159bd9dedSThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * linux/fs/namespace.c 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 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> 289caccd41SChristian Brauner #include <linux/proc_fs.h> 299ea459e1SAl Viro #include <linux/task_work.h> 309164bb4aSIngo Molnar #include <linux/sched/task.h> 31e262e32dSDavid Howells #include <uapi/linux/mount.h> 329bc61ab1SDavid Howells #include <linux/fs_context.h> 33037f11b4SAl Viro #include <linux/shmem_fs.h> 34bd303368SChristian Brauner #include <linux/mnt_idmapping.h> 359164bb4aSIngo Molnar 3607b20889SRam Pai #include "pnode.h" 37948730b0SAdrian Bunk #include "internal.h" 381da177e4SLinus Torvalds 39d2921684SEric W. Biederman /* Maximum number of mounts in a mount namespace */ 40ab171b95SLuis Chamberlain static unsigned int sysctl_mount_max __read_mostly = 100000; 41d2921684SEric W. Biederman 420818bf27SAl Viro static unsigned int m_hash_mask __read_mostly; 430818bf27SAl Viro static unsigned int m_hash_shift __read_mostly; 440818bf27SAl Viro static unsigned int mp_hash_mask __read_mostly; 450818bf27SAl Viro static unsigned int mp_hash_shift __read_mostly; 460818bf27SAl Viro 470818bf27SAl Viro static __initdata unsigned long mhash_entries; 480818bf27SAl Viro static int __init set_mhash_entries(char *str) 490818bf27SAl Viro { 500818bf27SAl Viro if (!str) 510818bf27SAl Viro return 0; 520818bf27SAl Viro mhash_entries = simple_strtoul(str, &str, 0); 530818bf27SAl Viro return 1; 540818bf27SAl Viro } 550818bf27SAl Viro __setup("mhash_entries=", set_mhash_entries); 560818bf27SAl Viro 570818bf27SAl Viro static __initdata unsigned long mphash_entries; 580818bf27SAl Viro static int __init set_mphash_entries(char *str) 590818bf27SAl Viro { 600818bf27SAl Viro if (!str) 610818bf27SAl Viro return 0; 620818bf27SAl Viro mphash_entries = simple_strtoul(str, &str, 0); 630818bf27SAl Viro return 1; 640818bf27SAl Viro } 650818bf27SAl Viro __setup("mphash_entries=", set_mphash_entries); 6613f14b4dSEric Dumazet 67c7999c36SAl Viro static u64 event; 6873cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 69719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 705addc5ddSAl Viro 7138129a13SAl Viro static struct hlist_head *mount_hashtable __read_mostly; 720818bf27SAl Viro static struct hlist_head *mountpoint_hashtable __read_mostly; 73e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 7459aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 754edbe133SAl Viro static HLIST_HEAD(unmounted); /* protected by namespace_sem */ 764edbe133SAl Viro static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */ 771da177e4SLinus Torvalds 78256c8aedSChristian Brauner struct mnt_idmap { 79256c8aedSChristian Brauner struct user_namespace *owner; 80256c8aedSChristian Brauner refcount_t count; 81256c8aedSChristian Brauner }; 82256c8aedSChristian Brauner 83256c8aedSChristian Brauner /* 84256c8aedSChristian Brauner * Carries the initial idmapping of 0:0:4294967295 which is an identity 85256c8aedSChristian Brauner * mapping. This means that {g,u}id 0 is mapped to {g,u}id 0, {g,u}id 1 is 86256c8aedSChristian Brauner * mapped to {g,u}id 1, [...], {g,u}id 1000 to {g,u}id 1000, [...]. 87256c8aedSChristian Brauner */ 88256c8aedSChristian Brauner struct mnt_idmap nop_mnt_idmap = { 89256c8aedSChristian Brauner .owner = &init_user_ns, 90256c8aedSChristian Brauner .count = REFCOUNT_INIT(1), 91256c8aedSChristian Brauner }; 92256c8aedSChristian Brauner EXPORT_SYMBOL_GPL(nop_mnt_idmap); 93256c8aedSChristian Brauner 942a186721SChristian Brauner struct mount_kattr { 952a186721SChristian Brauner unsigned int attr_set; 962a186721SChristian Brauner unsigned int attr_clr; 972a186721SChristian Brauner unsigned int propagation; 982a186721SChristian Brauner unsigned int lookup_flags; 992a186721SChristian Brauner bool recurse; 1009caccd41SChristian Brauner struct user_namespace *mnt_userns; 101256c8aedSChristian Brauner struct mnt_idmap *mnt_idmap; 1022a186721SChristian Brauner }; 1032a186721SChristian Brauner 104f87fd4c2SMiklos Szeredi /* /sys/fs */ 10500d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 10600d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 107f87fd4c2SMiklos Szeredi 10899b7db7bSNick Piggin /* 10999b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 11099b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 11199b7db7bSNick Piggin * up the tree. 11299b7db7bSNick Piggin * 11399b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 11499b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 11599b7db7bSNick Piggin */ 11648a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 11799b7db7bSNick Piggin 118d033cb67SChristian Brauner static inline void lock_mount_hash(void) 119d033cb67SChristian Brauner { 120d033cb67SChristian Brauner write_seqlock(&mount_lock); 121d033cb67SChristian Brauner } 122d033cb67SChristian Brauner 123d033cb67SChristian Brauner static inline void unlock_mount_hash(void) 124d033cb67SChristian Brauner { 125d033cb67SChristian Brauner write_sequnlock(&mount_lock); 126d033cb67SChristian Brauner } 127d033cb67SChristian Brauner 12838129a13SAl Viro static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry) 1291da177e4SLinus Torvalds { 1301da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 1311da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 1320818bf27SAl Viro tmp = tmp + (tmp >> m_hash_shift); 1330818bf27SAl Viro return &mount_hashtable[tmp & m_hash_mask]; 1340818bf27SAl Viro } 1350818bf27SAl Viro 1360818bf27SAl Viro static inline struct hlist_head *mp_hash(struct dentry *dentry) 1370818bf27SAl Viro { 1380818bf27SAl Viro unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES); 1390818bf27SAl Viro tmp = tmp + (tmp >> mp_hash_shift); 1400818bf27SAl Viro return &mountpoint_hashtable[tmp & mp_hash_mask]; 1411da177e4SLinus Torvalds } 1421da177e4SLinus Torvalds 143b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 14473cd49ecSMiklos Szeredi { 145169b480eSMatthew Wilcox int res = ida_alloc(&mnt_id_ida, GFP_KERNEL); 14673cd49ecSMiklos Szeredi 147169b480eSMatthew Wilcox if (res < 0) 14873cd49ecSMiklos Szeredi return res; 149169b480eSMatthew Wilcox mnt->mnt_id = res; 150169b480eSMatthew Wilcox return 0; 15173cd49ecSMiklos Szeredi } 15273cd49ecSMiklos Szeredi 153b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 15473cd49ecSMiklos Szeredi { 155169b480eSMatthew Wilcox ida_free(&mnt_id_ida, mnt->mnt_id); 15673cd49ecSMiklos Szeredi } 15773cd49ecSMiklos Szeredi 158719f5d7fSMiklos Szeredi /* 159719f5d7fSMiklos Szeredi * Allocate a new peer group ID 160719f5d7fSMiklos Szeredi */ 1614b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 162719f5d7fSMiklos Szeredi { 163169b480eSMatthew Wilcox int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL); 164f21f6220SAl Viro 165169b480eSMatthew Wilcox if (res < 0) 166f21f6220SAl Viro return res; 167169b480eSMatthew Wilcox mnt->mnt_group_id = res; 168169b480eSMatthew Wilcox return 0; 169719f5d7fSMiklos Szeredi } 170719f5d7fSMiklos Szeredi 171719f5d7fSMiklos Szeredi /* 172719f5d7fSMiklos Szeredi * Release a peer group ID 173719f5d7fSMiklos Szeredi */ 1744b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 175719f5d7fSMiklos Szeredi { 176169b480eSMatthew Wilcox ida_free(&mnt_group_ida, mnt->mnt_group_id); 17715169fe7SAl Viro mnt->mnt_group_id = 0; 178719f5d7fSMiklos Szeredi } 179719f5d7fSMiklos Szeredi 180b3e19d92SNick Piggin /* 181b3e19d92SNick Piggin * vfsmount lock must be held for read 182b3e19d92SNick Piggin */ 18383adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 184b3e19d92SNick Piggin { 185b3e19d92SNick Piggin #ifdef CONFIG_SMP 18668e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 187b3e19d92SNick Piggin #else 188b3e19d92SNick Piggin preempt_disable(); 18968e8a9feSAl Viro mnt->mnt_count += n; 190b3e19d92SNick Piggin preempt_enable(); 191b3e19d92SNick Piggin #endif 192b3e19d92SNick Piggin } 193b3e19d92SNick Piggin 194b3e19d92SNick Piggin /* 195b3e19d92SNick Piggin * vfsmount lock must be held for write 196b3e19d92SNick Piggin */ 197edf7ddbfSEric Biggers int mnt_get_count(struct mount *mnt) 198b3e19d92SNick Piggin { 199b3e19d92SNick Piggin #ifdef CONFIG_SMP 200edf7ddbfSEric Biggers int count = 0; 201b3e19d92SNick Piggin int cpu; 202b3e19d92SNick Piggin 203b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 20468e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 205b3e19d92SNick Piggin } 206b3e19d92SNick Piggin 207b3e19d92SNick Piggin return count; 208b3e19d92SNick Piggin #else 20968e8a9feSAl Viro return mnt->mnt_count; 210b3e19d92SNick Piggin #endif 211b3e19d92SNick Piggin } 212b3e19d92SNick Piggin 213256c8aedSChristian Brauner /** 214256c8aedSChristian Brauner * mnt_idmap_owner - retrieve owner of the mount's idmapping 215256c8aedSChristian Brauner * @idmap: mount idmapping 216256c8aedSChristian Brauner * 217256c8aedSChristian Brauner * This helper will go away once the conversion to use struct mnt_idmap 218256c8aedSChristian Brauner * everywhere has finished at which point the helper will be unexported. 219256c8aedSChristian Brauner * 220256c8aedSChristian Brauner * Only code that needs to perform permission checks based on the owner of the 221256c8aedSChristian Brauner * idmapping will get access to it. All other code will solely rely on 222256c8aedSChristian Brauner * idmappings. This will get us type safety so it's impossible to conflate 223256c8aedSChristian Brauner * filesystems idmappings with mount idmappings. 224256c8aedSChristian Brauner * 225256c8aedSChristian Brauner * Return: The owner of the idmapping. 226256c8aedSChristian Brauner */ 227256c8aedSChristian Brauner struct user_namespace *mnt_idmap_owner(const struct mnt_idmap *idmap) 228256c8aedSChristian Brauner { 229256c8aedSChristian Brauner return idmap->owner; 230256c8aedSChristian Brauner } 231256c8aedSChristian Brauner EXPORT_SYMBOL_GPL(mnt_idmap_owner); 232256c8aedSChristian Brauner 233256c8aedSChristian Brauner /** 234256c8aedSChristian Brauner * mnt_user_ns - retrieve owner of an idmapped mount 235256c8aedSChristian Brauner * @mnt: the relevant vfsmount 236256c8aedSChristian Brauner * 237256c8aedSChristian Brauner * This helper will go away once the conversion to use struct mnt_idmap 238256c8aedSChristian Brauner * everywhere has finished at which point the helper will be unexported. 239256c8aedSChristian Brauner * 240256c8aedSChristian Brauner * Only code that needs to perform permission checks based on the owner of the 241256c8aedSChristian Brauner * idmapping will get access to it. All other code will solely rely on 242256c8aedSChristian Brauner * idmappings. This will get us type safety so it's impossible to conflate 243256c8aedSChristian Brauner * filesystems idmappings with mount idmappings. 244256c8aedSChristian Brauner * 245256c8aedSChristian Brauner * Return: The owner of the idmapped. 246256c8aedSChristian Brauner */ 247256c8aedSChristian Brauner struct user_namespace *mnt_user_ns(const struct vfsmount *mnt) 248256c8aedSChristian Brauner { 249256c8aedSChristian Brauner struct mnt_idmap *idmap = mnt_idmap(mnt); 250256c8aedSChristian Brauner 251256c8aedSChristian Brauner /* Return the actual owner of the filesystem instead of the nop. */ 252256c8aedSChristian Brauner if (idmap == &nop_mnt_idmap && 253256c8aedSChristian Brauner !initial_idmapping(mnt->mnt_sb->s_user_ns)) 254256c8aedSChristian Brauner return mnt->mnt_sb->s_user_ns; 255256c8aedSChristian Brauner return mnt_idmap_owner(idmap); 256256c8aedSChristian Brauner } 257256c8aedSChristian Brauner EXPORT_SYMBOL_GPL(mnt_user_ns); 258256c8aedSChristian Brauner 259256c8aedSChristian Brauner /** 260256c8aedSChristian Brauner * alloc_mnt_idmap - allocate a new idmapping for the mount 261256c8aedSChristian Brauner * @mnt_userns: owning userns of the idmapping 262256c8aedSChristian Brauner * 263256c8aedSChristian Brauner * Allocate a new struct mnt_idmap which carries the idmapping of the mount. 264256c8aedSChristian Brauner * 265256c8aedSChristian Brauner * Return: On success a new idmap, on error an error pointer is returned. 266256c8aedSChristian Brauner */ 267256c8aedSChristian Brauner static struct mnt_idmap *alloc_mnt_idmap(struct user_namespace *mnt_userns) 268256c8aedSChristian Brauner { 269256c8aedSChristian Brauner struct mnt_idmap *idmap; 270256c8aedSChristian Brauner 271256c8aedSChristian Brauner idmap = kzalloc(sizeof(struct mnt_idmap), GFP_KERNEL_ACCOUNT); 272256c8aedSChristian Brauner if (!idmap) 273256c8aedSChristian Brauner return ERR_PTR(-ENOMEM); 274256c8aedSChristian Brauner 275256c8aedSChristian Brauner idmap->owner = get_user_ns(mnt_userns); 276256c8aedSChristian Brauner refcount_set(&idmap->count, 1); 277256c8aedSChristian Brauner return idmap; 278256c8aedSChristian Brauner } 279256c8aedSChristian Brauner 280256c8aedSChristian Brauner /** 281256c8aedSChristian Brauner * mnt_idmap_get - get a reference to an idmapping 282256c8aedSChristian Brauner * @idmap: the idmap to bump the reference on 283256c8aedSChristian Brauner * 284256c8aedSChristian Brauner * If @idmap is not the @nop_mnt_idmap bump the reference count. 285256c8aedSChristian Brauner * 286256c8aedSChristian Brauner * Return: @idmap with reference count bumped if @not_mnt_idmap isn't passed. 287256c8aedSChristian Brauner */ 288256c8aedSChristian Brauner static inline struct mnt_idmap *mnt_idmap_get(struct mnt_idmap *idmap) 289256c8aedSChristian Brauner { 290256c8aedSChristian Brauner if (idmap != &nop_mnt_idmap) 291256c8aedSChristian Brauner refcount_inc(&idmap->count); 292256c8aedSChristian Brauner 293256c8aedSChristian Brauner return idmap; 294256c8aedSChristian Brauner } 295256c8aedSChristian Brauner 296256c8aedSChristian Brauner /** 297256c8aedSChristian Brauner * mnt_idmap_put - put a reference to an idmapping 298256c8aedSChristian Brauner * @idmap: the idmap to put the reference on 299256c8aedSChristian Brauner * 300256c8aedSChristian Brauner * If this is a non-initial idmapping, put the reference count when a mount is 301256c8aedSChristian Brauner * released and free it if we're the last user. 302256c8aedSChristian Brauner */ 303256c8aedSChristian Brauner static inline void mnt_idmap_put(struct mnt_idmap *idmap) 304256c8aedSChristian Brauner { 305256c8aedSChristian Brauner if (idmap != &nop_mnt_idmap && refcount_dec_and_test(&idmap->count)) { 306256c8aedSChristian Brauner put_user_ns(idmap->owner); 307256c8aedSChristian Brauner kfree(idmap); 308256c8aedSChristian Brauner } 309256c8aedSChristian Brauner } 310256c8aedSChristian Brauner 311b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 3121da177e4SLinus Torvalds { 313c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 314c63181e6SAl Viro if (mnt) { 31573cd49ecSMiklos Szeredi int err; 31673cd49ecSMiklos Szeredi 317c63181e6SAl Viro err = mnt_alloc_id(mnt); 31888b38782SLi Zefan if (err) 31988b38782SLi Zefan goto out_free_cache; 32088b38782SLi Zefan 32188b38782SLi Zefan if (name) { 32279f6540bSVasily Averin mnt->mnt_devname = kstrdup_const(name, 32379f6540bSVasily Averin GFP_KERNEL_ACCOUNT); 324c63181e6SAl Viro if (!mnt->mnt_devname) 32588b38782SLi Zefan goto out_free_id; 32673cd49ecSMiklos Szeredi } 32773cd49ecSMiklos Szeredi 328b3e19d92SNick Piggin #ifdef CONFIG_SMP 329c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 330c63181e6SAl Viro if (!mnt->mnt_pcp) 331b3e19d92SNick Piggin goto out_free_devname; 332b3e19d92SNick Piggin 333c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 334b3e19d92SNick Piggin #else 335c63181e6SAl Viro mnt->mnt_count = 1; 336c63181e6SAl Viro mnt->mnt_writers = 0; 337b3e19d92SNick Piggin #endif 338b3e19d92SNick Piggin 33938129a13SAl Viro INIT_HLIST_NODE(&mnt->mnt_hash); 340c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 341c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 342c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 343c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 344c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 345c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 346c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 3470a5eb7c8SEric W. Biederman INIT_HLIST_NODE(&mnt->mnt_mp_list); 34899b19d16SEric W. Biederman INIT_LIST_HEAD(&mnt->mnt_umounting); 34956cbb429SAl Viro INIT_HLIST_HEAD(&mnt->mnt_stuck_children); 350256c8aedSChristian Brauner mnt->mnt.mnt_idmap = &nop_mnt_idmap; 3511da177e4SLinus Torvalds } 352c63181e6SAl Viro return mnt; 35388b38782SLi Zefan 354d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 355d3ef3d73Snpiggin@suse.de out_free_devname: 356fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 357d3ef3d73Snpiggin@suse.de #endif 35888b38782SLi Zefan out_free_id: 359c63181e6SAl Viro mnt_free_id(mnt); 36088b38782SLi Zefan out_free_cache: 361c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 36288b38782SLi Zefan return NULL; 3631da177e4SLinus Torvalds } 3641da177e4SLinus Torvalds 3658366025eSDave Hansen /* 3668366025eSDave Hansen * Most r/o checks on a fs are for operations that take 3678366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 3688366025eSDave Hansen * We must keep track of when those operations start 3698366025eSDave Hansen * (for permission checks) and when they end, so that 3708366025eSDave Hansen * we can determine when writes are able to occur to 3718366025eSDave Hansen * a filesystem. 3728366025eSDave Hansen */ 3733d733633SDave Hansen /* 3743d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 3753d733633SDave Hansen * @mnt: the mount to check for its write status 3763d733633SDave Hansen * 3773d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 3783d733633SDave Hansen * It does not guarantee that the filesystem will stay 3793d733633SDave Hansen * r/w, just that it is right *now*. This can not and 3803d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 3813d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 3823d733633SDave Hansen * r/w. 3833d733633SDave Hansen */ 38443f5e655SDavid Howells bool __mnt_is_readonly(struct vfsmount *mnt) 3853d733633SDave Hansen { 38643f5e655SDavid Howells return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb); 3873d733633SDave Hansen } 3883d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 3893d733633SDave Hansen 39083adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 3913d733633SDave Hansen { 392d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 39368e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 394d3ef3d73Snpiggin@suse.de #else 39568e8a9feSAl Viro mnt->mnt_writers++; 396d3ef3d73Snpiggin@suse.de #endif 3973d733633SDave Hansen } 3983d733633SDave Hansen 39983adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 4003d733633SDave Hansen { 401d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 40268e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 403d3ef3d73Snpiggin@suse.de #else 40468e8a9feSAl Viro mnt->mnt_writers--; 405d3ef3d73Snpiggin@suse.de #endif 406d3ef3d73Snpiggin@suse.de } 407d3ef3d73Snpiggin@suse.de 40883adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 409d3ef3d73Snpiggin@suse.de { 410d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 411d3ef3d73Snpiggin@suse.de unsigned int count = 0; 4123d733633SDave Hansen int cpu; 4133d733633SDave Hansen 4143d733633SDave Hansen for_each_possible_cpu(cpu) { 41568e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 4163d733633SDave Hansen } 4173d733633SDave Hansen 418d3ef3d73Snpiggin@suse.de return count; 419d3ef3d73Snpiggin@suse.de #else 420d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 421d3ef3d73Snpiggin@suse.de #endif 4223d733633SDave Hansen } 4233d733633SDave Hansen 4244ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 4254ed5e82fSMiklos Szeredi { 4264ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 4274ed5e82fSMiklos Szeredi return 1; 4284ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 4294ed5e82fSMiklos Szeredi smp_rmb(); 4304ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 4314ed5e82fSMiklos Szeredi } 4324ed5e82fSMiklos Szeredi 4333d733633SDave Hansen /* 434eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 435eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 436eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 437eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 4383d733633SDave Hansen */ 4398366025eSDave Hansen /** 440eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 44183adc753SAl Viro * @m: the mount on which to take a write 4428366025eSDave Hansen * 443eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 444eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 445eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 446eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 447eb04c282SJan Kara * called. This is effectively a refcount. 4488366025eSDave Hansen */ 449eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 4508366025eSDave Hansen { 45183adc753SAl Viro struct mount *mnt = real_mount(m); 4523d733633SDave Hansen int ret = 0; 4533d733633SDave Hansen 454d3ef3d73Snpiggin@suse.de preempt_disable(); 455c6653a83SNick Piggin mnt_inc_writers(mnt); 456d3ef3d73Snpiggin@suse.de /* 457c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 458d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 459d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 460d3ef3d73Snpiggin@suse.de */ 461d3ef3d73Snpiggin@suse.de smp_mb(); 4620f8821daSSebastian Andrzej Siewior might_lock(&mount_lock.lock); 4630f8821daSSebastian Andrzej Siewior while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) { 4640f8821daSSebastian Andrzej Siewior if (!IS_ENABLED(CONFIG_PREEMPT_RT)) { 465d3ef3d73Snpiggin@suse.de cpu_relax(); 4660f8821daSSebastian Andrzej Siewior } else { 4670f8821daSSebastian Andrzej Siewior /* 4680f8821daSSebastian Andrzej Siewior * This prevents priority inversion, if the task 4690f8821daSSebastian Andrzej Siewior * setting MNT_WRITE_HOLD got preempted on a remote 4700f8821daSSebastian Andrzej Siewior * CPU, and it prevents life lock if the task setting 4710f8821daSSebastian Andrzej Siewior * MNT_WRITE_HOLD has a lower priority and is bound to 4720f8821daSSebastian Andrzej Siewior * the same CPU as the task that is spinning here. 4730f8821daSSebastian Andrzej Siewior */ 4740f8821daSSebastian Andrzej Siewior preempt_enable(); 4750f8821daSSebastian Andrzej Siewior lock_mount_hash(); 4760f8821daSSebastian Andrzej Siewior unlock_mount_hash(); 4770f8821daSSebastian Andrzej Siewior preempt_disable(); 4780f8821daSSebastian Andrzej Siewior } 4790f8821daSSebastian Andrzej Siewior } 480d3ef3d73Snpiggin@suse.de /* 481d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 482d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 483d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 484d3ef3d73Snpiggin@suse.de */ 485d3ef3d73Snpiggin@suse.de smp_rmb(); 4864ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 487c6653a83SNick Piggin mnt_dec_writers(mnt); 4883d733633SDave Hansen ret = -EROFS; 4893d733633SDave Hansen } 490d3ef3d73Snpiggin@suse.de preempt_enable(); 491eb04c282SJan Kara 492eb04c282SJan Kara return ret; 493eb04c282SJan Kara } 494eb04c282SJan Kara 495eb04c282SJan Kara /** 496eb04c282SJan Kara * mnt_want_write - get write access to a mount 497eb04c282SJan Kara * @m: the mount on which to take a write 498eb04c282SJan Kara * 499eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 500eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 501eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 502eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 503eb04c282SJan Kara */ 504eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 505eb04c282SJan Kara { 506eb04c282SJan Kara int ret; 507eb04c282SJan Kara 508eb04c282SJan Kara sb_start_write(m->mnt_sb); 509eb04c282SJan Kara ret = __mnt_want_write(m); 510eb04c282SJan Kara if (ret) 511eb04c282SJan Kara sb_end_write(m->mnt_sb); 5123d733633SDave Hansen return ret; 5138366025eSDave Hansen } 5148366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 5158366025eSDave Hansen 5168366025eSDave Hansen /** 517eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 518eb04c282SJan Kara * @file: the file who's mount on which to take a write 519eb04c282SJan Kara * 52014e43bf4SEric Biggers * This is like __mnt_want_write, but if the file is already open for writing it 52114e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 52214e43bf4SEric Biggers * and instead only does the check for emergency r/o remounts. This must be 52314e43bf4SEric Biggers * paired with __mnt_drop_write_file. 524eb04c282SJan Kara */ 525eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 526eb04c282SJan Kara { 52714e43bf4SEric Biggers if (file->f_mode & FMODE_WRITER) { 52814e43bf4SEric Biggers /* 52914e43bf4SEric Biggers * Superblock may have become readonly while there are still 53014e43bf4SEric Biggers * writable fd's, e.g. due to a fs error with errors=remount-ro 53114e43bf4SEric Biggers */ 53214e43bf4SEric Biggers if (__mnt_is_readonly(file->f_path.mnt)) 53314e43bf4SEric Biggers return -EROFS; 53414e43bf4SEric Biggers return 0; 53514e43bf4SEric Biggers } 536eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 537eb04c282SJan Kara } 538eb04c282SJan Kara 539eb04c282SJan Kara /** 5407c6893e3SMiklos Szeredi * mnt_want_write_file - get write access to a file's mount 5417c6893e3SMiklos Szeredi * @file: the file who's mount on which to take a write 5427c6893e3SMiklos Szeredi * 54314e43bf4SEric Biggers * This is like mnt_want_write, but if the file is already open for writing it 54414e43bf4SEric Biggers * skips incrementing mnt_writers (since the open file already has a reference) 54514e43bf4SEric Biggers * and instead only does the freeze protection and the check for emergency r/o 54614e43bf4SEric Biggers * remounts. This must be paired with mnt_drop_write_file. 5477c6893e3SMiklos Szeredi */ 5487c6893e3SMiklos Szeredi int mnt_want_write_file(struct file *file) 5497c6893e3SMiklos Szeredi { 5507c6893e3SMiklos Szeredi int ret; 5517c6893e3SMiklos Szeredi 5527c6893e3SMiklos Szeredi sb_start_write(file_inode(file)->i_sb); 5537c6893e3SMiklos Szeredi ret = __mnt_want_write_file(file); 5547c6893e3SMiklos Szeredi if (ret) 5557c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 5567c6893e3SMiklos Szeredi return ret; 5577c6893e3SMiklos Szeredi } 55896029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 55996029c4eSnpiggin@suse.de 56096029c4eSnpiggin@suse.de /** 561eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 5628366025eSDave Hansen * @mnt: the mount on which to give up write access 5638366025eSDave Hansen * 5648366025eSDave Hansen * Tells the low-level filesystem that we are done 5658366025eSDave Hansen * performing writes to it. Must be matched with 566eb04c282SJan Kara * __mnt_want_write() call above. 5678366025eSDave Hansen */ 568eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 5698366025eSDave Hansen { 570d3ef3d73Snpiggin@suse.de preempt_disable(); 57183adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 572d3ef3d73Snpiggin@suse.de preempt_enable(); 5738366025eSDave Hansen } 574eb04c282SJan Kara 575eb04c282SJan Kara /** 576eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 577eb04c282SJan Kara * @mnt: the mount on which to give up write access 578eb04c282SJan Kara * 579eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 580eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 581eb04c282SJan Kara * mnt_want_write() call above. 582eb04c282SJan Kara */ 583eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 584eb04c282SJan Kara { 585eb04c282SJan Kara __mnt_drop_write(mnt); 586eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 587eb04c282SJan Kara } 5888366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 5898366025eSDave Hansen 590eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 591eb04c282SJan Kara { 59214e43bf4SEric Biggers if (!(file->f_mode & FMODE_WRITER)) 593eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 594eb04c282SJan Kara } 595eb04c282SJan Kara 5967c6893e3SMiklos Szeredi void mnt_drop_write_file(struct file *file) 5977c6893e3SMiklos Szeredi { 598a6795a58SMiklos Szeredi __mnt_drop_write_file(file); 5997c6893e3SMiklos Szeredi sb_end_write(file_inode(file)->i_sb); 6007c6893e3SMiklos Szeredi } 6012a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 6022a79f17eSAl Viro 603538f4f02SChristian Brauner /** 604538f4f02SChristian Brauner * mnt_hold_writers - prevent write access to the given mount 605538f4f02SChristian Brauner * @mnt: mnt to prevent write access to 606538f4f02SChristian Brauner * 607538f4f02SChristian Brauner * Prevents write access to @mnt if there are no active writers for @mnt. 608538f4f02SChristian Brauner * This function needs to be called and return successfully before changing 609538f4f02SChristian Brauner * properties of @mnt that need to remain stable for callers with write access 610538f4f02SChristian Brauner * to @mnt. 611538f4f02SChristian Brauner * 612538f4f02SChristian Brauner * After this functions has been called successfully callers must pair it with 613538f4f02SChristian Brauner * a call to mnt_unhold_writers() in order to stop preventing write access to 614538f4f02SChristian Brauner * @mnt. 615538f4f02SChristian Brauner * 616538f4f02SChristian Brauner * Context: This function expects lock_mount_hash() to be held serializing 617538f4f02SChristian Brauner * setting MNT_WRITE_HOLD. 618538f4f02SChristian Brauner * Return: On success 0 is returned. 619538f4f02SChristian Brauner * On error, -EBUSY is returned. 620538f4f02SChristian Brauner */ 621fbdc2f6cSChristian Brauner static inline int mnt_hold_writers(struct mount *mnt) 6228366025eSDave Hansen { 62383adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 624d3ef3d73Snpiggin@suse.de /* 625d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 626d3ef3d73Snpiggin@suse.de * should be visible before we do. 627d3ef3d73Snpiggin@suse.de */ 628d3ef3d73Snpiggin@suse.de smp_mb(); 629d3ef3d73Snpiggin@suse.de 630d3ef3d73Snpiggin@suse.de /* 631d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 632d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 633d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 634d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 635d3ef3d73Snpiggin@suse.de * 636d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 637d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 638d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 639d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 640d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 641d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 642d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 643d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 644d3ef3d73Snpiggin@suse.de * we're counting up here. 645d3ef3d73Snpiggin@suse.de */ 646c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 647fbdc2f6cSChristian Brauner return -EBUSY; 648fbdc2f6cSChristian Brauner 649fbdc2f6cSChristian Brauner return 0; 650fbdc2f6cSChristian Brauner } 651fbdc2f6cSChristian Brauner 652538f4f02SChristian Brauner /** 653538f4f02SChristian Brauner * mnt_unhold_writers - stop preventing write access to the given mount 654538f4f02SChristian Brauner * @mnt: mnt to stop preventing write access to 655538f4f02SChristian Brauner * 656538f4f02SChristian Brauner * Stop preventing write access to @mnt allowing callers to gain write access 657538f4f02SChristian Brauner * to @mnt again. 658538f4f02SChristian Brauner * 659538f4f02SChristian Brauner * This function can only be called after a successful call to 660538f4f02SChristian Brauner * mnt_hold_writers(). 661538f4f02SChristian Brauner * 662538f4f02SChristian Brauner * Context: This function expects lock_mount_hash() to be held. 663538f4f02SChristian Brauner */ 664fbdc2f6cSChristian Brauner static inline void mnt_unhold_writers(struct mount *mnt) 665fbdc2f6cSChristian Brauner { 666d3ef3d73Snpiggin@suse.de /* 667d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 668d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 669d3ef3d73Snpiggin@suse.de */ 670d3ef3d73Snpiggin@suse.de smp_wmb(); 67183adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 672fbdc2f6cSChristian Brauner } 673fbdc2f6cSChristian Brauner 674fbdc2f6cSChristian Brauner static int mnt_make_readonly(struct mount *mnt) 675fbdc2f6cSChristian Brauner { 676fbdc2f6cSChristian Brauner int ret; 677fbdc2f6cSChristian Brauner 678fbdc2f6cSChristian Brauner ret = mnt_hold_writers(mnt); 679fbdc2f6cSChristian Brauner if (!ret) 680fbdc2f6cSChristian Brauner mnt->mnt.mnt_flags |= MNT_READONLY; 681fbdc2f6cSChristian Brauner mnt_unhold_writers(mnt); 6823d733633SDave Hansen return ret; 6833d733633SDave Hansen } 6848366025eSDave Hansen 6854ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 6864ed5e82fSMiklos Szeredi { 6874ed5e82fSMiklos Szeredi struct mount *mnt; 6884ed5e82fSMiklos Szeredi int err = 0; 6894ed5e82fSMiklos Szeredi 6908e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 6918e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 6928e8b8796SMiklos Szeredi return -EBUSY; 6938e8b8796SMiklos Szeredi 694719ea2fbSAl Viro lock_mount_hash(); 6954ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 6964ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 697ad1844a0SChristian Brauner err = mnt_hold_writers(mnt); 698ad1844a0SChristian Brauner if (err) 6994ed5e82fSMiklos Szeredi break; 7004ed5e82fSMiklos Szeredi } 7014ed5e82fSMiklos Szeredi } 7028e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 7038e8b8796SMiklos Szeredi err = -EBUSY; 7048e8b8796SMiklos Szeredi 7054ed5e82fSMiklos Szeredi if (!err) { 7064ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 7074ed5e82fSMiklos Szeredi smp_wmb(); 7084ed5e82fSMiklos Szeredi } 7094ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 7104ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 7114ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 7124ed5e82fSMiklos Szeredi } 713719ea2fbSAl Viro unlock_mount_hash(); 7144ed5e82fSMiklos Szeredi 7154ed5e82fSMiklos Szeredi return err; 7164ed5e82fSMiklos Szeredi } 7174ed5e82fSMiklos Szeredi 718b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 7191da177e4SLinus Torvalds { 720256c8aedSChristian Brauner mnt_idmap_put(mnt_idmap(&mnt->mnt)); 721fcc139aeSAndrzej Hajda kfree_const(mnt->mnt_devname); 722d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 72368e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 724d3ef3d73Snpiggin@suse.de #endif 725b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 7261da177e4SLinus Torvalds } 7271da177e4SLinus Torvalds 7288ffcb32eSDavid Howells static void delayed_free_vfsmnt(struct rcu_head *head) 7298ffcb32eSDavid Howells { 7308ffcb32eSDavid Howells free_vfsmnt(container_of(head, struct mount, mnt_rcu)); 7318ffcb32eSDavid Howells } 7328ffcb32eSDavid Howells 73348a066e7SAl Viro /* call under rcu_read_lock */ 734294d71ffSAl Viro int __legitimize_mnt(struct vfsmount *bastard, unsigned seq) 73548a066e7SAl Viro { 73648a066e7SAl Viro struct mount *mnt; 73748a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 738294d71ffSAl Viro return 1; 73948a066e7SAl Viro if (bastard == NULL) 740294d71ffSAl Viro return 0; 74148a066e7SAl Viro mnt = real_mount(bastard); 74248a066e7SAl Viro mnt_add_count(mnt, 1); 743119e1ef8SAl Viro smp_mb(); // see mntput_no_expire() 74448a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 745294d71ffSAl Viro return 0; 74648a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 74748a066e7SAl Viro mnt_add_count(mnt, -1); 748294d71ffSAl Viro return 1; 74948a066e7SAl Viro } 750119e1ef8SAl Viro lock_mount_hash(); 751119e1ef8SAl Viro if (unlikely(bastard->mnt_flags & MNT_DOOMED)) { 752119e1ef8SAl Viro mnt_add_count(mnt, -1); 753119e1ef8SAl Viro unlock_mount_hash(); 754119e1ef8SAl Viro return 1; 755119e1ef8SAl Viro } 756119e1ef8SAl Viro unlock_mount_hash(); 757119e1ef8SAl Viro /* caller will mntput() */ 758294d71ffSAl Viro return -1; 759294d71ffSAl Viro } 760294d71ffSAl Viro 761294d71ffSAl Viro /* call under rcu_read_lock */ 7627e4745a0SAl Viro static bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 763294d71ffSAl Viro { 764294d71ffSAl Viro int res = __legitimize_mnt(bastard, seq); 765294d71ffSAl Viro if (likely(!res)) 766294d71ffSAl Viro return true; 767294d71ffSAl Viro if (unlikely(res < 0)) { 76848a066e7SAl Viro rcu_read_unlock(); 76948a066e7SAl Viro mntput(bastard); 77048a066e7SAl Viro rcu_read_lock(); 771294d71ffSAl Viro } 77248a066e7SAl Viro return false; 77348a066e7SAl Viro } 77448a066e7SAl Viro 7751da177e4SLinus Torvalds /* 776474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 77748a066e7SAl Viro * call under rcu_read_lock() 7781da177e4SLinus Torvalds */ 779474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 7801da177e4SLinus Torvalds { 78138129a13SAl Viro struct hlist_head *head = m_hash(mnt, dentry); 782474279dcSAl Viro struct mount *p; 7831da177e4SLinus Torvalds 78438129a13SAl Viro hlist_for_each_entry_rcu(p, head, mnt_hash) 785474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 786474279dcSAl Viro return p; 787474279dcSAl Viro return NULL; 7881da177e4SLinus Torvalds } 789474279dcSAl Viro 790474279dcSAl Viro /* 791f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 792f015f126SDavid Howells * 793f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 794f015f126SDavid Howells * following mounts: 795f015f126SDavid Howells * 796f015f126SDavid Howells * mount /dev/sda1 /mnt 797f015f126SDavid Howells * mount /dev/sda2 /mnt 798f015f126SDavid Howells * mount /dev/sda3 /mnt 799f015f126SDavid Howells * 800f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 801f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 802f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 803f015f126SDavid Howells * 804f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 805a05964f3SRam Pai */ 806ca71cf71SAl Viro struct vfsmount *lookup_mnt(const struct path *path) 807a05964f3SRam Pai { 808c7105365SAl Viro struct mount *child_mnt; 80948a066e7SAl Viro struct vfsmount *m; 81048a066e7SAl Viro unsigned seq; 81199b7db7bSNick Piggin 81248a066e7SAl Viro rcu_read_lock(); 81348a066e7SAl Viro do { 81448a066e7SAl Viro seq = read_seqbegin(&mount_lock); 815474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 81648a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 81748a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 81848a066e7SAl Viro rcu_read_unlock(); 81948a066e7SAl Viro return m; 820a05964f3SRam Pai } 821a05964f3SRam Pai 8229f6c61f9SMiklos Szeredi static inline void lock_ns_list(struct mnt_namespace *ns) 8239f6c61f9SMiklos Szeredi { 8249f6c61f9SMiklos Szeredi spin_lock(&ns->ns_lock); 8259f6c61f9SMiklos Szeredi } 8269f6c61f9SMiklos Szeredi 8279f6c61f9SMiklos Szeredi static inline void unlock_ns_list(struct mnt_namespace *ns) 8289f6c61f9SMiklos Szeredi { 8299f6c61f9SMiklos Szeredi spin_unlock(&ns->ns_lock); 8309f6c61f9SMiklos Szeredi } 8319f6c61f9SMiklos Szeredi 8329f6c61f9SMiklos Szeredi static inline bool mnt_is_cursor(struct mount *mnt) 8339f6c61f9SMiklos Szeredi { 8349f6c61f9SMiklos Szeredi return mnt->mnt.mnt_flags & MNT_CURSOR; 8359f6c61f9SMiklos Szeredi } 8369f6c61f9SMiklos Szeredi 8377af1364fSEric W. Biederman /* 8387af1364fSEric W. Biederman * __is_local_mountpoint - Test to see if dentry is a mountpoint in the 8397af1364fSEric W. Biederman * current mount namespace. 8407af1364fSEric W. Biederman * 8417af1364fSEric W. Biederman * The common case is dentries are not mountpoints at all and that 8427af1364fSEric W. Biederman * test is handled inline. For the slow case when we are actually 8437af1364fSEric W. Biederman * dealing with a mountpoint of some kind, walk through all of the 8447af1364fSEric W. Biederman * mounts in the current mount namespace and test to see if the dentry 8457af1364fSEric W. Biederman * is a mountpoint. 8467af1364fSEric W. Biederman * 8477af1364fSEric W. Biederman * The mount_hashtable is not usable in the context because we 8487af1364fSEric W. Biederman * need to identify all mounts that may be in the current mount 8497af1364fSEric W. Biederman * namespace not just a mount that happens to have some specified 8507af1364fSEric W. Biederman * parent mount. 8517af1364fSEric W. Biederman */ 8527af1364fSEric W. Biederman bool __is_local_mountpoint(struct dentry *dentry) 8537af1364fSEric W. Biederman { 8547af1364fSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 8557af1364fSEric W. Biederman struct mount *mnt; 8567af1364fSEric W. Biederman bool is_covered = false; 8577af1364fSEric W. Biederman 8587af1364fSEric W. Biederman down_read(&namespace_sem); 8599f6c61f9SMiklos Szeredi lock_ns_list(ns); 8607af1364fSEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 8619f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 8629f6c61f9SMiklos Szeredi continue; 8637af1364fSEric W. Biederman is_covered = (mnt->mnt_mountpoint == dentry); 8647af1364fSEric W. Biederman if (is_covered) 8657af1364fSEric W. Biederman break; 8667af1364fSEric W. Biederman } 8679f6c61f9SMiklos Szeredi unlock_ns_list(ns); 8687af1364fSEric W. Biederman up_read(&namespace_sem); 8695ad05cc8SNikolay Borisov 8707af1364fSEric W. Biederman return is_covered; 8717af1364fSEric W. Biederman } 8727af1364fSEric W. Biederman 873e2dfa935SEric W. Biederman static struct mountpoint *lookup_mountpoint(struct dentry *dentry) 87484d17192SAl Viro { 8750818bf27SAl Viro struct hlist_head *chain = mp_hash(dentry); 87684d17192SAl Viro struct mountpoint *mp; 87784d17192SAl Viro 8780818bf27SAl Viro hlist_for_each_entry(mp, chain, m_hash) { 87984d17192SAl Viro if (mp->m_dentry == dentry) { 88084d17192SAl Viro mp->m_count++; 88184d17192SAl Viro return mp; 88284d17192SAl Viro } 88384d17192SAl Viro } 884e2dfa935SEric W. Biederman return NULL; 885e2dfa935SEric W. Biederman } 886e2dfa935SEric W. Biederman 8873895dbf8SEric W. Biederman static struct mountpoint *get_mountpoint(struct dentry *dentry) 888e2dfa935SEric W. Biederman { 8893895dbf8SEric W. Biederman struct mountpoint *mp, *new = NULL; 890e2dfa935SEric W. Biederman int ret; 89184d17192SAl Viro 8923895dbf8SEric W. Biederman if (d_mountpoint(dentry)) { 8931e9c75fbSBenjamin Coddington /* might be worth a WARN_ON() */ 8941e9c75fbSBenjamin Coddington if (d_unlinked(dentry)) 8951e9c75fbSBenjamin Coddington return ERR_PTR(-ENOENT); 8963895dbf8SEric W. Biederman mountpoint: 8973895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 8983895dbf8SEric W. Biederman mp = lookup_mountpoint(dentry); 8993895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 9003895dbf8SEric W. Biederman if (mp) 9013895dbf8SEric W. Biederman goto done; 90284d17192SAl Viro } 903eed81007SMiklos Szeredi 9043895dbf8SEric W. Biederman if (!new) 9053895dbf8SEric W. Biederman new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 9063895dbf8SEric W. Biederman if (!new) 9073895dbf8SEric W. Biederman return ERR_PTR(-ENOMEM); 9083895dbf8SEric W. Biederman 9093895dbf8SEric W. Biederman 9103895dbf8SEric W. Biederman /* Exactly one processes may set d_mounted */ 9113895dbf8SEric W. Biederman ret = d_set_mounted(dentry); 9123895dbf8SEric W. Biederman 9133895dbf8SEric W. Biederman /* Someone else set d_mounted? */ 9143895dbf8SEric W. Biederman if (ret == -EBUSY) 9153895dbf8SEric W. Biederman goto mountpoint; 9163895dbf8SEric W. Biederman 9173895dbf8SEric W. Biederman /* The dentry is not available as a mountpoint? */ 9183895dbf8SEric W. Biederman mp = ERR_PTR(ret); 9193895dbf8SEric W. Biederman if (ret) 9203895dbf8SEric W. Biederman goto done; 9213895dbf8SEric W. Biederman 9223895dbf8SEric W. Biederman /* Add the new mountpoint to the hash table */ 9233895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 9244edbe133SAl Viro new->m_dentry = dget(dentry); 9253895dbf8SEric W. Biederman new->m_count = 1; 9263895dbf8SEric W. Biederman hlist_add_head(&new->m_hash, mp_hash(dentry)); 9273895dbf8SEric W. Biederman INIT_HLIST_HEAD(&new->m_list); 9283895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 9293895dbf8SEric W. Biederman 9303895dbf8SEric W. Biederman mp = new; 9313895dbf8SEric W. Biederman new = NULL; 9323895dbf8SEric W. Biederman done: 9333895dbf8SEric W. Biederman kfree(new); 93484d17192SAl Viro return mp; 93584d17192SAl Viro } 93684d17192SAl Viro 9374edbe133SAl Viro /* 9384edbe133SAl Viro * vfsmount lock must be held. Additionally, the caller is responsible 9394edbe133SAl Viro * for serializing calls for given disposal list. 9404edbe133SAl Viro */ 9414edbe133SAl Viro static void __put_mountpoint(struct mountpoint *mp, struct list_head *list) 94284d17192SAl Viro { 94384d17192SAl Viro if (!--mp->m_count) { 94484d17192SAl Viro struct dentry *dentry = mp->m_dentry; 9450a5eb7c8SEric W. Biederman BUG_ON(!hlist_empty(&mp->m_list)); 94684d17192SAl Viro spin_lock(&dentry->d_lock); 94784d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 94884d17192SAl Viro spin_unlock(&dentry->d_lock); 9494edbe133SAl Viro dput_to_list(dentry, list); 9500818bf27SAl Viro hlist_del(&mp->m_hash); 95184d17192SAl Viro kfree(mp); 95284d17192SAl Viro } 95384d17192SAl Viro } 95484d17192SAl Viro 9554edbe133SAl Viro /* called with namespace_lock and vfsmount lock */ 9564edbe133SAl Viro static void put_mountpoint(struct mountpoint *mp) 9574edbe133SAl Viro { 9584edbe133SAl Viro __put_mountpoint(mp, &ex_mountpoints); 9594edbe133SAl Viro } 9604edbe133SAl Viro 961143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 9621da177e4SLinus Torvalds { 9636b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 9641da177e4SLinus Torvalds } 9651da177e4SLinus Torvalds 96699b7db7bSNick Piggin /* 96799b7db7bSNick Piggin * vfsmount lock must be held for write 96899b7db7bSNick Piggin */ 9696b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 9705addc5ddSAl Viro { 9715addc5ddSAl Viro if (ns) { 9725addc5ddSAl Viro ns->event = ++event; 9735addc5ddSAl Viro wake_up_interruptible(&ns->poll); 9745addc5ddSAl Viro } 9755addc5ddSAl Viro } 9765addc5ddSAl Viro 97799b7db7bSNick Piggin /* 97899b7db7bSNick Piggin * vfsmount lock must be held for write 97999b7db7bSNick Piggin */ 9806b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 9815addc5ddSAl Viro { 9825addc5ddSAl Viro if (ns && ns->event != event) { 9835addc5ddSAl Viro ns->event = event; 9845addc5ddSAl Viro wake_up_interruptible(&ns->poll); 9855addc5ddSAl Viro } 9865addc5ddSAl Viro } 9875addc5ddSAl Viro 98899b7db7bSNick Piggin /* 98999b7db7bSNick Piggin * vfsmount lock must be held for write 99099b7db7bSNick Piggin */ 991e4e59906SAl Viro static struct mountpoint *unhash_mnt(struct mount *mnt) 9921da177e4SLinus Torvalds { 993e4e59906SAl Viro struct mountpoint *mp; 9940714a533SAl Viro mnt->mnt_parent = mnt; 995a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 9966b41d536SAl Viro list_del_init(&mnt->mnt_child); 99738129a13SAl Viro hlist_del_init_rcu(&mnt->mnt_hash); 9980a5eb7c8SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 999e4e59906SAl Viro mp = mnt->mnt_mp; 100084d17192SAl Viro mnt->mnt_mp = NULL; 1001e4e59906SAl Viro return mp; 10027bdb11deSEric W. Biederman } 10037bdb11deSEric W. Biederman 10047bdb11deSEric W. Biederman /* 10057bdb11deSEric W. Biederman * vfsmount lock must be held for write 10067bdb11deSEric W. Biederman */ 10076a46c573SEric W. Biederman static void umount_mnt(struct mount *mnt) 10086a46c573SEric W. Biederman { 1009e4e59906SAl Viro put_mountpoint(unhash_mnt(mnt)); 10106a46c573SEric W. Biederman } 10116a46c573SEric W. Biederman 10126a46c573SEric W. Biederman /* 10136a46c573SEric W. Biederman * vfsmount lock must be held for write 10146a46c573SEric W. Biederman */ 101584d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 101684d17192SAl Viro struct mountpoint *mp, 101744d964d6SAl Viro struct mount *child_mnt) 1018b90fa9aeSRam Pai { 101984d17192SAl Viro mp->m_count++; 10203a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 10214edbe133SAl Viro child_mnt->mnt_mountpoint = mp->m_dentry; 10223a2393d7SAl Viro child_mnt->mnt_parent = mnt; 102384d17192SAl Viro child_mnt->mnt_mp = mp; 10240a5eb7c8SEric W. Biederman hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list); 1025b90fa9aeSRam Pai } 1026b90fa9aeSRam Pai 10271064f874SEric W. Biederman static void __attach_mnt(struct mount *mnt, struct mount *parent) 10281064f874SEric W. Biederman { 10291064f874SEric W. Biederman hlist_add_head_rcu(&mnt->mnt_hash, 10301064f874SEric W. Biederman m_hash(&parent->mnt, mnt->mnt_mountpoint)); 10311064f874SEric W. Biederman list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 10321064f874SEric W. Biederman } 10331064f874SEric W. Biederman 103499b7db7bSNick Piggin /* 103599b7db7bSNick Piggin * vfsmount lock must be held for write 103699b7db7bSNick Piggin */ 103784d17192SAl Viro static void attach_mnt(struct mount *mnt, 103884d17192SAl Viro struct mount *parent, 103984d17192SAl Viro struct mountpoint *mp) 10401da177e4SLinus Torvalds { 104184d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 10421064f874SEric W. Biederman __attach_mnt(mnt, parent); 1043b90fa9aeSRam Pai } 1044b90fa9aeSRam Pai 10451064f874SEric W. Biederman void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt) 104612a5b529SAl Viro { 10471064f874SEric W. Biederman struct mountpoint *old_mp = mnt->mnt_mp; 10481064f874SEric W. Biederman struct mount *old_parent = mnt->mnt_parent; 10491064f874SEric W. Biederman 10501064f874SEric W. Biederman list_del_init(&mnt->mnt_child); 10511064f874SEric W. Biederman hlist_del_init(&mnt->mnt_mp_list); 10521064f874SEric W. Biederman hlist_del_init_rcu(&mnt->mnt_hash); 10531064f874SEric W. Biederman 10541064f874SEric W. Biederman attach_mnt(mnt, parent, mp); 10551064f874SEric W. Biederman 10561064f874SEric W. Biederman put_mountpoint(old_mp); 10571064f874SEric W. Biederman mnt_add_count(old_parent, -1); 105812a5b529SAl Viro } 105912a5b529SAl Viro 1060b90fa9aeSRam Pai /* 106199b7db7bSNick Piggin * vfsmount lock must be held for write 1062b90fa9aeSRam Pai */ 10631064f874SEric W. Biederman static void commit_tree(struct mount *mnt) 1064b90fa9aeSRam Pai { 10650714a533SAl Viro struct mount *parent = mnt->mnt_parent; 106683adc753SAl Viro struct mount *m; 1067b90fa9aeSRam Pai LIST_HEAD(head); 1068143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 1069b90fa9aeSRam Pai 10700714a533SAl Viro BUG_ON(parent == mnt); 1071b90fa9aeSRam Pai 10721a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 1073f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 1074143c8c91SAl Viro m->mnt_ns = n; 1075f03c6599SAl Viro 1076b90fa9aeSRam Pai list_splice(&head, n->list.prev); 1077b90fa9aeSRam Pai 1078d2921684SEric W. Biederman n->mounts += n->pending_mounts; 1079d2921684SEric W. Biederman n->pending_mounts = 0; 1080d2921684SEric W. Biederman 10811064f874SEric W. Biederman __attach_mnt(mnt, parent); 10826b3286edSKirill Korotaev touch_mnt_namespace(n); 10831da177e4SLinus Torvalds } 10841da177e4SLinus Torvalds 1085909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 10861da177e4SLinus Torvalds { 10876b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 10886b41d536SAl Viro if (next == &p->mnt_mounts) { 10891da177e4SLinus Torvalds while (1) { 1090909b0a88SAl Viro if (p == root) 10911da177e4SLinus Torvalds return NULL; 10926b41d536SAl Viro next = p->mnt_child.next; 10936b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 10941da177e4SLinus Torvalds break; 10950714a533SAl Viro p = p->mnt_parent; 10961da177e4SLinus Torvalds } 10971da177e4SLinus Torvalds } 10986b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 10991da177e4SLinus Torvalds } 11001da177e4SLinus Torvalds 1101315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 11029676f0c6SRam Pai { 11036b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 11046b41d536SAl Viro while (prev != &p->mnt_mounts) { 11056b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 11066b41d536SAl Viro prev = p->mnt_mounts.prev; 11079676f0c6SRam Pai } 11089676f0c6SRam Pai return p; 11099676f0c6SRam Pai } 11109676f0c6SRam Pai 11118f291889SAl Viro /** 11128f291889SAl Viro * vfs_create_mount - Create a mount for a configured superblock 11138f291889SAl Viro * @fc: The configuration context with the superblock attached 11148f291889SAl Viro * 11158f291889SAl Viro * Create a mount to an already configured superblock. If necessary, the 11168f291889SAl Viro * caller should invoke vfs_get_tree() before calling this. 11178f291889SAl Viro * 11188f291889SAl Viro * Note that this does not attach the mount to anything. 11198f291889SAl Viro */ 11208f291889SAl Viro struct vfsmount *vfs_create_mount(struct fs_context *fc) 11219d412a43SAl Viro { 1122b105e270SAl Viro struct mount *mnt; 11239d412a43SAl Viro 11248f291889SAl Viro if (!fc->root) 11258f291889SAl Viro return ERR_PTR(-EINVAL); 11269d412a43SAl Viro 11278f291889SAl Viro mnt = alloc_vfsmnt(fc->source ?: "none"); 11289d412a43SAl Viro if (!mnt) 11299d412a43SAl Viro return ERR_PTR(-ENOMEM); 11309d412a43SAl Viro 11318f291889SAl Viro if (fc->sb_flags & SB_KERNMOUNT) 1132b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 11339d412a43SAl Viro 11348f291889SAl Viro atomic_inc(&fc->root->d_sb->s_active); 11358f291889SAl Viro mnt->mnt.mnt_sb = fc->root->d_sb; 11368f291889SAl Viro mnt->mnt.mnt_root = dget(fc->root); 1137a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 11380714a533SAl Viro mnt->mnt_parent = mnt; 11398f291889SAl Viro 1140719ea2fbSAl Viro lock_mount_hash(); 11418f291889SAl Viro list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts); 1142719ea2fbSAl Viro unlock_mount_hash(); 1143b105e270SAl Viro return &mnt->mnt; 11449d412a43SAl Viro } 11458f291889SAl Viro EXPORT_SYMBOL(vfs_create_mount); 11468f291889SAl Viro 11478f291889SAl Viro struct vfsmount *fc_mount(struct fs_context *fc) 11488f291889SAl Viro { 11498f291889SAl Viro int err = vfs_get_tree(fc); 11508f291889SAl Viro if (!err) { 11518f291889SAl Viro up_write(&fc->root->d_sb->s_umount); 11528f291889SAl Viro return vfs_create_mount(fc); 11538f291889SAl Viro } 11548f291889SAl Viro return ERR_PTR(err); 11558f291889SAl Viro } 11568f291889SAl Viro EXPORT_SYMBOL(fc_mount); 11578f291889SAl Viro 11589bc61ab1SDavid Howells struct vfsmount *vfs_kern_mount(struct file_system_type *type, 11599bc61ab1SDavid Howells int flags, const char *name, 11609bc61ab1SDavid Howells void *data) 11611da177e4SLinus Torvalds { 11629bc61ab1SDavid Howells struct fs_context *fc; 11638f291889SAl Viro struct vfsmount *mnt; 11649bc61ab1SDavid Howells int ret = 0; 11659d412a43SAl Viro 11669d412a43SAl Viro if (!type) 11673e1aeb00SDavid Howells return ERR_PTR(-EINVAL); 11689d412a43SAl Viro 11699bc61ab1SDavid Howells fc = fs_context_for_mount(type, flags); 11709bc61ab1SDavid Howells if (IS_ERR(fc)) 11719bc61ab1SDavid Howells return ERR_CAST(fc); 11729bc61ab1SDavid Howells 11733e1aeb00SDavid Howells if (name) 11743e1aeb00SDavid Howells ret = vfs_parse_fs_string(fc, "source", 11753e1aeb00SDavid Howells name, strlen(name)); 11769bc61ab1SDavid Howells if (!ret) 11779bc61ab1SDavid Howells ret = parse_monolithic_mount_data(fc, data); 11789bc61ab1SDavid Howells if (!ret) 11798f291889SAl Viro mnt = fc_mount(fc); 11808f291889SAl Viro else 11818f291889SAl Viro mnt = ERR_PTR(ret); 11829d412a43SAl Viro 11839bc61ab1SDavid Howells put_fs_context(fc); 11848f291889SAl Viro return mnt; 11859d412a43SAl Viro } 11869d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 11879d412a43SAl Viro 118893faccbbSEric W. Biederman struct vfsmount * 118993faccbbSEric W. Biederman vfs_submount(const struct dentry *mountpoint, struct file_system_type *type, 119093faccbbSEric W. Biederman const char *name, void *data) 119193faccbbSEric W. Biederman { 119293faccbbSEric W. Biederman /* Until it is worked out how to pass the user namespace 119393faccbbSEric W. Biederman * through from the parent mount to the submount don't support 119493faccbbSEric W. Biederman * unprivileged mounts with submounts. 119593faccbbSEric W. Biederman */ 119693faccbbSEric W. Biederman if (mountpoint->d_sb->s_user_ns != &init_user_ns) 119793faccbbSEric W. Biederman return ERR_PTR(-EPERM); 119893faccbbSEric W. Biederman 1199e462ec50SDavid Howells return vfs_kern_mount(type, SB_SUBMOUNT, name, data); 120093faccbbSEric W. Biederman } 120193faccbbSEric W. Biederman EXPORT_SYMBOL_GPL(vfs_submount); 120293faccbbSEric W. Biederman 120387129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 120436341f64SRam Pai int flag) 12051da177e4SLinus Torvalds { 120687129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 1207be34d1a3SDavid Howells struct mount *mnt; 1208be34d1a3SDavid Howells int err; 12091da177e4SLinus Torvalds 1210be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 1211be34d1a3SDavid Howells if (!mnt) 1212be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 1213be34d1a3SDavid Howells 12147a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 121515169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 1216719f5d7fSMiklos Szeredi else 121715169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 1218719f5d7fSMiklos Szeredi 121915169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 1220be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 1221719f5d7fSMiklos Szeredi if (err) 1222719f5d7fSMiklos Szeredi goto out_free; 1223719f5d7fSMiklos Szeredi } 1224719f5d7fSMiklos Szeredi 122516a34adbSAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags; 122616a34adbSAl Viro mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL); 12275ff9d8a6SEric W. Biederman 12281da177e4SLinus Torvalds atomic_inc(&sb->s_active); 1229256c8aedSChristian Brauner mnt->mnt.mnt_idmap = mnt_idmap_get(mnt_idmap(&old->mnt)); 1230256c8aedSChristian Brauner 1231b105e270SAl Viro mnt->mnt.mnt_sb = sb; 1232b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 1233a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 12340714a533SAl Viro mnt->mnt_parent = mnt; 1235719ea2fbSAl Viro lock_mount_hash(); 123639f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 1237719ea2fbSAl Viro unlock_mount_hash(); 1238b90fa9aeSRam Pai 12397a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 12407a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 12416776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 124232301920SAl Viro mnt->mnt_master = old; 1243fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 12448aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 1245fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 12466776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 1247d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 12486776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 1249d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 12505235d448SAl Viro } else { 12515235d448SAl Viro CLEAR_MNT_SHARED(mnt); 12525afe0022SRam Pai } 1253b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 12540f0afb1dSAl Viro set_mnt_shared(mnt); 12551da177e4SLinus Torvalds 12561da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 12571da177e4SLinus Torvalds * as the original if that was on one */ 125836341f64SRam Pai if (flag & CL_EXPIRE) { 12596776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 12606776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 12611da177e4SLinus Torvalds } 1262be34d1a3SDavid Howells 1263cb338d06SAl Viro return mnt; 1264719f5d7fSMiklos Szeredi 1265719f5d7fSMiklos Szeredi out_free: 12668ffcb32eSDavid Howells mnt_free_id(mnt); 1267719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 1268be34d1a3SDavid Howells return ERR_PTR(err); 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 12719ea459e1SAl Viro static void cleanup_mnt(struct mount *mnt) 12729ea459e1SAl Viro { 127356cbb429SAl Viro struct hlist_node *p; 127456cbb429SAl Viro struct mount *m; 12759ea459e1SAl Viro /* 127656cbb429SAl Viro * The warning here probably indicates that somebody messed 127756cbb429SAl Viro * up a mnt_want/drop_write() pair. If this happens, the 127856cbb429SAl Viro * filesystem was probably unable to make r/w->r/o transitions. 12799ea459e1SAl Viro * The locking used to deal with mnt_count decrement provides barriers, 12809ea459e1SAl Viro * so mnt_get_writers() below is safe. 12819ea459e1SAl Viro */ 12829ea459e1SAl Viro WARN_ON(mnt_get_writers(mnt)); 12839ea459e1SAl Viro if (unlikely(mnt->mnt_pins.first)) 12849ea459e1SAl Viro mnt_pin_kill(mnt); 128556cbb429SAl Viro hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) { 128656cbb429SAl Viro hlist_del(&m->mnt_umount); 128756cbb429SAl Viro mntput(&m->mnt); 128856cbb429SAl Viro } 12899ea459e1SAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 12909ea459e1SAl Viro dput(mnt->mnt.mnt_root); 12919ea459e1SAl Viro deactivate_super(mnt->mnt.mnt_sb); 12929ea459e1SAl Viro mnt_free_id(mnt); 12939ea459e1SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt); 12949ea459e1SAl Viro } 12959ea459e1SAl Viro 12969ea459e1SAl Viro static void __cleanup_mnt(struct rcu_head *head) 12979ea459e1SAl Viro { 12989ea459e1SAl Viro cleanup_mnt(container_of(head, struct mount, mnt_rcu)); 12999ea459e1SAl Viro } 13009ea459e1SAl Viro 13019ea459e1SAl Viro static LLIST_HEAD(delayed_mntput_list); 13029ea459e1SAl Viro static void delayed_mntput(struct work_struct *unused) 13039ea459e1SAl Viro { 13049ea459e1SAl Viro struct llist_node *node = llist_del_all(&delayed_mntput_list); 130529785735SByungchul Park struct mount *m, *t; 13069ea459e1SAl Viro 130729785735SByungchul Park llist_for_each_entry_safe(m, t, node, mnt_llist) 130829785735SByungchul Park cleanup_mnt(m); 13099ea459e1SAl Viro } 13109ea459e1SAl Viro static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput); 13119ea459e1SAl Viro 1312900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 13137b7b1aceSAl Viro { 13144edbe133SAl Viro LIST_HEAD(list); 1315edf7ddbfSEric Biggers int count; 13164edbe133SAl Viro 131748a066e7SAl Viro rcu_read_lock(); 13189ea0a46cSAl Viro if (likely(READ_ONCE(mnt->mnt_ns))) { 13199ea0a46cSAl Viro /* 13209ea0a46cSAl Viro * Since we don't do lock_mount_hash() here, 13219ea0a46cSAl Viro * ->mnt_ns can change under us. However, if it's 13229ea0a46cSAl Viro * non-NULL, then there's a reference that won't 13239ea0a46cSAl Viro * be dropped until after an RCU delay done after 13249ea0a46cSAl Viro * turning ->mnt_ns NULL. So if we observe it 13259ea0a46cSAl Viro * non-NULL under rcu_read_lock(), the reference 13269ea0a46cSAl Viro * we are dropping is not the final one. 13279ea0a46cSAl Viro */ 1328aa9c0e07SAl Viro mnt_add_count(mnt, -1); 132948a066e7SAl Viro rcu_read_unlock(); 133099b7db7bSNick Piggin return; 1331b3e19d92SNick Piggin } 1332719ea2fbSAl Viro lock_mount_hash(); 1333119e1ef8SAl Viro /* 1334119e1ef8SAl Viro * make sure that if __legitimize_mnt() has not seen us grab 1335119e1ef8SAl Viro * mount_lock, we'll see their refcount increment here. 1336119e1ef8SAl Viro */ 1337119e1ef8SAl Viro smp_mb(); 13389ea0a46cSAl Viro mnt_add_count(mnt, -1); 1339edf7ddbfSEric Biggers count = mnt_get_count(mnt); 1340edf7ddbfSEric Biggers if (count != 0) { 1341edf7ddbfSEric Biggers WARN_ON(count < 0); 134248a066e7SAl Viro rcu_read_unlock(); 1343719ea2fbSAl Viro unlock_mount_hash(); 134499b7db7bSNick Piggin return; 134599b7db7bSNick Piggin } 134648a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 134748a066e7SAl Viro rcu_read_unlock(); 134848a066e7SAl Viro unlock_mount_hash(); 134948a066e7SAl Viro return; 135048a066e7SAl Viro } 135148a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 135248a066e7SAl Viro rcu_read_unlock(); 1353962830dfSAndi Kleen 135439f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 1355ce07d891SEric W. Biederman 1356ce07d891SEric W. Biederman if (unlikely(!list_empty(&mnt->mnt_mounts))) { 1357ce07d891SEric W. Biederman struct mount *p, *tmp; 1358ce07d891SEric W. Biederman list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts, mnt_child) { 13594edbe133SAl Viro __put_mountpoint(unhash_mnt(p), &list); 136056cbb429SAl Viro hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children); 1361ce07d891SEric W. Biederman } 1362ce07d891SEric W. Biederman } 1363719ea2fbSAl Viro unlock_mount_hash(); 13644edbe133SAl Viro shrink_dentry_list(&list); 1365649a795aSAl Viro 13669ea459e1SAl Viro if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) { 13679ea459e1SAl Viro struct task_struct *task = current; 13689ea459e1SAl Viro if (likely(!(task->flags & PF_KTHREAD))) { 13699ea459e1SAl Viro init_task_work(&mnt->mnt_rcu, __cleanup_mnt); 137091989c70SJens Axboe if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME)) 13719ea459e1SAl Viro return; 13729ea459e1SAl Viro } 13739ea459e1SAl Viro if (llist_add(&mnt->mnt_llist, &delayed_mntput_list)) 13749ea459e1SAl Viro schedule_delayed_work(&delayed_mntput_work, 1); 13759ea459e1SAl Viro return; 13769ea459e1SAl Viro } 13779ea459e1SAl Viro cleanup_mnt(mnt); 1378b3e19d92SNick Piggin } 1379b3e19d92SNick Piggin 1380b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 1381b3e19d92SNick Piggin { 1382b3e19d92SNick Piggin if (mnt) { 1383863d684fSAl Viro struct mount *m = real_mount(mnt); 1384b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 1385863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 1386863d684fSAl Viro m->mnt_expiry_mark = 0; 1387863d684fSAl Viro mntput_no_expire(m); 1388b3e19d92SNick Piggin } 1389b3e19d92SNick Piggin } 1390b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 1391b3e19d92SNick Piggin 1392b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 1393b3e19d92SNick Piggin { 1394b3e19d92SNick Piggin if (mnt) 139583adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 1396b3e19d92SNick Piggin return mnt; 1397b3e19d92SNick Piggin } 1398b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 1399b3e19d92SNick Piggin 1400*da27f796SRik van Riel /* 1401*da27f796SRik van Riel * Make a mount point inaccessible to new lookups. 1402*da27f796SRik van Riel * Because there may still be current users, the caller MUST WAIT 1403*da27f796SRik van Riel * for an RCU grace period before destroying the mount point. 1404*da27f796SRik van Riel */ 1405*da27f796SRik van Riel void mnt_make_shortterm(struct vfsmount *mnt) 1406*da27f796SRik van Riel { 1407*da27f796SRik van Riel if (mnt) 1408*da27f796SRik van Riel real_mount(mnt)->mnt_ns = NULL; 1409*da27f796SRik van Riel } 1410*da27f796SRik van Riel 14111f287bc4SRandy Dunlap /** 14121f287bc4SRandy Dunlap * path_is_mountpoint() - Check if path is a mount in the current namespace. 14131f287bc4SRandy Dunlap * @path: path to check 1414c6609c0aSIan Kent * 1415c6609c0aSIan Kent * d_mountpoint() can only be used reliably to establish if a dentry is 1416c6609c0aSIan Kent * not mounted in any namespace and that common case is handled inline. 1417c6609c0aSIan Kent * d_mountpoint() isn't aware of the possibility there may be multiple 1418c6609c0aSIan Kent * mounts using a given dentry in a different namespace. This function 1419c6609c0aSIan Kent * checks if the passed in path is a mountpoint rather than the dentry 1420c6609c0aSIan Kent * alone. 1421c6609c0aSIan Kent */ 1422c6609c0aSIan Kent bool path_is_mountpoint(const struct path *path) 1423c6609c0aSIan Kent { 1424c6609c0aSIan Kent unsigned seq; 1425c6609c0aSIan Kent bool res; 1426c6609c0aSIan Kent 1427c6609c0aSIan Kent if (!d_mountpoint(path->dentry)) 1428c6609c0aSIan Kent return false; 1429c6609c0aSIan Kent 1430c6609c0aSIan Kent rcu_read_lock(); 1431c6609c0aSIan Kent do { 1432c6609c0aSIan Kent seq = read_seqbegin(&mount_lock); 1433c6609c0aSIan Kent res = __path_is_mountpoint(path); 1434c6609c0aSIan Kent } while (read_seqretry(&mount_lock, seq)); 1435c6609c0aSIan Kent rcu_read_unlock(); 1436c6609c0aSIan Kent 1437c6609c0aSIan Kent return res; 1438c6609c0aSIan Kent } 1439c6609c0aSIan Kent EXPORT_SYMBOL(path_is_mountpoint); 1440c6609c0aSIan Kent 1441ca71cf71SAl Viro struct vfsmount *mnt_clone_internal(const struct path *path) 14427b7b1aceSAl Viro { 14433064c356SAl Viro struct mount *p; 14443064c356SAl Viro p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE); 14453064c356SAl Viro if (IS_ERR(p)) 14463064c356SAl Viro return ERR_CAST(p); 14473064c356SAl Viro p->mnt.mnt_flags |= MNT_INTERNAL; 14483064c356SAl Viro return &p->mnt; 14497b7b1aceSAl Viro } 14501da177e4SLinus Torvalds 1451a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 14529f6c61f9SMiklos Szeredi static struct mount *mnt_list_next(struct mnt_namespace *ns, 14539f6c61f9SMiklos Szeredi struct list_head *p) 14549f6c61f9SMiklos Szeredi { 14559f6c61f9SMiklos Szeredi struct mount *mnt, *ret = NULL; 14569f6c61f9SMiklos Szeredi 14579f6c61f9SMiklos Szeredi lock_ns_list(ns); 14589f6c61f9SMiklos Szeredi list_for_each_continue(p, &ns->list) { 14599f6c61f9SMiklos Szeredi mnt = list_entry(p, typeof(*mnt), mnt_list); 14609f6c61f9SMiklos Szeredi if (!mnt_is_cursor(mnt)) { 14619f6c61f9SMiklos Szeredi ret = mnt; 14629f6c61f9SMiklos Szeredi break; 14639f6c61f9SMiklos Szeredi } 14649f6c61f9SMiklos Szeredi } 14659f6c61f9SMiklos Szeredi unlock_ns_list(ns); 14669f6c61f9SMiklos Szeredi 14679f6c61f9SMiklos Szeredi return ret; 14689f6c61f9SMiklos Szeredi } 14699f6c61f9SMiklos Szeredi 14700226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 14711da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 14721da177e4SLinus Torvalds { 1473ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 14749f6c61f9SMiklos Szeredi struct list_head *prev; 14751da177e4SLinus Torvalds 1476390c6843SRam Pai down_read(&namespace_sem); 14779f6c61f9SMiklos Szeredi if (!*pos) { 14789f6c61f9SMiklos Szeredi prev = &p->ns->list; 14799f6c61f9SMiklos Szeredi } else { 14809f6c61f9SMiklos Szeredi prev = &p->cursor.mnt_list; 14819f6c61f9SMiklos Szeredi 14829f6c61f9SMiklos Szeredi /* Read after we'd reached the end? */ 14839f6c61f9SMiklos Szeredi if (list_empty(prev)) 14849f6c61f9SMiklos Szeredi return NULL; 1485c7999c36SAl Viro } 1486c7999c36SAl Viro 14879f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, prev); 14881da177e4SLinus Torvalds } 14891da177e4SLinus Torvalds 14901da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 14911da177e4SLinus Torvalds { 1492ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 14939f6c61f9SMiklos Szeredi struct mount *mnt = v; 1494b0765fb8SPavel Emelianov 14959f6c61f9SMiklos Szeredi ++*pos; 14969f6c61f9SMiklos Szeredi return mnt_list_next(p->ns, &mnt->mnt_list); 14971da177e4SLinus Torvalds } 14981da177e4SLinus Torvalds 14991da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 15001da177e4SLinus Torvalds { 15019f6c61f9SMiklos Szeredi struct proc_mounts *p = m->private; 15029f6c61f9SMiklos Szeredi struct mount *mnt = v; 15039f6c61f9SMiklos Szeredi 15049f6c61f9SMiklos Szeredi lock_ns_list(p->ns); 15059f6c61f9SMiklos Szeredi if (mnt) 15069f6c61f9SMiklos Szeredi list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list); 15079f6c61f9SMiklos Szeredi else 15089f6c61f9SMiklos Szeredi list_del_init(&p->cursor.mnt_list); 15099f6c61f9SMiklos Szeredi unlock_ns_list(p->ns); 1510390c6843SRam Pai up_read(&namespace_sem); 15111da177e4SLinus Torvalds } 15121da177e4SLinus Torvalds 15130226f492SAl Viro static int m_show(struct seq_file *m, void *v) 15149f5596afSAl Viro { 1515ede1bf0dSYann Droneaud struct proc_mounts *p = m->private; 15169f6c61f9SMiklos Szeredi struct mount *r = v; 15170226f492SAl Viro return p->show(m, &r->mnt); 15181da177e4SLinus Torvalds } 15191da177e4SLinus Torvalds 1520a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 15211da177e4SLinus Torvalds .start = m_start, 15221da177e4SLinus Torvalds .next = m_next, 15231da177e4SLinus Torvalds .stop = m_stop, 15240226f492SAl Viro .show = m_show, 1525b4629fe2SChuck Lever }; 15269f6c61f9SMiklos Szeredi 15279f6c61f9SMiklos Szeredi void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor) 15289f6c61f9SMiklos Szeredi { 15299f6c61f9SMiklos Szeredi down_read(&namespace_sem); 15309f6c61f9SMiklos Szeredi lock_ns_list(ns); 15319f6c61f9SMiklos Szeredi list_del(&cursor->mnt_list); 15329f6c61f9SMiklos Szeredi unlock_ns_list(ns); 15339f6c61f9SMiklos Szeredi up_read(&namespace_sem); 15349f6c61f9SMiklos Szeredi } 1535a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1536b4629fe2SChuck Lever 15371da177e4SLinus Torvalds /** 15381da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 15391f287bc4SRandy Dunlap * @m: root of mount tree 15401da177e4SLinus Torvalds * 15411da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 15421da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 15431da177e4SLinus Torvalds * busy. 15441da177e4SLinus Torvalds */ 1545909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 15461da177e4SLinus Torvalds { 1547909b0a88SAl Viro struct mount *mnt = real_mount(m); 154836341f64SRam Pai int actual_refs = 0; 154936341f64SRam Pai int minimum_refs = 0; 1550315fc83eSAl Viro struct mount *p; 1551909b0a88SAl Viro BUG_ON(!m); 15521da177e4SLinus Torvalds 1553b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1554719ea2fbSAl Viro lock_mount_hash(); 1555909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 155683adc753SAl Viro actual_refs += mnt_get_count(p); 15571da177e4SLinus Torvalds minimum_refs += 2; 15581da177e4SLinus Torvalds } 1559719ea2fbSAl Viro unlock_mount_hash(); 15601da177e4SLinus Torvalds 15611da177e4SLinus Torvalds if (actual_refs > minimum_refs) 15621da177e4SLinus Torvalds return 0; 1563e3474a8eSIan Kent 1564e3474a8eSIan Kent return 1; 15651da177e4SLinus Torvalds } 15661da177e4SLinus Torvalds 15671da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 15681da177e4SLinus Torvalds 15691da177e4SLinus Torvalds /** 15701da177e4SLinus Torvalds * may_umount - check if a mount point is busy 15711da177e4SLinus Torvalds * @mnt: root of mount 15721da177e4SLinus Torvalds * 15731da177e4SLinus Torvalds * This is called to check if a mount point has any 15741da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 15751da177e4SLinus Torvalds * mount has sub mounts this will return busy 15761da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 15771da177e4SLinus Torvalds * 15781da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 15791da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 15801da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 15811da177e4SLinus Torvalds */ 15821da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 15831da177e4SLinus Torvalds { 1584e3474a8eSIan Kent int ret = 1; 15858ad08d8aSAl Viro down_read(&namespace_sem); 1586719ea2fbSAl Viro lock_mount_hash(); 15871ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1588e3474a8eSIan Kent ret = 0; 1589719ea2fbSAl Viro unlock_mount_hash(); 15908ad08d8aSAl Viro up_read(&namespace_sem); 1591a05964f3SRam Pai return ret; 15921da177e4SLinus Torvalds } 15931da177e4SLinus Torvalds 15941da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 15951da177e4SLinus Torvalds 159697216be0SAl Viro static void namespace_unlock(void) 15971da177e4SLinus Torvalds { 1598a3b3c562SEric W. Biederman struct hlist_head head; 159956cbb429SAl Viro struct hlist_node *p; 160056cbb429SAl Viro struct mount *m; 16014edbe133SAl Viro LIST_HEAD(list); 160297216be0SAl Viro 1603a3b3c562SEric W. Biederman hlist_move_list(&unmounted, &head); 16044edbe133SAl Viro list_splice_init(&ex_mountpoints, &list); 1605a3b3c562SEric W. Biederman 160697216be0SAl Viro up_write(&namespace_sem); 1607a3b3c562SEric W. Biederman 16084edbe133SAl Viro shrink_dentry_list(&list); 16094edbe133SAl Viro 1610a3b3c562SEric W. Biederman if (likely(hlist_empty(&head))) 161197216be0SAl Viro return; 161297216be0SAl Viro 161322cb7405SNeilBrown synchronize_rcu_expedited(); 161448a066e7SAl Viro 161556cbb429SAl Viro hlist_for_each_entry_safe(m, p, &head, mnt_umount) { 161656cbb429SAl Viro hlist_del(&m->mnt_umount); 161756cbb429SAl Viro mntput(&m->mnt); 161856cbb429SAl Viro } 161970fbcdf4SRam Pai } 162070fbcdf4SRam Pai 162197216be0SAl Viro static inline void namespace_lock(void) 1622e3197d83SAl Viro { 162397216be0SAl Viro down_write(&namespace_sem); 1624e3197d83SAl Viro } 1625e3197d83SAl Viro 1626e819f152SEric W. Biederman enum umount_tree_flags { 1627e819f152SEric W. Biederman UMOUNT_SYNC = 1, 1628e819f152SEric W. Biederman UMOUNT_PROPAGATE = 2, 1629e0c9c0afSEric W. Biederman UMOUNT_CONNECTED = 4, 1630e819f152SEric W. Biederman }; 1631f2d0a123SEric W. Biederman 1632f2d0a123SEric W. Biederman static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how) 1633f2d0a123SEric W. Biederman { 1634f2d0a123SEric W. Biederman /* Leaving mounts connected is only valid for lazy umounts */ 1635f2d0a123SEric W. Biederman if (how & UMOUNT_SYNC) 1636f2d0a123SEric W. Biederman return true; 1637f2d0a123SEric W. Biederman 1638f2d0a123SEric W. Biederman /* A mount without a parent has nothing to be connected to */ 1639f2d0a123SEric W. Biederman if (!mnt_has_parent(mnt)) 1640f2d0a123SEric W. Biederman return true; 1641f2d0a123SEric W. Biederman 1642f2d0a123SEric W. Biederman /* Because the reference counting rules change when mounts are 1643f2d0a123SEric W. Biederman * unmounted and connected, umounted mounts may not be 1644f2d0a123SEric W. Biederman * connected to mounted mounts. 1645f2d0a123SEric W. Biederman */ 1646f2d0a123SEric W. Biederman if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT)) 1647f2d0a123SEric W. Biederman return true; 1648f2d0a123SEric W. Biederman 1649f2d0a123SEric W. Biederman /* Has it been requested that the mount remain connected? */ 1650f2d0a123SEric W. Biederman if (how & UMOUNT_CONNECTED) 1651f2d0a123SEric W. Biederman return false; 1652f2d0a123SEric W. Biederman 1653f2d0a123SEric W. Biederman /* Is the mount locked such that it needs to remain connected? */ 1654f2d0a123SEric W. Biederman if (IS_MNT_LOCKED(mnt)) 1655f2d0a123SEric W. Biederman return false; 1656f2d0a123SEric W. Biederman 1657f2d0a123SEric W. Biederman /* By default disconnect the mount */ 1658f2d0a123SEric W. Biederman return true; 1659f2d0a123SEric W. Biederman } 1660f2d0a123SEric W. Biederman 166199b7db7bSNick Piggin /* 166248a066e7SAl Viro * mount_lock must be held 166399b7db7bSNick Piggin * namespace_sem must be held for write 166499b7db7bSNick Piggin */ 1665e819f152SEric W. Biederman static void umount_tree(struct mount *mnt, enum umount_tree_flags how) 166670fbcdf4SRam Pai { 1667c003b26fSEric W. Biederman LIST_HEAD(tmp_list); 1668315fc83eSAl Viro struct mount *p; 166970fbcdf4SRam Pai 16705d88457eSEric W. Biederman if (how & UMOUNT_PROPAGATE) 16715d88457eSEric W. Biederman propagate_mount_unlock(mnt); 16725d88457eSEric W. Biederman 1673c003b26fSEric W. Biederman /* Gather the mounts to umount */ 1674590ce4bcSEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) { 1675590ce4bcSEric W. Biederman p->mnt.mnt_flags |= MNT_UMOUNT; 1676c003b26fSEric W. Biederman list_move(&p->mnt_list, &tmp_list); 1677590ce4bcSEric W. Biederman } 1678c003b26fSEric W. Biederman 1679411a938bSEric W. Biederman /* Hide the mounts from mnt_mounts */ 1680c003b26fSEric W. Biederman list_for_each_entry(p, &tmp_list, mnt_list) { 1681c003b26fSEric W. Biederman list_del_init(&p->mnt_child); 168238129a13SAl Viro } 168370fbcdf4SRam Pai 1684c003b26fSEric W. Biederman /* Add propogated mounts to the tmp_list */ 1685e819f152SEric W. Biederman if (how & UMOUNT_PROPAGATE) 16867b8a53fdSAl Viro propagate_umount(&tmp_list); 1687a05964f3SRam Pai 1688c003b26fSEric W. Biederman while (!list_empty(&tmp_list)) { 1689d2921684SEric W. Biederman struct mnt_namespace *ns; 1690ce07d891SEric W. Biederman bool disconnect; 1691c003b26fSEric W. Biederman p = list_first_entry(&tmp_list, struct mount, mnt_list); 16926776db3dSAl Viro list_del_init(&p->mnt_expire); 16931a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1694d2921684SEric W. Biederman ns = p->mnt_ns; 1695d2921684SEric W. Biederman if (ns) { 1696d2921684SEric W. Biederman ns->mounts--; 1697d2921684SEric W. Biederman __touch_mnt_namespace(ns); 1698d2921684SEric W. Biederman } 169963d37a84SAl Viro p->mnt_ns = NULL; 1700e819f152SEric W. Biederman if (how & UMOUNT_SYNC) 170148a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 170287b95ce0SAl Viro 1703f2d0a123SEric W. Biederman disconnect = disconnect_mount(p, how); 1704676da58dSAl Viro if (mnt_has_parent(p)) { 170581b6b061SAl Viro mnt_add_count(p->mnt_parent, -1); 1706ce07d891SEric W. Biederman if (!disconnect) { 1707ce07d891SEric W. Biederman /* Don't forget about p */ 1708ce07d891SEric W. Biederman list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts); 1709ce07d891SEric W. Biederman } else { 17106a46c573SEric W. Biederman umount_mnt(p); 17117c4b93d8SAl Viro } 1712ce07d891SEric W. Biederman } 17130f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 171419a1c409SAl Viro if (disconnect) 171519a1c409SAl Viro hlist_add_head(&p->mnt_umount, &unmounted); 171638129a13SAl Viro } 17171da177e4SLinus Torvalds } 17181da177e4SLinus Torvalds 1719b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1720c35038beSAl Viro 17218d0347f6SDavid Howells static int do_umount_root(struct super_block *sb) 17228d0347f6SDavid Howells { 17238d0347f6SDavid Howells int ret = 0; 17248d0347f6SDavid Howells 17258d0347f6SDavid Howells down_write(&sb->s_umount); 17268d0347f6SDavid Howells if (!sb_rdonly(sb)) { 17278d0347f6SDavid Howells struct fs_context *fc; 17288d0347f6SDavid Howells 17298d0347f6SDavid Howells fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY, 17308d0347f6SDavid Howells SB_RDONLY); 17318d0347f6SDavid Howells if (IS_ERR(fc)) { 17328d0347f6SDavid Howells ret = PTR_ERR(fc); 17338d0347f6SDavid Howells } else { 17348d0347f6SDavid Howells ret = parse_monolithic_mount_data(fc, NULL); 17358d0347f6SDavid Howells if (!ret) 17368d0347f6SDavid Howells ret = reconfigure_super(fc); 17378d0347f6SDavid Howells put_fs_context(fc); 17388d0347f6SDavid Howells } 17398d0347f6SDavid Howells } 17408d0347f6SDavid Howells up_write(&sb->s_umount); 17418d0347f6SDavid Howells return ret; 17428d0347f6SDavid Howells } 17438d0347f6SDavid Howells 17441ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 17451da177e4SLinus Torvalds { 17461ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 17471da177e4SLinus Torvalds int retval; 17481da177e4SLinus Torvalds 17491ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 17501da177e4SLinus Torvalds if (retval) 17511da177e4SLinus Torvalds return retval; 17521da177e4SLinus Torvalds 17531da177e4SLinus Torvalds /* 17541da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 17551da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 17561da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 17571da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 17581da177e4SLinus Torvalds */ 17591da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 17601ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 17611da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 17621da177e4SLinus Torvalds return -EINVAL; 17631da177e4SLinus Torvalds 1764b3e19d92SNick Piggin /* 1765b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1766b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1767b3e19d92SNick Piggin */ 1768719ea2fbSAl Viro lock_mount_hash(); 176983adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1770719ea2fbSAl Viro unlock_mount_hash(); 17711da177e4SLinus Torvalds return -EBUSY; 1772b3e19d92SNick Piggin } 1773719ea2fbSAl Viro unlock_mount_hash(); 17741da177e4SLinus Torvalds 1775863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 17761da177e4SLinus Torvalds return -EAGAIN; 17771da177e4SLinus Torvalds } 17781da177e4SLinus Torvalds 17791da177e4SLinus Torvalds /* 17801da177e4SLinus Torvalds * If we may have to abort operations to get out of this 17811da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 17821da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 17831da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 17841da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 17851da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 17861da177e4SLinus Torvalds * about for the moment. 17871da177e4SLinus Torvalds */ 17881da177e4SLinus Torvalds 178942faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 179042faad99SAl Viro sb->s_op->umount_begin(sb); 179142faad99SAl Viro } 17921da177e4SLinus Torvalds 17931da177e4SLinus Torvalds /* 17941da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 17951da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 17961da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 17971da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 17981da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 17991da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 18001da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 18011da177e4SLinus Torvalds */ 18021ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 18031da177e4SLinus Torvalds /* 18041da177e4SLinus Torvalds * Special case for "unmounting" root ... 18051da177e4SLinus Torvalds * we just try to remount it readonly. 18061da177e4SLinus Torvalds */ 1807bc6155d1SEric W. Biederman if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) 1808a1480dccSAndy Lutomirski return -EPERM; 18098d0347f6SDavid Howells return do_umount_root(sb); 18101da177e4SLinus Torvalds } 18111da177e4SLinus Torvalds 181297216be0SAl Viro namespace_lock(); 1813719ea2fbSAl Viro lock_mount_hash(); 18141da177e4SLinus Torvalds 181525d202edSEric W. Biederman /* Recheck MNT_LOCKED with the locks held */ 181625d202edSEric W. Biederman retval = -EINVAL; 181725d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 181825d202edSEric W. Biederman goto out; 181925d202edSEric W. Biederman 182025d202edSEric W. Biederman event++; 182148a066e7SAl Viro if (flags & MNT_DETACH) { 182248a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 1823e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE); 182448a066e7SAl Viro retval = 0; 182548a066e7SAl Viro } else { 1826b54b9be7SAl Viro shrink_submounts(mnt); 18271da177e4SLinus Torvalds retval = -EBUSY; 182848a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 18291a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1830e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 18311da177e4SLinus Torvalds retval = 0; 18321da177e4SLinus Torvalds } 183348a066e7SAl Viro } 183425d202edSEric W. Biederman out: 1835719ea2fbSAl Viro unlock_mount_hash(); 1836e3197d83SAl Viro namespace_unlock(); 18371da177e4SLinus Torvalds return retval; 18381da177e4SLinus Torvalds } 18391da177e4SLinus Torvalds 18401da177e4SLinus Torvalds /* 184180b5dce8SEric W. Biederman * __detach_mounts - lazily unmount all mounts on the specified dentry 184280b5dce8SEric W. Biederman * 184380b5dce8SEric W. Biederman * During unlink, rmdir, and d_drop it is possible to loose the path 184480b5dce8SEric W. Biederman * to an existing mountpoint, and wind up leaking the mount. 184580b5dce8SEric W. Biederman * detach_mounts allows lazily unmounting those mounts instead of 184680b5dce8SEric W. Biederman * leaking them. 184780b5dce8SEric W. Biederman * 184880b5dce8SEric W. Biederman * The caller may hold dentry->d_inode->i_mutex. 184980b5dce8SEric W. Biederman */ 185080b5dce8SEric W. Biederman void __detach_mounts(struct dentry *dentry) 185180b5dce8SEric W. Biederman { 185280b5dce8SEric W. Biederman struct mountpoint *mp; 185380b5dce8SEric W. Biederman struct mount *mnt; 185480b5dce8SEric W. Biederman 185580b5dce8SEric W. Biederman namespace_lock(); 18563895dbf8SEric W. Biederman lock_mount_hash(); 185780b5dce8SEric W. Biederman mp = lookup_mountpoint(dentry); 1858adc9b5c0SAl Viro if (!mp) 185980b5dce8SEric W. Biederman goto out_unlock; 186080b5dce8SEric W. Biederman 1861e06b933eSAndrey Ulanov event++; 186280b5dce8SEric W. Biederman while (!hlist_empty(&mp->m_list)) { 186380b5dce8SEric W. Biederman mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list); 1864ce07d891SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_UMOUNT) { 1865fe78fcc8SEric W. Biederman umount_mnt(mnt); 186656cbb429SAl Viro hlist_add_head(&mnt->mnt_umount, &unmounted); 1867ce07d891SEric W. Biederman } 1868e0c9c0afSEric W. Biederman else umount_tree(mnt, UMOUNT_CONNECTED); 186980b5dce8SEric W. Biederman } 187080b5dce8SEric W. Biederman put_mountpoint(mp); 187180b5dce8SEric W. Biederman out_unlock: 18723895dbf8SEric W. Biederman unlock_mount_hash(); 187380b5dce8SEric W. Biederman namespace_unlock(); 187480b5dce8SEric W. Biederman } 187580b5dce8SEric W. Biederman 187680b5dce8SEric W. Biederman /* 18779b40bc90SAl Viro * Is the caller allowed to modify his namespace? 18789b40bc90SAl Viro */ 1879a5f85d78SAl Viro bool may_mount(void) 18809b40bc90SAl Viro { 18819b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 18829b40bc90SAl Viro } 18839b40bc90SAl Viro 1884f7e33bdbSJeff Layton static void warn_mandlock(void) 18859e8925b6SJeff Layton { 1886f7e33bdbSJeff Layton pr_warn_once("=======================================================\n" 1887f7e33bdbSJeff Layton "WARNING: The mand mount option has been deprecated and\n" 1888f7e33bdbSJeff Layton " and is ignored by this kernel. Remove the mand\n" 1889f7e33bdbSJeff Layton " option from the mount to silence this warning.\n" 1890f7e33bdbSJeff Layton "=======================================================\n"); 18919e8925b6SJeff Layton } 18929e8925b6SJeff Layton 189325ccd24fSChristoph Hellwig static int can_umount(const struct path *path, int flags) 18941da177e4SLinus Torvalds { 189525ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 18961da177e4SLinus Torvalds 18979b40bc90SAl Viro if (!may_mount()) 18989b40bc90SAl Viro return -EPERM; 189941525f56SChristoph Hellwig if (path->dentry != path->mnt->mnt_root) 190025ccd24fSChristoph Hellwig return -EINVAL; 1901143c8c91SAl Viro if (!check_mnt(mnt)) 190225ccd24fSChristoph Hellwig return -EINVAL; 190325d202edSEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */ 190425ccd24fSChristoph Hellwig return -EINVAL; 1905b2f5d4dcSEric W. Biederman if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN)) 190625ccd24fSChristoph Hellwig return -EPERM; 190725ccd24fSChristoph Hellwig return 0; 190825ccd24fSChristoph Hellwig } 19091da177e4SLinus Torvalds 1910a0a6df9aSAl Viro // caller is responsible for flags being sane 191125ccd24fSChristoph Hellwig int path_umount(struct path *path, int flags) 191225ccd24fSChristoph Hellwig { 191325ccd24fSChristoph Hellwig struct mount *mnt = real_mount(path->mnt); 191425ccd24fSChristoph Hellwig int ret; 191525ccd24fSChristoph Hellwig 191625ccd24fSChristoph Hellwig ret = can_umount(path, flags); 191725ccd24fSChristoph Hellwig if (!ret) 191825ccd24fSChristoph Hellwig ret = do_umount(mnt, flags); 191925ccd24fSChristoph Hellwig 1920429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 192141525f56SChristoph Hellwig dput(path->dentry); 1922900148dcSAl Viro mntput_no_expire(mnt); 192325ccd24fSChristoph Hellwig return ret; 19241da177e4SLinus Torvalds } 19251da177e4SLinus Torvalds 192609267defSChristoph Hellwig static int ksys_umount(char __user *name, int flags) 192741525f56SChristoph Hellwig { 192841525f56SChristoph Hellwig int lookup_flags = LOOKUP_MOUNTPOINT; 192941525f56SChristoph Hellwig struct path path; 193041525f56SChristoph Hellwig int ret; 193141525f56SChristoph Hellwig 1932a0a6df9aSAl Viro // basic validity checks done first 1933a0a6df9aSAl Viro if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1934a0a6df9aSAl Viro return -EINVAL; 1935a0a6df9aSAl Viro 193641525f56SChristoph Hellwig if (!(flags & UMOUNT_NOFOLLOW)) 193741525f56SChristoph Hellwig lookup_flags |= LOOKUP_FOLLOW; 193841525f56SChristoph Hellwig ret = user_path_at(AT_FDCWD, name, lookup_flags, &path); 193941525f56SChristoph Hellwig if (ret) 194041525f56SChristoph Hellwig return ret; 194141525f56SChristoph Hellwig return path_umount(&path, flags); 194241525f56SChristoph Hellwig } 194341525f56SChristoph Hellwig 19443a18ef5cSDominik Brodowski SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 19453a18ef5cSDominik Brodowski { 19463a18ef5cSDominik Brodowski return ksys_umount(name, flags); 19473a18ef5cSDominik Brodowski } 19483a18ef5cSDominik Brodowski 19491da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 19501da177e4SLinus Torvalds 19511da177e4SLinus Torvalds /* 19521da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 19531da177e4SLinus Torvalds */ 1954bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 19551da177e4SLinus Torvalds { 19563a18ef5cSDominik Brodowski return ksys_umount(name, 0); 19571da177e4SLinus Torvalds } 19581da177e4SLinus Torvalds 19591da177e4SLinus Torvalds #endif 19601da177e4SLinus Torvalds 19614ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 19628823c079SEric W. Biederman { 19634ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 1964e149ed2bSAl Viro return dentry->d_op == &ns_dentry_operations && 1965e149ed2bSAl Viro dentry->d_fsdata == &mntns_operations; 19664ce5d2b1SEric W. Biederman } 19674ce5d2b1SEric W. Biederman 1968213921f9SEric Biggers static struct mnt_namespace *to_mnt_ns(struct ns_common *ns) 196958be2825SAl Viro { 197058be2825SAl Viro return container_of(ns, struct mnt_namespace, ns); 197158be2825SAl Viro } 197258be2825SAl Viro 1973303cc571SChristian Brauner struct ns_common *from_mnt_ns(struct mnt_namespace *mnt) 1974303cc571SChristian Brauner { 1975303cc571SChristian Brauner return &mnt->ns; 1976303cc571SChristian Brauner } 1977303cc571SChristian Brauner 19784ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 19794ce5d2b1SEric W. Biederman { 19804ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 19814ce5d2b1SEric W. Biederman * mount namespace loop? 19824ce5d2b1SEric W. Biederman */ 19834ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 19844ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 19854ce5d2b1SEric W. Biederman return false; 19864ce5d2b1SEric W. Biederman 1987f77c8014SAl Viro mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode)); 19888823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 19898823c079SEric W. Biederman } 19908823c079SEric W. Biederman 199187129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 199236341f64SRam Pai int flag) 19931da177e4SLinus Torvalds { 199484d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 19951da177e4SLinus Torvalds 19964ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 19974ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 19984ce5d2b1SEric W. Biederman 19994ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 2000be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 20019676f0c6SRam Pai 200236341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 2003be34d1a3SDavid Howells if (IS_ERR(q)) 2004be34d1a3SDavid Howells return q; 2005be34d1a3SDavid Howells 2006a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 20071da177e4SLinus Torvalds 20081da177e4SLinus Torvalds p = mnt; 20096b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 2010315fc83eSAl Viro struct mount *s; 20117ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 20121da177e4SLinus Torvalds continue; 20131da177e4SLinus Torvalds 2014909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 20154ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 20164ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 2017df7342b2SEric W. Biederman if (s->mnt.mnt_flags & MNT_LOCKED) { 2018df7342b2SEric W. Biederman /* Both unbindable and locked. */ 2019df7342b2SEric W. Biederman q = ERR_PTR(-EPERM); 2020df7342b2SEric W. Biederman goto out; 2021df7342b2SEric W. Biederman } else { 20224ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 20234ce5d2b1SEric W. Biederman continue; 20244ce5d2b1SEric W. Biederman } 2025df7342b2SEric W. Biederman } 20264ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 20274ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 20289676f0c6SRam Pai s = skip_mnt_tree(s); 20299676f0c6SRam Pai continue; 20309676f0c6SRam Pai } 20310714a533SAl Viro while (p != s->mnt_parent) { 20320714a533SAl Viro p = p->mnt_parent; 20330714a533SAl Viro q = q->mnt_parent; 20341da177e4SLinus Torvalds } 203587129cc0SAl Viro p = s; 203684d17192SAl Viro parent = q; 203787129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 2038be34d1a3SDavid Howells if (IS_ERR(q)) 2039be34d1a3SDavid Howells goto out; 2040719ea2fbSAl Viro lock_mount_hash(); 20411a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 20421064f874SEric W. Biederman attach_mnt(q, parent, p->mnt_mp); 2043719ea2fbSAl Viro unlock_mount_hash(); 20441da177e4SLinus Torvalds } 20451da177e4SLinus Torvalds } 20461da177e4SLinus Torvalds return res; 2047be34d1a3SDavid Howells out: 20481da177e4SLinus Torvalds if (res) { 2049719ea2fbSAl Viro lock_mount_hash(); 2050e819f152SEric W. Biederman umount_tree(res, UMOUNT_SYNC); 2051719ea2fbSAl Viro unlock_mount_hash(); 20521da177e4SLinus Torvalds } 2053be34d1a3SDavid Howells return q; 20541da177e4SLinus Torvalds } 20551da177e4SLinus Torvalds 2056be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 2057be34d1a3SDavid Howells 2058ca71cf71SAl Viro struct vfsmount *collect_mounts(const struct path *path) 20598aec0809SAl Viro { 2060cb338d06SAl Viro struct mount *tree; 206197216be0SAl Viro namespace_lock(); 2062cd4a4017SEric W. Biederman if (!check_mnt(real_mount(path->mnt))) 2063cd4a4017SEric W. Biederman tree = ERR_PTR(-EINVAL); 2064cd4a4017SEric W. Biederman else 206587129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 206687129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 2067328e6d90SAl Viro namespace_unlock(); 2068be34d1a3SDavid Howells if (IS_ERR(tree)) 206952e220d3SDan Carpenter return ERR_CAST(tree); 2070be34d1a3SDavid Howells return &tree->mnt; 20718aec0809SAl Viro } 20728aec0809SAl Viro 2073a07b2000SAl Viro static void free_mnt_ns(struct mnt_namespace *); 2074a07b2000SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool); 2075a07b2000SAl Viro 2076a07b2000SAl Viro void dissolve_on_fput(struct vfsmount *mnt) 2077a07b2000SAl Viro { 2078a07b2000SAl Viro struct mnt_namespace *ns; 2079a07b2000SAl Viro namespace_lock(); 2080a07b2000SAl Viro lock_mount_hash(); 2081a07b2000SAl Viro ns = real_mount(mnt)->mnt_ns; 208244dfd84aSDavid Howells if (ns) { 208344dfd84aSDavid Howells if (is_anon_ns(ns)) 2084a07b2000SAl Viro umount_tree(real_mount(mnt), UMOUNT_CONNECTED); 208544dfd84aSDavid Howells else 208644dfd84aSDavid Howells ns = NULL; 208744dfd84aSDavid Howells } 2088a07b2000SAl Viro unlock_mount_hash(); 2089a07b2000SAl Viro namespace_unlock(); 209044dfd84aSDavid Howells if (ns) 2091a07b2000SAl Viro free_mnt_ns(ns); 2092a07b2000SAl Viro } 2093a07b2000SAl Viro 20948aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 20958aec0809SAl Viro { 209697216be0SAl Viro namespace_lock(); 2097719ea2fbSAl Viro lock_mount_hash(); 20989c8e0a1bSEric W. Biederman umount_tree(real_mount(mnt), 0); 2099719ea2fbSAl Viro unlock_mount_hash(); 21003ab6abeeSAl Viro namespace_unlock(); 21018aec0809SAl Viro } 21028aec0809SAl Viro 2103427215d8SMiklos Szeredi static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 2104427215d8SMiklos Szeredi { 2105427215d8SMiklos Szeredi struct mount *child; 2106427215d8SMiklos Szeredi 2107427215d8SMiklos Szeredi list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 2108427215d8SMiklos Szeredi if (!is_subdir(child->mnt_mountpoint, dentry)) 2109427215d8SMiklos Szeredi continue; 2110427215d8SMiklos Szeredi 2111427215d8SMiklos Szeredi if (child->mnt.mnt_flags & MNT_LOCKED) 2112427215d8SMiklos Szeredi return true; 2113427215d8SMiklos Szeredi } 2114427215d8SMiklos Szeredi return false; 2115427215d8SMiklos Szeredi } 2116427215d8SMiklos Szeredi 2117c771d683SMiklos Szeredi /** 2118c771d683SMiklos Szeredi * clone_private_mount - create a private clone of a path 21191f287bc4SRandy Dunlap * @path: path to clone 2120c771d683SMiklos Szeredi * 21211f287bc4SRandy Dunlap * This creates a new vfsmount, which will be the clone of @path. The new mount 21221f287bc4SRandy Dunlap * will not be attached anywhere in the namespace and will be private (i.e. 21231f287bc4SRandy Dunlap * changes to the originating mount won't be propagated into this). 2124c771d683SMiklos Szeredi * 2125c771d683SMiklos Szeredi * Release with mntput(). 2126c771d683SMiklos Szeredi */ 2127ca71cf71SAl Viro struct vfsmount *clone_private_mount(const struct path *path) 2128c771d683SMiklos Szeredi { 2129c771d683SMiklos Szeredi struct mount *old_mnt = real_mount(path->mnt); 2130c771d683SMiklos Szeredi struct mount *new_mnt; 2131c771d683SMiklos Szeredi 2132427215d8SMiklos Szeredi down_read(&namespace_sem); 2133c771d683SMiklos Szeredi if (IS_MNT_UNBINDABLE(old_mnt)) 2134427215d8SMiklos Szeredi goto invalid; 2135427215d8SMiklos Szeredi 2136427215d8SMiklos Szeredi if (!check_mnt(old_mnt)) 2137427215d8SMiklos Szeredi goto invalid; 2138427215d8SMiklos Szeredi 2139427215d8SMiklos Szeredi if (has_locked_children(old_mnt, path->dentry)) 2140427215d8SMiklos Szeredi goto invalid; 2141c771d683SMiklos Szeredi 2142c771d683SMiklos Szeredi new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE); 2143427215d8SMiklos Szeredi up_read(&namespace_sem); 2144427215d8SMiklos Szeredi 2145c771d683SMiklos Szeredi if (IS_ERR(new_mnt)) 2146c771d683SMiklos Szeredi return ERR_CAST(new_mnt); 2147c771d683SMiklos Szeredi 2148df820f8dSMiklos Szeredi /* Longterm mount to be removed by kern_unmount*() */ 2149df820f8dSMiklos Szeredi new_mnt->mnt_ns = MNT_NS_INTERNAL; 2150df820f8dSMiklos Szeredi 2151c771d683SMiklos Szeredi return &new_mnt->mnt; 2152427215d8SMiklos Szeredi 2153427215d8SMiklos Szeredi invalid: 2154427215d8SMiklos Szeredi up_read(&namespace_sem); 2155427215d8SMiklos Szeredi return ERR_PTR(-EINVAL); 2156c771d683SMiklos Szeredi } 2157c771d683SMiklos Szeredi EXPORT_SYMBOL_GPL(clone_private_mount); 2158c771d683SMiklos Szeredi 21591f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 21601f707137SAl Viro struct vfsmount *root) 21611f707137SAl Viro { 21621a4eeaf2SAl Viro struct mount *mnt; 21631f707137SAl Viro int res = f(root, arg); 21641f707137SAl Viro if (res) 21651f707137SAl Viro return res; 21661a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 21671a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 21681f707137SAl Viro if (res) 21691f707137SAl Viro return res; 21701f707137SAl Viro } 21711f707137SAl Viro return 0; 21721f707137SAl Viro } 21731f707137SAl Viro 21743bd045ccSAl Viro static void lock_mnt_tree(struct mount *mnt) 21753bd045ccSAl Viro { 21763bd045ccSAl Viro struct mount *p; 21773bd045ccSAl Viro 21783bd045ccSAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 21793bd045ccSAl Viro int flags = p->mnt.mnt_flags; 21803bd045ccSAl Viro /* Don't allow unprivileged users to change mount flags */ 21813bd045ccSAl Viro flags |= MNT_LOCK_ATIME; 21823bd045ccSAl Viro 21833bd045ccSAl Viro if (flags & MNT_READONLY) 21843bd045ccSAl Viro flags |= MNT_LOCK_READONLY; 21853bd045ccSAl Viro 21863bd045ccSAl Viro if (flags & MNT_NODEV) 21873bd045ccSAl Viro flags |= MNT_LOCK_NODEV; 21883bd045ccSAl Viro 21893bd045ccSAl Viro if (flags & MNT_NOSUID) 21903bd045ccSAl Viro flags |= MNT_LOCK_NOSUID; 21913bd045ccSAl Viro 21923bd045ccSAl Viro if (flags & MNT_NOEXEC) 21933bd045ccSAl Viro flags |= MNT_LOCK_NOEXEC; 21943bd045ccSAl Viro /* Don't allow unprivileged users to reveal what is under a mount */ 21953bd045ccSAl Viro if (list_empty(&p->mnt_expire)) 21963bd045ccSAl Viro flags |= MNT_LOCKED; 21973bd045ccSAl Viro p->mnt.mnt_flags = flags; 21983bd045ccSAl Viro } 21993bd045ccSAl Viro } 22003bd045ccSAl Viro 22014b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 2202719f5d7fSMiklos Szeredi { 2203315fc83eSAl Viro struct mount *p; 2204719f5d7fSMiklos Szeredi 2205909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 2206fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 22074b8b21f4SAl Viro mnt_release_group_id(p); 2208719f5d7fSMiklos Szeredi } 2209719f5d7fSMiklos Szeredi } 2210719f5d7fSMiklos Szeredi 22114b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 2212719f5d7fSMiklos Szeredi { 2213315fc83eSAl Viro struct mount *p; 2214719f5d7fSMiklos Szeredi 2215909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 2216fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 22174b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 2218719f5d7fSMiklos Szeredi if (err) { 22194b8b21f4SAl Viro cleanup_group_ids(mnt, p); 2220719f5d7fSMiklos Szeredi return err; 2221719f5d7fSMiklos Szeredi } 2222719f5d7fSMiklos Szeredi } 2223719f5d7fSMiklos Szeredi } 2224719f5d7fSMiklos Szeredi 2225719f5d7fSMiklos Szeredi return 0; 2226719f5d7fSMiklos Szeredi } 2227719f5d7fSMiklos Szeredi 2228d2921684SEric W. Biederman int count_mounts(struct mnt_namespace *ns, struct mount *mnt) 2229d2921684SEric W. Biederman { 2230d2921684SEric W. Biederman unsigned int max = READ_ONCE(sysctl_mount_max); 2231124f75f8SAl Viro unsigned int mounts = 0; 2232d2921684SEric W. Biederman struct mount *p; 2233d2921684SEric W. Biederman 2234124f75f8SAl Viro if (ns->mounts >= max) 2235124f75f8SAl Viro return -ENOSPC; 2236124f75f8SAl Viro max -= ns->mounts; 2237124f75f8SAl Viro if (ns->pending_mounts >= max) 2238124f75f8SAl Viro return -ENOSPC; 2239124f75f8SAl Viro max -= ns->pending_mounts; 2240124f75f8SAl Viro 2241d2921684SEric W. Biederman for (p = mnt; p; p = next_mnt(p, mnt)) 2242d2921684SEric W. Biederman mounts++; 2243d2921684SEric W. Biederman 2244124f75f8SAl Viro if (mounts > max) 2245d2921684SEric W. Biederman return -ENOSPC; 2246d2921684SEric W. Biederman 2247124f75f8SAl Viro ns->pending_mounts += mounts; 2248d2921684SEric W. Biederman return 0; 2249d2921684SEric W. Biederman } 2250d2921684SEric W. Biederman 2251b90fa9aeSRam Pai /* 2252b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 2253b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 225421444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 225521444403SRam Pai * store the parent mount and mountpoint dentry. 225621444403SRam Pai * (done when source_mnt is moved) 2257b90fa9aeSRam Pai * 2258b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 2259b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 22609676f0c6SRam Pai * --------------------------------------------------------------------------- 2261b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 22629676f0c6SRam Pai * |************************************************************************** 22639676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 22649676f0c6SRam Pai * | dest | | | | | 22659676f0c6SRam Pai * | | | | | | | 22669676f0c6SRam Pai * | v | | | | | 22679676f0c6SRam Pai * |************************************************************************** 22689676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 22695afe0022SRam Pai * | | | | | | 22709676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 22719676f0c6SRam Pai * *************************************************************************** 2272b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 2273b90fa9aeSRam Pai * destination mount. 2274b90fa9aeSRam Pai * 2275b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 2276b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 2277b90fa9aeSRam Pai * the peer group of the source mount. 2278b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 2279b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 2280b90fa9aeSRam Pai * mount. 22815afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 22825afe0022SRam Pai * of the destination mount and the cloned mount is made slave 22835afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 22845afe0022SRam Pai * is marked as 'shared and slave'. 22855afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 22865afe0022SRam Pai * source mount. 22875afe0022SRam Pai * 22889676f0c6SRam Pai * --------------------------------------------------------------------------- 228921444403SRam Pai * | MOVE MOUNT OPERATION | 22909676f0c6SRam Pai * |************************************************************************** 22919676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 22929676f0c6SRam Pai * | dest | | | | | 22939676f0c6SRam Pai * | | | | | | | 22949676f0c6SRam Pai * | v | | | | | 22959676f0c6SRam Pai * |************************************************************************** 22969676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 22975afe0022SRam Pai * | | | | | | 22989676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 22999676f0c6SRam Pai * *************************************************************************** 23005afe0022SRam Pai * 23015afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 23025afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 230321444403SRam Pai * (+*) the mount is moved to the destination. 23045afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 23055afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 23065afe0022SRam Pai * the mount is marked as 'shared and slave'. 23075afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 2308b90fa9aeSRam Pai * 2309b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 2310b90fa9aeSRam Pai * applied to each mount in the tree. 2311b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 2312b90fa9aeSRam Pai * in allocations. 2313b90fa9aeSRam Pai */ 23140fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 231584d17192SAl Viro struct mount *dest_mnt, 231684d17192SAl Viro struct mountpoint *dest_mp, 23172763d119SAl Viro bool moving) 2318b90fa9aeSRam Pai { 23193bd045ccSAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 232038129a13SAl Viro HLIST_HEAD(tree_list); 2321d2921684SEric W. Biederman struct mnt_namespace *ns = dest_mnt->mnt_ns; 23221064f874SEric W. Biederman struct mountpoint *smp; 2323315fc83eSAl Viro struct mount *child, *p; 232438129a13SAl Viro struct hlist_node *n; 2325719f5d7fSMiklos Szeredi int err; 2326b90fa9aeSRam Pai 23271064f874SEric W. Biederman /* Preallocate a mountpoint in case the new mounts need 23281064f874SEric W. Biederman * to be tucked under other mounts. 23291064f874SEric W. Biederman */ 23301064f874SEric W. Biederman smp = get_mountpoint(source_mnt->mnt.mnt_root); 23311064f874SEric W. Biederman if (IS_ERR(smp)) 23321064f874SEric W. Biederman return PTR_ERR(smp); 23331064f874SEric W. Biederman 2334d2921684SEric W. Biederman /* Is there space to add these mounts to the mount namespace? */ 23352763d119SAl Viro if (!moving) { 2336d2921684SEric W. Biederman err = count_mounts(ns, source_mnt); 2337d2921684SEric W. Biederman if (err) 2338d2921684SEric W. Biederman goto out; 2339d2921684SEric W. Biederman } 2340d2921684SEric W. Biederman 2341fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 23420fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 2343719f5d7fSMiklos Szeredi if (err) 2344719f5d7fSMiklos Szeredi goto out; 234584d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 2346f2ebb3a9SAl Viro lock_mount_hash(); 2347719f5d7fSMiklos Szeredi if (err) 2348719f5d7fSMiklos Szeredi goto out_cleanup_ids; 2349909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 23500f0afb1dSAl Viro set_mnt_shared(p); 23510b1b901bSAl Viro } else { 23520b1b901bSAl Viro lock_mount_hash(); 2353b90fa9aeSRam Pai } 23542763d119SAl Viro if (moving) { 23552763d119SAl Viro unhash_mnt(source_mnt); 235684d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 2357143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 235821444403SRam Pai } else { 235944dfd84aSDavid Howells if (source_mnt->mnt_ns) { 236044dfd84aSDavid Howells /* move from anon - the caller will destroy */ 236144dfd84aSDavid Howells list_del_init(&source_mnt->mnt_ns->list); 236244dfd84aSDavid Howells } 236384d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 23641064f874SEric W. Biederman commit_tree(source_mnt); 236521444403SRam Pai } 2366b90fa9aeSRam Pai 236738129a13SAl Viro hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) { 23681d6a32acSAl Viro struct mount *q; 236938129a13SAl Viro hlist_del_init(&child->mnt_hash); 23701064f874SEric W. Biederman q = __lookup_mnt(&child->mnt_parent->mnt, 23711d6a32acSAl Viro child->mnt_mountpoint); 23721064f874SEric W. Biederman if (q) 23731064f874SEric W. Biederman mnt_change_mountpoint(child, smp, q); 23743bd045ccSAl Viro /* Notice when we are propagating across user namespaces */ 23753bd045ccSAl Viro if (child->mnt_parent->mnt_ns->user_ns != user_ns) 23763bd045ccSAl Viro lock_mnt_tree(child); 2377d728cf79SChristian Brauner child->mnt.mnt_flags &= ~MNT_LOCKED; 23781064f874SEric W. Biederman commit_tree(child); 2379b90fa9aeSRam Pai } 23801064f874SEric W. Biederman put_mountpoint(smp); 2381719ea2fbSAl Viro unlock_mount_hash(); 238299b7db7bSNick Piggin 2383b90fa9aeSRam Pai return 0; 2384719f5d7fSMiklos Szeredi 2385719f5d7fSMiklos Szeredi out_cleanup_ids: 2386f2ebb3a9SAl Viro while (!hlist_empty(&tree_list)) { 2387f2ebb3a9SAl Viro child = hlist_entry(tree_list.first, struct mount, mnt_hash); 2388d2921684SEric W. Biederman child->mnt_parent->mnt_ns->pending_mounts = 0; 2389e819f152SEric W. Biederman umount_tree(child, UMOUNT_SYNC); 2390f2ebb3a9SAl Viro } 2391f2ebb3a9SAl Viro unlock_mount_hash(); 23920fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 2393719f5d7fSMiklos Szeredi out: 2394d2921684SEric W. Biederman ns->pending_mounts = 0; 23951064f874SEric W. Biederman 23961064f874SEric W. Biederman read_seqlock_excl(&mount_lock); 23971064f874SEric W. Biederman put_mountpoint(smp); 23981064f874SEric W. Biederman read_sequnlock_excl(&mount_lock); 23991064f874SEric W. Biederman 2400719f5d7fSMiklos Szeredi return err; 2401b90fa9aeSRam Pai } 2402b90fa9aeSRam Pai 240384d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 2404b12cea91SAl Viro { 2405b12cea91SAl Viro struct vfsmount *mnt; 240684d17192SAl Viro struct dentry *dentry = path->dentry; 2407b12cea91SAl Viro retry: 24085955102cSAl Viro inode_lock(dentry->d_inode); 240984d17192SAl Viro if (unlikely(cant_mount(dentry))) { 24105955102cSAl Viro inode_unlock(dentry->d_inode); 241184d17192SAl Viro return ERR_PTR(-ENOENT); 2412b12cea91SAl Viro } 241397216be0SAl Viro namespace_lock(); 2414b12cea91SAl Viro mnt = lookup_mnt(path); 241584d17192SAl Viro if (likely(!mnt)) { 24163895dbf8SEric W. Biederman struct mountpoint *mp = get_mountpoint(dentry); 241784d17192SAl Viro if (IS_ERR(mp)) { 241897216be0SAl Viro namespace_unlock(); 24195955102cSAl Viro inode_unlock(dentry->d_inode); 242084d17192SAl Viro return mp; 242184d17192SAl Viro } 242284d17192SAl Viro return mp; 242384d17192SAl Viro } 242497216be0SAl Viro namespace_unlock(); 24255955102cSAl Viro inode_unlock(path->dentry->d_inode); 2426b12cea91SAl Viro path_put(path); 2427b12cea91SAl Viro path->mnt = mnt; 242884d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 2429b12cea91SAl Viro goto retry; 2430b12cea91SAl Viro } 2431b12cea91SAl Viro 243284d17192SAl Viro static void unlock_mount(struct mountpoint *where) 2433b12cea91SAl Viro { 243484d17192SAl Viro struct dentry *dentry = where->m_dentry; 24353895dbf8SEric W. Biederman 24363895dbf8SEric W. Biederman read_seqlock_excl(&mount_lock); 243784d17192SAl Viro put_mountpoint(where); 24383895dbf8SEric W. Biederman read_sequnlock_excl(&mount_lock); 24393895dbf8SEric W. Biederman 2440328e6d90SAl Viro namespace_unlock(); 24415955102cSAl Viro inode_unlock(dentry->d_inode); 2442b12cea91SAl Viro } 2443b12cea91SAl Viro 244484d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 24451da177e4SLinus Torvalds { 2446e462ec50SDavid Howells if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER) 24471da177e4SLinus Torvalds return -EINVAL; 24481da177e4SLinus Torvalds 2449e36cb0b8SDavid Howells if (d_is_dir(mp->m_dentry) != 2450e36cb0b8SDavid Howells d_is_dir(mnt->mnt.mnt_root)) 24511da177e4SLinus Torvalds return -ENOTDIR; 24521da177e4SLinus Torvalds 24532763d119SAl Viro return attach_recursive_mnt(mnt, p, mp, false); 24541da177e4SLinus Torvalds } 24551da177e4SLinus Torvalds 24561da177e4SLinus Torvalds /* 24577a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 24587a2e8a8fSValerie Aurora */ 24597a2e8a8fSValerie Aurora 2460e462ec50SDavid Howells static int flags_to_propagation_type(int ms_flags) 24617a2e8a8fSValerie Aurora { 2462e462ec50SDavid Howells int type = ms_flags & ~(MS_REC | MS_SILENT); 24637a2e8a8fSValerie Aurora 24647a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 24657a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 24667a2e8a8fSValerie Aurora return 0; 24677a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 24687a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 24697a2e8a8fSValerie Aurora return 0; 24707a2e8a8fSValerie Aurora return type; 24717a2e8a8fSValerie Aurora } 24727a2e8a8fSValerie Aurora 24737a2e8a8fSValerie Aurora /* 247407b20889SRam Pai * recursively change the type of the mountpoint. 247507b20889SRam Pai */ 2476e462ec50SDavid Howells static int do_change_type(struct path *path, int ms_flags) 247707b20889SRam Pai { 2478315fc83eSAl Viro struct mount *m; 24794b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 2480e462ec50SDavid Howells int recurse = ms_flags & MS_REC; 24817a2e8a8fSValerie Aurora int type; 2482719f5d7fSMiklos Szeredi int err = 0; 248307b20889SRam Pai 24842d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 248507b20889SRam Pai return -EINVAL; 248607b20889SRam Pai 2487e462ec50SDavid Howells type = flags_to_propagation_type(ms_flags); 24887a2e8a8fSValerie Aurora if (!type) 24897a2e8a8fSValerie Aurora return -EINVAL; 24907a2e8a8fSValerie Aurora 249197216be0SAl Viro namespace_lock(); 2492719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 2493719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 2494719f5d7fSMiklos Szeredi if (err) 2495719f5d7fSMiklos Szeredi goto out_unlock; 2496719f5d7fSMiklos Szeredi } 2497719f5d7fSMiklos Szeredi 2498719ea2fbSAl Viro lock_mount_hash(); 2499909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 25000f0afb1dSAl Viro change_mnt_propagation(m, type); 2501719ea2fbSAl Viro unlock_mount_hash(); 2502719f5d7fSMiklos Szeredi 2503719f5d7fSMiklos Szeredi out_unlock: 250497216be0SAl Viro namespace_unlock(); 2505719f5d7fSMiklos Szeredi return err; 250607b20889SRam Pai } 250707b20889SRam Pai 2508a07b2000SAl Viro static struct mount *__do_loopback(struct path *old_path, int recurse) 2509a07b2000SAl Viro { 2510a07b2000SAl Viro struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt); 2511a07b2000SAl Viro 2512a07b2000SAl Viro if (IS_MNT_UNBINDABLE(old)) 2513a07b2000SAl Viro return mnt; 2514a07b2000SAl Viro 2515a07b2000SAl Viro if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations) 2516a07b2000SAl Viro return mnt; 2517a07b2000SAl Viro 2518a07b2000SAl Viro if (!recurse && has_locked_children(old, old_path->dentry)) 2519a07b2000SAl Viro return mnt; 2520a07b2000SAl Viro 2521a07b2000SAl Viro if (recurse) 2522a07b2000SAl Viro mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE); 2523a07b2000SAl Viro else 2524a07b2000SAl Viro mnt = clone_mnt(old, old_path->dentry, 0); 2525a07b2000SAl Viro 2526a07b2000SAl Viro if (!IS_ERR(mnt)) 2527a07b2000SAl Viro mnt->mnt.mnt_flags &= ~MNT_LOCKED; 2528a07b2000SAl Viro 2529a07b2000SAl Viro return mnt; 2530a07b2000SAl Viro } 2531a07b2000SAl Viro 253207b20889SRam Pai /* 25331da177e4SLinus Torvalds * do loopback mount. 25341da177e4SLinus Torvalds */ 2535808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 25362dafe1c4SEric Sandeen int recurse) 25371da177e4SLinus Torvalds { 25382d92ab3cSAl Viro struct path old_path; 2539a07b2000SAl Viro struct mount *mnt = NULL, *parent; 254084d17192SAl Viro struct mountpoint *mp; 254157eccb83SAl Viro int err; 25421da177e4SLinus Torvalds if (!old_name || !*old_name) 25431da177e4SLinus Torvalds return -EINVAL; 2544815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 25451da177e4SLinus Torvalds if (err) 25461da177e4SLinus Torvalds return err; 25471da177e4SLinus Torvalds 25488823c079SEric W. Biederman err = -EINVAL; 25494ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 25508823c079SEric W. Biederman goto out; 25518823c079SEric W. Biederman 255284d17192SAl Viro mp = lock_mount(path); 2553a07b2000SAl Viro if (IS_ERR(mp)) { 255484d17192SAl Viro err = PTR_ERR(mp); 25559676f0c6SRam Pai goto out; 2556a07b2000SAl Viro } 25579676f0c6SRam Pai 255884d17192SAl Viro parent = real_mount(path->mnt); 2559e149ed2bSAl Viro if (!check_mnt(parent)) 2560e149ed2bSAl Viro goto out2; 2561e149ed2bSAl Viro 2562a07b2000SAl Viro mnt = __do_loopback(&old_path, recurse); 2563be34d1a3SDavid Howells if (IS_ERR(mnt)) { 2564be34d1a3SDavid Howells err = PTR_ERR(mnt); 2565e9c5d8a5SAndrey Vagin goto out2; 2566be34d1a3SDavid Howells } 2567ccd48bc7SAl Viro 256884d17192SAl Viro err = graft_tree(mnt, parent, mp); 25691da177e4SLinus Torvalds if (err) { 2570719ea2fbSAl Viro lock_mount_hash(); 2571e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_SYNC); 2572719ea2fbSAl Viro unlock_mount_hash(); 25735b83d2c5SRam Pai } 2574b12cea91SAl Viro out2: 257584d17192SAl Viro unlock_mount(mp); 2576ccd48bc7SAl Viro out: 25772d92ab3cSAl Viro path_put(&old_path); 25781da177e4SLinus Torvalds return err; 25791da177e4SLinus Torvalds } 25801da177e4SLinus Torvalds 2581a07b2000SAl Viro static struct file *open_detached_copy(struct path *path, bool recursive) 2582a07b2000SAl Viro { 2583a07b2000SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 2584a07b2000SAl Viro struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true); 2585a07b2000SAl Viro struct mount *mnt, *p; 2586a07b2000SAl Viro struct file *file; 2587a07b2000SAl Viro 2588a07b2000SAl Viro if (IS_ERR(ns)) 2589a07b2000SAl Viro return ERR_CAST(ns); 2590a07b2000SAl Viro 2591a07b2000SAl Viro namespace_lock(); 2592a07b2000SAl Viro mnt = __do_loopback(path, recursive); 2593a07b2000SAl Viro if (IS_ERR(mnt)) { 2594a07b2000SAl Viro namespace_unlock(); 2595a07b2000SAl Viro free_mnt_ns(ns); 2596a07b2000SAl Viro return ERR_CAST(mnt); 2597a07b2000SAl Viro } 2598a07b2000SAl Viro 2599a07b2000SAl Viro lock_mount_hash(); 2600a07b2000SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2601a07b2000SAl Viro p->mnt_ns = ns; 2602a07b2000SAl Viro ns->mounts++; 2603a07b2000SAl Viro } 2604a07b2000SAl Viro ns->root = mnt; 2605a07b2000SAl Viro list_add_tail(&ns->list, &mnt->mnt_list); 2606a07b2000SAl Viro mntget(&mnt->mnt); 2607a07b2000SAl Viro unlock_mount_hash(); 2608a07b2000SAl Viro namespace_unlock(); 2609a07b2000SAl Viro 2610a07b2000SAl Viro mntput(path->mnt); 2611a07b2000SAl Viro path->mnt = &mnt->mnt; 2612a07b2000SAl Viro file = dentry_open(path, O_PATH, current_cred()); 2613a07b2000SAl Viro if (IS_ERR(file)) 2614a07b2000SAl Viro dissolve_on_fput(path->mnt); 2615a07b2000SAl Viro else 2616a07b2000SAl Viro file->f_mode |= FMODE_NEED_UNMOUNT; 2617a07b2000SAl Viro return file; 2618a07b2000SAl Viro } 2619a07b2000SAl Viro 26202658ce09SBen Dooks SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags) 2621a07b2000SAl Viro { 2622a07b2000SAl Viro struct file *file; 2623a07b2000SAl Viro struct path path; 2624a07b2000SAl Viro int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 2625a07b2000SAl Viro bool detached = flags & OPEN_TREE_CLONE; 2626a07b2000SAl Viro int error; 2627a07b2000SAl Viro int fd; 2628a07b2000SAl Viro 2629a07b2000SAl Viro BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC); 2630a07b2000SAl Viro 2631a07b2000SAl Viro if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE | 2632a07b2000SAl Viro AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE | 2633a07b2000SAl Viro OPEN_TREE_CLOEXEC)) 2634a07b2000SAl Viro return -EINVAL; 2635a07b2000SAl Viro 2636a07b2000SAl Viro if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE) 2637a07b2000SAl Viro return -EINVAL; 2638a07b2000SAl Viro 2639a07b2000SAl Viro if (flags & AT_NO_AUTOMOUNT) 2640a07b2000SAl Viro lookup_flags &= ~LOOKUP_AUTOMOUNT; 2641a07b2000SAl Viro if (flags & AT_SYMLINK_NOFOLLOW) 2642a07b2000SAl Viro lookup_flags &= ~LOOKUP_FOLLOW; 2643a07b2000SAl Viro if (flags & AT_EMPTY_PATH) 2644a07b2000SAl Viro lookup_flags |= LOOKUP_EMPTY; 2645a07b2000SAl Viro 2646a07b2000SAl Viro if (detached && !may_mount()) 2647a07b2000SAl Viro return -EPERM; 2648a07b2000SAl Viro 2649a07b2000SAl Viro fd = get_unused_fd_flags(flags & O_CLOEXEC); 2650a07b2000SAl Viro if (fd < 0) 2651a07b2000SAl Viro return fd; 2652a07b2000SAl Viro 2653a07b2000SAl Viro error = user_path_at(dfd, filename, lookup_flags, &path); 2654a07b2000SAl Viro if (unlikely(error)) { 2655a07b2000SAl Viro file = ERR_PTR(error); 2656a07b2000SAl Viro } else { 2657a07b2000SAl Viro if (detached) 2658a07b2000SAl Viro file = open_detached_copy(&path, flags & AT_RECURSIVE); 2659a07b2000SAl Viro else 2660a07b2000SAl Viro file = dentry_open(&path, O_PATH, current_cred()); 2661a07b2000SAl Viro path_put(&path); 2662a07b2000SAl Viro } 2663a07b2000SAl Viro if (IS_ERR(file)) { 2664a07b2000SAl Viro put_unused_fd(fd); 2665a07b2000SAl Viro return PTR_ERR(file); 2666a07b2000SAl Viro } 2667a07b2000SAl Viro fd_install(fd, file); 2668a07b2000SAl Viro return fd; 2669a07b2000SAl Viro } 2670a07b2000SAl Viro 267143f5e655SDavid Howells /* 267243f5e655SDavid Howells * Don't allow locked mount flags to be cleared. 267343f5e655SDavid Howells * 267443f5e655SDavid Howells * No locks need to be held here while testing the various MNT_LOCK 267543f5e655SDavid Howells * flags because those flags can never be cleared once they are set. 267643f5e655SDavid Howells */ 267743f5e655SDavid Howells static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags) 26782e4b7fcdSDave Hansen { 267943f5e655SDavid Howells unsigned int fl = mnt->mnt.mnt_flags; 26802e4b7fcdSDave Hansen 268143f5e655SDavid Howells if ((fl & MNT_LOCK_READONLY) && 268243f5e655SDavid Howells !(mnt_flags & MNT_READONLY)) 268343f5e655SDavid Howells return false; 268443f5e655SDavid Howells 268543f5e655SDavid Howells if ((fl & MNT_LOCK_NODEV) && 268643f5e655SDavid Howells !(mnt_flags & MNT_NODEV)) 268743f5e655SDavid Howells return false; 268843f5e655SDavid Howells 268943f5e655SDavid Howells if ((fl & MNT_LOCK_NOSUID) && 269043f5e655SDavid Howells !(mnt_flags & MNT_NOSUID)) 269143f5e655SDavid Howells return false; 269243f5e655SDavid Howells 269343f5e655SDavid Howells if ((fl & MNT_LOCK_NOEXEC) && 269443f5e655SDavid Howells !(mnt_flags & MNT_NOEXEC)) 269543f5e655SDavid Howells return false; 269643f5e655SDavid Howells 269743f5e655SDavid Howells if ((fl & MNT_LOCK_ATIME) && 269843f5e655SDavid Howells ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK))) 269943f5e655SDavid Howells return false; 270043f5e655SDavid Howells 270143f5e655SDavid Howells return true; 270243f5e655SDavid Howells } 270343f5e655SDavid Howells 270443f5e655SDavid Howells static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags) 270543f5e655SDavid Howells { 270643f5e655SDavid Howells bool readonly_request = (mnt_flags & MNT_READONLY); 270743f5e655SDavid Howells 270843f5e655SDavid Howells if (readonly_request == __mnt_is_readonly(&mnt->mnt)) 27092e4b7fcdSDave Hansen return 0; 27102e4b7fcdSDave Hansen 27112e4b7fcdSDave Hansen if (readonly_request) 271243f5e655SDavid Howells return mnt_make_readonly(mnt); 271343f5e655SDavid Howells 271468847c94SChristian Brauner mnt->mnt.mnt_flags &= ~MNT_READONLY; 271568847c94SChristian Brauner return 0; 271643f5e655SDavid Howells } 271743f5e655SDavid Howells 271843f5e655SDavid Howells static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags) 271943f5e655SDavid Howells { 272043f5e655SDavid Howells mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK; 272143f5e655SDavid Howells mnt->mnt.mnt_flags = mnt_flags; 272243f5e655SDavid Howells touch_mnt_namespace(mnt->mnt_ns); 272343f5e655SDavid Howells } 272443f5e655SDavid Howells 2725f8b92ba6SDeepa Dinamani static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt) 2726f8b92ba6SDeepa Dinamani { 2727f8b92ba6SDeepa Dinamani struct super_block *sb = mnt->mnt_sb; 2728f8b92ba6SDeepa Dinamani 2729f8b92ba6SDeepa Dinamani if (!__mnt_is_readonly(mnt) && 2730a128b054SAnthony Iliopoulos (!(sb->s_iflags & SB_I_TS_EXPIRY_WARNED)) && 2731f8b92ba6SDeepa Dinamani (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) { 2732f8b92ba6SDeepa Dinamani char *buf = (char *)__get_free_page(GFP_KERNEL); 2733f8b92ba6SDeepa Dinamani char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM); 2734f8b92ba6SDeepa Dinamani struct tm tm; 2735f8b92ba6SDeepa Dinamani 2736f8b92ba6SDeepa Dinamani time64_to_tm(sb->s_time_max, 0, &tm); 2737f8b92ba6SDeepa Dinamani 27380ecee669SEric Biggers pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n", 27390ecee669SEric Biggers sb->s_type->name, 27400ecee669SEric Biggers is_mounted(mnt) ? "remounted" : "mounted", 27410ecee669SEric Biggers mntpath, 2742f8b92ba6SDeepa Dinamani tm.tm_year+1900, (unsigned long long)sb->s_time_max); 2743f8b92ba6SDeepa Dinamani 2744f8b92ba6SDeepa Dinamani free_page((unsigned long)buf); 2745a128b054SAnthony Iliopoulos sb->s_iflags |= SB_I_TS_EXPIRY_WARNED; 2746f8b92ba6SDeepa Dinamani } 2747f8b92ba6SDeepa Dinamani } 2748f8b92ba6SDeepa Dinamani 274943f5e655SDavid Howells /* 275043f5e655SDavid Howells * Handle reconfiguration of the mountpoint only without alteration of the 275143f5e655SDavid Howells * superblock it refers to. This is triggered by specifying MS_REMOUNT|MS_BIND 275243f5e655SDavid Howells * to mount(2). 275343f5e655SDavid Howells */ 275443f5e655SDavid Howells static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags) 275543f5e655SDavid Howells { 275643f5e655SDavid Howells struct super_block *sb = path->mnt->mnt_sb; 275743f5e655SDavid Howells struct mount *mnt = real_mount(path->mnt); 275843f5e655SDavid Howells int ret; 275943f5e655SDavid Howells 276043f5e655SDavid Howells if (!check_mnt(mnt)) 276143f5e655SDavid Howells return -EINVAL; 276243f5e655SDavid Howells 276343f5e655SDavid Howells if (path->dentry != mnt->mnt.mnt_root) 276443f5e655SDavid Howells return -EINVAL; 276543f5e655SDavid Howells 276643f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 276743f5e655SDavid Howells return -EPERM; 276843f5e655SDavid Howells 2769e58ace1aSChristian Brauner /* 2770e58ace1aSChristian Brauner * We're only checking whether the superblock is read-only not 2771e58ace1aSChristian Brauner * changing it, so only take down_read(&sb->s_umount). 2772e58ace1aSChristian Brauner */ 2773e58ace1aSChristian Brauner down_read(&sb->s_umount); 277468847c94SChristian Brauner lock_mount_hash(); 277543f5e655SDavid Howells ret = change_mount_ro_state(mnt, mnt_flags); 277643f5e655SDavid Howells if (ret == 0) 277743f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 277868847c94SChristian Brauner unlock_mount_hash(); 2779e58ace1aSChristian Brauner up_read(&sb->s_umount); 2780f8b92ba6SDeepa Dinamani 2781f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2782f8b92ba6SDeepa Dinamani 278343f5e655SDavid Howells return ret; 27842e4b7fcdSDave Hansen } 27852e4b7fcdSDave Hansen 27861da177e4SLinus Torvalds /* 27871da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 27881da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 27891da177e4SLinus Torvalds * on it - tough luck. 27901da177e4SLinus Torvalds */ 2791e462ec50SDavid Howells static int do_remount(struct path *path, int ms_flags, int sb_flags, 2792e462ec50SDavid Howells int mnt_flags, void *data) 27931da177e4SLinus Torvalds { 27941da177e4SLinus Torvalds int err; 27952d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 2796143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 27978d0347f6SDavid Howells struct fs_context *fc; 27981da177e4SLinus Torvalds 2799143c8c91SAl Viro if (!check_mnt(mnt)) 28001da177e4SLinus Torvalds return -EINVAL; 28011da177e4SLinus Torvalds 28022d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 28031da177e4SLinus Torvalds return -EINVAL; 28041da177e4SLinus Torvalds 280543f5e655SDavid Howells if (!can_change_locked_flags(mnt, mnt_flags)) 280607b64558SEric W. Biederman return -EPERM; 28079566d674SEric W. Biederman 28088d0347f6SDavid Howells fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK); 28098d0347f6SDavid Howells if (IS_ERR(fc)) 28108d0347f6SDavid Howells return PTR_ERR(fc); 2811ff36fe2cSEric Paris 2812b330966fSMiklos Szeredi fc->oldapi = true; 28138d0347f6SDavid Howells err = parse_monolithic_mount_data(fc, data); 28148d0347f6SDavid Howells if (!err) { 28151da177e4SLinus Torvalds down_write(&sb->s_umount); 281657eccb83SAl Viro err = -EPERM; 281743f5e655SDavid Howells if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) { 28188d0347f6SDavid Howells err = reconfigure_super(fc); 281968847c94SChristian Brauner if (!err) { 282068847c94SChristian Brauner lock_mount_hash(); 282143f5e655SDavid Howells set_mount_attributes(mnt, mnt_flags); 282268847c94SChristian Brauner unlock_mount_hash(); 282368847c94SChristian Brauner } 28240e55a7ccSDan Williams } 28256339dab8SAl Viro up_write(&sb->s_umount); 28268d0347f6SDavid Howells } 2827f8b92ba6SDeepa Dinamani 2828f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(path, &mnt->mnt); 2829f8b92ba6SDeepa Dinamani 28308d0347f6SDavid Howells put_fs_context(fc); 28311da177e4SLinus Torvalds return err; 28321da177e4SLinus Torvalds } 28331da177e4SLinus Torvalds 2834cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 28359676f0c6SRam Pai { 2836315fc83eSAl Viro struct mount *p; 2837909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 2838fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 28399676f0c6SRam Pai return 1; 28409676f0c6SRam Pai } 28419676f0c6SRam Pai return 0; 28429676f0c6SRam Pai } 28439676f0c6SRam Pai 284444dfd84aSDavid Howells /* 284544dfd84aSDavid Howells * Check that there aren't references to earlier/same mount namespaces in the 284644dfd84aSDavid Howells * specified subtree. Such references can act as pins for mount namespaces 284744dfd84aSDavid Howells * that aren't checked by the mount-cycle checking code, thereby allowing 284844dfd84aSDavid Howells * cycles to be made. 284944dfd84aSDavid Howells */ 285044dfd84aSDavid Howells static bool check_for_nsfs_mounts(struct mount *subtree) 285144dfd84aSDavid Howells { 285244dfd84aSDavid Howells struct mount *p; 285344dfd84aSDavid Howells bool ret = false; 285444dfd84aSDavid Howells 285544dfd84aSDavid Howells lock_mount_hash(); 285644dfd84aSDavid Howells for (p = subtree; p; p = next_mnt(p, subtree)) 285744dfd84aSDavid Howells if (mnt_ns_loop(p->mnt.mnt_root)) 285844dfd84aSDavid Howells goto out; 285944dfd84aSDavid Howells 286044dfd84aSDavid Howells ret = true; 286144dfd84aSDavid Howells out: 286244dfd84aSDavid Howells unlock_mount_hash(); 286344dfd84aSDavid Howells return ret; 286444dfd84aSDavid Howells } 286544dfd84aSDavid Howells 28669ffb14efSPavel Tikhomirov static int do_set_group(struct path *from_path, struct path *to_path) 28679ffb14efSPavel Tikhomirov { 28689ffb14efSPavel Tikhomirov struct mount *from, *to; 28699ffb14efSPavel Tikhomirov int err; 28709ffb14efSPavel Tikhomirov 28719ffb14efSPavel Tikhomirov from = real_mount(from_path->mnt); 28729ffb14efSPavel Tikhomirov to = real_mount(to_path->mnt); 28739ffb14efSPavel Tikhomirov 28749ffb14efSPavel Tikhomirov namespace_lock(); 28759ffb14efSPavel Tikhomirov 28769ffb14efSPavel Tikhomirov err = -EINVAL; 28779ffb14efSPavel Tikhomirov /* To and From must be mounted */ 28789ffb14efSPavel Tikhomirov if (!is_mounted(&from->mnt)) 28799ffb14efSPavel Tikhomirov goto out; 28809ffb14efSPavel Tikhomirov if (!is_mounted(&to->mnt)) 28819ffb14efSPavel Tikhomirov goto out; 28829ffb14efSPavel Tikhomirov 28839ffb14efSPavel Tikhomirov err = -EPERM; 28849ffb14efSPavel Tikhomirov /* We should be allowed to modify mount namespaces of both mounts */ 28859ffb14efSPavel Tikhomirov if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN)) 28869ffb14efSPavel Tikhomirov goto out; 28879ffb14efSPavel Tikhomirov if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN)) 28889ffb14efSPavel Tikhomirov goto out; 28899ffb14efSPavel Tikhomirov 28909ffb14efSPavel Tikhomirov err = -EINVAL; 28919ffb14efSPavel Tikhomirov /* To and From paths should be mount roots */ 28929ffb14efSPavel Tikhomirov if (from_path->dentry != from_path->mnt->mnt_root) 28939ffb14efSPavel Tikhomirov goto out; 28949ffb14efSPavel Tikhomirov if (to_path->dentry != to_path->mnt->mnt_root) 28959ffb14efSPavel Tikhomirov goto out; 28969ffb14efSPavel Tikhomirov 28979ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed across same superblock */ 28989ffb14efSPavel Tikhomirov if (from->mnt.mnt_sb != to->mnt.mnt_sb) 28999ffb14efSPavel Tikhomirov goto out; 29009ffb14efSPavel Tikhomirov 29019ffb14efSPavel Tikhomirov /* From mount root should be wider than To mount root */ 29029ffb14efSPavel Tikhomirov if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root)) 29039ffb14efSPavel Tikhomirov goto out; 29049ffb14efSPavel Tikhomirov 29059ffb14efSPavel Tikhomirov /* From mount should not have locked children in place of To's root */ 29069ffb14efSPavel Tikhomirov if (has_locked_children(from, to->mnt.mnt_root)) 29079ffb14efSPavel Tikhomirov goto out; 29089ffb14efSPavel Tikhomirov 29099ffb14efSPavel Tikhomirov /* Setting sharing groups is only allowed on private mounts */ 29109ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to)) 29119ffb14efSPavel Tikhomirov goto out; 29129ffb14efSPavel Tikhomirov 29139ffb14efSPavel Tikhomirov /* From should not be private */ 29149ffb14efSPavel Tikhomirov if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from)) 29159ffb14efSPavel Tikhomirov goto out; 29169ffb14efSPavel Tikhomirov 29179ffb14efSPavel Tikhomirov if (IS_MNT_SLAVE(from)) { 29189ffb14efSPavel Tikhomirov struct mount *m = from->mnt_master; 29199ffb14efSPavel Tikhomirov 29209ffb14efSPavel Tikhomirov list_add(&to->mnt_slave, &m->mnt_slave_list); 29219ffb14efSPavel Tikhomirov to->mnt_master = m; 29229ffb14efSPavel Tikhomirov } 29239ffb14efSPavel Tikhomirov 29249ffb14efSPavel Tikhomirov if (IS_MNT_SHARED(from)) { 29259ffb14efSPavel Tikhomirov to->mnt_group_id = from->mnt_group_id; 29269ffb14efSPavel Tikhomirov list_add(&to->mnt_share, &from->mnt_share); 29279ffb14efSPavel Tikhomirov lock_mount_hash(); 29289ffb14efSPavel Tikhomirov set_mnt_shared(to); 29299ffb14efSPavel Tikhomirov unlock_mount_hash(); 29309ffb14efSPavel Tikhomirov } 29319ffb14efSPavel Tikhomirov 29329ffb14efSPavel Tikhomirov err = 0; 29339ffb14efSPavel Tikhomirov out: 29349ffb14efSPavel Tikhomirov namespace_unlock(); 29359ffb14efSPavel Tikhomirov return err; 29369ffb14efSPavel Tikhomirov } 29379ffb14efSPavel Tikhomirov 29382db154b3SDavid Howells static int do_move_mount(struct path *old_path, struct path *new_path) 29391da177e4SLinus Torvalds { 294044dfd84aSDavid Howells struct mnt_namespace *ns; 2941676da58dSAl Viro struct mount *p; 29420fb54e50SAl Viro struct mount *old; 29432763d119SAl Viro struct mount *parent; 29442763d119SAl Viro struct mountpoint *mp, *old_mp; 294557eccb83SAl Viro int err; 294644dfd84aSDavid Howells bool attached; 29471da177e4SLinus Torvalds 29482db154b3SDavid Howells mp = lock_mount(new_path); 294984d17192SAl Viro if (IS_ERR(mp)) 29502db154b3SDavid Howells return PTR_ERR(mp); 2951cc53ce53SDavid Howells 29522db154b3SDavid Howells old = real_mount(old_path->mnt); 29532db154b3SDavid Howells p = real_mount(new_path->mnt); 29542763d119SAl Viro parent = old->mnt_parent; 295544dfd84aSDavid Howells attached = mnt_has_parent(old); 29562763d119SAl Viro old_mp = old->mnt_mp; 295744dfd84aSDavid Howells ns = old->mnt_ns; 2958143c8c91SAl Viro 29591da177e4SLinus Torvalds err = -EINVAL; 296044dfd84aSDavid Howells /* The mountpoint must be in our namespace. */ 296144dfd84aSDavid Howells if (!check_mnt(p)) 29622db154b3SDavid Howells goto out; 29631da177e4SLinus Torvalds 2964570d7a98SEric Biggers /* The thing moved must be mounted... */ 2965570d7a98SEric Biggers if (!is_mounted(&old->mnt)) 296644dfd84aSDavid Howells goto out; 296744dfd84aSDavid Howells 2968570d7a98SEric Biggers /* ... and either ours or the root of anon namespace */ 2969570d7a98SEric Biggers if (!(attached ? check_mnt(old) : is_anon_ns(ns))) 29702db154b3SDavid Howells goto out; 29711da177e4SLinus Torvalds 29722db154b3SDavid Howells if (old->mnt.mnt_flags & MNT_LOCKED) 29732db154b3SDavid Howells goto out; 29742db154b3SDavid Howells 29752db154b3SDavid Howells if (old_path->dentry != old_path->mnt->mnt_root) 29762db154b3SDavid Howells goto out; 29772db154b3SDavid Howells 29782db154b3SDavid Howells if (d_is_dir(new_path->dentry) != 29792db154b3SDavid Howells d_is_dir(old_path->dentry)) 29802db154b3SDavid Howells goto out; 298121444403SRam Pai /* 298221444403SRam Pai * Don't move a mount residing in a shared parent. 298321444403SRam Pai */ 29842763d119SAl Viro if (attached && IS_MNT_SHARED(parent)) 29852db154b3SDavid Howells goto out; 29869676f0c6SRam Pai /* 29879676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 29889676f0c6SRam Pai * mount which is shared. 29899676f0c6SRam Pai */ 2990fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 29912db154b3SDavid Howells goto out; 29921da177e4SLinus Torvalds err = -ELOOP; 299344dfd84aSDavid Howells if (!check_for_nsfs_mounts(old)) 299444dfd84aSDavid Howells goto out; 2995fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 2996676da58dSAl Viro if (p == old) 29972db154b3SDavid Howells goto out; 29981da177e4SLinus Torvalds 29992db154b3SDavid Howells err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp, 30002763d119SAl Viro attached); 30014ac91378SJan Blunck if (err) 30022db154b3SDavid Howells goto out; 30031da177e4SLinus Torvalds 30041da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 30051da177e4SLinus Torvalds * automatically */ 30066776db3dSAl Viro list_del_init(&old->mnt_expire); 30072763d119SAl Viro if (attached) 30082763d119SAl Viro put_mountpoint(old_mp); 30091da177e4SLinus Torvalds out: 30102db154b3SDavid Howells unlock_mount(mp); 301144dfd84aSDavid Howells if (!err) { 30122763d119SAl Viro if (attached) 30132763d119SAl Viro mntput_no_expire(parent); 30142763d119SAl Viro else 301544dfd84aSDavid Howells free_mnt_ns(ns); 301644dfd84aSDavid Howells } 30172db154b3SDavid Howells return err; 30182db154b3SDavid Howells } 30192db154b3SDavid Howells 30202db154b3SDavid Howells static int do_move_mount_old(struct path *path, const char *old_name) 30212db154b3SDavid Howells { 30222db154b3SDavid Howells struct path old_path; 30232db154b3SDavid Howells int err; 30242db154b3SDavid Howells 30252db154b3SDavid Howells if (!old_name || !*old_name) 30262db154b3SDavid Howells return -EINVAL; 30272db154b3SDavid Howells 30282db154b3SDavid Howells err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 30292db154b3SDavid Howells if (err) 30302db154b3SDavid Howells return err; 30312db154b3SDavid Howells 30322db154b3SDavid Howells err = do_move_mount(&old_path, path); 30332d92ab3cSAl Viro path_put(&old_path); 30341da177e4SLinus Torvalds return err; 30351da177e4SLinus Torvalds } 30361da177e4SLinus Torvalds 30379d412a43SAl Viro /* 30389d412a43SAl Viro * add a mount into a namespace's mount tree 30399d412a43SAl Viro */ 30408f11538eSAl Viro static int do_add_mount(struct mount *newmnt, struct mountpoint *mp, 30411e2d8464SAl Viro const struct path *path, int mnt_flags) 30429d412a43SAl Viro { 30438f11538eSAl Viro struct mount *parent = real_mount(path->mnt); 30449d412a43SAl Viro 3045f2ebb3a9SAl Viro mnt_flags &= ~MNT_INTERNAL_FLAGS; 30469d412a43SAl Viro 304784d17192SAl Viro if (unlikely(!check_mnt(parent))) { 3048156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 3049156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 30508f11538eSAl Viro return -EINVAL; 3051156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 305284d17192SAl Viro if (!parent->mnt_ns) 30538f11538eSAl Viro return -EINVAL; 3054156cacb1SAl Viro } 30559d412a43SAl Viro 30569d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 305795bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 30589d412a43SAl Viro path->mnt->mnt_root == path->dentry) 30598f11538eSAl Viro return -EBUSY; 30609d412a43SAl Viro 3061e36cb0b8SDavid Howells if (d_is_symlink(newmnt->mnt.mnt_root)) 30628f11538eSAl Viro return -EINVAL; 30639d412a43SAl Viro 306495bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 30658f11538eSAl Viro return graft_tree(newmnt, parent, mp); 30669d412a43SAl Viro } 3067b1e75df4SAl Viro 3068132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags); 3069132e4608SDavid Howells 3070132e4608SDavid Howells /* 3071132e4608SDavid Howells * Create a new mount using a superblock configuration and request it 3072132e4608SDavid Howells * be added to the namespace tree. 3073132e4608SDavid Howells */ 3074132e4608SDavid Howells static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint, 3075132e4608SDavid Howells unsigned int mnt_flags) 3076132e4608SDavid Howells { 3077132e4608SDavid Howells struct vfsmount *mnt; 30788f11538eSAl Viro struct mountpoint *mp; 3079132e4608SDavid Howells struct super_block *sb = fc->root->d_sb; 3080132e4608SDavid Howells int error; 3081132e4608SDavid Howells 3082c9ce29edSAl Viro error = security_sb_kern_mount(sb); 3083c9ce29edSAl Viro if (!error && mount_too_revealing(sb, &mnt_flags)) 3084c9ce29edSAl Viro error = -EPERM; 3085c9ce29edSAl Viro 3086c9ce29edSAl Viro if (unlikely(error)) { 3087c9ce29edSAl Viro fc_drop_locked(fc); 3088c9ce29edSAl Viro return error; 3089132e4608SDavid Howells } 3090132e4608SDavid Howells 3091132e4608SDavid Howells up_write(&sb->s_umount); 3092132e4608SDavid Howells 3093132e4608SDavid Howells mnt = vfs_create_mount(fc); 3094132e4608SDavid Howells if (IS_ERR(mnt)) 3095132e4608SDavid Howells return PTR_ERR(mnt); 3096132e4608SDavid Howells 3097f8b92ba6SDeepa Dinamani mnt_warn_timestamp_expiry(mountpoint, mnt); 3098f8b92ba6SDeepa Dinamani 30998f11538eSAl Viro mp = lock_mount(mountpoint); 31008f11538eSAl Viro if (IS_ERR(mp)) { 31018f11538eSAl Viro mntput(mnt); 31028f11538eSAl Viro return PTR_ERR(mp); 31038f11538eSAl Viro } 31048f11538eSAl Viro error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags); 31058f11538eSAl Viro unlock_mount(mp); 31060ecee669SEric Biggers if (error < 0) 31070ecee669SEric Biggers mntput(mnt); 3108f8b92ba6SDeepa Dinamani return error; 3109f8b92ba6SDeepa Dinamani } 3110f8b92ba6SDeepa Dinamani 31111da177e4SLinus Torvalds /* 31121da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 31131da177e4SLinus Torvalds * namespace's tree 31141da177e4SLinus Torvalds */ 3115e462ec50SDavid Howells static int do_new_mount(struct path *path, const char *fstype, int sb_flags, 3116808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 31171da177e4SLinus Torvalds { 31180c55cfc4SEric W. Biederman struct file_system_type *type; 3119a0c9a8b8SAl Viro struct fs_context *fc; 3120a0c9a8b8SAl Viro const char *subtype = NULL; 3121a0c9a8b8SAl Viro int err = 0; 31221da177e4SLinus Torvalds 31230c55cfc4SEric W. Biederman if (!fstype) 31241da177e4SLinus Torvalds return -EINVAL; 31251da177e4SLinus Torvalds 31260c55cfc4SEric W. Biederman type = get_fs_type(fstype); 31270c55cfc4SEric W. Biederman if (!type) 31280c55cfc4SEric W. Biederman return -ENODEV; 31290c55cfc4SEric W. Biederman 3130a0c9a8b8SAl Viro if (type->fs_flags & FS_HAS_SUBTYPE) { 3131a0c9a8b8SAl Viro subtype = strchr(fstype, '.'); 3132a0c9a8b8SAl Viro if (subtype) { 3133a0c9a8b8SAl Viro subtype++; 3134a0c9a8b8SAl Viro if (!*subtype) { 31350c55cfc4SEric W. Biederman put_filesystem(type); 3136a0c9a8b8SAl Viro return -EINVAL; 3137a0c9a8b8SAl Viro } 3138a0c9a8b8SAl Viro } 31398654df4eSEric W. Biederman } 31408654df4eSEric W. Biederman 3141a0c9a8b8SAl Viro fc = fs_context_for_mount(type, sb_flags); 3142a0c9a8b8SAl Viro put_filesystem(type); 3143a0c9a8b8SAl Viro if (IS_ERR(fc)) 3144a0c9a8b8SAl Viro return PTR_ERR(fc); 3145a0c9a8b8SAl Viro 31463e1aeb00SDavid Howells if (subtype) 31473e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "subtype", 31483e1aeb00SDavid Howells subtype, strlen(subtype)); 31493e1aeb00SDavid Howells if (!err && name) 31503e1aeb00SDavid Howells err = vfs_parse_fs_string(fc, "source", name, strlen(name)); 3151a0c9a8b8SAl Viro if (!err) 3152a0c9a8b8SAl Viro err = parse_monolithic_mount_data(fc, data); 3153c3aabf07SAl Viro if (!err && !mount_capable(fc)) 3154c3aabf07SAl Viro err = -EPERM; 3155a0c9a8b8SAl Viro if (!err) 3156a0c9a8b8SAl Viro err = vfs_get_tree(fc); 3157132e4608SDavid Howells if (!err) 3158132e4608SDavid Howells err = do_new_mount_fc(fc, path, mnt_flags); 3159a0c9a8b8SAl Viro 3160a0c9a8b8SAl Viro put_fs_context(fc); 316115f9a3f3SAl Viro return err; 31621da177e4SLinus Torvalds } 31631da177e4SLinus Torvalds 31641e2d8464SAl Viro int finish_automount(struct vfsmount *m, const struct path *path) 316519a167afSAl Viro { 316626df6034SAl Viro struct dentry *dentry = path->dentry; 31678f11538eSAl Viro struct mountpoint *mp; 316825e195aaSAl Viro struct mount *mnt; 316919a167afSAl Viro int err; 317025e195aaSAl Viro 317125e195aaSAl Viro if (!m) 317225e195aaSAl Viro return 0; 317325e195aaSAl Viro if (IS_ERR(m)) 317425e195aaSAl Viro return PTR_ERR(m); 317525e195aaSAl Viro 317625e195aaSAl Viro mnt = real_mount(m); 317719a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 317819a167afSAl Viro * expired before we get a chance to add it 317919a167afSAl Viro */ 31806776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 318119a167afSAl Viro 318219a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 318326df6034SAl Viro m->mnt_root == dentry) { 3184b1e75df4SAl Viro err = -ELOOP; 318526df6034SAl Viro goto discard; 318619a167afSAl Viro } 318719a167afSAl Viro 318826df6034SAl Viro /* 318926df6034SAl Viro * we don't want to use lock_mount() - in this case finding something 319026df6034SAl Viro * that overmounts our mountpoint to be means "quitely drop what we've 319126df6034SAl Viro * got", not "try to mount it on top". 319226df6034SAl Viro */ 319326df6034SAl Viro inode_lock(dentry->d_inode); 319426df6034SAl Viro namespace_lock(); 319526df6034SAl Viro if (unlikely(cant_mount(dentry))) { 319626df6034SAl Viro err = -ENOENT; 319726df6034SAl Viro goto discard_locked; 319826df6034SAl Viro } 319926df6034SAl Viro rcu_read_lock(); 320026df6034SAl Viro if (unlikely(__lookup_mnt(path->mnt, dentry))) { 320126df6034SAl Viro rcu_read_unlock(); 320226df6034SAl Viro err = 0; 320326df6034SAl Viro goto discard_locked; 320426df6034SAl Viro } 320526df6034SAl Viro rcu_read_unlock(); 320626df6034SAl Viro mp = get_mountpoint(dentry); 32078f11538eSAl Viro if (IS_ERR(mp)) { 32088f11538eSAl Viro err = PTR_ERR(mp); 320926df6034SAl Viro goto discard_locked; 32108f11538eSAl Viro } 321126df6034SAl Viro 32128f11538eSAl Viro err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 32138f11538eSAl Viro unlock_mount(mp); 321426df6034SAl Viro if (unlikely(err)) 321526df6034SAl Viro goto discard; 321626df6034SAl Viro mntput(m); 3217b1e75df4SAl Viro return 0; 321826df6034SAl Viro 321926df6034SAl Viro discard_locked: 322026df6034SAl Viro namespace_unlock(); 322126df6034SAl Viro inode_unlock(dentry->d_inode); 322226df6034SAl Viro discard: 3223b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 32246776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 322597216be0SAl Viro namespace_lock(); 32266776db3dSAl Viro list_del_init(&mnt->mnt_expire); 322797216be0SAl Viro namespace_unlock(); 322819a167afSAl Viro } 3229b1e75df4SAl Viro mntput(m); 3230b1e75df4SAl Viro mntput(m); 323119a167afSAl Viro return err; 323219a167afSAl Viro } 323319a167afSAl Viro 3234ea5b778aSDavid Howells /** 3235ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 3236ea5b778aSDavid Howells * @mnt: The mount to list. 3237ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 3238ea5b778aSDavid Howells */ 3239ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 3240ea5b778aSDavid Howells { 324197216be0SAl Viro namespace_lock(); 3242ea5b778aSDavid Howells 32436776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 3244ea5b778aSDavid Howells 324597216be0SAl Viro namespace_unlock(); 3246ea5b778aSDavid Howells } 3247ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 3248ea5b778aSDavid Howells 3249ea5b778aSDavid Howells /* 32501da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 32511da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 32521da177e4SLinus Torvalds * here 32531da177e4SLinus Torvalds */ 32541da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 32551da177e4SLinus Torvalds { 3256761d5c38SAl Viro struct mount *mnt, *next; 32571da177e4SLinus Torvalds LIST_HEAD(graveyard); 32581da177e4SLinus Torvalds 32591da177e4SLinus Torvalds if (list_empty(mounts)) 32601da177e4SLinus Torvalds return; 32611da177e4SLinus Torvalds 326297216be0SAl Viro namespace_lock(); 3263719ea2fbSAl Viro lock_mount_hash(); 32641da177e4SLinus Torvalds 32651da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 32661da177e4SLinus Torvalds * following criteria: 32671da177e4SLinus Torvalds * - only referenced by its parent vfsmount 32681da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 32691da177e4SLinus Torvalds * cleared by mntput()) 32701da177e4SLinus Torvalds */ 32716776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 3272863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 32731ab59738SAl Viro propagate_mount_busy(mnt, 1)) 32741da177e4SLinus Torvalds continue; 32756776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 32761da177e4SLinus Torvalds } 3277bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 32786776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 3279143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 3280e819f152SEric W. Biederman umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3281bcc5c7d2SAl Viro } 3282719ea2fbSAl Viro unlock_mount_hash(); 32833ab6abeeSAl Viro namespace_unlock(); 32841da177e4SLinus Torvalds } 32851da177e4SLinus Torvalds 32861da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 32871da177e4SLinus Torvalds 32881da177e4SLinus Torvalds /* 32895528f911STrond Myklebust * Ripoff of 'select_parent()' 32905528f911STrond Myklebust * 32915528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 32925528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 32935528f911STrond Myklebust */ 3294692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 32955528f911STrond Myklebust { 3296692afc31SAl Viro struct mount *this_parent = parent; 32975528f911STrond Myklebust struct list_head *next; 32985528f911STrond Myklebust int found = 0; 32995528f911STrond Myklebust 33005528f911STrond Myklebust repeat: 33016b41d536SAl Viro next = this_parent->mnt_mounts.next; 33025528f911STrond Myklebust resume: 33036b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 33045528f911STrond Myklebust struct list_head *tmp = next; 33056b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 33065528f911STrond Myklebust 33075528f911STrond Myklebust next = tmp->next; 3308692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 33095528f911STrond Myklebust continue; 33105528f911STrond Myklebust /* 33115528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 33125528f911STrond Myklebust */ 33136b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 33145528f911STrond Myklebust this_parent = mnt; 33155528f911STrond Myklebust goto repeat; 33165528f911STrond Myklebust } 33175528f911STrond Myklebust 33181ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 33196776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 33205528f911STrond Myklebust found++; 33215528f911STrond Myklebust } 33225528f911STrond Myklebust } 33235528f911STrond Myklebust /* 33245528f911STrond Myklebust * All done at this level ... ascend and resume the search 33255528f911STrond Myklebust */ 33265528f911STrond Myklebust if (this_parent != parent) { 33276b41d536SAl Viro next = this_parent->mnt_child.next; 33280714a533SAl Viro this_parent = this_parent->mnt_parent; 33295528f911STrond Myklebust goto resume; 33305528f911STrond Myklebust } 33315528f911STrond Myklebust return found; 33325528f911STrond Myklebust } 33335528f911STrond Myklebust 33345528f911STrond Myklebust /* 33355528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 33365528f911STrond Myklebust * submounts of a specific parent mountpoint 333799b7db7bSNick Piggin * 333848a066e7SAl Viro * mount_lock must be held for write 33395528f911STrond Myklebust */ 3340b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 33415528f911STrond Myklebust { 33425528f911STrond Myklebust LIST_HEAD(graveyard); 3343761d5c38SAl Viro struct mount *m; 33445528f911STrond Myklebust 33455528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 3346c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 3347bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 3348761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 33496776db3dSAl Viro mnt_expire); 3350143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 3351e819f152SEric W. Biederman umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC); 3352bcc5c7d2SAl Viro } 3353bcc5c7d2SAl Viro } 33545528f911STrond Myklebust } 33555528f911STrond Myklebust 3356028abd92SChristoph Hellwig static void *copy_mount_options(const void __user * data) 33571da177e4SLinus Torvalds { 3358b40ef869SAl Viro char *copy; 3359d563d678SCatalin Marinas unsigned left, offset; 33601da177e4SLinus Torvalds 33611da177e4SLinus Torvalds if (!data) 3362b40ef869SAl Viro return NULL; 33631da177e4SLinus Torvalds 3364b40ef869SAl Viro copy = kmalloc(PAGE_SIZE, GFP_KERNEL); 3365b40ef869SAl Viro if (!copy) 3366b40ef869SAl Viro return ERR_PTR(-ENOMEM); 33671da177e4SLinus Torvalds 3368d563d678SCatalin Marinas left = copy_from_user(copy, data, PAGE_SIZE); 33691da177e4SLinus Torvalds 3370d563d678SCatalin Marinas /* 3371d563d678SCatalin Marinas * Not all architectures have an exact copy_from_user(). Resort to 3372d563d678SCatalin Marinas * byte at a time. 3373d563d678SCatalin Marinas */ 3374d563d678SCatalin Marinas offset = PAGE_SIZE - left; 3375d563d678SCatalin Marinas while (left) { 3376d563d678SCatalin Marinas char c; 3377d563d678SCatalin Marinas if (get_user(c, (const char __user *)data + offset)) 3378d563d678SCatalin Marinas break; 3379d563d678SCatalin Marinas copy[offset] = c; 3380d563d678SCatalin Marinas left--; 3381d563d678SCatalin Marinas offset++; 3382d563d678SCatalin Marinas } 3383d563d678SCatalin Marinas 3384d563d678SCatalin Marinas if (left == PAGE_SIZE) { 3385b40ef869SAl Viro kfree(copy); 3386b40ef869SAl Viro return ERR_PTR(-EFAULT); 33871da177e4SLinus Torvalds } 3388d563d678SCatalin Marinas 3389b40ef869SAl Viro return copy; 33901da177e4SLinus Torvalds } 33911da177e4SLinus Torvalds 3392028abd92SChristoph Hellwig static char *copy_mount_string(const void __user *data) 3393eca6f534SVegard Nossum { 3394fbdb4401SChandan Rajendra return data ? strndup_user(data, PATH_MAX) : NULL; 3395eca6f534SVegard Nossum } 3396eca6f534SVegard Nossum 33971da177e4SLinus Torvalds /* 33981da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 33991da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 34001da177e4SLinus Torvalds * 34011da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 34021da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 34031da177e4SLinus Torvalds * information (or be NULL). 34041da177e4SLinus Torvalds * 34051da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 34061da177e4SLinus Torvalds * When the flags word was introduced its top half was required 34071da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 34081da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 34091da177e4SLinus Torvalds * and must be discarded. 34101da177e4SLinus Torvalds */ 3411c60166f0SChristoph Hellwig int path_mount(const char *dev_name, struct path *path, 3412808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 34131da177e4SLinus Torvalds { 3414e462ec50SDavid Howells unsigned int mnt_flags = 0, sb_flags; 3415a1e6aaa3SChristoph Hellwig int ret; 34161da177e4SLinus Torvalds 34171da177e4SLinus Torvalds /* Discard magic */ 34181da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 34191da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 34201da177e4SLinus Torvalds 34211da177e4SLinus Torvalds /* Basic sanity checks */ 34221da177e4SLinus Torvalds if (data_page) 34231da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 34241da177e4SLinus Torvalds 3425e462ec50SDavid Howells if (flags & MS_NOUSER) 3426e462ec50SDavid Howells return -EINVAL; 3427e462ec50SDavid Howells 3428a1e6aaa3SChristoph Hellwig ret = security_sb_mount(dev_name, path, type_page, flags, data_page); 3429a1e6aaa3SChristoph Hellwig if (ret) 3430a1e6aaa3SChristoph Hellwig return ret; 3431a1e6aaa3SChristoph Hellwig if (!may_mount()) 3432a1e6aaa3SChristoph Hellwig return -EPERM; 3433f7e33bdbSJeff Layton if (flags & SB_MANDLOCK) 3434f7e33bdbSJeff Layton warn_mandlock(); 3435a27ab9f2STetsuo Handa 3436613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 3437613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 34380a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 34390a1c01c9SMatthew Garrett 34401da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 34411da177e4SLinus Torvalds if (flags & MS_NOSUID) 34421da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 34431da177e4SLinus Torvalds if (flags & MS_NODEV) 34441da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 34451da177e4SLinus Torvalds if (flags & MS_NOEXEC) 34461da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 3447fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 3448fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 3449fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 3450fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 3451d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 3452d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 3453a9e5b732SDavid Howells if (flags & MS_RDONLY) 34542e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 3455dab741e0SMattias Nissler if (flags & MS_NOSYMFOLLOW) 3456dab741e0SMattias Nissler mnt_flags |= MNT_NOSYMFOLLOW; 3457fc33a7bbSChristoph Hellwig 3458ffbc6f0eSEric W. Biederman /* The default atime for remount is preservation */ 3459ffbc6f0eSEric W. Biederman if ((flags & MS_REMOUNT) && 3460ffbc6f0eSEric W. Biederman ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME | 3461ffbc6f0eSEric W. Biederman MS_STRICTATIME)) == 0)) { 3462ffbc6f0eSEric W. Biederman mnt_flags &= ~MNT_ATIME_MASK; 3463a1e6aaa3SChristoph Hellwig mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK; 3464ffbc6f0eSEric W. Biederman } 3465ffbc6f0eSEric W. Biederman 3466e462ec50SDavid Howells sb_flags = flags & (SB_RDONLY | 3467e462ec50SDavid Howells SB_SYNCHRONOUS | 3468e462ec50SDavid Howells SB_MANDLOCK | 3469e462ec50SDavid Howells SB_DIRSYNC | 3470e462ec50SDavid Howells SB_SILENT | 3471917086ffSMimi Zohar SB_POSIXACL | 3472d7ee9469SMarkus Trippelsdorf SB_LAZYTIME | 3473917086ffSMimi Zohar SB_I_VERSION); 34741da177e4SLinus Torvalds 347543f5e655SDavid Howells if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND)) 3476a1e6aaa3SChristoph Hellwig return do_reconfigure_mnt(path, mnt_flags); 3477a1e6aaa3SChristoph Hellwig if (flags & MS_REMOUNT) 3478a1e6aaa3SChristoph Hellwig return do_remount(path, flags, sb_flags, mnt_flags, data_page); 3479a1e6aaa3SChristoph Hellwig if (flags & MS_BIND) 3480a1e6aaa3SChristoph Hellwig return do_loopback(path, dev_name, flags & MS_REC); 3481a1e6aaa3SChristoph Hellwig if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 3482a1e6aaa3SChristoph Hellwig return do_change_type(path, flags); 3483a1e6aaa3SChristoph Hellwig if (flags & MS_MOVE) 3484a1e6aaa3SChristoph Hellwig return do_move_mount_old(path, dev_name); 3485a1e6aaa3SChristoph Hellwig 3486a1e6aaa3SChristoph Hellwig return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name, 34871da177e4SLinus Torvalds data_page); 3488a1e6aaa3SChristoph Hellwig } 3489a1e6aaa3SChristoph Hellwig 3490a1e6aaa3SChristoph Hellwig long do_mount(const char *dev_name, const char __user *dir_name, 3491a1e6aaa3SChristoph Hellwig const char *type_page, unsigned long flags, void *data_page) 3492a1e6aaa3SChristoph Hellwig { 3493a1e6aaa3SChristoph Hellwig struct path path; 3494a1e6aaa3SChristoph Hellwig int ret; 3495a1e6aaa3SChristoph Hellwig 3496a1e6aaa3SChristoph Hellwig ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path); 3497a1e6aaa3SChristoph Hellwig if (ret) 3498a1e6aaa3SChristoph Hellwig return ret; 3499a1e6aaa3SChristoph Hellwig ret = path_mount(dev_name, &path, type_page, flags, data_page); 35002d92ab3cSAl Viro path_put(&path); 3501a1e6aaa3SChristoph Hellwig return ret; 35021da177e4SLinus Torvalds } 35031da177e4SLinus Torvalds 3504537f7ccbSEric W. Biederman static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns) 3505537f7ccbSEric W. Biederman { 3506537f7ccbSEric W. Biederman return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES); 3507537f7ccbSEric W. Biederman } 3508537f7ccbSEric W. Biederman 3509537f7ccbSEric W. Biederman static void dec_mnt_namespaces(struct ucounts *ucounts) 3510537f7ccbSEric W. Biederman { 3511537f7ccbSEric W. Biederman dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES); 3512537f7ccbSEric W. Biederman } 3513537f7ccbSEric W. Biederman 3514771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 3515771b1371SEric W. Biederman { 351674e83122SAl Viro if (!is_anon_ns(ns)) 35176344c433SAl Viro ns_free_inum(&ns->ns); 3518537f7ccbSEric W. Biederman dec_mnt_namespaces(ns->ucounts); 3519771b1371SEric W. Biederman put_user_ns(ns->user_ns); 3520771b1371SEric W. Biederman kfree(ns); 3521771b1371SEric W. Biederman } 3522771b1371SEric W. Biederman 35238823c079SEric W. Biederman /* 35248823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 35258823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 35268823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 35278823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 35288823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 35298823c079SEric W. Biederman */ 35308823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 35318823c079SEric W. Biederman 353274e83122SAl Viro static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon) 3533cf8d2c11STrond Myklebust { 3534cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 3535537f7ccbSEric W. Biederman struct ucounts *ucounts; 353698f842e6SEric W. Biederman int ret; 3537cf8d2c11STrond Myklebust 3538537f7ccbSEric W. Biederman ucounts = inc_mnt_namespaces(user_ns); 3539537f7ccbSEric W. Biederman if (!ucounts) 3540df75e774SEric W. Biederman return ERR_PTR(-ENOSPC); 3541537f7ccbSEric W. Biederman 354230acd0bdSVasily Averin new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT); 3543537f7ccbSEric W. Biederman if (!new_ns) { 3544537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 3545cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 3546537f7ccbSEric W. Biederman } 354774e83122SAl Viro if (!anon) { 35486344c433SAl Viro ret = ns_alloc_inum(&new_ns->ns); 354998f842e6SEric W. Biederman if (ret) { 355098f842e6SEric W. Biederman kfree(new_ns); 3551537f7ccbSEric W. Biederman dec_mnt_namespaces(ucounts); 355298f842e6SEric W. Biederman return ERR_PTR(ret); 355398f842e6SEric W. Biederman } 355474e83122SAl Viro } 355533c42940SAl Viro new_ns->ns.ops = &mntns_operations; 355674e83122SAl Viro if (!anon) 35578823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 35581a7b8969SKirill Tkhai refcount_set(&new_ns->ns.count, 1); 3559cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 3560cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 35619f6c61f9SMiklos Szeredi spin_lock_init(&new_ns->ns_lock); 3562771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 3563537f7ccbSEric W. Biederman new_ns->ucounts = ucounts; 3564cf8d2c11STrond Myklebust return new_ns; 3565cf8d2c11STrond Myklebust } 3566cf8d2c11STrond Myklebust 35670766f788SEmese Revfy __latent_entropy 35689559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 35699559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 35701da177e4SLinus Torvalds { 35716b3286edSKirill Korotaev struct mnt_namespace *new_ns; 35727f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 3573315fc83eSAl Viro struct mount *p, *q; 35749559f689SAl Viro struct mount *old; 3575cb338d06SAl Viro struct mount *new; 35767a472ef4SEric W. Biederman int copy_flags; 35771da177e4SLinus Torvalds 35789559f689SAl Viro BUG_ON(!ns); 35799559f689SAl Viro 35809559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 35819559f689SAl Viro get_mnt_ns(ns); 35829559f689SAl Viro return ns; 35839559f689SAl Viro } 35849559f689SAl Viro 35859559f689SAl Viro old = ns->root; 35869559f689SAl Viro 358774e83122SAl Viro new_ns = alloc_mnt_ns(user_ns, false); 3588cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 3589cf8d2c11STrond Myklebust return new_ns; 35901da177e4SLinus Torvalds 359197216be0SAl Viro namespace_lock(); 35921da177e4SLinus Torvalds /* First pass: copy the tree topology */ 35934ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 35949559f689SAl Viro if (user_ns != ns->user_ns) 35953bd045ccSAl Viro copy_flags |= CL_SHARED_TO_SLAVE; 35967a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 3597be34d1a3SDavid Howells if (IS_ERR(new)) { 3598328e6d90SAl Viro namespace_unlock(); 3599771b1371SEric W. Biederman free_mnt_ns(new_ns); 3600be34d1a3SDavid Howells return ERR_CAST(new); 36011da177e4SLinus Torvalds } 36023bd045ccSAl Viro if (user_ns != ns->user_ns) { 36033bd045ccSAl Viro lock_mount_hash(); 36043bd045ccSAl Viro lock_mnt_tree(new); 36053bd045ccSAl Viro unlock_mount_hash(); 36063bd045ccSAl Viro } 3607be08d6d2SAl Viro new_ns->root = new; 36081a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 36091da177e4SLinus Torvalds 36101da177e4SLinus Torvalds /* 36111da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 36121da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 36131da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 36141da177e4SLinus Torvalds */ 3615909b0a88SAl Viro p = old; 3616cb338d06SAl Viro q = new; 36171da177e4SLinus Torvalds while (p) { 3618143c8c91SAl Viro q->mnt_ns = new_ns; 3619d2921684SEric W. Biederman new_ns->mounts++; 36209559f689SAl Viro if (new_fs) { 36219559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 36229559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 3623315fc83eSAl Viro rootmnt = &p->mnt; 36241da177e4SLinus Torvalds } 36259559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 36269559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 3627315fc83eSAl Viro pwdmnt = &p->mnt; 36281da177e4SLinus Torvalds } 36291da177e4SLinus Torvalds } 3630909b0a88SAl Viro p = next_mnt(p, old); 3631909b0a88SAl Viro q = next_mnt(q, new); 36324ce5d2b1SEric W. Biederman if (!q) 36334ce5d2b1SEric W. Biederman break; 363461d8e426SAl Viro // an mntns binding we'd skipped? 36354ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 363661d8e426SAl Viro p = next_mnt(skip_mnt_tree(p), old); 36371da177e4SLinus Torvalds } 3638328e6d90SAl Viro namespace_unlock(); 36391da177e4SLinus Torvalds 36401da177e4SLinus Torvalds if (rootmnt) 3641f03c6599SAl Viro mntput(rootmnt); 36421da177e4SLinus Torvalds if (pwdmnt) 3643f03c6599SAl Viro mntput(pwdmnt); 36441da177e4SLinus Torvalds 3645741a2951SJANAK DESAI return new_ns; 3646741a2951SJANAK DESAI } 3647741a2951SJANAK DESAI 364874e83122SAl Viro struct dentry *mount_subtree(struct vfsmount *m, const char *name) 3649cf8d2c11STrond Myklebust { 36501a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 3651ea441d11SAl Viro struct mnt_namespace *ns; 3652d31da0f0SAl Viro struct super_block *s; 3653ea441d11SAl Viro struct path path; 3654ea441d11SAl Viro int err; 3655ea441d11SAl Viro 365674e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, true); 365774e83122SAl Viro if (IS_ERR(ns)) { 365874e83122SAl Viro mntput(m); 3659ea441d11SAl Viro return ERR_CAST(ns); 366074e83122SAl Viro } 366174e83122SAl Viro mnt->mnt_ns = ns; 366274e83122SAl Viro ns->root = mnt; 366374e83122SAl Viro ns->mounts++; 366474e83122SAl Viro list_add(&mnt->mnt_list, &ns->list); 3665ea441d11SAl Viro 366674e83122SAl Viro err = vfs_path_lookup(m->mnt_root, m, 3667ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 3668ea441d11SAl Viro 3669ea441d11SAl Viro put_mnt_ns(ns); 3670ea441d11SAl Viro 3671ea441d11SAl Viro if (err) 3672ea441d11SAl Viro return ERR_PTR(err); 3673ea441d11SAl Viro 3674ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 3675d31da0f0SAl Viro s = path.mnt->mnt_sb; 3676d31da0f0SAl Viro atomic_inc(&s->s_active); 3677ea441d11SAl Viro mntput(path.mnt); 3678ea441d11SAl Viro /* lock the sucker */ 3679d31da0f0SAl Viro down_write(&s->s_umount); 3680ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 3681ea441d11SAl Viro return path.dentry; 3682ea441d11SAl Viro } 3683ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 3684ea441d11SAl Viro 3685cccaa5e3SDominik Brodowski SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 3686cccaa5e3SDominik Brodowski char __user *, type, unsigned long, flags, void __user *, data) 36871da177e4SLinus Torvalds { 3688eca6f534SVegard Nossum int ret; 3689eca6f534SVegard Nossum char *kernel_type; 3690eca6f534SVegard Nossum char *kernel_dev; 3691b40ef869SAl Viro void *options; 36921da177e4SLinus Torvalds 3693b8850d1fSTim Gardner kernel_type = copy_mount_string(type); 3694b8850d1fSTim Gardner ret = PTR_ERR(kernel_type); 3695b8850d1fSTim Gardner if (IS_ERR(kernel_type)) 3696eca6f534SVegard Nossum goto out_type; 36971da177e4SLinus Torvalds 3698b8850d1fSTim Gardner kernel_dev = copy_mount_string(dev_name); 3699b8850d1fSTim Gardner ret = PTR_ERR(kernel_dev); 3700b8850d1fSTim Gardner if (IS_ERR(kernel_dev)) 3701eca6f534SVegard Nossum goto out_dev; 37021da177e4SLinus Torvalds 3703b40ef869SAl Viro options = copy_mount_options(data); 3704b40ef869SAl Viro ret = PTR_ERR(options); 3705b40ef869SAl Viro if (IS_ERR(options)) 3706eca6f534SVegard Nossum goto out_data; 37071da177e4SLinus Torvalds 3708b40ef869SAl Viro ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options); 3709eca6f534SVegard Nossum 3710b40ef869SAl Viro kfree(options); 3711eca6f534SVegard Nossum out_data: 3712eca6f534SVegard Nossum kfree(kernel_dev); 3713eca6f534SVegard Nossum out_dev: 3714eca6f534SVegard Nossum kfree(kernel_type); 3715eca6f534SVegard Nossum out_type: 3716eca6f534SVegard Nossum return ret; 37171da177e4SLinus Torvalds } 37181da177e4SLinus Torvalds 37195b490500SChristian Brauner #define FSMOUNT_VALID_FLAGS \ 37205b490500SChristian Brauner (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \ 3721dd8b477fSChristian Brauner MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME | \ 3722dd8b477fSChristian Brauner MOUNT_ATTR_NOSYMFOLLOW) 37235b490500SChristian Brauner 37249caccd41SChristian Brauner #define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP) 37252a186721SChristian Brauner 37262a186721SChristian Brauner #define MOUNT_SETATTR_PROPAGATION_FLAGS \ 37272a186721SChristian Brauner (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED) 37282a186721SChristian Brauner 37295b490500SChristian Brauner static unsigned int attr_flags_to_mnt_flags(u64 attr_flags) 37305b490500SChristian Brauner { 37315b490500SChristian Brauner unsigned int mnt_flags = 0; 37325b490500SChristian Brauner 37335b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_RDONLY) 37345b490500SChristian Brauner mnt_flags |= MNT_READONLY; 37355b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOSUID) 37365b490500SChristian Brauner mnt_flags |= MNT_NOSUID; 37375b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODEV) 37385b490500SChristian Brauner mnt_flags |= MNT_NODEV; 37395b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NOEXEC) 37405b490500SChristian Brauner mnt_flags |= MNT_NOEXEC; 37415b490500SChristian Brauner if (attr_flags & MOUNT_ATTR_NODIRATIME) 37425b490500SChristian Brauner mnt_flags |= MNT_NODIRATIME; 3743dd8b477fSChristian Brauner if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW) 3744dd8b477fSChristian Brauner mnt_flags |= MNT_NOSYMFOLLOW; 37455b490500SChristian Brauner 37465b490500SChristian Brauner return mnt_flags; 37475b490500SChristian Brauner } 37485b490500SChristian Brauner 37491da177e4SLinus Torvalds /* 375093766fbdSDavid Howells * Create a kernel mount representation for a new, prepared superblock 375193766fbdSDavid Howells * (specified by fs_fd) and attach to an open_tree-like file descriptor. 375293766fbdSDavid Howells */ 375393766fbdSDavid Howells SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags, 375493766fbdSDavid Howells unsigned int, attr_flags) 375593766fbdSDavid Howells { 375693766fbdSDavid Howells struct mnt_namespace *ns; 375793766fbdSDavid Howells struct fs_context *fc; 375893766fbdSDavid Howells struct file *file; 375993766fbdSDavid Howells struct path newmount; 376093766fbdSDavid Howells struct mount *mnt; 376193766fbdSDavid Howells struct fd f; 376293766fbdSDavid Howells unsigned int mnt_flags = 0; 376393766fbdSDavid Howells long ret; 376493766fbdSDavid Howells 376593766fbdSDavid Howells if (!may_mount()) 376693766fbdSDavid Howells return -EPERM; 376793766fbdSDavid Howells 376893766fbdSDavid Howells if ((flags & ~(FSMOUNT_CLOEXEC)) != 0) 376993766fbdSDavid Howells return -EINVAL; 377093766fbdSDavid Howells 37715b490500SChristian Brauner if (attr_flags & ~FSMOUNT_VALID_FLAGS) 377293766fbdSDavid Howells return -EINVAL; 377393766fbdSDavid Howells 37745b490500SChristian Brauner mnt_flags = attr_flags_to_mnt_flags(attr_flags); 377593766fbdSDavid Howells 377693766fbdSDavid Howells switch (attr_flags & MOUNT_ATTR__ATIME) { 377793766fbdSDavid Howells case MOUNT_ATTR_STRICTATIME: 377893766fbdSDavid Howells break; 377993766fbdSDavid Howells case MOUNT_ATTR_NOATIME: 378093766fbdSDavid Howells mnt_flags |= MNT_NOATIME; 378193766fbdSDavid Howells break; 378293766fbdSDavid Howells case MOUNT_ATTR_RELATIME: 378393766fbdSDavid Howells mnt_flags |= MNT_RELATIME; 378493766fbdSDavid Howells break; 378593766fbdSDavid Howells default: 378693766fbdSDavid Howells return -EINVAL; 378793766fbdSDavid Howells } 378893766fbdSDavid Howells 378993766fbdSDavid Howells f = fdget(fs_fd); 379093766fbdSDavid Howells if (!f.file) 379193766fbdSDavid Howells return -EBADF; 379293766fbdSDavid Howells 379393766fbdSDavid Howells ret = -EINVAL; 379493766fbdSDavid Howells if (f.file->f_op != &fscontext_fops) 379593766fbdSDavid Howells goto err_fsfd; 379693766fbdSDavid Howells 379793766fbdSDavid Howells fc = f.file->private_data; 379893766fbdSDavid Howells 379993766fbdSDavid Howells ret = mutex_lock_interruptible(&fc->uapi_mutex); 380093766fbdSDavid Howells if (ret < 0) 380193766fbdSDavid Howells goto err_fsfd; 380293766fbdSDavid Howells 380393766fbdSDavid Howells /* There must be a valid superblock or we can't mount it */ 380493766fbdSDavid Howells ret = -EINVAL; 380593766fbdSDavid Howells if (!fc->root) 380693766fbdSDavid Howells goto err_unlock; 380793766fbdSDavid Howells 380893766fbdSDavid Howells ret = -EPERM; 380993766fbdSDavid Howells if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) { 381093766fbdSDavid Howells pr_warn("VFS: Mount too revealing\n"); 381193766fbdSDavid Howells goto err_unlock; 381293766fbdSDavid Howells } 381393766fbdSDavid Howells 381493766fbdSDavid Howells ret = -EBUSY; 381593766fbdSDavid Howells if (fc->phase != FS_CONTEXT_AWAITING_MOUNT) 381693766fbdSDavid Howells goto err_unlock; 381793766fbdSDavid Howells 3818f7e33bdbSJeff Layton if (fc->sb_flags & SB_MANDLOCK) 3819f7e33bdbSJeff Layton warn_mandlock(); 382093766fbdSDavid Howells 382193766fbdSDavid Howells newmount.mnt = vfs_create_mount(fc); 382293766fbdSDavid Howells if (IS_ERR(newmount.mnt)) { 382393766fbdSDavid Howells ret = PTR_ERR(newmount.mnt); 382493766fbdSDavid Howells goto err_unlock; 382593766fbdSDavid Howells } 382693766fbdSDavid Howells newmount.dentry = dget(fc->root); 382793766fbdSDavid Howells newmount.mnt->mnt_flags = mnt_flags; 382893766fbdSDavid Howells 382993766fbdSDavid Howells /* We've done the mount bit - now move the file context into more or 383093766fbdSDavid Howells * less the same state as if we'd done an fspick(). We don't want to 383193766fbdSDavid Howells * do any memory allocation or anything like that at this point as we 383293766fbdSDavid Howells * don't want to have to handle any errors incurred. 383393766fbdSDavid Howells */ 383493766fbdSDavid Howells vfs_clean_context(fc); 383593766fbdSDavid Howells 383693766fbdSDavid Howells ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true); 383793766fbdSDavid Howells if (IS_ERR(ns)) { 383893766fbdSDavid Howells ret = PTR_ERR(ns); 383993766fbdSDavid Howells goto err_path; 384093766fbdSDavid Howells } 384193766fbdSDavid Howells mnt = real_mount(newmount.mnt); 384293766fbdSDavid Howells mnt->mnt_ns = ns; 384393766fbdSDavid Howells ns->root = mnt; 384493766fbdSDavid Howells ns->mounts = 1; 384593766fbdSDavid Howells list_add(&mnt->mnt_list, &ns->list); 38461b0b9cc8SEric Biggers mntget(newmount.mnt); 384793766fbdSDavid Howells 384893766fbdSDavid Howells /* Attach to an apparent O_PATH fd with a note that we need to unmount 384993766fbdSDavid Howells * it, not just simply put it. 385093766fbdSDavid Howells */ 385193766fbdSDavid Howells file = dentry_open(&newmount, O_PATH, fc->cred); 385293766fbdSDavid Howells if (IS_ERR(file)) { 385393766fbdSDavid Howells dissolve_on_fput(newmount.mnt); 385493766fbdSDavid Howells ret = PTR_ERR(file); 385593766fbdSDavid Howells goto err_path; 385693766fbdSDavid Howells } 385793766fbdSDavid Howells file->f_mode |= FMODE_NEED_UNMOUNT; 385893766fbdSDavid Howells 385993766fbdSDavid Howells ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0); 386093766fbdSDavid Howells if (ret >= 0) 386193766fbdSDavid Howells fd_install(ret, file); 386293766fbdSDavid Howells else 386393766fbdSDavid Howells fput(file); 386493766fbdSDavid Howells 386593766fbdSDavid Howells err_path: 386693766fbdSDavid Howells path_put(&newmount); 386793766fbdSDavid Howells err_unlock: 386893766fbdSDavid Howells mutex_unlock(&fc->uapi_mutex); 386993766fbdSDavid Howells err_fsfd: 387093766fbdSDavid Howells fdput(f); 387193766fbdSDavid Howells return ret; 387293766fbdSDavid Howells } 387393766fbdSDavid Howells 387493766fbdSDavid Howells /* 387593766fbdSDavid Howells * Move a mount from one place to another. In combination with 387693766fbdSDavid Howells * fsopen()/fsmount() this is used to install a new mount and in combination 387793766fbdSDavid Howells * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy 387893766fbdSDavid Howells * a mount subtree. 38792db154b3SDavid Howells * 38802db154b3SDavid Howells * Note the flags value is a combination of MOVE_MOUNT_* flags. 38812db154b3SDavid Howells */ 38822db154b3SDavid Howells SYSCALL_DEFINE5(move_mount, 38832658ce09SBen Dooks int, from_dfd, const char __user *, from_pathname, 38842658ce09SBen Dooks int, to_dfd, const char __user *, to_pathname, 38852db154b3SDavid Howells unsigned int, flags) 38862db154b3SDavid Howells { 38872db154b3SDavid Howells struct path from_path, to_path; 38882db154b3SDavid Howells unsigned int lflags; 38892db154b3SDavid Howells int ret = 0; 38902db154b3SDavid Howells 38912db154b3SDavid Howells if (!may_mount()) 38922db154b3SDavid Howells return -EPERM; 38932db154b3SDavid Howells 38942db154b3SDavid Howells if (flags & ~MOVE_MOUNT__MASK) 38952db154b3SDavid Howells return -EINVAL; 38962db154b3SDavid Howells 38972db154b3SDavid Howells /* If someone gives a pathname, they aren't permitted to move 38982db154b3SDavid Howells * from an fd that requires unmount as we can't get at the flag 38992db154b3SDavid Howells * to clear it afterwards. 39002db154b3SDavid Howells */ 39012db154b3SDavid Howells lflags = 0; 39022db154b3SDavid Howells if (flags & MOVE_MOUNT_F_SYMLINKS) lflags |= LOOKUP_FOLLOW; 39032db154b3SDavid Howells if (flags & MOVE_MOUNT_F_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 39042db154b3SDavid Howells if (flags & MOVE_MOUNT_F_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 39052db154b3SDavid Howells 39062db154b3SDavid Howells ret = user_path_at(from_dfd, from_pathname, lflags, &from_path); 39072db154b3SDavid Howells if (ret < 0) 39082db154b3SDavid Howells return ret; 39092db154b3SDavid Howells 39102db154b3SDavid Howells lflags = 0; 39112db154b3SDavid Howells if (flags & MOVE_MOUNT_T_SYMLINKS) lflags |= LOOKUP_FOLLOW; 39122db154b3SDavid Howells if (flags & MOVE_MOUNT_T_AUTOMOUNTS) lflags |= LOOKUP_AUTOMOUNT; 39132db154b3SDavid Howells if (flags & MOVE_MOUNT_T_EMPTY_PATH) lflags |= LOOKUP_EMPTY; 39142db154b3SDavid Howells 39152db154b3SDavid Howells ret = user_path_at(to_dfd, to_pathname, lflags, &to_path); 39162db154b3SDavid Howells if (ret < 0) 39172db154b3SDavid Howells goto out_from; 39182db154b3SDavid Howells 39192db154b3SDavid Howells ret = security_move_mount(&from_path, &to_path); 39202db154b3SDavid Howells if (ret < 0) 39212db154b3SDavid Howells goto out_to; 39222db154b3SDavid Howells 39239ffb14efSPavel Tikhomirov if (flags & MOVE_MOUNT_SET_GROUP) 39249ffb14efSPavel Tikhomirov ret = do_set_group(&from_path, &to_path); 39259ffb14efSPavel Tikhomirov else 39262db154b3SDavid Howells ret = do_move_mount(&from_path, &to_path); 39272db154b3SDavid Howells 39282db154b3SDavid Howells out_to: 39292db154b3SDavid Howells path_put(&to_path); 39302db154b3SDavid Howells out_from: 39312db154b3SDavid Howells path_put(&from_path); 39322db154b3SDavid Howells return ret; 39332db154b3SDavid Howells } 39342db154b3SDavid Howells 39352db154b3SDavid Howells /* 3936afac7cbaSAl Viro * Return true if path is reachable from root 3937afac7cbaSAl Viro * 393848a066e7SAl Viro * namespace_sem or mount_lock is held 3939afac7cbaSAl Viro */ 3940643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 3941afac7cbaSAl Viro const struct path *root) 3942afac7cbaSAl Viro { 3943643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 3944a73324daSAl Viro dentry = mnt->mnt_mountpoint; 39450714a533SAl Viro mnt = mnt->mnt_parent; 3946afac7cbaSAl Viro } 3947643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 3948afac7cbaSAl Viro } 3949afac7cbaSAl Viro 3950640eb7e7SMickaël Salaün bool path_is_under(const struct path *path1, const struct path *path2) 3951afac7cbaSAl Viro { 395225ab4c9bSYaowei Bai bool res; 395348a066e7SAl Viro read_seqlock_excl(&mount_lock); 3954643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 395548a066e7SAl Viro read_sequnlock_excl(&mount_lock); 3956afac7cbaSAl Viro return res; 3957afac7cbaSAl Viro } 3958afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 3959afac7cbaSAl Viro 3960afac7cbaSAl Viro /* 39611da177e4SLinus Torvalds * pivot_root Semantics: 39621da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 39631da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 39641da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 39651da177e4SLinus Torvalds * 39661da177e4SLinus Torvalds * Restrictions: 39671da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 39681da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 39691da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 39701da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 39711da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 39721da177e4SLinus Torvalds * 39734a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 39740c1bc6b8SMauro Carvalho Chehab * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives 39754a0d11faSNeil Brown * in this situation. 39764a0d11faSNeil Brown * 39771da177e4SLinus Torvalds * Notes: 39781da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 39791da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 39801da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 39811da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 39821da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 39831da177e4SLinus Torvalds * first. 39841da177e4SLinus Torvalds */ 39853480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 39863480b257SHeiko Carstens const char __user *, put_old) 39871da177e4SLinus Torvalds { 39882763d119SAl Viro struct path new, old, root; 39892763d119SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent; 399084d17192SAl Viro struct mountpoint *old_mp, *root_mp; 39911da177e4SLinus Torvalds int error; 39921da177e4SLinus Torvalds 39939b40bc90SAl Viro if (!may_mount()) 39941da177e4SLinus Torvalds return -EPERM; 39951da177e4SLinus Torvalds 3996ce6595a2SAl Viro error = user_path_at(AT_FDCWD, new_root, 3997ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new); 39981da177e4SLinus Torvalds if (error) 39991da177e4SLinus Torvalds goto out0; 40001da177e4SLinus Torvalds 4001ce6595a2SAl Viro error = user_path_at(AT_FDCWD, put_old, 4002ce6595a2SAl Viro LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old); 40031da177e4SLinus Torvalds if (error) 40041da177e4SLinus Torvalds goto out1; 40051da177e4SLinus Torvalds 40062d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 4007b12cea91SAl Viro if (error) 4008b12cea91SAl Viro goto out2; 40091da177e4SLinus Torvalds 4010f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 401184d17192SAl Viro old_mp = lock_mount(&old); 401284d17192SAl Viro error = PTR_ERR(old_mp); 401384d17192SAl Viro if (IS_ERR(old_mp)) 4014b12cea91SAl Viro goto out3; 4015b12cea91SAl Viro 40161da177e4SLinus Torvalds error = -EINVAL; 4017419148daSAl Viro new_mnt = real_mount(new.mnt); 4018419148daSAl Viro root_mnt = real_mount(root.mnt); 401984d17192SAl Viro old_mnt = real_mount(old.mnt); 40202763d119SAl Viro ex_parent = new_mnt->mnt_parent; 40212763d119SAl Viro root_parent = root_mnt->mnt_parent; 402284d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 40232763d119SAl Viro IS_MNT_SHARED(ex_parent) || 40242763d119SAl Viro IS_MNT_SHARED(root_parent)) 4025b12cea91SAl Viro goto out4; 4026143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 4027b12cea91SAl Viro goto out4; 40285ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 40295ff9d8a6SEric W. Biederman goto out4; 40301da177e4SLinus Torvalds error = -ENOENT; 4031f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 4032b12cea91SAl Viro goto out4; 40331da177e4SLinus Torvalds error = -EBUSY; 403484d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 4035b12cea91SAl Viro goto out4; /* loop, on the same file system */ 40361da177e4SLinus Torvalds error = -EINVAL; 40378c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 4038b12cea91SAl Viro goto out4; /* not a mountpoint */ 4039676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 4040b12cea91SAl Viro goto out4; /* not attached */ 40412d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 4042b12cea91SAl Viro goto out4; /* not a mountpoint */ 4043676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 4044b12cea91SAl Viro goto out4; /* not attached */ 40454ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 404684d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 4047b12cea91SAl Viro goto out4; 40480d082601SEric W. Biederman /* make certain new is below the root */ 40490d082601SEric W. Biederman if (!is_path_reachable(new_mnt, new.dentry, &root)) 40500d082601SEric W. Biederman goto out4; 4051719ea2fbSAl Viro lock_mount_hash(); 40522763d119SAl Viro umount_mnt(new_mnt); 40532763d119SAl Viro root_mp = unhash_mnt(root_mnt); /* we'll need its mountpoint */ 40545ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 40555ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 40565ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 40575ff9d8a6SEric W. Biederman } 40584ac91378SJan Blunck /* mount old root on put_old */ 405984d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 40604ac91378SJan Blunck /* mount new_root on / */ 40612763d119SAl Viro attach_mnt(new_mnt, root_parent, root_mp); 40622763d119SAl Viro mnt_add_count(root_parent, -1); 40636b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 40644fed655cSEric W. Biederman /* A moved mount should not expire automatically */ 40654fed655cSEric W. Biederman list_del_init(&new_mnt->mnt_expire); 40663895dbf8SEric W. Biederman put_mountpoint(root_mp); 4067719ea2fbSAl Viro unlock_mount_hash(); 40682d8f3038SAl Viro chroot_fs_refs(&root, &new); 40691da177e4SLinus Torvalds error = 0; 4070b12cea91SAl Viro out4: 407184d17192SAl Viro unlock_mount(old_mp); 40722763d119SAl Viro if (!error) 40732763d119SAl Viro mntput_no_expire(ex_parent); 4074b12cea91SAl Viro out3: 40758c3ee42eSAl Viro path_put(&root); 4076b12cea91SAl Viro out2: 40772d8f3038SAl Viro path_put(&old); 40781da177e4SLinus Torvalds out1: 40792d8f3038SAl Viro path_put(&new); 40801da177e4SLinus Torvalds out0: 40811da177e4SLinus Torvalds return error; 40821da177e4SLinus Torvalds } 40831da177e4SLinus Torvalds 40842a186721SChristian Brauner static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt) 40852a186721SChristian Brauner { 40862a186721SChristian Brauner unsigned int flags = mnt->mnt.mnt_flags; 40872a186721SChristian Brauner 40882a186721SChristian Brauner /* flags to clear */ 40892a186721SChristian Brauner flags &= ~kattr->attr_clr; 40902a186721SChristian Brauner /* flags to raise */ 40912a186721SChristian Brauner flags |= kattr->attr_set; 40922a186721SChristian Brauner 40932a186721SChristian Brauner return flags; 40942a186721SChristian Brauner } 40952a186721SChristian Brauner 40969caccd41SChristian Brauner static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 40979caccd41SChristian Brauner { 40989caccd41SChristian Brauner struct vfsmount *m = &mnt->mnt; 4099bd303368SChristian Brauner struct user_namespace *fs_userns = m->mnt_sb->s_user_ns; 41009caccd41SChristian Brauner 4101256c8aedSChristian Brauner if (!kattr->mnt_idmap) 41029caccd41SChristian Brauner return 0; 41039caccd41SChristian Brauner 41049caccd41SChristian Brauner /* 4105bd303368SChristian Brauner * Creating an idmapped mount with the filesystem wide idmapping 4106bd303368SChristian Brauner * doesn't make sense so block that. We don't allow mushy semantics. 4107bd303368SChristian Brauner */ 4108256c8aedSChristian Brauner if (mnt_idmap_owner(kattr->mnt_idmap) == fs_userns) 4109bd303368SChristian Brauner return -EINVAL; 4110bd303368SChristian Brauner 4111bd303368SChristian Brauner /* 41129caccd41SChristian Brauner * Once a mount has been idmapped we don't allow it to change its 41139caccd41SChristian Brauner * mapping. It makes things simpler and callers can just create 41149caccd41SChristian Brauner * another bind-mount they can idmap if they want to. 41159caccd41SChristian Brauner */ 4116bb49e9e7SChristian Brauner if (is_idmapped_mnt(m)) 41179caccd41SChristian Brauner return -EPERM; 41189caccd41SChristian Brauner 41199caccd41SChristian Brauner /* The underlying filesystem doesn't support idmapped mounts yet. */ 41209caccd41SChristian Brauner if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP)) 41219caccd41SChristian Brauner return -EINVAL; 41229caccd41SChristian Brauner 41239caccd41SChristian Brauner /* We're not controlling the superblock. */ 4124bd303368SChristian Brauner if (!ns_capable(fs_userns, CAP_SYS_ADMIN)) 41259caccd41SChristian Brauner return -EPERM; 41269caccd41SChristian Brauner 41279caccd41SChristian Brauner /* Mount has already been visible in the filesystem hierarchy. */ 41289caccd41SChristian Brauner if (!is_anon_ns(mnt->mnt_ns)) 41299caccd41SChristian Brauner return -EINVAL; 41309caccd41SChristian Brauner 41319caccd41SChristian Brauner return 0; 41329caccd41SChristian Brauner } 41339caccd41SChristian Brauner 4134a26f788bSChristian Brauner /** 4135a26f788bSChristian Brauner * mnt_allow_writers() - check whether the attribute change allows writers 4136a26f788bSChristian Brauner * @kattr: the new mount attributes 4137a26f788bSChristian Brauner * @mnt: the mount to which @kattr will be applied 4138a26f788bSChristian Brauner * 4139a26f788bSChristian Brauner * Check whether thew new mount attributes in @kattr allow concurrent writers. 4140a26f788bSChristian Brauner * 4141a26f788bSChristian Brauner * Return: true if writers need to be held, false if not 4142a26f788bSChristian Brauner */ 4143a26f788bSChristian Brauner static inline bool mnt_allow_writers(const struct mount_kattr *kattr, 4144a26f788bSChristian Brauner const struct mount *mnt) 41452a186721SChristian Brauner { 4146e1bbcd27SChristian Brauner return (!(kattr->attr_set & MNT_READONLY) || 4147e1bbcd27SChristian Brauner (mnt->mnt.mnt_flags & MNT_READONLY)) && 4148256c8aedSChristian Brauner !kattr->mnt_idmap; 41492a186721SChristian Brauner } 41502a186721SChristian Brauner 415187bb5b60SChristian Brauner static int mount_setattr_prepare(struct mount_kattr *kattr, struct mount *mnt) 41522a186721SChristian Brauner { 4153e257039fSAl Viro struct mount *m; 4154e257039fSAl Viro int err; 41552a186721SChristian Brauner 4156e257039fSAl Viro for (m = mnt; m; m = next_mnt(m, mnt)) { 4157e257039fSAl Viro if (!can_change_locked_flags(m, recalc_flags(kattr, m))) { 4158e257039fSAl Viro err = -EPERM; 4159e257039fSAl Viro break; 41602a186721SChristian Brauner } 41612a186721SChristian Brauner 416287bb5b60SChristian Brauner err = can_idmap_mount(kattr, m); 416387bb5b60SChristian Brauner if (err) 4164e257039fSAl Viro break; 41652a186721SChristian Brauner 4166e257039fSAl Viro if (!mnt_allow_writers(kattr, m)) { 416787bb5b60SChristian Brauner err = mnt_hold_writers(m); 416887bb5b60SChristian Brauner if (err) 4169e257039fSAl Viro break; 41702a186721SChristian Brauner } 41712a186721SChristian Brauner 4172e257039fSAl Viro if (!kattr->recurse) 417387bb5b60SChristian Brauner return 0; 41742a186721SChristian Brauner } 41752a186721SChristian Brauner 4176e257039fSAl Viro if (err) { 4177e257039fSAl Viro struct mount *p; 4178e257039fSAl Viro 41790014edaeSChristian Brauner /* 41800014edaeSChristian Brauner * If we had to call mnt_hold_writers() MNT_WRITE_HOLD will 41810014edaeSChristian Brauner * be set in @mnt_flags. The loop unsets MNT_WRITE_HOLD for all 41820014edaeSChristian Brauner * mounts and needs to take care to include the first mount. 41830014edaeSChristian Brauner */ 41840014edaeSChristian Brauner for (p = mnt; p; p = next_mnt(p, mnt)) { 4185e257039fSAl Viro /* If we had to hold writers unblock them. */ 4186e257039fSAl Viro if (p->mnt.mnt_flags & MNT_WRITE_HOLD) 4187e257039fSAl Viro mnt_unhold_writers(p); 41880014edaeSChristian Brauner 41890014edaeSChristian Brauner /* 41900014edaeSChristian Brauner * We're done once the first mount we changed got 41910014edaeSChristian Brauner * MNT_WRITE_HOLD unset. 41920014edaeSChristian Brauner */ 41930014edaeSChristian Brauner if (p == m) 41940014edaeSChristian Brauner break; 4195e257039fSAl Viro } 4196e257039fSAl Viro } 4197e257039fSAl Viro return err; 41982a186721SChristian Brauner } 41992a186721SChristian Brauner 42009caccd41SChristian Brauner static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt) 42019caccd41SChristian Brauner { 4202256c8aedSChristian Brauner if (!kattr->mnt_idmap) 42039caccd41SChristian Brauner return; 42049caccd41SChristian Brauner 4205bd303368SChristian Brauner /* 4206256c8aedSChristian Brauner * Pairs with smp_load_acquire() in mnt_idmap(). 4207256c8aedSChristian Brauner * 4208256c8aedSChristian Brauner * Since we only allow a mount to change the idmapping once and 4209256c8aedSChristian Brauner * verified this in can_idmap_mount() we know that the mount has 4210256c8aedSChristian Brauner * @nop_mnt_idmap attached to it. So there's no need to drop any 4211256c8aedSChristian Brauner * references. 4212bd303368SChristian Brauner */ 4213256c8aedSChristian Brauner smp_store_release(&mnt->mnt.mnt_idmap, mnt_idmap_get(kattr->mnt_idmap)); 42149caccd41SChristian Brauner } 42159caccd41SChristian Brauner 4216e257039fSAl Viro static void mount_setattr_commit(struct mount_kattr *kattr, struct mount *mnt) 42172a186721SChristian Brauner { 4218e257039fSAl Viro struct mount *m; 42192a186721SChristian Brauner 4220e257039fSAl Viro for (m = mnt; m; m = next_mnt(m, mnt)) { 42212a186721SChristian Brauner unsigned int flags; 42222a186721SChristian Brauner 42239caccd41SChristian Brauner do_idmap_mount(kattr, m); 42242a186721SChristian Brauner flags = recalc_flags(kattr, m); 42252a186721SChristian Brauner WRITE_ONCE(m->mnt.mnt_flags, flags); 42262a186721SChristian Brauner 422703b6abeeSChristian Brauner /* If we had to hold writers unblock them. */ 422803b6abeeSChristian Brauner if (m->mnt.mnt_flags & MNT_WRITE_HOLD) 42292a186721SChristian Brauner mnt_unhold_writers(m); 42302a186721SChristian Brauner 4231e257039fSAl Viro if (kattr->propagation) 42322a186721SChristian Brauner change_mnt_propagation(m, kattr->propagation); 4233e257039fSAl Viro if (!kattr->recurse) 42342a186721SChristian Brauner break; 4235e257039fSAl Viro } 42362a186721SChristian Brauner touch_mnt_namespace(mnt->mnt_ns); 42372a186721SChristian Brauner } 42382a186721SChristian Brauner 42392a186721SChristian Brauner static int do_mount_setattr(struct path *path, struct mount_kattr *kattr) 42402a186721SChristian Brauner { 424187bb5b60SChristian Brauner struct mount *mnt = real_mount(path->mnt); 42422a186721SChristian Brauner int err = 0; 42432a186721SChristian Brauner 42442a186721SChristian Brauner if (path->dentry != mnt->mnt.mnt_root) 42452a186721SChristian Brauner return -EINVAL; 42462a186721SChristian Brauner 4247256c8aedSChristian Brauner if (kattr->mnt_userns) { 4248256c8aedSChristian Brauner struct mnt_idmap *mnt_idmap; 4249256c8aedSChristian Brauner 4250256c8aedSChristian Brauner mnt_idmap = alloc_mnt_idmap(kattr->mnt_userns); 4251256c8aedSChristian Brauner if (IS_ERR(mnt_idmap)) 4252256c8aedSChristian Brauner return PTR_ERR(mnt_idmap); 4253256c8aedSChristian Brauner kattr->mnt_idmap = mnt_idmap; 4254256c8aedSChristian Brauner } 4255256c8aedSChristian Brauner 42562a186721SChristian Brauner if (kattr->propagation) { 42572a186721SChristian Brauner /* 42582a186721SChristian Brauner * Only take namespace_lock() if we're actually changing 42592a186721SChristian Brauner * propagation. 42602a186721SChristian Brauner */ 42612a186721SChristian Brauner namespace_lock(); 42622a186721SChristian Brauner if (kattr->propagation == MS_SHARED) { 42632a186721SChristian Brauner err = invent_group_ids(mnt, kattr->recurse); 42642a186721SChristian Brauner if (err) { 42652a186721SChristian Brauner namespace_unlock(); 42662a186721SChristian Brauner return err; 42672a186721SChristian Brauner } 42682a186721SChristian Brauner } 42692a186721SChristian Brauner } 42702a186721SChristian Brauner 427187bb5b60SChristian Brauner err = -EINVAL; 42722a186721SChristian Brauner lock_mount_hash(); 42732a186721SChristian Brauner 427487bb5b60SChristian Brauner /* Ensure that this isn't anything purely vfs internal. */ 427587bb5b60SChristian Brauner if (!is_mounted(&mnt->mnt)) 427687bb5b60SChristian Brauner goto out; 42772a186721SChristian Brauner 427887bb5b60SChristian Brauner /* 427987bb5b60SChristian Brauner * If this is an attached mount make sure it's located in the callers 428087bb5b60SChristian Brauner * mount namespace. If it's not don't let the caller interact with it. 428187bb5b60SChristian Brauner * If this is a detached mount make sure it has an anonymous mount 428287bb5b60SChristian Brauner * namespace attached to it, i.e. we've created it via OPEN_TREE_CLONE. 428387bb5b60SChristian Brauner */ 428487bb5b60SChristian Brauner if (!(mnt_has_parent(mnt) ? check_mnt(mnt) : is_anon_ns(mnt->mnt_ns))) 428587bb5b60SChristian Brauner goto out; 428687bb5b60SChristian Brauner 428787bb5b60SChristian Brauner /* 428887bb5b60SChristian Brauner * First, we get the mount tree in a shape where we can change mount 428987bb5b60SChristian Brauner * properties without failure. If we succeeded to do so we commit all 429087bb5b60SChristian Brauner * changes and if we failed we clean up. 429187bb5b60SChristian Brauner */ 429287bb5b60SChristian Brauner err = mount_setattr_prepare(kattr, mnt); 4293e257039fSAl Viro if (!err) 4294e257039fSAl Viro mount_setattr_commit(kattr, mnt); 429587bb5b60SChristian Brauner 429687bb5b60SChristian Brauner out: 42972a186721SChristian Brauner unlock_mount_hash(); 42982a186721SChristian Brauner 42992a186721SChristian Brauner if (kattr->propagation) { 43002a186721SChristian Brauner namespace_unlock(); 43012a186721SChristian Brauner if (err) 43022a186721SChristian Brauner cleanup_group_ids(mnt, NULL); 43032a186721SChristian Brauner } 43042a186721SChristian Brauner 43052a186721SChristian Brauner return err; 43062a186721SChristian Brauner } 43072a186721SChristian Brauner 43089caccd41SChristian Brauner static int build_mount_idmapped(const struct mount_attr *attr, size_t usize, 43099caccd41SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 43109caccd41SChristian Brauner { 43119caccd41SChristian Brauner int err = 0; 43129caccd41SChristian Brauner struct ns_common *ns; 43139caccd41SChristian Brauner struct user_namespace *mnt_userns; 43149caccd41SChristian Brauner struct file *file; 43159caccd41SChristian Brauner 43169caccd41SChristian Brauner if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP)) 43179caccd41SChristian Brauner return 0; 43189caccd41SChristian Brauner 43199caccd41SChristian Brauner /* 43209caccd41SChristian Brauner * We currently do not support clearing an idmapped mount. If this ever 43219caccd41SChristian Brauner * is a use-case we can revisit this but for now let's keep it simple 43229caccd41SChristian Brauner * and not allow it. 43239caccd41SChristian Brauner */ 43249caccd41SChristian Brauner if (attr->attr_clr & MOUNT_ATTR_IDMAP) 43259caccd41SChristian Brauner return -EINVAL; 43269caccd41SChristian Brauner 43279caccd41SChristian Brauner if (attr->userns_fd > INT_MAX) 43289caccd41SChristian Brauner return -EINVAL; 43299caccd41SChristian Brauner 43309caccd41SChristian Brauner file = fget(attr->userns_fd); 43319caccd41SChristian Brauner if (!file) 43329caccd41SChristian Brauner return -EBADF; 43339caccd41SChristian Brauner 43349caccd41SChristian Brauner if (!proc_ns_file(file)) { 43359caccd41SChristian Brauner err = -EINVAL; 43369caccd41SChristian Brauner goto out_fput; 43379caccd41SChristian Brauner } 43389caccd41SChristian Brauner 43399caccd41SChristian Brauner ns = get_proc_ns(file_inode(file)); 43409caccd41SChristian Brauner if (ns->ops->type != CLONE_NEWUSER) { 43419caccd41SChristian Brauner err = -EINVAL; 43429caccd41SChristian Brauner goto out_fput; 43439caccd41SChristian Brauner } 43449caccd41SChristian Brauner 43459caccd41SChristian Brauner /* 4346bd303368SChristian Brauner * The initial idmapping cannot be used to create an idmapped 4347bd303368SChristian Brauner * mount. We use the initial idmapping as an indicator of a mount 4348bd303368SChristian Brauner * that is not idmapped. It can simply be passed into helpers that 4349bd303368SChristian Brauner * are aware of idmapped mounts as a convenient shortcut. A user 4350bd303368SChristian Brauner * can just create a dedicated identity mapping to achieve the same 4351bd303368SChristian Brauner * result. 43529caccd41SChristian Brauner */ 43539caccd41SChristian Brauner mnt_userns = container_of(ns, struct user_namespace, ns); 4354bd303368SChristian Brauner if (initial_idmapping(mnt_userns)) { 43559caccd41SChristian Brauner err = -EPERM; 43569caccd41SChristian Brauner goto out_fput; 43579caccd41SChristian Brauner } 4358bf1ac16eSSeth Forshee 4359bf1ac16eSSeth Forshee /* We're not controlling the target namespace. */ 4360bf1ac16eSSeth Forshee if (!ns_capable(mnt_userns, CAP_SYS_ADMIN)) { 4361bf1ac16eSSeth Forshee err = -EPERM; 4362bf1ac16eSSeth Forshee goto out_fput; 4363bf1ac16eSSeth Forshee } 4364bf1ac16eSSeth Forshee 43659caccd41SChristian Brauner kattr->mnt_userns = get_user_ns(mnt_userns); 43669caccd41SChristian Brauner 43679caccd41SChristian Brauner out_fput: 43689caccd41SChristian Brauner fput(file); 43699caccd41SChristian Brauner return err; 43709caccd41SChristian Brauner } 43719caccd41SChristian Brauner 43729caccd41SChristian Brauner static int build_mount_kattr(const struct mount_attr *attr, size_t usize, 43732a186721SChristian Brauner struct mount_kattr *kattr, unsigned int flags) 43742a186721SChristian Brauner { 43752a186721SChristian Brauner unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW; 43762a186721SChristian Brauner 43772a186721SChristian Brauner if (flags & AT_NO_AUTOMOUNT) 43782a186721SChristian Brauner lookup_flags &= ~LOOKUP_AUTOMOUNT; 43792a186721SChristian Brauner if (flags & AT_SYMLINK_NOFOLLOW) 43802a186721SChristian Brauner lookup_flags &= ~LOOKUP_FOLLOW; 43812a186721SChristian Brauner if (flags & AT_EMPTY_PATH) 43822a186721SChristian Brauner lookup_flags |= LOOKUP_EMPTY; 43832a186721SChristian Brauner 43842a186721SChristian Brauner *kattr = (struct mount_kattr) { 43852a186721SChristian Brauner .lookup_flags = lookup_flags, 43862a186721SChristian Brauner .recurse = !!(flags & AT_RECURSIVE), 43872a186721SChristian Brauner }; 43882a186721SChristian Brauner 43892a186721SChristian Brauner if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS) 43902a186721SChristian Brauner return -EINVAL; 43912a186721SChristian Brauner if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1) 43922a186721SChristian Brauner return -EINVAL; 43932a186721SChristian Brauner kattr->propagation = attr->propagation; 43942a186721SChristian Brauner 43952a186721SChristian Brauner if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS) 43962a186721SChristian Brauner return -EINVAL; 43972a186721SChristian Brauner 43982a186721SChristian Brauner kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set); 43992a186721SChristian Brauner kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr); 44002a186721SChristian Brauner 44012a186721SChristian Brauner /* 44022a186721SChristian Brauner * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap, 44032a186721SChristian Brauner * users wanting to transition to a different atime setting cannot 44042a186721SChristian Brauner * simply specify the atime setting in @attr_set, but must also 44052a186721SChristian Brauner * specify MOUNT_ATTR__ATIME in the @attr_clr field. 44062a186721SChristian Brauner * So ensure that MOUNT_ATTR__ATIME can't be partially set in 44072a186721SChristian Brauner * @attr_clr and that @attr_set can't have any atime bits set if 44082a186721SChristian Brauner * MOUNT_ATTR__ATIME isn't set in @attr_clr. 44092a186721SChristian Brauner */ 44102a186721SChristian Brauner if (attr->attr_clr & MOUNT_ATTR__ATIME) { 44112a186721SChristian Brauner if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME) 44122a186721SChristian Brauner return -EINVAL; 44132a186721SChristian Brauner 44142a186721SChristian Brauner /* 44152a186721SChristian Brauner * Clear all previous time settings as they are mutually 44162a186721SChristian Brauner * exclusive. 44172a186721SChristian Brauner */ 44182a186721SChristian Brauner kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME; 44192a186721SChristian Brauner switch (attr->attr_set & MOUNT_ATTR__ATIME) { 44202a186721SChristian Brauner case MOUNT_ATTR_RELATIME: 44212a186721SChristian Brauner kattr->attr_set |= MNT_RELATIME; 44222a186721SChristian Brauner break; 44232a186721SChristian Brauner case MOUNT_ATTR_NOATIME: 44242a186721SChristian Brauner kattr->attr_set |= MNT_NOATIME; 44252a186721SChristian Brauner break; 44262a186721SChristian Brauner case MOUNT_ATTR_STRICTATIME: 44272a186721SChristian Brauner break; 44282a186721SChristian Brauner default: 44292a186721SChristian Brauner return -EINVAL; 44302a186721SChristian Brauner } 44312a186721SChristian Brauner } else { 44322a186721SChristian Brauner if (attr->attr_set & MOUNT_ATTR__ATIME) 44332a186721SChristian Brauner return -EINVAL; 44342a186721SChristian Brauner } 44352a186721SChristian Brauner 44369caccd41SChristian Brauner return build_mount_idmapped(attr, usize, kattr, flags); 44379caccd41SChristian Brauner } 44389caccd41SChristian Brauner 44399caccd41SChristian Brauner static void finish_mount_kattr(struct mount_kattr *kattr) 44409caccd41SChristian Brauner { 44419caccd41SChristian Brauner put_user_ns(kattr->mnt_userns); 44429caccd41SChristian Brauner kattr->mnt_userns = NULL; 4443256c8aedSChristian Brauner 4444256c8aedSChristian Brauner if (kattr->mnt_idmap) 4445256c8aedSChristian Brauner mnt_idmap_put(kattr->mnt_idmap); 44462a186721SChristian Brauner } 44472a186721SChristian Brauner 44482a186721SChristian Brauner SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path, 44492a186721SChristian Brauner unsigned int, flags, struct mount_attr __user *, uattr, 44502a186721SChristian Brauner size_t, usize) 44512a186721SChristian Brauner { 44522a186721SChristian Brauner int err; 44532a186721SChristian Brauner struct path target; 44542a186721SChristian Brauner struct mount_attr attr; 44552a186721SChristian Brauner struct mount_kattr kattr; 44562a186721SChristian Brauner 44572a186721SChristian Brauner BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0); 44582a186721SChristian Brauner 44592a186721SChristian Brauner if (flags & ~(AT_EMPTY_PATH | 44602a186721SChristian Brauner AT_RECURSIVE | 44612a186721SChristian Brauner AT_SYMLINK_NOFOLLOW | 44622a186721SChristian Brauner AT_NO_AUTOMOUNT)) 44632a186721SChristian Brauner return -EINVAL; 44642a186721SChristian Brauner 44652a186721SChristian Brauner if (unlikely(usize > PAGE_SIZE)) 44662a186721SChristian Brauner return -E2BIG; 44672a186721SChristian Brauner if (unlikely(usize < MOUNT_ATTR_SIZE_VER0)) 44682a186721SChristian Brauner return -EINVAL; 44692a186721SChristian Brauner 44702a186721SChristian Brauner if (!may_mount()) 44712a186721SChristian Brauner return -EPERM; 44722a186721SChristian Brauner 44732a186721SChristian Brauner err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize); 44742a186721SChristian Brauner if (err) 44752a186721SChristian Brauner return err; 44762a186721SChristian Brauner 44772a186721SChristian Brauner /* Don't bother walking through the mounts if this is a nop. */ 44782a186721SChristian Brauner if (attr.attr_set == 0 && 44792a186721SChristian Brauner attr.attr_clr == 0 && 44802a186721SChristian Brauner attr.propagation == 0) 44812a186721SChristian Brauner return 0; 44822a186721SChristian Brauner 44839caccd41SChristian Brauner err = build_mount_kattr(&attr, usize, &kattr, flags); 44842a186721SChristian Brauner if (err) 44852a186721SChristian Brauner return err; 44862a186721SChristian Brauner 44872a186721SChristian Brauner err = user_path_at(dfd, path, kattr.lookup_flags, &target); 4488012e3322SChristian Brauner if (!err) { 44892a186721SChristian Brauner err = do_mount_setattr(&target, &kattr); 44902a186721SChristian Brauner path_put(&target); 4491012e3322SChristian Brauner } 4492012e3322SChristian Brauner finish_mount_kattr(&kattr); 44932a186721SChristian Brauner return err; 44942a186721SChristian Brauner } 44952a186721SChristian Brauner 44961da177e4SLinus Torvalds static void __init init_mount_tree(void) 44971da177e4SLinus Torvalds { 44981da177e4SLinus Torvalds struct vfsmount *mnt; 449974e83122SAl Viro struct mount *m; 45006b3286edSKirill Korotaev struct mnt_namespace *ns; 4501ac748a09SJan Blunck struct path root; 45021da177e4SLinus Torvalds 4503fd3e007fSAl Viro mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL); 45041da177e4SLinus Torvalds if (IS_ERR(mnt)) 45051da177e4SLinus Torvalds panic("Can't create rootfs"); 4506b3e19d92SNick Piggin 450774e83122SAl Viro ns = alloc_mnt_ns(&init_user_ns, false); 45083b22edc5STrond Myklebust if (IS_ERR(ns)) 45091da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 451074e83122SAl Viro m = real_mount(mnt); 451174e83122SAl Viro m->mnt_ns = ns; 451274e83122SAl Viro ns->root = m; 451374e83122SAl Viro ns->mounts = 1; 451474e83122SAl Viro list_add(&m->mnt_list, &ns->list); 45156b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 45166b3286edSKirill Korotaev get_mnt_ns(ns); 45171da177e4SLinus Torvalds 4518be08d6d2SAl Viro root.mnt = mnt; 4519be08d6d2SAl Viro root.dentry = mnt->mnt_root; 4520da362b09SEric W. Biederman mnt->mnt_flags |= MNT_LOCKED; 4521ac748a09SJan Blunck 4522ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 4523ac748a09SJan Blunck set_fs_root(current->fs, &root); 45241da177e4SLinus Torvalds } 45251da177e4SLinus Torvalds 452674bf17cfSDenis Cheng void __init mnt_init(void) 45271da177e4SLinus Torvalds { 452815a67dd8SRandy Dunlap int err; 45291da177e4SLinus Torvalds 45307d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 453179f6540bSVasily Averin 0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL); 45321da177e4SLinus Torvalds 45330818bf27SAl Viro mount_hashtable = alloc_large_system_hash("Mount-cache", 453438129a13SAl Viro sizeof(struct hlist_head), 45350818bf27SAl Viro mhash_entries, 19, 45363d375d78SPavel Tatashin HASH_ZERO, 45370818bf27SAl Viro &m_hash_shift, &m_hash_mask, 0, 0); 45380818bf27SAl Viro mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache", 45390818bf27SAl Viro sizeof(struct hlist_head), 45400818bf27SAl Viro mphash_entries, 19, 45413d375d78SPavel Tatashin HASH_ZERO, 45420818bf27SAl Viro &mp_hash_shift, &mp_hash_mask, 0, 0); 45431da177e4SLinus Torvalds 454484d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 45451da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 45461da177e4SLinus Torvalds 45474b93dc9bSTejun Heo kernfs_init(); 45484b93dc9bSTejun Heo 454915a67dd8SRandy Dunlap err = sysfs_init(); 455015a67dd8SRandy Dunlap if (err) 455115a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 45528e24eea7SHarvey Harrison __func__, err); 455300d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 455400d26666SGreg Kroah-Hartman if (!fs_kobj) 45558e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 4556037f11b4SAl Viro shmem_init(); 45571da177e4SLinus Torvalds init_rootfs(); 45581da177e4SLinus Torvalds init_mount_tree(); 45591da177e4SLinus Torvalds } 45601da177e4SLinus Torvalds 4561616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 45621da177e4SLinus Torvalds { 45631a7b8969SKirill Tkhai if (!refcount_dec_and_test(&ns->ns.count)) 4564616511d0STrond Myklebust return; 45657b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 4566771b1371SEric W. Biederman free_mnt_ns(ns); 45671da177e4SLinus Torvalds } 45689d412a43SAl Viro 4569d911b458SDavid Howells struct vfsmount *kern_mount(struct file_system_type *type) 45709d412a43SAl Viro { 4571423e0ab0STim Chen struct vfsmount *mnt; 4572d911b458SDavid Howells mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL); 4573423e0ab0STim Chen if (!IS_ERR(mnt)) { 4574423e0ab0STim Chen /* 4575423e0ab0STim Chen * it is a longterm mount, don't release mnt until 4576423e0ab0STim Chen * we unmount before file sys is unregistered 4577423e0ab0STim Chen */ 4578f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 4579423e0ab0STim Chen } 4580423e0ab0STim Chen return mnt; 45819d412a43SAl Viro } 4582d911b458SDavid Howells EXPORT_SYMBOL_GPL(kern_mount); 4583423e0ab0STim Chen 4584423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 4585423e0ab0STim Chen { 4586423e0ab0STim Chen /* release long term mount so mount point can be released */ 4587*da27f796SRik van Riel if (!IS_ERR(mnt)) { 4588*da27f796SRik van Riel mnt_make_shortterm(mnt); 458948a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 4590423e0ab0STim Chen mntput(mnt); 4591423e0ab0STim Chen } 4592423e0ab0STim Chen } 4593423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 459402125a82SAl Viro 4595df820f8dSMiklos Szeredi void kern_unmount_array(struct vfsmount *mnt[], unsigned int num) 4596df820f8dSMiklos Szeredi { 4597df820f8dSMiklos Szeredi unsigned int i; 4598df820f8dSMiklos Szeredi 4599df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4600*da27f796SRik van Riel mnt_make_shortterm(mnt[i]); 4601df820f8dSMiklos Szeredi synchronize_rcu_expedited(); 4602df820f8dSMiklos Szeredi for (i = 0; i < num; i++) 4603df820f8dSMiklos Szeredi mntput(mnt[i]); 4604df820f8dSMiklos Szeredi } 4605df820f8dSMiklos Szeredi EXPORT_SYMBOL(kern_unmount_array); 4606df820f8dSMiklos Szeredi 460702125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 460802125a82SAl Viro { 4609143c8c91SAl Viro return check_mnt(real_mount(mnt)); 461002125a82SAl Viro } 46118823c079SEric W. Biederman 46123151527eSEric W. Biederman bool current_chrooted(void) 46133151527eSEric W. Biederman { 46143151527eSEric W. Biederman /* Does the current process have a non-standard root */ 46153151527eSEric W. Biederman struct path ns_root; 46163151527eSEric W. Biederman struct path fs_root; 46173151527eSEric W. Biederman bool chrooted; 46183151527eSEric W. Biederman 46193151527eSEric W. Biederman /* Find the namespace root */ 46203151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 46213151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 46223151527eSEric W. Biederman path_get(&ns_root); 46233151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 46243151527eSEric W. Biederman ; 46253151527eSEric W. Biederman 46263151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 46273151527eSEric W. Biederman 46283151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 46293151527eSEric W. Biederman 46303151527eSEric W. Biederman path_put(&fs_root); 46313151527eSEric W. Biederman path_put(&ns_root); 46323151527eSEric W. Biederman 46333151527eSEric W. Biederman return chrooted; 46343151527eSEric W. Biederman } 46353151527eSEric W. Biederman 4636132e4608SDavid Howells static bool mnt_already_visible(struct mnt_namespace *ns, 4637132e4608SDavid Howells const struct super_block *sb, 46388654df4eSEric W. Biederman int *new_mnt_flags) 463987a8ebd6SEric W. Biederman { 46408c6cf9ccSEric W. Biederman int new_flags = *new_mnt_flags; 464187a8ebd6SEric W. Biederman struct mount *mnt; 4642e51db735SEric W. Biederman bool visible = false; 464387a8ebd6SEric W. Biederman 464444bb4385SAl Viro down_read(&namespace_sem); 46459f6c61f9SMiklos Szeredi lock_ns_list(ns); 464687a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 4647e51db735SEric W. Biederman struct mount *child; 464877b1a97dSEric W. Biederman int mnt_flags; 464977b1a97dSEric W. Biederman 46509f6c61f9SMiklos Szeredi if (mnt_is_cursor(mnt)) 46519f6c61f9SMiklos Szeredi continue; 46529f6c61f9SMiklos Szeredi 4653132e4608SDavid Howells if (mnt->mnt.mnt_sb->s_type != sb->s_type) 4654e51db735SEric W. Biederman continue; 4655e51db735SEric W. Biederman 46567e96c1b0SEric W. Biederman /* This mount is not fully visible if it's root directory 46577e96c1b0SEric W. Biederman * is not the root directory of the filesystem. 46587e96c1b0SEric W. Biederman */ 46597e96c1b0SEric W. Biederman if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root) 46607e96c1b0SEric W. Biederman continue; 46617e96c1b0SEric W. Biederman 4662a1935c17SEric W. Biederman /* A local view of the mount flags */ 466377b1a97dSEric W. Biederman mnt_flags = mnt->mnt.mnt_flags; 466477b1a97dSEric W. Biederman 4665695e9df0SEric W. Biederman /* Don't miss readonly hidden in the superblock flags */ 4666bc98a42cSDavid Howells if (sb_rdonly(mnt->mnt.mnt_sb)) 4667695e9df0SEric W. Biederman mnt_flags |= MNT_LOCK_READONLY; 4668695e9df0SEric W. Biederman 46698c6cf9ccSEric W. Biederman /* Verify the mount flags are equal to or more permissive 46708c6cf9ccSEric W. Biederman * than the proposed new mount. 46718c6cf9ccSEric W. Biederman */ 467277b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_READONLY) && 46738c6cf9ccSEric W. Biederman !(new_flags & MNT_READONLY)) 46748c6cf9ccSEric W. Biederman continue; 467577b1a97dSEric W. Biederman if ((mnt_flags & MNT_LOCK_ATIME) && 467677b1a97dSEric W. Biederman ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK))) 46778c6cf9ccSEric W. Biederman continue; 46788c6cf9ccSEric W. Biederman 4679ceeb0e5dSEric W. Biederman /* This mount is not fully visible if there are any 4680ceeb0e5dSEric W. Biederman * locked child mounts that cover anything except for 4681ceeb0e5dSEric W. Biederman * empty directories. 4682e51db735SEric W. Biederman */ 4683e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 4684e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 4685ceeb0e5dSEric W. Biederman /* Only worry about locked mounts */ 4686d71ed6c9SEric W. Biederman if (!(child->mnt.mnt_flags & MNT_LOCKED)) 4687ceeb0e5dSEric W. Biederman continue; 46887236c85eSEric W. Biederman /* Is the directory permanetly empty? */ 46897236c85eSEric W. Biederman if (!is_empty_dir_inode(inode)) 4690e51db735SEric W. Biederman goto next; 469187a8ebd6SEric W. Biederman } 46928c6cf9ccSEric W. Biederman /* Preserve the locked attributes */ 469377b1a97dSEric W. Biederman *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \ 46948c6cf9ccSEric W. Biederman MNT_LOCK_ATIME); 4695e51db735SEric W. Biederman visible = true; 4696e51db735SEric W. Biederman goto found; 4697e51db735SEric W. Biederman next: ; 469887a8ebd6SEric W. Biederman } 4699e51db735SEric W. Biederman found: 47009f6c61f9SMiklos Szeredi unlock_ns_list(ns); 470144bb4385SAl Viro up_read(&namespace_sem); 4702e51db735SEric W. Biederman return visible; 470387a8ebd6SEric W. Biederman } 470487a8ebd6SEric W. Biederman 4705132e4608SDavid Howells static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags) 47068654df4eSEric W. Biederman { 4707a1935c17SEric W. Biederman const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV; 47088654df4eSEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 47098654df4eSEric W. Biederman unsigned long s_iflags; 47108654df4eSEric W. Biederman 47118654df4eSEric W. Biederman if (ns->user_ns == &init_user_ns) 47128654df4eSEric W. Biederman return false; 47138654df4eSEric W. Biederman 47148654df4eSEric W. Biederman /* Can this filesystem be too revealing? */ 4715132e4608SDavid Howells s_iflags = sb->s_iflags; 47168654df4eSEric W. Biederman if (!(s_iflags & SB_I_USERNS_VISIBLE)) 47178654df4eSEric W. Biederman return false; 47188654df4eSEric W. Biederman 4719a1935c17SEric W. Biederman if ((s_iflags & required_iflags) != required_iflags) { 4720a1935c17SEric W. Biederman WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n", 4721a1935c17SEric W. Biederman required_iflags); 4722a1935c17SEric W. Biederman return true; 4723a1935c17SEric W. Biederman } 4724a1935c17SEric W. Biederman 4725132e4608SDavid Howells return !mnt_already_visible(ns, sb, new_mnt_flags); 47268654df4eSEric W. Biederman } 47278654df4eSEric W. Biederman 4728380cf5baSAndy Lutomirski bool mnt_may_suid(struct vfsmount *mnt) 4729380cf5baSAndy Lutomirski { 4730380cf5baSAndy Lutomirski /* 4731380cf5baSAndy Lutomirski * Foreign mounts (accessed via fchdir or through /proc 4732380cf5baSAndy Lutomirski * symlinks) are always treated as if they are nosuid. This 4733380cf5baSAndy Lutomirski * prevents namespaces from trusting potentially unsafe 4734380cf5baSAndy Lutomirski * suid/sgid bits, file caps, or security labels that originate 4735380cf5baSAndy Lutomirski * in other namespaces. 4736380cf5baSAndy Lutomirski */ 4737380cf5baSAndy Lutomirski return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) && 4738380cf5baSAndy Lutomirski current_in_userns(mnt->mnt_sb->s_user_ns); 4739380cf5baSAndy Lutomirski } 4740380cf5baSAndy Lutomirski 474164964528SAl Viro static struct ns_common *mntns_get(struct task_struct *task) 47428823c079SEric W. Biederman { 474358be2825SAl Viro struct ns_common *ns = NULL; 47448823c079SEric W. Biederman struct nsproxy *nsproxy; 47458823c079SEric W. Biederman 4746728dba3aSEric W. Biederman task_lock(task); 4747728dba3aSEric W. Biederman nsproxy = task->nsproxy; 47488823c079SEric W. Biederman if (nsproxy) { 474958be2825SAl Viro ns = &nsproxy->mnt_ns->ns; 475058be2825SAl Viro get_mnt_ns(to_mnt_ns(ns)); 47518823c079SEric W. Biederman } 4752728dba3aSEric W. Biederman task_unlock(task); 47538823c079SEric W. Biederman 47548823c079SEric W. Biederman return ns; 47558823c079SEric W. Biederman } 47568823c079SEric W. Biederman 475764964528SAl Viro static void mntns_put(struct ns_common *ns) 47588823c079SEric W. Biederman { 475958be2825SAl Viro put_mnt_ns(to_mnt_ns(ns)); 47608823c079SEric W. Biederman } 47618823c079SEric W. Biederman 4762f2a8d52eSChristian Brauner static int mntns_install(struct nsset *nsset, struct ns_common *ns) 47638823c079SEric W. Biederman { 4764f2a8d52eSChristian Brauner struct nsproxy *nsproxy = nsset->nsproxy; 4765f2a8d52eSChristian Brauner struct fs_struct *fs = nsset->fs; 47664f757f3cSAl Viro struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns; 4767f2a8d52eSChristian Brauner struct user_namespace *user_ns = nsset->cred->user_ns; 47688823c079SEric W. Biederman struct path root; 47694f757f3cSAl Viro int err; 47708823c079SEric W. Biederman 47710c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 4772f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_CHROOT) || 4773f2a8d52eSChristian Brauner !ns_capable(user_ns, CAP_SYS_ADMIN)) 4774ae11e0f1SZhao Hongjiang return -EPERM; 47758823c079SEric W. Biederman 477674e83122SAl Viro if (is_anon_ns(mnt_ns)) 477774e83122SAl Viro return -EINVAL; 477874e83122SAl Viro 47798823c079SEric W. Biederman if (fs->users != 1) 47808823c079SEric W. Biederman return -EINVAL; 47818823c079SEric W. Biederman 47828823c079SEric W. Biederman get_mnt_ns(mnt_ns); 47834f757f3cSAl Viro old_mnt_ns = nsproxy->mnt_ns; 47848823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 47858823c079SEric W. Biederman 47868823c079SEric W. Biederman /* Find the root */ 47874f757f3cSAl Viro err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt, 47884f757f3cSAl Viro "/", LOOKUP_DOWN, &root); 47894f757f3cSAl Viro if (err) { 47904f757f3cSAl Viro /* revert to old namespace */ 47914f757f3cSAl Viro nsproxy->mnt_ns = old_mnt_ns; 47924f757f3cSAl Viro put_mnt_ns(mnt_ns); 47934f757f3cSAl Viro return err; 47944f757f3cSAl Viro } 47958823c079SEric W. Biederman 47964068367cSAndrei Vagin put_mnt_ns(old_mnt_ns); 47974068367cSAndrei Vagin 47988823c079SEric W. Biederman /* Update the pwd and root */ 47998823c079SEric W. Biederman set_fs_pwd(fs, &root); 48008823c079SEric W. Biederman set_fs_root(fs, &root); 48018823c079SEric W. Biederman 48028823c079SEric W. Biederman path_put(&root); 48038823c079SEric W. Biederman return 0; 48048823c079SEric W. Biederman } 48058823c079SEric W. Biederman 4806bcac25a5SAndrey Vagin static struct user_namespace *mntns_owner(struct ns_common *ns) 4807bcac25a5SAndrey Vagin { 4808bcac25a5SAndrey Vagin return to_mnt_ns(ns)->user_ns; 4809bcac25a5SAndrey Vagin } 4810bcac25a5SAndrey Vagin 48118823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 48128823c079SEric W. Biederman .name = "mnt", 48138823c079SEric W. Biederman .type = CLONE_NEWNS, 48148823c079SEric W. Biederman .get = mntns_get, 48158823c079SEric W. Biederman .put = mntns_put, 48168823c079SEric W. Biederman .install = mntns_install, 4817bcac25a5SAndrey Vagin .owner = mntns_owner, 48188823c079SEric W. Biederman }; 4819ab171b95SLuis Chamberlain 4820ab171b95SLuis Chamberlain #ifdef CONFIG_SYSCTL 4821ab171b95SLuis Chamberlain static struct ctl_table fs_namespace_sysctls[] = { 4822ab171b95SLuis Chamberlain { 4823ab171b95SLuis Chamberlain .procname = "mount-max", 4824ab171b95SLuis Chamberlain .data = &sysctl_mount_max, 4825ab171b95SLuis Chamberlain .maxlen = sizeof(unsigned int), 4826ab171b95SLuis Chamberlain .mode = 0644, 4827ab171b95SLuis Chamberlain .proc_handler = proc_dointvec_minmax, 4828ab171b95SLuis Chamberlain .extra1 = SYSCTL_ONE, 4829ab171b95SLuis Chamberlain }, 4830ab171b95SLuis Chamberlain { } 4831ab171b95SLuis Chamberlain }; 4832ab171b95SLuis Chamberlain 4833ab171b95SLuis Chamberlain static int __init init_fs_namespace_sysctls(void) 4834ab171b95SLuis Chamberlain { 4835ab171b95SLuis Chamberlain register_sysctl_init("fs", fs_namespace_sysctls); 4836ab171b95SLuis Chamberlain return 0; 4837ab171b95SLuis Chamberlain } 4838ab171b95SLuis Chamberlain fs_initcall(init_fs_namespace_sysctls); 4839ab171b95SLuis Chamberlain 4840ab171b95SLuis Chamberlain #endif /* CONFIG_SYSCTL */ 4841