Torvalds } 1335719ea2fbSAl Viro unlock_mount_hash(); 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds if (actual_refs > minimum_refs) 13381da177e4SLinus Torvalds return 0; 1339e3474a8eSIan Kent 1340e3474a8eSIan Kent return 1; 13411da177e4SLinus Torvalds } 13421da177e4SLinus Torvalds 13431da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 13441da177e4SLinus Torvalds 13451da177e4SLinus Torvalds /** 13461da177e4SLinus Torvalds * may_umount - check if a mount point is busy 13471da177e4SLinus Torvalds * @mnt: root of mount 13481da177e4SLinus Torvalds * 13491da177e4SLinus Torvalds * This is called to check if a mount point has any 13501da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 13511da177e4SLinus Torvalds * mount has sub mounts this will return busy 13521da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 13531da177e4SLinus Torvalds * 13541da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 13551da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 13561da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 13571da177e4SLinus Torvalds */ 13581da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 13591da177e4SLinus Torvalds { 1360e3474a8eSIan Kent int ret = 1; 13618ad08d8aSAl Viro down_read(&namespace_sem); 1362719ea2fbSAl Viro lock_mount_hash(); 13631ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1364e3474a8eSIan Kent ret = 0; 1365719ea2fbSAl Viro unlock_mount_hash(); 13668ad08d8aSAl Viro up_read(&namespace_sem); 1367a05964f3SRam Pai return ret; 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 13711da177e4SLinus Torvalds 137238129a13SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 1373e3197d83SAl Viro 137497216be0SAl Viro static void namespace_unlock(void) 13751da177e4SLinus Torvalds { 1376a3b3c562SEric W. Biederman struct hlist_head head; 137797216be0SAl Viro 1378a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 1379a3b3c562SEric W. Biederman 138097216be0SAl Viro up_write(&namespace_sem); 1381a3b3c562SEric W. Biederman 1382a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 138397216be0SAl Viro return; 138497216be0SAl Viro 138522cb7405SNeilBrown synchronize_rcu_expedited(); 138648a066e7SAl Viro 138787b95ce0SAl Viro group_pin_kill(&head); 138870fbcdf4SRam Pai } 138970fbcdf4SRam Pai 139097216be0SAl Viro static inline void namespace_lock(void) 1391e3197d83SAl Viro { 139297216be0SAl Viro down_write(&namespace_sem); 1393e3197d83SAl Viro } 1394e3197d83SAl Viro 1395e819f152SEric W. Biederman enum umount_tree_flags { 1396e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1397e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1398e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1399e819f152SEric W. Biederman }; 1400f2d0a123SEric W. Biederman 1401f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1402f2d0a123SEric W. Biederman { 1403f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1404f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1405f2d0a123SEric W. Biederman return true; 1406f2d0a123SEric W. Biederman 1407f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1408f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1409f2d0a123SEric W. Biederman return true; 1410f2d0a123SEric W. Biederman 1411f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1412f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1413f2d0a123SEric W. Biederman * connected to mounted mounts. 1414f2d0a123SEric W. Biederman */ 1415f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1416f2d0a123SEric W. Biederman return true; 1417f2d0a123SEric W. Biederman 1418f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1419f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1420f2d0a123SEric W. Biederman return false; 1421f2d0a123SEric W. Biederman 1422f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1423f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1424f2d0a123SEric W. Biederman return false; 1425f2d0a123SEric W. Biederman 1426f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1427f2d0a123SEric W. Biederman return true; 1428f2d0a123SEric W. Biederman } 1429f2d0a123SEric W. Biederman 143099b7db7bSNick Piggin /* 143148a066e7SAl Viro * mount_lock must be held 143299b7db7bSNick Piggin * namespace_sem must be held for write 143399b7db7bSNick Piggin */ 1434e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 143570fbcdf4SRam Pai { 1436c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1437315fc83eSAl Viro struct mount *p; 143870fbcdf4SRam Pai 14395d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 14405d88457eSEric W. Biederman propagate_mount_unlock(mnt); 14415d88457eSEric W. Biederman 1442c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1443590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1444590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1445c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1446590ce4bcSEric W. Biederman } 1447c003b26fSEric W. Biederman 1448411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1449c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1450c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 145138129a13SAl Viro } 145270fbcdf4SRam Pai 1453c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1454e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 14557b8a53fdSAl Viro propagate_umount(&tmp_list); 1456a05964f3SRam Pai 1457c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1458d2921684SEric W. Biederman struct mnt_namespace *ns; 1459ce07d891SEric W. Biederman bool disconnect; 1460c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 14616776db3dSAl Viro list_del_init(&p->mnt_expire); 14621a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1463d2921684SEric W. Biederman ns = p->mnt_ns; 1464d2921684SEric W. Biederman if (ns) { 1465d2921684SEric W. Biederman ns->mounts--; 1466d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1467d2921684SEric W. Biederman } 146863d37a84SAl Viro p->mnt_ns = NULL; 1469e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 147048a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 147187b95ce0SAl Viro 1472f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1473ce07d891SEric W. Biederman 1474ce07d891SEric W. Biederman pin_insert_group(&p->mnt_umount, &p->mnt_parent->mnt, 1475ce07d891SEric W. Biederman disconnect ? &unmounted : NULL); 1476676da58dSAl Viro if (mnt_has_parent(p)) { 147781b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1478ce07d891SEric W. Biederman if (!disconnect) { 1479ce07d891SEric W. Biederman /* Don't forget about p */ 1480ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1481ce07d891SEric W. Biederman } else { 14826a46c573SEric W. Biederman umount_mnt(p); 14837c4b93d8SAl Viro } 1484ce07d891SEric W. Biederman } 14850f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 148638129a13SAl Viro } 14871da177e4SLinus Torvalds } 14881da177e4SLinus Torvalds 1489b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1490c35038beSAl Viro 14918d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 14928d0347f6SDavid Howells { 14938d0347f6SDavid Howells int ret = 0; 14948d0347f6SDavid Howells 14958d0347f6SDavid Howells down_write(&sb->s_umount); 14968d0347f6SDavid Howells if (!sb_rdonly(sb)) { 14978d0347f6SDavid Howells struct fs_context *fc; 14988d0347f6SDavid Howells 14998d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 15008d0347f6SDavid Howells SB_RDONLY); 15018d0347f6SDavid Howells if (IS_ERR(fc)) { 15028d0347f6SDavid Howells ret = PTR_ERR(fc); 15038d0347f6SDavid Howells } else { 15048d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 15058d0347f6SDavid Howells if (!ret) 15068d0347f6SDavid Howells ret = reconfigure_super(fc); 15078d0347f6SDavid Howells put_fs_context(fc); 15088d0347f6SDavid Howells } 15098d0347f6SDavid Howells } 15108d0347f6SDavid Howells up_write(&sb->s_umount); 15118d0347f6SDavid Howells return ret; 15128d0347f6SDavid Howells } 15138d0347f6SDavid Howells 15141ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 15151da177e4SLinus Torvalds { 15161ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 15171da177e4SLinus Torvalds int retval; 15181da177e4SLinus Torvalds 15191ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 15201da177e4SLinus Torvalds if (retval) 15211da177e4SLinus Torvalds return retval; 15221da177e4SLinus Torvalds 15231da177e4SLinus Torvalds /* 15241da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 15251da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 15261da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 15271da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 15281da177e4SLinus Torvalds */ 15291da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 15301ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 15311da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 15321da177e4SLinus Torvalds return -EINVAL; 15331da177e4SLinus Torvalds 1534b3e19d92SNick Piggin /* 1535b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1536b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1537b3e19d92SNick Piggin */ 1538719ea2fbSAl Viro lock_mount_hash(); 153983adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1540719ea2fbSAl Viro unlock_mount_hash(); 15411da177e4SLinus Torvalds return -EBUSY; 1542b3e19d92SNick Piggin } 1543719ea2fbSAl Viro unlock_mount_hash(); 15441da177e4SLinus Torvalds 1545863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 15461da177e4SLinus Torvalds return -EAGAIN; 15471da177e4SLinus Torvalds } 15481da177e4SLinus Torvalds 15491da177e4SLinus Torvalds /* 15501da177e4SLinus Torvalds * If we may have to abort operations to get out of this 15511da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 15521da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 15531da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 15541da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 15551da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 15561da177e4SLinus Torvalds * about for the moment. 15571da177e4SLinus Torvalds */ 15581da177e4SLinus Torvalds 155942faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 156042faad99SAl Viro sb->s_op->umount_begin(sb); 156142faad99SAl Viro } 15621da177e4SLinus Torvalds 15631da177e4SLinus Torvalds /* 15641da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 15651da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 15661da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 15671da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 15681da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 15691da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 15701da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 15711da177e4SLinus Torvalds */ 15721ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 15731da177e4SLinus Torvalds /* 15741da177e4SLinus Torvalds * Special case for "unmounting" root ... 15751da177e4SLinus Torvalds * we just try to remount it readonly. 15761da177e4SLinus Torvalds */ 1577bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1578a1480dccSAndy Lutomirski return -EPERM; 15798d0347f6SDavid Howells return do_umount_root(sb); 15801da177e4SLinus Torvalds } 15811da177e4SLinus Torvalds 158297216be0SAl Viro namespace_lock(); 1583719ea2fbSAl Viro lock_mount_hash(); 15841da177e4SLinus Torvalds 158525d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 158625d202edSEric W. Biederman retval = -EINVAL; 158725d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 158825d202edSEric W. Biederman goto out; 158925d202edSEric W. Biederman 159025d202edSEric W. Biederman event++; 159148a066e7SAl Viro if (flags & MNT_DETACH) { 159248a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1593e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 159448a066e7SAl Viro retval = 0; 159548a066e7SAl Viro } else { 1596b54b9be7SAl Viro shrink_submounts(mnt); 15971da177e4SLinus Torvalds retval = -EBUSY; 159848a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 15991a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1600e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 16011da177e4SLinus Torvalds retval = 0; 16021da177e4SLinus Torvalds } 160348a066e7SAl Viro } 160425d202edSEric W. Biederman out: 1605719ea2fbSAl Viro unlock_mount_hash(); 1606e3197d83SAl Viro namespace_unlock(); 16071da177e4SLinus Torvalds return retval; 16081da177e4SLinus Torvalds } 16091da177e4SLinus Torvalds 16101da177e4SLinus Torvalds /* 161180b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 161280b5dce8SEric W. Biederman * 161380b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 161480b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 161580b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 161680b5dce8SEric W. Biederman * leaking them. 161780b5dce8SEric W. Biederman * 161880b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 161980b5dce8SEric W. Biederman */ 162080b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 162180b5dce8SEric W. Biederman { 162280b5dce8SEric W. Biederman struct mountpoint *mp; 162380b5dce8SEric W. Biederman struct mount *mnt; 162480b5dce8SEric W. Biederman 162580b5dce8SEric W. Biederman namespace_lock(); 16263895dbf8SEric W. Biederman lock_mount_hash(); 162780b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1628f53e5797SEric W. Biederman if (IS_ERR_OR_NULL(mp)) 162980b5dce8SEric W. Biederman goto out_unlock; 163080b5dce8SEric W. Biederman 1631e06b933eSAndrey Ulanov event++; 163280b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 163380b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1634ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1635fe78fcc8SEric W. Biederman hlist_add_head(&mnt->mnt_umount.s_list, &unmounted); 1636fe78fcc8SEric W. Biederman umount_mnt(mnt); 1637ce07d891SEric W. Biederman } 1638e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 163980b5dce8SEric W. Biederman } 164080b5dce8SEric W. Biederman put_mountpoint(mp); 164180b5dce8SEric W. Biederman out_unlock: 16423895dbf8SEric W. Biederman unlock_mount_hash(); 164380b5dce8SEric W. Biederman namespace_unlock(); 164480b5dce8SEric W. Biederman } 164580b5dce8SEric W. Biederman 164680b5dce8SEric W. Biederman /* 16479b40bc90SAl Viro * Is the caller allowed to modify his namespace? 16489b40bc90SAl Viro */ 16499b40bc90SAl Viro static inline bool may_mount(void) 16509b40bc90SAl Viro { 16519b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 16529b40bc90SAl Viro } 16539b40bc90SAl Viro 16549e8925b6SJeff Layton static inline bool may_mandlock(void) 16559e8925b6SJeff Layton { 16569e8925b6SJeff Layton #ifndef CONFIG_MANDATORY_FILE_LOCKING 16579e8925b6SJeff Layton return false; 16589e8925b6SJeff Layton #endif 165995ace754SEric W. Biederman return capable(CAP_SYS_ADMIN); 16609e8925b6SJeff Layton } 16619e8925b6SJeff Layton 16629b40bc90SAl Viro /* 16631da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 16641da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 16651da177e4SLinus Torvalds * 16661da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 16671da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 16681da177e4SLinus Torvalds */ 16691da177e4SLinus Torvalds 16703a18ef5cSDominik Brodowski int ksys_umount(char __user *name, int flags) 16711da177e4SLinus Torvalds { 16722d8f3038SAl Viro struct path path; 1673900148dcSAl Viro struct mount *mnt; 16741da177e4SLinus Torvalds int retval; 1675db1f05bbSMiklos Szeredi int lookup_flags = 0; 16761da177e4SLinus Torvalds 1677db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1678db1f05bbSMiklos Szeredi return -EINVAL; 1679db1f05bbSMiklos Szeredi 16809b40bc90SAl Viro if (!may_mount()) 16819b40bc90SAl Viro return -EPERM; 16829b40bc90SAl Viro 1683db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1684db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1685db1f05bbSMiklos Szeredi 168657d46577SRichard Guy Briggs lookup_flags |= LOOKUP_NO_EVAL; 168757d46577SRichard Guy Briggs 1688197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 16891da177e4SLinus Torvalds if (retval) 16901da177e4SLinus Torvalds goto out; 1691900148dcSAl Viro mnt = real_mount(path.mnt); 16921da177e4SLinus Torvalds retval = -EINVAL; 16932d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 16941da177e4SLinus Torvalds goto dput_and_out; 1695143c8c91SAl Viro if (!check_mnt(mnt)) 16961da177e4SLinus Torvalds goto dput_and_out; 169725d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 16985ff9d8a6SEric W. Biederman goto dput_and_out; 1699b2f5d4dcSEric W. Biederman retval = -EPERM; 1700b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 1701b2f5d4dcSEric W. Biederman goto dput_and_out; 17021da177e4SLinus Torvalds 1703900148dcSAl Viro retval = do_umount(mnt, flags); 17041da177e4SLinus Torvalds dput_and_out: 1705429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 17062d8f3038SAl Viro dput(path.dentry); 1707900148dcSAl Viro mntput_no_expire(mnt); 17081da177e4SLinus Torvalds out: 17091da177e4SLinus Torvalds return retval; 17101da177e4SLinus Torvalds } 17111da177e4SLinus Torvalds 17123a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 17133a18ef5cSDominik Brodowski { 17143a18ef5cSDominik Brodowski return ksys_umount(name, flags); 17153a18ef5cSDominik Brodowski } 17163a18ef5cSDominik Brodowski 17171da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 17181da177e4SLinus Torvalds 17191da177e4SLinus Torvalds /* 17201da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 17211da177e4SLinus Torvalds */ 1722bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 17231da177e4SLinus Torvalds { 17243a18ef5cSDominik Brodowski return ksys_umount(name, 0); 17251da177e4SLinus Torvalds } 17261da177e4SLinus Torvalds 17271da177e4SLinus Torvalds #endif 17281da177e4SLinus Torvalds 17294ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 17308823c079SEric W. Biederman { 17314ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1732e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1733e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 17344ce5d2b1SEric W. Biederman } 17354ce5d2b1SEric W. Biederman 173658be2825SAl Viro struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 173758be2825SAl Viro { 173858be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 173958be2825SAl Viro } 174058be2825SAl Viro 17414ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 17424ce5d2b1SEric W. Biederman { 17434ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 17444ce5d2b1SEric W. Biederman * mount namespace loop? 17454ce5d2b1SEric W. Biederman */ 17464ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 17474ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 17484ce5d2b1SEric W. Biederman return false; 17494ce5d2b1SEric W. Biederman 1750f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 17518823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 17528823c079SEric W. Biederman } 17538823c079SEric W. Biederman 175487129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 175536341f64SRam Pai int flag) 17561da177e4SLinus Torvalds { 175784d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 17581da177e4SLinus Torvalds 17594ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 17604ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 17614ce5d2b1SEric W. Biederman 17624ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1763be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 17649676f0c6SRam Pai 176536341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1766be34d1a3SDavid Howells if (IS_ERR(q)) 1767be34d1a3SDavid Howells return q; 1768be34d1a3SDavid Howells 1769a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 17701da177e4SLinus Torvalds 17711da177e4SLinus Torvalds p = mnt; 17726b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1773315fc83eSAl Viro struct mount *s; 17747ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 17751da177e4SLinus Torvalds continue; 17761da177e4SLinus Torvalds 1777909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 17784ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 17794ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 1780df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 1781df7342b2SEric W. Biederman /* Both unbindable and locked. */ 1782df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 1783df7342b2SEric W. Biederman goto out; 1784df7342b2SEric W. Biederman } else { 17854ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 17864ce5d2b1SEric W. Biederman continue; 17874ce5d2b1SEric W. Biederman } 1788df7342b2SEric W. Biederman } 17894ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 17904ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 17919676f0c6SRam Pai s = skip_mnt_tree(s); 17929676f0c6SRam Pai continue; 17939676f0c6SRam Pai } 17940714a533SAl Viro while (p != s->mnt_parent) { 17950714a533SAl Viro p = p->mnt_parent; 17960714a533SAl Viro q = q->mnt_parent; 17971da177e4SLinus Torvalds } 179887129cc0SAl Viro p = s; 179984d17192SAl Viro parent = q; 180087129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1801be34d1a3SDavid Howells if (IS_ERR(q)) 1802be34d1a3SDavid Howells goto out; 1803719ea2fbSAl Viro lock_mount_hash(); 18041a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 18051064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 1806719ea2fbSAl Viro unlock_mount_hash(); 18071da177e4SLinus Torvalds } 18081da177e4SLinus Torvalds } 18091da177e4SLinus Torvalds return res; 1810be34d1a3SDavid Howells out: 18111da177e4SLinus Torvalds if (res) { 1812719ea2fbSAl Viro lock_mount_hash(); 1813e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 1814719ea2fbSAl Viro unlock_mount_hash(); 18151da177e4SLinus Torvalds } 1816be34d1a3SDavid Howells return q; 18171da177e4SLinus Torvalds } 18181da177e4SLinus Torvalds 1819be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1820be34d1a3SDavid Howells 1821ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 18228aec0809SAl Viro { 1823cb338d06SAl Viro struct mount *tree; 182497216be0SAl Viro namespace_lock(); 1825cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 1826cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 1827cd4a4017SEric W. Biederman else 182887129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 182987129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1830328e6d90SAl Viro namespace_unlock(); 1831be34d1a3SDavid Howells if (IS_ERR(tree)) 183252e220d3SDan Carpenter return ERR_CAST(tree); 1833be34d1a3SDavid Howells return &tree->mnt; 18348aec0809SAl Viro } 18358aec0809SAl Viro 1836a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 1837a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 1838a07b2000SAl Viro 1839a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 1840a07b2000SAl Viro { 1841a07b2000SAl Viro struct mnt_namespace *ns; 1842a07b2000SAl Viro namespace_lock(); 1843a07b2000SAl Viro lock_mount_hash(); 1844a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 1845a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 1846a07b2000SAl Viro unlock_mount_hash(); 1847a07b2000SAl Viro namespace_unlock(); 1848a07b2000SAl Viro free_mnt_ns(ns); 1849a07b2000SAl Viro } 1850a07b2000SAl Viro 18518aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 18528aec0809SAl Viro { 185397216be0SAl Viro namespace_lock(); 1854719ea2fbSAl Viro lock_mount_hash(); 18559c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 1856719ea2fbSAl Viro unlock_mount_hash(); 18573ab6abeeSAl Viro namespace_unlock(); 18588aec0809SAl Viro } 18598aec0809SAl Viro 1860c771d683SMiklos Szeredi /** 1861c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 1862c771d683SMiklos Szeredi * 1863c771d683SMiklos Szeredi * This creates a new vfsmount, which will be the clone of @path. The new will 1864c771d683SMiklos Szeredi * not be attached anywhere in the namespace and will be private (i.e. changes 1865c771d683SMiklos Szeredi * to the originating mount won't be propagated into this). 1866c771d683SMiklos Szeredi * 1867c771d683SMiklos Szeredi * Release with mntput(). 1868c771d683SMiklos Szeredi */ 1869ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 1870c771d683SMiklos Szeredi { 1871c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 1872c771d683SMiklos Szeredi struct mount *new_mnt; 1873c771d683SMiklos Szeredi 1874c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 1875c771d683SMiklos Szeredi return ERR_PTR(-EINVAL); 1876c771d683SMiklos Szeredi 1877c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 1878c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 1879c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 1880c771d683SMiklos Szeredi 1881c771d683SMiklos Szeredi return &new_mnt->mnt; 1882c771d683SMiklos Szeredi } 1883c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 1884c771d683SMiklos Szeredi 18851f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 18861f707137SAl Viro struct vfsmount *root) 18871f707137SAl Viro { 18881a4eeaf2SAl Viro struct mount *mnt; 18891f707137SAl Viro int res = f(root, arg); 18901f707137SAl Viro if (res) 18911f707137SAl Viro return res; 18921a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 18931a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 18941f707137SAl Viro if (res) 18951f707137SAl Viro return res; 18961f707137SAl Viro } 18971f707137SAl Viro return 0; 18981f707137SAl Viro } 18991f707137SAl Viro 19003bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 19013bd045ccSAl Viro { 19023bd045ccSAl Viro struct mount *p; 19033bd045ccSAl Viro 19043bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 19053bd045ccSAl Viro int flags = p->mnt.mnt_flags; 19063bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 19073bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 19083bd045ccSAl Viro 19093bd045ccSAl Viro if (flags & MNT_READONLY) 19103bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 19113bd045ccSAl Viro 19123bd045ccSAl Viro if (flags & MNT_NODEV) 19133bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 19143bd045ccSAl Viro 19153bd045ccSAl Viro if (flags & MNT_NOSUID) 19163bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 19173bd045ccSAl Viro 19183bd045ccSAl Viro if (flags & MNT_NOEXEC) 19193bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 19203bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 19213bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 19223bd045ccSAl Viro flags |= MNT_LOCKED; 19233bd045ccSAl Viro p->mnt.mnt_flags = flags; 19243bd045ccSAl Viro } 19253bd045ccSAl Viro } 19263bd045ccSAl Viro 19274b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1928719f5d7fSMiklos Szeredi { 1929315fc83eSAl Viro struct mount *p; 1930719f5d7fSMiklos Szeredi 1931909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1932fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 19334b8b21f4SAl Viro mnt_release_group_id(p); 1934719f5d7fSMiklos Szeredi } 1935719f5d7fSMiklos Szeredi } 1936719f5d7fSMiklos Szeredi 19374b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1938719f5d7fSMiklos Szeredi { 1939315fc83eSAl Viro struct mount *p; 1940719f5d7fSMiklos Szeredi 1941909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1942fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 19434b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1944719f5d7fSMiklos Szeredi if (err) { 19454b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1946719f5d7fSMiklos Szeredi return err; 1947719f5d7fSMiklos Szeredi } 1948719f5d7fSMiklos Szeredi } 1949719f5d7fSMiklos Szeredi } 1950719f5d7fSMiklos Szeredi 1951719f5d7fSMiklos Szeredi return 0; 1952719f5d7fSMiklos Szeredi } 1953719f5d7fSMiklos Szeredi 1954d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 1955d2921684SEric W. Biederman { 1956d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 1957d2921684SEric W. Biederman unsigned int mounts = 0, old, pending, sum; 1958d2921684SEric W. Biederman struct mount *p; 1959d2921684SEric W. Biederman 1960d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 1961d2921684SEric W. Biederman mounts++; 1962d2921684SEric W. Biederman 1963d2921684SEric W. Biederman old = ns->mounts; 1964d2921684SEric W. Biederman pending = ns->pending_mounts; 1965d2921684SEric W. Biederman sum = old + pending; 1966d2921684SEric W. Biederman if ((old > sum) || 1967d2921684SEric W. Biederman (pending > sum) || 1968d2921684SEric W. Biederman (max < sum) || 1969d2921684SEric W. Biederman (mounts > (max - sum))) 1970d2921684SEric W. Biederman return -ENOSPC; 1971d2921684SEric W. Biederman 1972d2921684SEric W. Biederman ns->pending_mounts = pending + mounts; 1973d2921684SEric W. Biederman return 0; 1974d2921684SEric W. Biederman } 1975d2921684SEric W. Biederman 1976b90fa9aeSRam Pai /* 1977b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1978b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 197921444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 198021444403SRam Pai * store the parent mount and mountpoint dentry. 198121444403SRam Pai * (done when source_mnt is moved) 1982b90fa9aeSRam Pai * 1983b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1984b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 19859676f0c6SRam Pai * --------------------------------------------------------------------------- 1986b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 19879676f0c6SRam Pai * |************************************************************************** 19889676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 19899676f0c6SRam Pai * | dest | | | | | 19909676f0c6SRam Pai * | | | | | | | 19919676f0c6SRam Pai * | v | | | | | 19929676f0c6SRam Pai * |************************************************************************** 19939676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 19945afe0022SRam Pai * | | | | | | 19959676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 19969676f0c6SRam Pai * *************************************************************************** 1997b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1998b90fa9aeSRam Pai * destination mount. 1999b90fa9aeSRam Pai * 2000b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2001b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2002b90fa9aeSRam Pai * the peer group of the source mount. 2003b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2004b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2005b90fa9aeSRam Pai * mount. 20065afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 20075afe0022SRam Pai * of the destination mount and the cloned mount is made slave 20085afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 20095afe0022SRam Pai * is marked as 'shared and slave'. 20105afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 20115afe0022SRam Pai * source mount. 20125afe0022SRam Pai * 20139676f0c6SRam Pai * --------------------------------------------------------------------------- 201421444403SRam Pai * | MOVE MOUNT OPERATION | 20159676f0c6SRam Pai * |************************************************************************** 20169676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 20179676f0c6SRam Pai * | dest | | | | | 20189676f0c6SRam Pai * | | | | | | | 20199676f0c6SRam Pai * | v | | | | | 20209676f0c6SRam Pai * |************************************************************************** 20219676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 20225afe0022SRam Pai * | | | | | | 20239676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 20249676f0c6SRam Pai * *************************************************************************** 20255afe0022SRam Pai * 20265afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 20275afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 202821444403SRam Pai * (+*) the mount is moved to the destination. 20295afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 20305afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 20315afe0022SRam Pai * the mount is marked as 'shared and slave'. 20325afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2033b90fa9aeSRam Pai * 2034b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2035b90fa9aeSRam Pai * applied to each mount in the tree. 2036b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2037b90fa9aeSRam Pai * in allocations. 2038b90fa9aeSRam Pai */ 20390fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 204084d17192SAl Viro struct mount *dest_mnt, 204184d17192SAl Viro struct mountpoint *dest_mp, 204284d17192SAl Viro struct path *parent_path) 2043b90fa9aeSRam Pai { 20443bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 204538129a13SAl Viro HLIST_HEAD(tree_list); 2046d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 20471064f874SEric W. Biederman struct mountpoint *smp; 2048315fc83eSAl Viro struct mount *child, *p; 204938129a13SAl Viro struct hlist_node *n; 2050719f5d7fSMiklos Szeredi int err; 2051b90fa9aeSRam Pai 20521064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 20531064f874SEric W. Biederman * to be tucked under other mounts. 20541064f874SEric W. Biederman */ 20551064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 20561064f874SEric W. Biederman if (IS_ERR(smp)) 20571064f874SEric W. Biederman return PTR_ERR(smp); 20581064f874SEric W. Biederman 2059d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 2060d2921684SEric W. Biederman if (!parent_path) { 2061d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2062d2921684SEric W. Biederman if (err) 2063d2921684SEric W. Biederman goto out; 2064d2921684SEric W. Biederman } 2065d2921684SEric W. Biederman 2066fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 20670fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2068719f5d7fSMiklos Szeredi if (err) 2069719f5d7fSMiklos Szeredi goto out; 207084d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2071f2ebb3a9SAl Viro lock_mount_hash(); 2072719f5d7fSMiklos Szeredi if (err) 2073719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2074909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 20750f0afb1dSAl Viro set_mnt_shared(p); 20760b1b901bSAl Viro } else { 20770b1b901bSAl Viro lock_mount_hash(); 2078b90fa9aeSRam Pai } 20791a390689SAl Viro if (parent_path) { 20800fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 208184d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2082143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 208321444403SRam Pai } else { 208484d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 20851064f874SEric W. Biederman commit_tree(source_mnt); 208621444403SRam Pai } 2087b90fa9aeSRam Pai 208838129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 20891d6a32acSAl Viro struct mount *q; 209038129a13SAl Viro hlist_del_init(&child->mnt_hash); 20911064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 20921d6a32acSAl Viro child->mnt_mountpoint); 20931064f874SEric W. Biederman if (q) 20941064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 20953bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 20963bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 20973bd045ccSAl Viro lock_mnt_tree(child); 20981064f874SEric W. Biederman commit_tree(child); 2099b90fa9aeSRam Pai } 21001064f874SEric W. Biederman put_mountpoint(smp); 2101719ea2fbSAl Viro unlock_mount_hash(); 210299b7db7bSNick Piggin 2103b90fa9aeSRam Pai return 0; 2104719f5d7fSMiklos Szeredi 2105719f5d7fSMiklos Szeredi out_cleanup_ids: 2106f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2107f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2108d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2109e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2110f2ebb3a9SAl Viro } 2111f2ebb3a9SAl Viro unlock_mount_hash(); 21120fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2113719f5d7fSMiklos Szeredi out: 2114d2921684SEric W. Biederman ns->pending_mounts = 0; 21151064f874SEric W. Biederman 21161064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 21171064f874SEric W. Biederman put_mountpoint(smp); 21181064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 21191064f874SEric W. Biederman 2120719f5d7fSMiklos Szeredi return err; 2121b90fa9aeSRam Pai } 2122b90fa9aeSRam Pai 212384d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2124b12cea91SAl Viro { 2125b12cea91SAl Viro struct vfsmount *mnt; 212684d17192SAl Viro struct dentry *dentry = path->dentry; 2127b12cea91SAl Viro retry: 21285955102cSAl Viro inode_lock(dentry->d_inode); 212984d17192SAl Viro if (unlikely(cant_mount(dentry))) { 21305955102cSAl Viro inode_unlock(dentry->d_inode); 213184d17192SAl Viro return ERR_PTR(-ENOENT); 2132b12cea91SAl Viro } 213397216be0SAl Viro namespace_lock(); 2134b12cea91SAl Viro mnt = lookup_mnt(path); 213584d17192SAl Viro if (likely(!mnt)) { 21363895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 213784d17192SAl Viro if (IS_ERR(mp)) { 213897216be0SAl Viro namespace_unlock(); 21395955102cSAl Viro inode_unlock(dentry->d_inode); 214084d17192SAl Viro return mp; 214184d17192SAl Viro } 214284d17192SAl Viro return mp; 214384d17192SAl Viro } 214497216be0SAl Viro namespace_unlock(); 21455955102cSAl Viro inode_unlock(path->dentry->d_inode); 2146b12cea91SAl Viro path_put(path); 2147b12cea91SAl Viro path->mnt = mnt; 214884d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2149b12cea91SAl Viro goto retry; 2150b12cea91SAl Viro } 2151b12cea91SAl Viro 215284d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2153b12cea91SAl Viro { 215484d17192SAl Viro struct dentry *dentry = where->m_dentry; 21553895dbf8SEric W. Biederman 21563895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 215784d17192SAl Viro put_mountpoint(where); 21583895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 21593895dbf8SEric W. Biederman 2160328e6d90SAl Viro namespace_unlock(); 21615955102cSAl Viro inode_unlock(dentry->d_inode); 2162b12cea91SAl Viro } 2163b12cea91SAl Viro 216484d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 21651da177e4SLinus Torvalds { 2166e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 21671da177e4SLinus Torvalds return -EINVAL; 21681da177e4SLinus Torvalds 2169e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2170e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 21711da177e4SLinus Torvalds return -ENOTDIR; 21721da177e4SLinus Torvalds 217384d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 21741da177e4SLinus Torvalds } 21751da177e4SLinus Torvalds 21761da177e4SLinus Torvalds /* 21777a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 21787a2e8a8fSValerie Aurora */ 21797a2e8a8fSValerie Aurora 2180e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 21817a2e8a8fSValerie Aurora { 2182e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 21837a2e8a8fSValerie Aurora 21847a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 21857a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 21867a2e8a8fSValerie Aurora return 0; 21877a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 21887a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 21897a2e8a8fSValerie Aurora return 0; 21907a2e8a8fSValerie Aurora return type; 21917a2e8a8fSValerie Aurora } 21927a2e8a8fSValerie Aurora 21937a2e8a8fSValerie Aurora /* 219407b20889SRam Pai * recursively change the type of the mountpoint. 219507b20889SRam Pai */ 2196e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 219707b20889SRam Pai { 2198315fc83eSAl Viro struct mount *m; 21994b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2200e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 22017a2e8a8fSValerie Aurora int type; 2202719f5d7fSMiklos Szeredi int err = 0; 220307b20889SRam Pai 22042d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 220507b20889SRam Pai return -EINVAL; 220607b20889SRam Pai 2207e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 22087a2e8a8fSValerie Aurora if (!type) 22097a2e8a8fSValerie Aurora return -EINVAL; 22107a2e8a8fSValerie Aurora 221197216be0SAl Viro namespace_lock(); 2212719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2213719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2214719f5d7fSMiklos Szeredi if (err) 2215719f5d7fSMiklos Szeredi goto out_unlock; 2216719f5d7fSMiklos Szeredi } 2217719f5d7fSMiklos Szeredi 2218719ea2fbSAl Viro lock_mount_hash(); 2219909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 22200f0afb1dSAl Viro change_mnt_propagation(m, type); 2221719ea2fbSAl Viro unlock_mount_hash(); 2222719f5d7fSMiklos Szeredi 2223719f5d7fSMiklos Szeredi out_unlock: 222497216be0SAl Viro namespace_unlock(); 2225719f5d7fSMiklos Szeredi return err; 222607b20889SRam Pai } 222707b20889SRam Pai 22285ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 22295ff9d8a6SEric W. Biederman { 22305ff9d8a6SEric W. Biederman struct mount *child; 22315ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 22325ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 22335ff9d8a6SEric W. Biederman continue; 22345ff9d8a6SEric W. Biederman 22355ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 22365ff9d8a6SEric W. Biederman return true; 22375ff9d8a6SEric W. Biederman } 22385ff9d8a6SEric W. Biederman return false; 22395ff9d8a6SEric W. Biederman } 22405ff9d8a6SEric W. Biederman 2241a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2242a07b2000SAl Viro { 2243a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2244a07b2000SAl Viro 2245a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2246a07b2000SAl Viro return mnt; 2247a07b2000SAl Viro 2248a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2249a07b2000SAl Viro return mnt; 2250a07b2000SAl Viro 2251a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2252a07b2000SAl Viro return mnt; 2253a07b2000SAl Viro 2254a07b2000SAl Viro if (recurse) 2255a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2256a07b2000SAl Viro else 2257a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2258a07b2000SAl Viro 2259a07b2000SAl Viro if (!IS_ERR(mnt)) 2260a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2261a07b2000SAl Viro 2262a07b2000SAl Viro return mnt; 2263a07b2000SAl Viro } 2264a07b2000SAl Viro 226507b20889SRam Pai /* 22661da177e4SLinus Torvalds * do loopback mount. 22671da177e4SLinus Torvalds */ 2268808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 22692dafe1c4SEric Sandeen int recurse) 22701da177e4SLinus Torvalds { 22712d92ab3cSAl Viro struct path old_path; 2272a07b2000SAl Viro struct mount *mnt = NULL, *parent; 227384d17192SAl Viro struct mountpoint *mp; 227457eccb83SAl Viro int err; 22751da177e4SLinus Torvalds if (!old_name || !*old_name) 22761da177e4SLinus Torvalds return -EINVAL; 2277815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 22781da177e4SLinus Torvalds if (err) 22791da177e4SLinus Torvalds return err; 22801da177e4SLinus Torvalds 22818823c079SEric W. Biederman err = -EINVAL; 22824ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 22838823c079SEric W. Biederman goto out; 22848823c079SEric W. Biederman 228584d17192SAl Viro mp = lock_mount(path); 2286a07b2000SAl Viro if (IS_ERR(mp)) { 228784d17192SAl Viro err = PTR_ERR(mp); 22889676f0c6SRam Pai goto out; 2289a07b2000SAl Viro } 22909676f0c6SRam Pai 229184d17192SAl Viro parent = real_mount(path->mnt); 2292e149ed2bSAl Viro if (!check_mnt(parent)) 2293e149ed2bSAl Viro goto out2; 2294e149ed2bSAl Viro 2295a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2296be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2297be34d1a3SDavid Howells err = PTR_ERR(mnt); 2298e9c5d8a5SAndrey Vagin goto out2; 2299be34d1a3SDavid Howells } 2300ccd48bc7SAl Viro 230184d17192SAl Viro err = graft_tree(mnt, parent, mp); 23021da177e4SLinus Torvalds if (err) { 2303719ea2fbSAl Viro lock_mount_hash(); 2304e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2305719ea2fbSAl Viro unlock_mount_hash(); 23065b83d2c5SRam Pai } 2307b12cea91SAl Viro out2: 230884d17192SAl Viro unlock_mount(mp); 2309ccd48bc7SAl Viro out: 23102d92ab3cSAl Viro path_put(&old_path); 23111da177e4SLinus Torvalds return err; 23121da177e4SLinus Torvalds } 23131da177e4SLinus Torvalds 2314a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2315a07b2000SAl Viro { 2316a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2317a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2318a07b2000SAl Viro struct mount *mnt, *p; 2319a07b2000SAl Viro struct file *file; 2320a07b2000SAl Viro 2321a07b2000SAl Viro if (IS_ERR(ns)) 2322a07b2000SAl Viro return ERR_CAST(ns); 2323a07b2000SAl Viro 2324a07b2000SAl Viro namespace_lock(); 2325a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2326a07b2000SAl Viro if (IS_ERR(mnt)) { 2327a07b2000SAl Viro namespace_unlock(); 2328a07b2000SAl Viro free_mnt_ns(ns); 2329a07b2000SAl Viro return ERR_CAST(mnt); 2330a07b2000SAl Viro } 2331a07b2000SAl Viro 2332a07b2000SAl Viro lock_mount_hash(); 2333a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2334a07b2000SAl Viro p->mnt_ns = ns; 2335a07b2000SAl Viro ns->mounts++; 2336a07b2000SAl Viro } 2337a07b2000SAl Viro ns->root = mnt; 2338a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2339a07b2000SAl Viro mntget(&mnt->mnt); 2340a07b2000SAl Viro unlock_mount_hash(); 2341a07b2000SAl Viro namespace_unlock(); 2342a07b2000SAl Viro 2343a07b2000SAl Viro mntput(path->mnt); 2344a07b2000SAl Viro path->mnt = &mnt->mnt; 2345a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2346a07b2000SAl Viro if (IS_ERR(file)) 2347a07b2000SAl Viro dissolve_on_fput(path->mnt); 2348a07b2000SAl Viro else 2349a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2350a07b2000SAl Viro return file; 2351a07b2000SAl Viro } 2352a07b2000SAl Viro 2353a07b2000SAl Viro SYSCALL_DEFINE3(open_tree, int, dfd, const char *, filename, unsigned, flags) 2354a07b2000SAl Viro { 2355a07b2000SAl Viro struct file *file; 2356a07b2000SAl Viro struct path path; 2357a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2358a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2359a07b2000SAl Viro int error; 2360a07b2000SAl Viro int fd; 2361a07b2000SAl Viro 2362a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2363a07b2000SAl Viro 2364a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2365a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2366a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2367a07b2000SAl Viro return -EINVAL; 2368a07b2000SAl Viro 2369a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2370a07b2000SAl Viro return -EINVAL; 2371a07b2000SAl Viro 2372a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2373a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2374a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2375a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2376a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2377a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2378a07b2000SAl Viro 2379a07b2000SAl Viro if (detached && !may_mount()) 2380a07b2000SAl Viro return -EPERM; 2381a07b2000SAl Viro 2382a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2383a07b2000SAl Viro if (fd < 0) 2384a07b2000SAl Viro return fd; 2385a07b2000SAl Viro 2386a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2387a07b2000SAl Viro if (unlikely(error)) { 2388a07b2000SAl Viro file = ERR_PTR(error); 2389a07b2000SAl Viro } else { 2390a07b2000SAl Viro if (detached) 2391a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2392a07b2000SAl Viro else 2393a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2394a07b2000SAl Viro path_put(&path); 2395a07b2000SAl Viro } 2396a07b2000SAl Viro if (IS_ERR(file)) { 2397a07b2000SAl Viro put_unused_fd(fd); 2398a07b2000SAl Viro return PTR_ERR(file); 2399a07b2000SAl Viro } 2400a07b2000SAl Viro fd_install(fd, file); 2401a07b2000SAl Viro return fd; 2402a07b2000SAl Viro } 2403a07b2000SAl Viro 240443f5e655SDavid Howells /* 240543f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 240643f5e655SDavid Howells * 240743f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 240843f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 240943f5e655SDavid Howells */ 241043f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 24112e4b7fcdSDave Hansen { 241243f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 24132e4b7fcdSDave Hansen 241443f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 241543f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 241643f5e655SDavid Howells return false; 241743f5e655SDavid Howells 241843f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 241943f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 242043f5e655SDavid Howells return false; 242143f5e655SDavid Howells 242243f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 242343f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 242443f5e655SDavid Howells return false; 242543f5e655SDavid Howells 242643f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 242743f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 242843f5e655SDavid Howells return false; 242943f5e655SDavid Howells 243043f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 243143f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 243243f5e655SDavid Howells return false; 243343f5e655SDavid Howells 243443f5e655SDavid Howells return true; 243543f5e655SDavid Howells } 243643f5e655SDavid Howells 243743f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 243843f5e655SDavid Howells { 243943f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 244043f5e655SDavid Howells 244143f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 24422e4b7fcdSDave Hansen return 0; 24432e4b7fcdSDave Hansen 24442e4b7fcdSDave Hansen if (readonly_request) 244543f5e655SDavid Howells return mnt_make_readonly(mnt); 244643f5e655SDavid Howells 244743f5e655SDavid Howells return __mnt_unmake_readonly(mnt); 244843f5e655SDavid Howells } 244943f5e655SDavid Howells 245043f5e655SDavid Howells /* 245143f5e655SDavid Howells * Update the user-settable attributes on a mount. The caller must hold 245243f5e655SDavid Howells * sb->s_umount for writing. 245343f5e655SDavid Howells */ 245443f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 245543f5e655SDavid Howells { 245643f5e655SDavid Howells lock_mount_hash(); 245743f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 245843f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 245943f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 246043f5e655SDavid Howells unlock_mount_hash(); 246143f5e655SDavid Howells } 246243f5e655SDavid Howells 246343f5e655SDavid Howells /* 246443f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 246543f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 246643f5e655SDavid Howells * to mount(2). 246743f5e655SDavid Howells */ 246843f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 246943f5e655SDavid Howells { 247043f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 247143f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 247243f5e655SDavid Howells int ret; 247343f5e655SDavid Howells 247443f5e655SDavid Howells if (!check_mnt(mnt)) 247543f5e655SDavid Howells return -EINVAL; 247643f5e655SDavid Howells 247743f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 247843f5e655SDavid Howells return -EINVAL; 247943f5e655SDavid Howells 248043f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 248143f5e655SDavid Howells return -EPERM; 248243f5e655SDavid Howells 248343f5e655SDavid Howells down_write(&sb->s_umount); 248443f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 248543f5e655SDavid Howells if (ret == 0) 248643f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 248743f5e655SDavid Howells up_write(&sb->s_umount); 248843f5e655SDavid Howells return ret; 24892e4b7fcdSDave Hansen } 24902e4b7fcdSDave Hansen 24911da177e4SLinus Torvalds /* 24921da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 24931da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 24941da177e4SLinus Torvalds * on it - tough luck. 24951da177e4SLinus Torvalds */ 2496e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2497e462ec50SDavid Howells int mnt_flags, void *data) 24981da177e4SLinus Torvalds { 24991da177e4SLinus Torvalds int err; 25002d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2501143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 25028d0347f6SDavid Howells struct fs_context *fc; 25031da177e4SLinus Torvalds 2504143c8c91SAl Viro if (!check_mnt(mnt)) 25051da177e4SLinus Torvalds return -EINVAL; 25061da177e4SLinus Torvalds 25072d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 25081da177e4SLinus Torvalds return -EINVAL; 25091da177e4SLinus Torvalds 251043f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 251107b64558SEric W. Biederman return -EPERM; 25129566d674SEric W. Biederman 25138d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 25148d0347f6SDavid Howells if (IS_ERR(fc)) 25158d0347f6SDavid Howells return PTR_ERR(fc); 2516ff36fe2cSEric Paris 25178d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 25188d0347f6SDavid Howells if (!err) { 25191da177e4SLinus Torvalds down_write(&sb->s_umount); 252057eccb83SAl Viro err = -EPERM; 252143f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 25228d0347f6SDavid Howells err = reconfigure_super(fc); 252343f5e655SDavid Howells if (!err) 252443f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 25250e55a7ccSDan Williams } 25266339dab8SAl Viro up_write(&sb->s_umount); 25278d0347f6SDavid Howells } 25288d0347f6SDavid Howells put_fs_context(fc); 25291da177e4SLinus Torvalds return err; 25301da177e4SLinus Torvalds } 25311da177e4SLinus Torvalds 2532cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 25339676f0c6SRam Pai { 2534315fc83eSAl Viro struct mount *p; 2535909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2536fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 25379676f0c6SRam Pai return 1; 25389676f0c6SRam Pai } 25399676f0c6SRam Pai return 0; 25409676f0c6SRam Pai } 25419676f0c6SRam Pai 2542808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 25431da177e4SLinus Torvalds { 25442d92ab3cSAl Viro struct path old_path, parent_path; 2545676da58dSAl Viro struct mount *p; 25460fb54e50SAl Viro struct mount *old; 254784d17192SAl Viro struct mountpoint *mp; 254857eccb83SAl Viro int err; 25491da177e4SLinus Torvalds if (!old_name || !*old_name) 25501da177e4SLinus Torvalds return -EINVAL; 25512d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 25521da177e4SLinus Torvalds if (err) 25531da177e4SLinus Torvalds return err; 25541da177e4SLinus Torvalds 255584d17192SAl Viro mp = lock_mount(path); 255684d17192SAl Viro err = PTR_ERR(mp); 255784d17192SAl Viro if (IS_ERR(mp)) 2558cc53ce53SDavid Howells goto out; 2559cc53ce53SDavid Howells 2560143c8c91SAl Viro old = real_mount(old_path.mnt); 2561fc7be130SAl Viro p = real_mount(path->mnt); 2562143c8c91SAl Viro 25631da177e4SLinus Torvalds err = -EINVAL; 2564fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 25651da177e4SLinus Torvalds goto out1; 25661da177e4SLinus Torvalds 25675ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 25685ff9d8a6SEric W. Biederman goto out1; 25695ff9d8a6SEric W. Biederman 25701da177e4SLinus Torvalds err = -EINVAL; 25712d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 257221444403SRam Pai goto out1; 25731da177e4SLinus Torvalds 2574676da58dSAl Viro if (!mnt_has_parent(old)) 257521444403SRam Pai goto out1; 25761da177e4SLinus Torvalds 2577e36cb0b8SDavid Howells if (d_is_dir(path->dentry) != 2578e36cb0b8SDavid Howells d_is_dir(old_path.dentry)) 257921444403SRam Pai goto out1; 258021444403SRam Pai /* 258121444403SRam Pai * Don't move a mount residing in a shared parent. 258221444403SRam Pai */ 2583fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 258421444403SRam Pai goto out1; 25859676f0c6SRam Pai /* 25869676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 25879676f0c6SRam Pai * mount which is shared. 25889676f0c6SRam Pai */ 2589fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 25909676f0c6SRam Pai goto out1; 25911da177e4SLinus Torvalds err = -ELOOP; 2592fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2593676da58dSAl Viro if (p == old) 259421444403SRam Pai goto out1; 25951da177e4SLinus Torvalds 259684d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 25974ac91378SJan Blunck if (err) 259821444403SRam Pai goto out1; 25991da177e4SLinus Torvalds 26001da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 26011da177e4SLinus Torvalds * automatically */ 26026776db3dSAl Viro list_del_init(&old->mnt_expire); 26031da177e4SLinus Torvalds out1: 260484d17192SAl Viro unlock_mount(mp); 26051da177e4SLinus Torvalds out: 26061da177e4SLinus Torvalds if (!err) 26071a390689SAl Viro path_put(&parent_path); 26082d92ab3cSAl Viro path_put(&old_path); 26091da177e4SLinus Torvalds return err; 26101da177e4SLinus Torvalds } 26111da177e4SLinus Torvalds 26129d412a43SAl Viro /* 26139d412a43SAl Viro * add a mount into a namespace's mount tree 26149d412a43SAl Viro */ 261595bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 26169d412a43SAl Viro { 261784d17192SAl Viro struct mountpoint *mp; 261884d17192SAl Viro struct mount *parent; 26199d412a43SAl Viro int err; 26209d412a43SAl Viro 2621f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 26229d412a43SAl Viro 262384d17192SAl Viro mp = lock_mount(path); 262484d17192SAl Viro if (IS_ERR(mp)) 262584d17192SAl Viro return PTR_ERR(mp); 26269d412a43SAl Viro 262784d17192SAl Viro parent = real_mount(path->mnt); 26289d412a43SAl Viro err = -EINVAL; 262984d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2630156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2631156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 26329d412a43SAl Viro goto unlock; 2633156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 263484d17192SAl Viro if (!parent->mnt_ns) 2635156cacb1SAl Viro goto unlock; 2636156cacb1SAl Viro } 26379d412a43SAl Viro 26389d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 26399d412a43SAl Viro err = -EBUSY; 264095bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 26419d412a43SAl Viro path->mnt->mnt_root == path->dentry) 26429d412a43SAl Viro goto unlock; 26439d412a43SAl Viro 26449d412a43SAl Viro err = -EINVAL; 2645e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 26469d412a43SAl Viro goto unlock; 26479d412a43SAl Viro 264895bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 264984d17192SAl Viro err = graft_tree(newmnt, parent, mp); 26509d412a43SAl Viro 26519d412a43SAl Viro unlock: 265284d17192SAl Viro unlock_mount(mp); 26539d412a43SAl Viro return err; 26549d412a43SAl Viro } 2655b1e75df4SAl Viro 2656132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 2657132e4608SDavid Howells 2658132e4608SDavid Howells /* 2659132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 2660132e4608SDavid Howells * be added to the namespace tree. 2661132e4608SDavid Howells */ 2662132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 2663132e4608SDavid Howells unsigned int mnt_flags) 2664132e4608SDavid Howells { 2665132e4608SDavid Howells struct vfsmount *mnt; 2666132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 2667132e4608SDavid Howells int error; 2668132e4608SDavid Howells 2669c9ce29edSAl Viro error = security_sb_kern_mount(sb); 2670c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 2671c9ce29edSAl Viro error = -EPERM; 2672c9ce29edSAl Viro 2673c9ce29edSAl Viro if (unlikely(error)) { 2674c9ce29edSAl Viro fc_drop_locked(fc); 2675c9ce29edSAl Viro return error; 2676132e4608SDavid Howells } 2677132e4608SDavid Howells 2678132e4608SDavid Howells up_write(&sb->s_umount); 2679132e4608SDavid Howells 2680132e4608SDavid Howells mnt = vfs_create_mount(fc); 2681132e4608SDavid Howells if (IS_ERR(mnt)) 2682132e4608SDavid Howells return PTR_ERR(mnt); 2683132e4608SDavid Howells 2684132e4608SDavid Howells error = do_add_mount(real_mount(mnt), mountpoint, mnt_flags); 2685132e4608SDavid Howells if (error < 0) 2686132e4608SDavid Howells mntput(mnt); 2687132e4608SDavid Howells return error; 2688132e4608SDavid Howells } 26891b852bceSEric W. Biederman 26901da177e4SLinus Torvalds /* 26911da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 26921da177e4SLinus Torvalds * namespace's tree 26931da177e4SLinus Torvalds */ 2694e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 2695808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 26961da177e4SLinus Torvalds { 26970c55cfc4SEric W. Biederman struct file_system_type *type; 2698a0c9a8b8SAl Viro struct fs_context *fc; 2699a0c9a8b8SAl Viro const char *subtype = NULL; 2700a0c9a8b8SAl Viro int err = 0; 27011da177e4SLinus Torvalds 27020c55cfc4SEric W. Biederman if (!fstype) 27031da177e4SLinus Torvalds return -EINVAL; 27041da177e4SLinus Torvalds 27050c55cfc4SEric W. Biederman type = get_fs_type(fstype); 27060c55cfc4SEric W. Biederman if (!type) 27070c55cfc4SEric W. Biederman return -ENODEV; 27080c55cfc4SEric W. Biederman 2709a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 2710a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 2711a0c9a8b8SAl Viro if (subtype) { 2712a0c9a8b8SAl Viro subtype++; 2713a0c9a8b8SAl Viro if (!*subtype) { 27140c55cfc4SEric W. Biederman put_filesystem(type); 2715a0c9a8b8SAl Viro return -EINVAL; 2716a0c9a8b8SAl Viro } 2717a0c9a8b8SAl Viro } else { 2718a0c9a8b8SAl Viro subtype = ""; 2719a0c9a8b8SAl Viro } 27208654df4eSEric W. Biederman } 27218654df4eSEric W. Biederman 2722a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 2723a0c9a8b8SAl Viro put_filesystem(type); 2724a0c9a8b8SAl Viro if (IS_ERR(fc)) 2725a0c9a8b8SAl Viro return PTR_ERR(fc); 2726a0c9a8b8SAl Viro 27273e1aeb00SDavid Howells if (subtype) 27283e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 27293e1aeb00SDavid Howells subtype, strlen(subtype)); 27303e1aeb00SDavid Howells if (!err && name) 27313e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 2732a0c9a8b8SAl Viro if (!err) 2733a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 2734a0c9a8b8SAl Viro if (!err) 2735a0c9a8b8SAl Viro err = vfs_get_tree(fc); 2736132e4608SDavid Howells if (!err) 2737132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 2738a0c9a8b8SAl Viro 2739a0c9a8b8SAl Viro put_fs_context(fc); 274015f9a3f3SAl Viro return err; 27411da177e4SLinus Torvalds } 27421da177e4SLinus Torvalds 274319a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 274419a167afSAl Viro { 27456776db3dSAl Viro struct mount *mnt = real_mount(m); 274619a167afSAl Viro int err; 274719a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 274819a167afSAl Viro * expired before we get a chance to add it 274919a167afSAl Viro */ 27506776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 275119a167afSAl Viro 275219a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 275319a167afSAl Viro m->mnt_root == path->dentry) { 2754b1e75df4SAl Viro err = -ELOOP; 2755b1e75df4SAl Viro goto fail; 275619a167afSAl Viro } 275719a167afSAl Viro 275895bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2759b1e75df4SAl Viro if (!err) 2760b1e75df4SAl Viro return 0; 2761b1e75df4SAl Viro fail: 2762b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 27636776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 276497216be0SAl Viro namespace_lock(); 27656776db3dSAl Viro list_del_init(&mnt->mnt_expire); 276697216be0SAl Viro namespace_unlock(); 276719a167afSAl Viro } 2768b1e75df4SAl Viro mntput(m); 2769b1e75df4SAl Viro mntput(m); 277019a167afSAl Viro return err; 277119a167afSAl Viro } 277219a167afSAl Viro 2773ea5b778aSDavid Howells /** 2774ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2775ea5b778aSDavid Howells * @mnt: The mount to list. 2776ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2777ea5b778aSDavid Howells */ 2778ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2779ea5b778aSDavid Howells { 278097216be0SAl Viro namespace_lock(); 2781ea5b778aSDavid Howells 27826776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2783ea5b778aSDavid Howells 278497216be0SAl Viro namespace_unlock(); 2785ea5b778aSDavid Howells } 2786ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2787ea5b778aSDavid Howells 2788ea5b778aSDavid Howells /* 27891da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 27901da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 27911da177e4SLinus Torvalds * here 27921da177e4SLinus Torvalds */ 27931da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 27941da177e4SLinus Torvalds { 2795761d5c38SAl Viro struct mount *mnt, *next; 27961da177e4SLinus Torvalds LIST_HEAD(graveyard); 27971da177e4SLinus Torvalds 27981da177e4SLinus Torvalds if (list_empty(mounts)) 27991da177e4SLinus Torvalds return; 28001da177e4SLinus Torvalds 280197216be0SAl Viro namespace_lock(); 2802719ea2fbSAl Viro lock_mount_hash(); 28031da177e4SLinus Torvalds 28041da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 28051da177e4SLinus Torvalds * following criteria: 28061da177e4SLinus Torvalds * - only referenced by its parent vfsmount 28071da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 28081da177e4SLinus Torvalds * cleared by mntput()) 28091da177e4SLinus Torvalds */ 28106776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2811863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 28121ab59738SAl Viro propagate_mount_busy(mnt, 1)) 28131da177e4SLinus Torvalds continue; 28146776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 28151da177e4SLinus Torvalds } 2816bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 28176776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2818143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2819e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2820bcc5c7d2SAl Viro } 2821719ea2fbSAl Viro unlock_mount_hash(); 28223ab6abeeSAl Viro namespace_unlock(); 28231da177e4SLinus Torvalds } 28241da177e4SLinus Torvalds 28251da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 28261da177e4SLinus Torvalds 28271da177e4SLinus Torvalds /* 28285528f911STrond Myklebust * Ripoff of 'select_parent()' 28295528f911STrond Myklebust * 28305528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 28315528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 28325528f911STrond Myklebust */ 2833692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 28345528f911STrond Myklebust { 2835692afc31SAl Viro struct mount *this_parent = parent; 28365528f911STrond Myklebust struct list_head *next; 28375528f911STrond Myklebust int found = 0; 28385528f911STrond Myklebust 28395528f911STrond Myklebust repeat: 28406b41d536SAl Viro next = this_parent->mnt_mounts.next; 28415528f911STrond Myklebust resume: 28426b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 28435528f911STrond Myklebust struct list_head *tmp = next; 28446b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 28455528f911STrond Myklebust 28465528f911STrond Myklebust next = tmp->next; 2847692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 28485528f911STrond Myklebust continue; 28495528f911STrond Myklebust /* 28505528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 28515528f911STrond Myklebust */ 28526b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 28535528f911STrond Myklebust this_parent = mnt; 28545528f911STrond Myklebust goto repeat; 28555528f911STrond Myklebust } 28565528f911STrond Myklebust 28571ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 28586776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 28595528f911STrond Myklebust found++; 28605528f911STrond Myklebust } 28615528f911STrond Myklebust } 28625528f911STrond Myklebust /* 28635528f911STrond Myklebust * All done at this level ... ascend and resume the search 28645528f911STrond Myklebust */ 28655528f911STrond Myklebust if (this_parent != parent) { 28666b41d536SAl Viro next = this_parent->mnt_child.next; 28670714a533SAl Viro this_parent = this_parent->mnt_parent; 28685528f911STrond Myklebust goto resume; 28695528f911STrond Myklebust } 28705528f911STrond Myklebust return found; 28715528f911STrond Myklebust } 28725528f911STrond Myklebust 28735528f911STrond Myklebust /* 28745528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 28755528f911STrond Myklebust * submounts of a specific parent mountpoint 287699b7db7bSNick Piggin * 287748a066e7SAl Viro * mount_lock must be held for write 28785528f911STrond Myklebust */ 2879b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 28805528f911STrond Myklebust { 28815528f911STrond Myklebust LIST_HEAD(graveyard); 2882761d5c38SAl Viro struct mount *m; 28835528f911STrond Myklebust 28845528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2885c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2886bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2887761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 28886776db3dSAl Viro mnt_expire); 2889143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2890e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 2891bcc5c7d2SAl Viro } 2892bcc5c7d2SAl Viro } 28935528f911STrond Myklebust } 28945528f911STrond Myklebust 28955528f911STrond Myklebust /* 28961da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 28971da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 28981da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 28991da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 29001da177e4SLinus Torvalds */ 2901b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2902b58fed8bSRam Pai unsigned long n) 29031da177e4SLinus Torvalds { 29041da177e4SLinus Torvalds char *t = to; 29051da177e4SLinus Torvalds const char __user *f = from; 29061da177e4SLinus Torvalds char c; 29071da177e4SLinus Torvalds 290896d4f267SLinus Torvalds if (!access_ok(from, n)) 29091da177e4SLinus Torvalds return n; 29101da177e4SLinus Torvalds 29111da177e4SLinus Torvalds while (n) { 29121da177e4SLinus Torvalds if (__get_user(c, f)) { 29131da177e4SLinus Torvalds memset(t, 0, n); 29141da177e4SLinus Torvalds break; 29151da177e4SLinus Torvalds } 29161da177e4SLinus Torvalds *t++ = c; 29171da177e4SLinus Torvalds f++; 29181da177e4SLinus Torvalds n--; 29191da177e4SLinus Torvalds } 29201da177e4SLinus Torvalds return n; 29211da177e4SLinus Torvalds } 29221da177e4SLinus Torvalds 2923b40ef869SAl Viro void *copy_mount_options(const void __user * data) 29241da177e4SLinus Torvalds { 29251da177e4SLinus Torvalds int i; 29261da177e4SLinus Torvalds unsigned long size; 2927b40ef869SAl Viro char *copy; 29281da177e4SLinus Torvalds 29291da177e4SLinus Torvalds if (!data) 2930b40ef869SAl Viro return NULL; 29311da177e4SLinus Torvalds 2932b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 2933b40ef869SAl Viro if (!copy) 2934b40ef869SAl Viro return ERR_PTR(-ENOMEM); 29351da177e4SLinus Torvalds 29361da177e4SLinus Torvalds /* We only care that *some* data at the address the user 29371da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 29381da177e4SLinus Torvalds * the remainder of the page. 29391da177e4SLinus Torvalds */ 29401da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 29411da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 29421da177e4SLinus Torvalds if (size > PAGE_SIZE) 29431da177e4SLinus Torvalds size = PAGE_SIZE; 29441da177e4SLinus Torvalds 2945b40ef869SAl Viro i = size - exact_copy_from_user(copy, data, size); 29461da177e4SLinus Torvalds if (!i) { 2947b40ef869SAl Viro kfree(copy); 2948b40ef869SAl Viro return ERR_PTR(-EFAULT); 29491da177e4SLinus Torvalds } 29501da177e4SLinus Torvalds if (i != PAGE_SIZE) 2951b40ef869SAl Viro memset(copy + i, 0, PAGE_SIZE - i); 2952b40ef869SAl Viro return copy; 29531da177e4SLinus Torvalds } 29541da177e4SLinus Torvalds 2955b8850d1fSTim Gardner char *copy_mount_string(const void __user *data) 2956eca6f534SVegard Nossum { 2957fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 2958eca6f534SVegard Nossum } 2959eca6f534SVegard Nossum 29601da177e4SLinus Torvalds /* 29611da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 29621da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 29631da177e4SLinus Torvalds * 29641da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 29651da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 29661da177e4SLinus Torvalds * information (or be NULL). 29671da177e4SLinus Torvalds * 29681da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 29691da177e4SLinus Torvalds * When the flags word was introduced its top half was required 29701da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 29711da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 29721da177e4SLinus Torvalds * and must be discarded. 29731da177e4SLinus Torvalds */ 29745e6123f3SSeunghun Lee long do_mount(const char *dev_name, const char __user *dir_name, 2975808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 29761da177e4SLinus Torvalds { 29772d92ab3cSAl Viro struct path path; 2978e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 29791da177e4SLinus Torvalds int retval = 0; 29801da177e4SLinus Torvalds 29811da177e4SLinus Torvalds /* Discard magic */ 29821da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 29831da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 29841da177e4SLinus Torvalds 29851da177e4SLinus Torvalds /* Basic sanity checks */ 29861da177e4SLinus Torvalds if (data_page) 29871da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 29881da177e4SLinus Torvalds 2989e462ec50SDavid Howells if (flags & MS_NOUSER) 2990e462ec50SDavid Howells return -EINVAL; 2991e462ec50SDavid Howells 2992a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 29935e6123f3SSeunghun Lee retval = user_path(dir_name, &path); 2994a27ab9f2STetsuo Handa if (retval) 2995a27ab9f2STetsuo Handa return retval; 2996a27ab9f2STetsuo Handa 2997a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2998a27ab9f2STetsuo Handa type_page, flags, data_page); 29990d5cadb8SAl Viro if (!retval && !may_mount()) 30000d5cadb8SAl Viro retval = -EPERM; 3001e462ec50SDavid Howells if (!retval && (flags & SB_MANDLOCK) && !may_mandlock()) 30029e8925b6SJeff Layton retval = -EPERM; 3003a27ab9f2STetsuo Handa if (retval) 3004a27ab9f2STetsuo Handa goto dput_out; 3005a27ab9f2STetsuo Handa 3006613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3007613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 30080a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 30090a1c01c9SMatthew Garrett 30101da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 30111da177e4SLinus Torvalds if (flags & MS_NOSUID) 30121da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 30131da177e4SLinus Torvalds if (flags & MS_NODEV) 30141da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 30151da177e4SLinus Torvalds if (flags & MS_NOEXEC) 30161da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3017fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3018fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3019fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3020fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3021d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3022d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3023a9e5b732SDavid Howells if (flags & MS_RDONLY) 30242e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3025fc33a7bbSChristoph Hellwig 3026ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3027ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3028ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3029ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3030ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3031ffbc6f0eSEric W. Biederman mnt_flags |= path.mnt->mnt_flags & MNT_ATIME_MASK; 3032ffbc6f0eSEric W. Biederman } 3033ffbc6f0eSEric W. Biederman 3034e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3035e462ec50SDavid Howells SB_SYNCHRONOUS | 3036e462ec50SDavid Howells SB_MANDLOCK | 3037e462ec50SDavid Howells SB_DIRSYNC | 3038e462ec50SDavid Howells SB_SILENT | 3039917086ffSMimi Zohar SB_POSIXACL | 3040d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3041917086ffSMimi Zohar SB_I_VERSION); 30421da177e4SLinus Torvalds 304343f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 304443f5e655SDavid Howells retval = do_reconfigure_mnt(&path, mnt_flags); 304543f5e655SDavid Howells else if (flags & MS_REMOUNT) 3046e462ec50SDavid Howells retval = do_remount(&path, flags, sb_flags, mnt_flags, 30471da177e4SLinus Torvalds data_page); 30481da177e4SLinus Torvalds else if (flags & MS_BIND) 30492d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 30509676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 30512d92ab3cSAl Viro retval = do_change_type(&path, flags); 30521da177e4SLinus Torvalds else if (flags & MS_MOVE) 30532d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 30541da177e4SLinus Torvalds else 3055e462ec50SDavid Howells retval = do_new_mount(&path, type_page, sb_flags, mnt_flags, 30561da177e4SLinus Torvalds dev_name, data_page); 30571da177e4SLinus Torvalds dput_out: 30582d92ab3cSAl Viro path_put(&path); 30591da177e4SLinus Torvalds return retval; 30601da177e4SLinus Torvalds } 30611da177e4SLinus Torvalds 3062537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3063537f7ccbSEric W. Biederman { 3064537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3065537f7ccbSEric W. Biederman } 3066537f7ccbSEric W. Biederman 3067537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3068537f7ccbSEric W. Biederman { 3069537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3070537f7ccbSEric W. Biederman } 3071537f7ccbSEric W. Biederman 3072771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3073771b1371SEric W. Biederman { 307474e83122SAl Viro if (!is_anon_ns(ns)) 30756344c433SAl Viro ns_free_inum(&ns->ns); 3076537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3077771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3078771b1371SEric W. Biederman kfree(ns); 3079771b1371SEric W. Biederman } 3080771b1371SEric W. Biederman 30818823c079SEric W. Biederman /* 30828823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 30838823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 30848823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 30858823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 30868823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 30878823c079SEric W. Biederman */ 30888823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 30898823c079SEric W. Biederman 309074e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3091cf8d2c11STrond Myklebust { 3092cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3093537f7ccbSEric W. Biederman struct ucounts *ucounts; 309498f842e6SEric W. Biederman int ret; 3095cf8d2c11STrond Myklebust 3096537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3097537f7ccbSEric W. Biederman if (!ucounts) 3098df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3099537f7ccbSEric W. Biederman 310074e83122SAl Viro new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 3101537f7ccbSEric W. Biederman if (!new_ns) { 3102537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3103cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3104537f7ccbSEric W. Biederman } 310574e83122SAl Viro if (!anon) { 31066344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 310798f842e6SEric W. Biederman if (ret) { 310898f842e6SEric W. Biederman kfree(new_ns); 3109537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 311098f842e6SEric W. Biederman return ERR_PTR(ret); 311198f842e6SEric W. Biederman } 311274e83122SAl Viro } 311333c42940SAl Viro new_ns->ns.ops = &mntns_operations; 311474e83122SAl Viro if (!anon) 31158823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 3116cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 3117cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3118cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 3119771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3120537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3121cf8d2c11STrond Myklebust return new_ns; 3122cf8d2c11STrond Myklebust } 3123cf8d2c11STrond Myklebust 31240766f788SEmese Revfy __latent_entropy 31259559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 31269559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 31271da177e4SLinus Torvalds { 31286b3286edSKirill Korotaev struct mnt_namespace *new_ns; 31297f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3130315fc83eSAl Viro struct mount *p, *q; 31319559f689SAl Viro struct mount *old; 3132cb338d06SAl Viro struct mount *new; 31337a472ef4SEric W. Biederman int copy_flags; 31341da177e4SLinus Torvalds 31359559f689SAl Viro BUG_ON(!ns); 31369559f689SAl Viro 31379559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 31389559f689SAl Viro get_mnt_ns(ns); 31399559f689SAl Viro return ns; 31409559f689SAl Viro } 31419559f689SAl Viro 31429559f689SAl Viro old = ns->root; 31439559f689SAl Viro 314474e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3145cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3146cf8d2c11STrond Myklebust return new_ns; 31471da177e4SLinus Torvalds 314897216be0SAl Viro namespace_lock(); 31491da177e4SLinus Torvalds /* First pass: copy the tree topology */ 31504ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 31519559f689SAl Viro if (user_ns != ns->user_ns) 31523bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 31537a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3154be34d1a3SDavid Howells if (IS_ERR(new)) { 3155328e6d90SAl Viro namespace_unlock(); 3156771b1371SEric W. Biederman free_mnt_ns(new_ns); 3157be34d1a3SDavid Howells return ERR_CAST(new); 31581da177e4SLinus Torvalds } 31593bd045ccSAl Viro if (user_ns != ns->user_ns) { 31603bd045ccSAl Viro lock_mount_hash(); 31613bd045ccSAl Viro lock_mnt_tree(new); 31623bd045ccSAl Viro unlock_mount_hash(); 31633bd045ccSAl Viro } 3164be08d6d2SAl Viro new_ns->root = new; 31651a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 31661da177e4SLinus Torvalds 31671da177e4SLinus Torvalds /* 31681da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 31691da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 31701da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 31711da177e4SLinus Torvalds */ 3172909b0a88SAl Viro p = old; 3173cb338d06SAl Viro q = new; 31741da177e4SLinus Torvalds while (p) { 3175143c8c91SAl Viro q->mnt_ns = new_ns; 3176d2921684SEric W. Biederman new_ns->mounts++; 31779559f689SAl Viro if (new_fs) { 31789559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 31799559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3180315fc83eSAl Viro rootmnt = &p->mnt; 31811da177e4SLinus Torvalds } 31829559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 31839559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3184315fc83eSAl Viro pwdmnt = &p->mnt; 31851da177e4SLinus Torvalds } 31861da177e4SLinus Torvalds } 3187909b0a88SAl Viro p = next_mnt(p, old); 3188909b0a88SAl Viro q = next_mnt(q, new); 31894ce5d2b1SEric W. Biederman if (!q) 31904ce5d2b1SEric W. Biederman break; 31914ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 31924ce5d2b1SEric W. Biederman p = next_mnt(p, old); 31931da177e4SLinus Torvalds } 3194328e6d90SAl Viro namespace_unlock(); 31951da177e4SLinus Torvalds 31961da177e4SLinus Torvalds if (rootmnt) 3197f03c6599SAl Viro mntput(rootmnt); 31981da177e4SLinus Torvalds if (pwdmnt) 3199f03c6599SAl Viro mntput(pwdmnt); 32001da177e4SLinus Torvalds 3201741a2951SJANAK DESAI return new_ns; 3202741a2951SJANAK DESAI } 3203741a2951SJANAK DESAI 320474e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3205cf8d2c11STrond Myklebust { 32061a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3207ea441d11SAl Viro struct mnt_namespace *ns; 3208d31da0f0SAl Viro struct super_block *s; 3209ea441d11SAl Viro struct path path; 3210ea441d11SAl Viro int err; 3211ea441d11SAl Viro 321274e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 321374e83122SAl Viro if (IS_ERR(ns)) { 321474e83122SAl Viro mntput(m); 3215ea441d11SAl Viro return ERR_CAST(ns); 321674e83122SAl Viro } 321774e83122SAl Viro mnt->mnt_ns = ns; 321874e83122SAl Viro ns->root = mnt; 321974e83122SAl Viro ns->mounts++; 322074e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3221ea441d11SAl Viro 322274e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3223ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3224ea441d11SAl Viro 3225ea441d11SAl Viro put_mnt_ns(ns); 3226ea441d11SAl Viro 3227ea441d11SAl Viro if (err) 3228ea441d11SAl Viro return ERR_PTR(err); 3229ea441d11SAl Viro 3230ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3231d31da0f0SAl Viro s = path.mnt->mnt_sb; 3232d31da0f0SAl Viro atomic_inc(&s->s_active); 3233ea441d11SAl Viro mntput(path.mnt); 3234ea441d11SAl Viro /* lock the sucker */ 3235d31da0f0SAl Viro down_write(&s->s_umount); 3236ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3237ea441d11SAl Viro return path.dentry; 3238ea441d11SAl Viro } 3239ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3240ea441d11SAl Viro 3241312db1aaSDominik Brodowski int ksys_mount(char __user *dev_name, char __user *dir_name, char __user *type, 3242312db1aaSDominik Brodowski unsigned long flags, void __user *data) 32431da177e4SLinus Torvalds { 3244eca6f534SVegard Nossum int ret; 3245eca6f534SVegard Nossum char *kernel_type; 3246eca6f534SVegard Nossum char *kernel_dev; 3247b40ef869SAl Viro void *options; 32481da177e4SLinus Torvalds 3249b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3250b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3251b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3252eca6f534SVegard Nossum goto out_type; 32531da177e4SLinus Torvalds 3254b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3255b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3256b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3257eca6f534SVegard Nossum goto out_dev; 32581da177e4SLinus Torvalds 3259b40ef869SAl Viro options = copy_mount_options(data); 3260b40ef869SAl Viro ret = PTR_ERR(options); 3261b40ef869SAl Viro if (IS_ERR(options)) 3262eca6f534SVegard Nossum goto out_data; 32631da177e4SLinus Torvalds 3264b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3265eca6f534SVegard Nossum 3266b40ef869SAl Viro kfree(options); 3267eca6f534SVegard Nossum out_data: 3268eca6f534SVegard Nossum kfree(kernel_dev); 3269eca6f534SVegard Nossum out_dev: 3270eca6f534SVegard Nossum kfree(kernel_type); 3271eca6f534SVegard Nossum out_type: 3272eca6f534SVegard Nossum return ret; 32731da177e4SLinus Torvalds } 32741da177e4SLinus Torvalds 3275312db1aaSDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3276312db1aaSDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 3277312db1aaSDominik Brodowski { 3278312db1aaSDominik Brodowski return ksys_mount(dev_name, dir_name, type, flags, data); 3279312db1aaSDominik Brodowski } 3280312db1aaSDominik Brodowski 32811da177e4SLinus Torvalds /* 3282afac7cbaSAl Viro * Return true if path is reachable from root 3283afac7cbaSAl Viro * 328448a066e7SAl Viro * namespace_sem or mount_lock is held 3285afac7cbaSAl Viro */ 3286643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3287afac7cbaSAl Viro const struct path *root) 3288afac7cbaSAl Viro { 3289643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3290a73324daSAl Viro dentry = mnt->mnt_mountpoint; 32910714a533SAl Viro mnt = mnt->mnt_parent; 3292afac7cbaSAl Viro } 3293643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3294afac7cbaSAl Viro } 3295afac7cbaSAl Viro 3296640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3297afac7cbaSAl Viro { 329825ab4c9bSYaowei Bai bool res; 329948a066e7SAl Viro read_seqlock_excl(&mount_lock); 3300643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 330148a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3302afac7cbaSAl Viro return res; 3303afac7cbaSAl Viro } 3304afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3305afac7cbaSAl Viro 3306afac7cbaSAl Viro /* 33071da177e4SLinus Torvalds * pivot_root Semantics: 33081da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 33091da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 33101da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 33111da177e4SLinus Torvalds * 33121da177e4SLinus Torvalds * Restrictions: 33131da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 33141da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 33151da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 33161da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 33171da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 33181da177e4SLinus Torvalds * 33194a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 33204a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 33214a0d11faSNeil Brown * in this situation. 33224a0d11faSNeil Brown * 33231da177e4SLinus Torvalds * Notes: 33241da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 33251da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 33261da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 33271da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 33281da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 33291da177e4SLinus Torvalds * first. 33301da177e4SLinus Torvalds */ 33313480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 33323480b257SHeiko Carstens const char __user *, put_old) 33331da177e4SLinus Torvalds { 33342d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 333584d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 333684d17192SAl Viro struct mountpoint *old_mp, *root_mp; 33371da177e4SLinus Torvalds int error; 33381da177e4SLinus Torvalds 33399b40bc90SAl Viro if (!may_mount()) 33401da177e4SLinus Torvalds return -EPERM; 33411da177e4SLinus Torvalds 33422d8f3038SAl Viro error = user_path_dir(new_root, &new); 33431da177e4SLinus Torvalds if (error) 33441da177e4SLinus Torvalds goto out0; 33451da177e4SLinus Torvalds 33462d8f3038SAl Viro error = user_path_dir(put_old, &old); 33471da177e4SLinus Torvalds if (error) 33481da177e4SLinus Torvalds goto out1; 33491da177e4SLinus Torvalds 33502d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 3351b12cea91SAl Viro if (error) 3352b12cea91SAl Viro goto out2; 33531da177e4SLinus Torvalds 3354f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 335584d17192SAl Viro old_mp = lock_mount(&old); 335684d17192SAl Viro error = PTR_ERR(old_mp); 335784d17192SAl Viro if (IS_ERR(old_mp)) 3358b12cea91SAl Viro goto out3; 3359b12cea91SAl Viro 33601da177e4SLinus Torvalds error = -EINVAL; 3361419148daSAl Viro new_mnt = real_mount(new.mnt); 3362419148daSAl Viro root_mnt = real_mount(root.mnt); 336384d17192SAl Viro old_mnt = real_mount(old.mnt); 336484d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 3365fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 3366fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 3367b12cea91SAl Viro goto out4; 3368143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 3369b12cea91SAl Viro goto out4; 33705ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 33715ff9d8a6SEric W. Biederman goto out4; 33721da177e4SLinus Torvalds error = -ENOENT; 3373f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 3374b12cea91SAl Viro goto out4; 33751da177e4SLinus Torvalds error = -EBUSY; 337684d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 3377b12cea91SAl Viro goto out4; /* loop, on the same file system */ 33781da177e4SLinus Torvalds error = -EINVAL; 33798c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 3380b12cea91SAl Viro goto out4; /* not a mountpoint */ 3381676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 3382b12cea91SAl Viro goto out4; /* not attached */ 338384d17192SAl Viro root_mp = root_mnt->mnt_mp; 33842d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 3385b12cea91SAl Viro goto out4; /* not a mountpoint */ 3386676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 3387b12cea91SAl Viro goto out4; /* not attached */ 33884ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 338984d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 3390b12cea91SAl Viro goto out4; 33910d082601SEric W. Biederman /* make certain new is below the root */ 33920d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 33930d082601SEric W. Biederman goto out4; 339484d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 3395719ea2fbSAl Viro lock_mount_hash(); 3396419148daSAl Viro detach_mnt(new_mnt, &parent_path); 3397419148daSAl Viro detach_mnt(root_mnt, &root_parent); 33985ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 33995ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 34005ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 34015ff9d8a6SEric W. Biederman } 34024ac91378SJan Blunck /* mount old root on put_old */ 340384d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 34044ac91378SJan Blunck /* mount new_root on / */ 340584d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 34066b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 34074fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 34084fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 34093895dbf8SEric W. Biederman put_mountpoint(root_mp); 3410719ea2fbSAl Viro unlock_mount_hash(); 34112d8f3038SAl Viro chroot_fs_refs(&root, &new); 34121da177e4SLinus Torvalds error = 0; 3413b12cea91SAl Viro out4: 341484d17192SAl Viro unlock_mount(old_mp); 3415b12cea91SAl Viro if (!error) { 34161a390689SAl Viro path_put(&root_parent); 34171a390689SAl Viro path_put(&parent_path); 3418b12cea91SAl Viro } 3419b12cea91SAl Viro out3: 34208c3ee42eSAl Viro path_put(&root); 3421b12cea91SAl Viro out2: 34222d8f3038SAl Viro path_put(&old); 34231da177e4SLinus Torvalds out1: 34242d8f3038SAl Viro path_put(&new); 34251da177e4SLinus Torvalds out0: 34261da177e4SLinus Torvalds return error; 34271da177e4SLinus Torvalds } 34281da177e4SLinus Torvalds 34291da177e4SLinus Torvalds static void __init init_mount_tree(void) 34301da177e4SLinus Torvalds { 34311da177e4SLinus Torvalds struct vfsmount *mnt; 343274e83122SAl Viro struct mount *m; 34336b3286edSKirill Korotaev struct mnt_namespace *ns; 3434ac748a09SJan Blunck struct path root; 34350c55cfc4SEric W. Biederman struct file_system_type *type; 34361da177e4SLinus Torvalds 34370c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 34380c55cfc4SEric W. Biederman if (!type) 34390c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 34400c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 34410c55cfc4SEric W. Biederman put_filesystem(type); 34421da177e4SLinus Torvalds if (IS_ERR(mnt)) 34431da177e4SLinus Torvalds panic("Can't create rootfs"); 3444b3e19d92SNick Piggin 344574e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 34463b22edc5STrond Myklebust if (IS_ERR(ns)) 34471da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 344874e83122SAl Viro m = real_mount(mnt); 344974e83122SAl Viro m->mnt_ns = ns; 345074e83122SAl Viro ns->root = m; 345174e83122SAl Viro ns->mounts = 1; 345274e83122SAl Viro list_add(&m->mnt_list, &ns->list); 34536b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 34546b3286edSKirill Korotaev get_mnt_ns(ns); 34551da177e4SLinus Torvalds 3456be08d6d2SAl Viro root.mnt = mnt; 3457be08d6d2SAl Viro root.dentry = mnt->mnt_root; 3458da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 3459ac748a09SJan Blunck 3460ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 3461ac748a09SJan Blunck set_fs_root(current->fs, &root); 34621da177e4SLinus Torvalds } 34631da177e4SLinus Torvalds 346474bf17cfSDenis Cheng void __init mnt_init(void) 34651da177e4SLinus Torvalds { 346615a67dd8SRandy Dunlap int err; 34671da177e4SLinus Torvalds 34687d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 346920c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 34701da177e4SLinus Torvalds 34710818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 347238129a13SAl Viro sizeof(struct hlist_head), 34730818bf27SAl Viro mhash_entries, 19, 34743d375d78SPavel Tatashin HASH_ZERO, 34750818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 34760818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 34770818bf27SAl Viro sizeof(struct hlist_head), 34780818bf27SAl Viro mphash_entries, 19, 34793d375d78SPavel Tatashin HASH_ZERO, 34800818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 34811da177e4SLinus Torvalds 348284d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 34831da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 34841da177e4SLinus Torvalds 34854b93dc9bSTejun Heo kernfs_init(); 34864b93dc9bSTejun Heo 348715a67dd8SRandy Dunlap err = sysfs_init(); 348815a67dd8SRandy Dunlap if (err) 348915a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 34908e24eea7SHarvey Harrison __func__, err); 349100d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 349200d26666SGreg Kroah-Hartman if (!fs_kobj) 34938e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 34941da177e4SLinus Torvalds init_rootfs(); 34951da177e4SLinus Torvalds init_mount_tree(); 34961da177e4SLinus Torvalds } 34971da177e4SLinus Torvalds 3498616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 34991da177e4SLinus Torvalds { 3500d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 3501616511d0STrond Myklebust return; 35027b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 3503771b1371SEric W. Biederman free_mnt_ns(ns); 35041da177e4SLinus Torvalds } 35059d412a43SAl Viro 3506d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 35079d412a43SAl Viro { 3508423e0ab0STim Chen struct vfsmount *mnt; 3509d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 3510423e0ab0STim Chen if (!IS_ERR(mnt)) { 3511423e0ab0STim Chen /* 3512423e0ab0STim Chen * it is a longterm mount, don't release mnt until 3513423e0ab0STim Chen * we unmount before file sys is unregistered 3514423e0ab0STim Chen */ 3515f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 3516423e0ab0STim Chen } 3517423e0ab0STim Chen return mnt; 35189d412a43SAl Viro } 3519d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 3520423e0ab0STim Chen 3521423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 3522423e0ab0STim Chen { 3523423e0ab0STim Chen /* release long term mount so mount point can be released */ 3524423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 3525f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 352648a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 3527423e0ab0STim Chen mntput(mnt); 3528423e0ab0STim Chen } 3529423e0ab0STim Chen } 3530423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 353102125a82SAl Viro 353202125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 353302125a82SAl Viro { 3534143c8c91SAl Viro return check_mnt(real_mount(mnt)); 353502125a82SAl Viro } 35368823c079SEric W. Biederman 35373151527eSEric W. Biederman bool current_chrooted(void) 35383151527eSEric W. Biederman { 35393151527eSEric W. Biederman /* Does the current process have a non-standard root */ 35403151527eSEric W. Biederman struct path ns_root; 35413151527eSEric W. Biederman struct path fs_root; 35423151527eSEric W. Biederman bool chrooted; 35433151527eSEric W. Biederman 35443151527eSEric W. Biederman /* Find the namespace root */ 35453151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 35463151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 35473151527eSEric W. Biederman path_get(&ns_root); 35483151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 35493151527eSEric W. Biederman ; 35503151527eSEric W. Biederman 35513151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 35523151527eSEric W. Biederman 35533151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 35543151527eSEric W. Biederman 35553151527eSEric W. Biederman path_put(&fs_root); 35563151527eSEric W. Biederman path_put(&ns_root); 35573151527eSEric W. Biederman 35583151527eSEric W. Biederman return chrooted; 35593151527eSEric W. Biederman } 35603151527eSEric W. Biederman 3561132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 3562132e4608SDavid Howells const struct super_block *sb, 35638654df4eSEric W. Biederman int *new_mnt_flags) 356487a8ebd6SEric W. Biederman { 35658c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 356687a8ebd6SEric W. Biederman struct mount *mnt; 3567e51db735SEric W. Biederman bool visible = false; 356887a8ebd6SEric W. Biederman 356944bb4385SAl Viro down_read(&namespace_sem); 357087a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 3571e51db735SEric W. Biederman struct mount *child; 357277b1a97dSEric W. Biederman int mnt_flags; 357377b1a97dSEric W. Biederman 3574132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 3575e51db735SEric W. Biederman continue; 3576e51db735SEric W. Biederman 35777e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 35787e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 35797e96c1b0SEric W. Biederman */ 35807e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 35817e96c1b0SEric W. Biederman continue; 35827e96c1b0SEric W. Biederman 3583a1935c17SEric W. Biederman /* A local view of the mount flags */ 358477b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 358577b1a97dSEric W. Biederman 3586695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 3587bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 3588695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 3589695e9df0SEric W. Biederman 35908c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 35918c6cf9ccSEric W. Biederman * than the proposed new mount. 35928c6cf9ccSEric W. Biederman */ 359377b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 35948c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 35958c6cf9ccSEric W. Biederman continue; 359677b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 359777b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 35988c6cf9ccSEric W. Biederman continue; 35998c6cf9ccSEric W. Biederman 3600ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 3601ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 3602ceeb0e5dSEric W. Biederman * empty directories. 3603e51db735SEric W. Biederman */ 3604e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 3605e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 3606ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 3607d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 3608ceeb0e5dSEric W. Biederman continue; 36097236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 36107236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 3611e51db735SEric W. Biederman goto next; 361287a8ebd6SEric W. Biederman } 36138c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 361477b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 36158c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 3616e51db735SEric W. Biederman visible = true; 3617e51db735SEric W. Biederman goto found; 3618e51db735SEric W. Biederman next: ; 361987a8ebd6SEric W. Biederman } 3620e51db735SEric W. Biederman found: 362144bb4385SAl Viro up_read(&namespace_sem); 3622e51db735SEric W. Biederman return visible; 362387a8ebd6SEric W. Biederman } 362487a8ebd6SEric W. Biederman 3625132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 36268654df4eSEric W. Biederman { 3627a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 36288654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 36298654df4eSEric W. Biederman unsigned long s_iflags; 36308654df4eSEric W. Biederman 36318654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 36328654df4eSEric W. Biederman return false; 36338654df4eSEric W. Biederman 36348654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 3635132e4608SDavid Howells s_iflags = sb->s_iflags; 36368654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 36378654df4eSEric W. Biederman return false; 36388654df4eSEric W. Biederman 3639a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 3640a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 3641a1935c17SEric W. Biederman required_iflags); 3642a1935c17SEric W. Biederman return true; 3643a1935c17SEric W. Biederman } 3644a1935c17SEric W. Biederman 3645132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 36468654df4eSEric W. Biederman } 36478654df4eSEric W. Biederman 3648380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 3649380cf5baSAndy Lutomirski { 3650380cf5baSAndy Lutomirski /* 3651380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 3652380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 3653380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 3654380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 3655380cf5baSAndy Lutomirski * in other namespaces. 3656380cf5baSAndy Lutomirski */ 3657380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 3658380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 3659380cf5baSAndy Lutomirski } 3660380cf5baSAndy Lutomirski 366164964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 36628823c079SEric W. Biederman { 366358be2825SAl Viro struct ns_common *ns = NULL; 36648823c079SEric W. Biederman struct nsproxy *nsproxy; 36658823c079SEric W. Biederman 3666728dba3aSEric W. Biederman task_lock(task); 3667728dba3aSEric W. Biederman nsproxy = task->nsproxy; 36688823c079SEric W. Biederman if (nsproxy) { 366958be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 367058be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 36718823c079SEric W. Biederman } 3672728dba3aSEric W. Biederman task_unlock(task); 36738823c079SEric W. Biederman 36748823c079SEric W. Biederman return ns; 36758823c079SEric W. Biederman } 36768823c079SEric W. Biederman 367764964528SAl Viro static void mntns_put(struct ns_common *ns) 36788823c079SEric W. Biederman { 367958be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 36808823c079SEric W. Biederman } 36818823c079SEric W. Biederman 368264964528SAl Viro static int mntns_install(struct nsproxy *nsproxy, struct ns_common *ns) 36838823c079SEric W. Biederman { 36848823c079SEric W. Biederman struct fs_struct *fs = current->fs; 36854f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 36868823c079SEric W. Biederman struct path root; 36874f757f3cSAl Viro int err; 36888823c079SEric W. Biederman 36890c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 3690c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 3691c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 3692ae11e0f1SZhao Hongjiang return -EPERM; 36938823c079SEric W. Biederman 369474e83122SAl Viro if (is_anon_ns(mnt_ns)) 369574e83122SAl Viro return -EINVAL; 369674e83122SAl Viro 36978823c079SEric W. Biederman if (fs->users != 1) 36988823c079SEric W. Biederman return -EINVAL; 36998823c079SEric W. Biederman 37008823c079SEric W. Biederman get_mnt_ns(mnt_ns); 37014f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 37028823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 37038823c079SEric W. Biederman 37048823c079SEric W. Biederman /* Find the root */ 37054f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 37064f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 37074f757f3cSAl Viro if (err) { 37084f757f3cSAl Viro /* revert to old namespace */ 37094f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 37104f757f3cSAl Viro put_mnt_ns(mnt_ns); 37114f757f3cSAl Viro return err; 37124f757f3cSAl Viro } 37138823c079SEric W. Biederman 37144068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 37154068367cSAndrei Vagin 37168823c079SEric W. Biederman /* Update the pwd and root */ 37178823c079SEric W. Biederman set_fs_pwd(fs, &root); 37188823c079SEric W. Biederman set_fs_root(fs, &root); 37198823c079SEric W. Biederman 37208823c079SEric W. Biederman path_put(&root); 37218823c079SEric W. Biederman return 0; 37228823c079SEric W. Biederman } 37238823c079SEric W. Biederman 3724bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 3725bcac25a5SAndrey Vagin { 3726bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 3727bcac25a5SAndrey Vagin } 3728bcac25a5SAndrey Vagin 37298823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 37308823c079SEric W. Biederman .name = "mnt", 37318823c079SEric W. Biederman .type = CLONE_NEWNS, 37328823c079SEric W. Biederman .get = mntns_get, 37338823c079SEric W. Biederman .put = mntns_put, 37348823c079SEric W. Biederman .install = mntns_install, 3735bcac25a5SAndrey Vagin .owner = mntns_owner, 37368823c079SEric W. Biederman }; 3737