11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 1873cd49ecSMiklos Szeredi #include <linux/idr.h> 19d10577a8SAl Viro #include <linux/acct.h> /* acct_auto_close_mnt */ 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23d10577a8SAl Viro #include <linux/uaccess.h> 240bb80f24SDavid Howells #include <linux/proc_ns.h> 2520b4fb48SLinus Torvalds #include <linux/magic.h> 2607b20889SRam Pai #include "pnode.h" 27948730b0SAdrian Bunk #include "internal.h" 281da177e4SLinus Torvalds 2913f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 3013f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 3113f14b4dSEric Dumazet 325addc5ddSAl Viro static int event; 3373cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 34719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 3599b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 36f21f6220SAl Viro static int mnt_id_start = 0; 37f21f6220SAl Viro static int mnt_group_start = 1; 385addc5ddSAl Viro 39fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 4084d17192SAl Viro static struct list_head *mountpoint_hashtable __read_mostly; 41e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 4259aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 431da177e4SLinus Torvalds 44f87fd4c2SMiklos Szeredi /* /sys/fs */ 4500d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 4600d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 47f87fd4c2SMiklos Szeredi 4899b7db7bSNick Piggin /* 4999b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 5099b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 5199b7db7bSNick Piggin * up the tree. 5299b7db7bSNick Piggin * 5399b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 5499b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 5599b7db7bSNick Piggin */ 5699b7db7bSNick Piggin DEFINE_BRLOCK(vfsmount_lock); 5799b7db7bSNick Piggin 581da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 591da177e4SLinus Torvalds { 601da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 611da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 6213f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 6313f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 641da177e4SLinus Torvalds } 651da177e4SLinus Torvalds 6699b7db7bSNick Piggin /* 6799b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 6899b7db7bSNick Piggin * serialize with freeing. 6999b7db7bSNick Piggin */ 70b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 7173cd49ecSMiklos Szeredi { 7273cd49ecSMiklos Szeredi int res; 7373cd49ecSMiklos Szeredi 7473cd49ecSMiklos Szeredi retry: 7573cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 7699b7db7bSNick Piggin spin_lock(&mnt_id_lock); 7715169fe7SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 78f21f6220SAl Viro if (!res) 7915169fe7SAl Viro mnt_id_start = mnt->mnt_id + 1; 8099b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 8173cd49ecSMiklos Szeredi if (res == -EAGAIN) 8273cd49ecSMiklos Szeredi goto retry; 8373cd49ecSMiklos Szeredi 8473cd49ecSMiklos Szeredi return res; 8573cd49ecSMiklos Szeredi } 8673cd49ecSMiklos Szeredi 87b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 8873cd49ecSMiklos Szeredi { 8915169fe7SAl Viro int id = mnt->mnt_id; 9099b7db7bSNick Piggin spin_lock(&mnt_id_lock); 91f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 92f21f6220SAl Viro if (mnt_id_start > id) 93f21f6220SAl Viro mnt_id_start = id; 9499b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 9573cd49ecSMiklos Szeredi } 9673cd49ecSMiklos Szeredi 97719f5d7fSMiklos Szeredi /* 98719f5d7fSMiklos Szeredi * Allocate a new peer group ID 99719f5d7fSMiklos Szeredi * 100719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 101719f5d7fSMiklos Szeredi */ 1024b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 103719f5d7fSMiklos Szeredi { 104f21f6220SAl Viro int res; 105f21f6220SAl Viro 106719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 107719f5d7fSMiklos Szeredi return -ENOMEM; 108719f5d7fSMiklos Szeredi 109f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 110f21f6220SAl Viro mnt_group_start, 11115169fe7SAl Viro &mnt->mnt_group_id); 112f21f6220SAl Viro if (!res) 11315169fe7SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 114f21f6220SAl Viro 115f21f6220SAl Viro return res; 116719f5d7fSMiklos Szeredi } 117719f5d7fSMiklos Szeredi 118719f5d7fSMiklos Szeredi /* 119719f5d7fSMiklos Szeredi * Release a peer group ID 120719f5d7fSMiklos Szeredi */ 1214b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 122719f5d7fSMiklos Szeredi { 12315169fe7SAl Viro int id = mnt->mnt_group_id; 124f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 125f21f6220SAl Viro if (mnt_group_start > id) 126f21f6220SAl Viro mnt_group_start = id; 12715169fe7SAl Viro mnt->mnt_group_id = 0; 128719f5d7fSMiklos Szeredi } 129719f5d7fSMiklos Szeredi 130b3e19d92SNick Piggin /* 131b3e19d92SNick Piggin * vfsmount lock must be held for read 132b3e19d92SNick Piggin */ 13383adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 134b3e19d92SNick Piggin { 135b3e19d92SNick Piggin #ifdef CONFIG_SMP 13668e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 137b3e19d92SNick Piggin #else 138b3e19d92SNick Piggin preempt_disable(); 13968e8a9feSAl Viro mnt->mnt_count += n; 140b3e19d92SNick Piggin preempt_enable(); 141b3e19d92SNick Piggin #endif 142b3e19d92SNick Piggin } 143b3e19d92SNick Piggin 144b3e19d92SNick Piggin /* 145b3e19d92SNick Piggin * vfsmount lock must be held for write 146b3e19d92SNick Piggin */ 14783adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 148b3e19d92SNick Piggin { 149b3e19d92SNick Piggin #ifdef CONFIG_SMP 150f03c6599SAl Viro unsigned int count = 0; 151b3e19d92SNick Piggin int cpu; 152b3e19d92SNick Piggin 153b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 15468e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 155b3e19d92SNick Piggin } 156b3e19d92SNick Piggin 157b3e19d92SNick Piggin return count; 158b3e19d92SNick Piggin #else 15968e8a9feSAl Viro return mnt->mnt_count; 160b3e19d92SNick Piggin #endif 161b3e19d92SNick Piggin } 162b3e19d92SNick Piggin 163b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1641da177e4SLinus Torvalds { 165c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 166c63181e6SAl Viro if (mnt) { 16773cd49ecSMiklos Szeredi int err; 16873cd49ecSMiklos Szeredi 169c63181e6SAl Viro err = mnt_alloc_id(mnt); 17088b38782SLi Zefan if (err) 17188b38782SLi Zefan goto out_free_cache; 17288b38782SLi Zefan 17388b38782SLi Zefan if (name) { 174c63181e6SAl Viro mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 175c63181e6SAl Viro if (!mnt->mnt_devname) 17688b38782SLi Zefan goto out_free_id; 17773cd49ecSMiklos Szeredi } 17873cd49ecSMiklos Szeredi 179b3e19d92SNick Piggin #ifdef CONFIG_SMP 180c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 181c63181e6SAl Viro if (!mnt->mnt_pcp) 182b3e19d92SNick Piggin goto out_free_devname; 183b3e19d92SNick Piggin 184c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 185b3e19d92SNick Piggin #else 186c63181e6SAl Viro mnt->mnt_count = 1; 187c63181e6SAl Viro mnt->mnt_writers = 0; 188b3e19d92SNick Piggin #endif 189b3e19d92SNick Piggin 190c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_hash); 191c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 192c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 193c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 194c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 195c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 196c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 197c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 1982504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 1992504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 2002504c5d6SAndreas Gruenbacher #endif 2011da177e4SLinus Torvalds } 202c63181e6SAl Viro return mnt; 20388b38782SLi Zefan 204d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 205d3ef3d73Snpiggin@suse.de out_free_devname: 206c63181e6SAl Viro kfree(mnt->mnt_devname); 207d3ef3d73Snpiggin@suse.de #endif 20888b38782SLi Zefan out_free_id: 209c63181e6SAl Viro mnt_free_id(mnt); 21088b38782SLi Zefan out_free_cache: 211c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 21288b38782SLi Zefan return NULL; 2131da177e4SLinus Torvalds } 2141da177e4SLinus Torvalds 2158366025eSDave Hansen /* 2168366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2178366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2188366025eSDave Hansen * We must keep track of when those operations start 2198366025eSDave Hansen * (for permission checks) and when they end, so that 2208366025eSDave Hansen * we can determine when writes are able to occur to 2218366025eSDave Hansen * a filesystem. 2228366025eSDave Hansen */ 2233d733633SDave Hansen /* 2243d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2253d733633SDave Hansen * @mnt: the mount to check for its write status 2263d733633SDave Hansen * 2273d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2283d733633SDave Hansen * It does not guarantee that the filesystem will stay 2293d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2303d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2313d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2323d733633SDave Hansen * r/w. 2333d733633SDave Hansen */ 2343d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2353d733633SDave Hansen { 2362e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2372e4b7fcdSDave Hansen return 1; 2382e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2392e4b7fcdSDave Hansen return 1; 2402e4b7fcdSDave Hansen return 0; 2413d733633SDave Hansen } 2423d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2433d733633SDave Hansen 24483adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2453d733633SDave Hansen { 246d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 24768e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 248d3ef3d73Snpiggin@suse.de #else 24968e8a9feSAl Viro mnt->mnt_writers++; 250d3ef3d73Snpiggin@suse.de #endif 2513d733633SDave Hansen } 2523d733633SDave Hansen 25383adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2543d733633SDave Hansen { 255d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 25668e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 257d3ef3d73Snpiggin@suse.de #else 25868e8a9feSAl Viro mnt->mnt_writers--; 259d3ef3d73Snpiggin@suse.de #endif 260d3ef3d73Snpiggin@suse.de } 261d3ef3d73Snpiggin@suse.de 26283adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 263d3ef3d73Snpiggin@suse.de { 264d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 265d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2663d733633SDave Hansen int cpu; 2673d733633SDave Hansen 2683d733633SDave Hansen for_each_possible_cpu(cpu) { 26968e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2703d733633SDave Hansen } 2713d733633SDave Hansen 272d3ef3d73Snpiggin@suse.de return count; 273d3ef3d73Snpiggin@suse.de #else 274d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 275d3ef3d73Snpiggin@suse.de #endif 2763d733633SDave Hansen } 2773d733633SDave Hansen 2784ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2794ed5e82fSMiklos Szeredi { 2804ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2814ed5e82fSMiklos Szeredi return 1; 2824ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2834ed5e82fSMiklos Szeredi smp_rmb(); 2844ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2854ed5e82fSMiklos Szeredi } 2864ed5e82fSMiklos Szeredi 2873d733633SDave Hansen /* 288eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 289eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 290eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 291eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 2923d733633SDave Hansen */ 2938366025eSDave Hansen /** 294eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 29583adc753SAl Viro * @m: the mount on which to take a write 2968366025eSDave Hansen * 297eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 298eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 299eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 300eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 301eb04c282SJan Kara * called. This is effectively a refcount. 3028366025eSDave Hansen */ 303eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3048366025eSDave Hansen { 30583adc753SAl Viro struct mount *mnt = real_mount(m); 3063d733633SDave Hansen int ret = 0; 3073d733633SDave Hansen 308d3ef3d73Snpiggin@suse.de preempt_disable(); 309c6653a83SNick Piggin mnt_inc_writers(mnt); 310d3ef3d73Snpiggin@suse.de /* 311c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 312d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 313d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 314d3ef3d73Snpiggin@suse.de */ 315d3ef3d73Snpiggin@suse.de smp_mb(); 3161e75529eSMiao Xie while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 317d3ef3d73Snpiggin@suse.de cpu_relax(); 318d3ef3d73Snpiggin@suse.de /* 319d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 320d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 321d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 322d3ef3d73Snpiggin@suse.de */ 323d3ef3d73Snpiggin@suse.de smp_rmb(); 3244ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 325c6653a83SNick Piggin mnt_dec_writers(mnt); 3263d733633SDave Hansen ret = -EROFS; 3273d733633SDave Hansen } 328d3ef3d73Snpiggin@suse.de preempt_enable(); 329eb04c282SJan Kara 330eb04c282SJan Kara return ret; 331eb04c282SJan Kara } 332eb04c282SJan Kara 333eb04c282SJan Kara /** 334eb04c282SJan Kara * mnt_want_write - get write access to a mount 335eb04c282SJan Kara * @m: the mount on which to take a write 336eb04c282SJan Kara * 337eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 338eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 339eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 340eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 341eb04c282SJan Kara */ 342eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 343eb04c282SJan Kara { 344eb04c282SJan Kara int ret; 345eb04c282SJan Kara 346eb04c282SJan Kara sb_start_write(m->mnt_sb); 347eb04c282SJan Kara ret = __mnt_want_write(m); 348eb04c282SJan Kara if (ret) 349eb04c282SJan Kara sb_end_write(m->mnt_sb); 3503d733633SDave Hansen return ret; 3518366025eSDave Hansen } 3528366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3538366025eSDave Hansen 3548366025eSDave Hansen /** 35596029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 35696029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 35796029c4eSnpiggin@suse.de * 35896029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 35996029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 36096029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 36196029c4eSnpiggin@suse.de * on it. This allows some optimisation. 36296029c4eSnpiggin@suse.de * 36396029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 36496029c4eSnpiggin@suse.de * drop the reference. 36596029c4eSnpiggin@suse.de */ 36696029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 36796029c4eSnpiggin@suse.de { 36896029c4eSnpiggin@suse.de /* superblock may be r/o */ 36996029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 37096029c4eSnpiggin@suse.de return -EROFS; 37196029c4eSnpiggin@suse.de preempt_disable(); 37283adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 37396029c4eSnpiggin@suse.de preempt_enable(); 37496029c4eSnpiggin@suse.de return 0; 37596029c4eSnpiggin@suse.de } 37696029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 37796029c4eSnpiggin@suse.de 37896029c4eSnpiggin@suse.de /** 379eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 380eb04c282SJan Kara * @file: the file who's mount on which to take a write 381eb04c282SJan Kara * 382eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 383eb04c282SJan Kara * do some optimisations if the file is open for write already 384eb04c282SJan Kara */ 385eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 386eb04c282SJan Kara { 387496ad9aaSAl Viro struct inode *inode = file_inode(file); 388eb04c282SJan Kara 389eb04c282SJan Kara if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) 390eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 391eb04c282SJan Kara else 392eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 393eb04c282SJan Kara } 394eb04c282SJan Kara 395eb04c282SJan Kara /** 39696029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 39796029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 39896029c4eSnpiggin@suse.de * 39996029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 40096029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 40196029c4eSnpiggin@suse.de */ 40296029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 40396029c4eSnpiggin@suse.de { 404eb04c282SJan Kara int ret; 405eb04c282SJan Kara 406eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 407eb04c282SJan Kara ret = __mnt_want_write_file(file); 408eb04c282SJan Kara if (ret) 409eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 410eb04c282SJan Kara return ret; 41196029c4eSnpiggin@suse.de } 41296029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 41396029c4eSnpiggin@suse.de 41496029c4eSnpiggin@suse.de /** 415eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4168366025eSDave Hansen * @mnt: the mount on which to give up write access 4178366025eSDave Hansen * 4188366025eSDave Hansen * Tells the low-level filesystem that we are done 4198366025eSDave Hansen * performing writes to it. Must be matched with 420eb04c282SJan Kara * __mnt_want_write() call above. 4218366025eSDave Hansen */ 422eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4238366025eSDave Hansen { 424d3ef3d73Snpiggin@suse.de preempt_disable(); 42583adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 426d3ef3d73Snpiggin@suse.de preempt_enable(); 4278366025eSDave Hansen } 428eb04c282SJan Kara 429eb04c282SJan Kara /** 430eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 431eb04c282SJan Kara * @mnt: the mount on which to give up write access 432eb04c282SJan Kara * 433eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 434eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 435eb04c282SJan Kara * mnt_want_write() call above. 436eb04c282SJan Kara */ 437eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 438eb04c282SJan Kara { 439eb04c282SJan Kara __mnt_drop_write(mnt); 440eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 441eb04c282SJan Kara } 4428366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4438366025eSDave Hansen 444eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 445eb04c282SJan Kara { 446eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 447eb04c282SJan Kara } 448eb04c282SJan Kara 4492a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4502a79f17eSAl Viro { 4512a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4522a79f17eSAl Viro } 4532a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4542a79f17eSAl Viro 45583adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4568366025eSDave Hansen { 4573d733633SDave Hansen int ret = 0; 4583d733633SDave Hansen 459962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 46083adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 461d3ef3d73Snpiggin@suse.de /* 462d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 463d3ef3d73Snpiggin@suse.de * should be visible before we do. 464d3ef3d73Snpiggin@suse.de */ 465d3ef3d73Snpiggin@suse.de smp_mb(); 466d3ef3d73Snpiggin@suse.de 467d3ef3d73Snpiggin@suse.de /* 468d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 469d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 470d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 471d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 472d3ef3d73Snpiggin@suse.de * 473d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 474d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 475d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 476d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 477d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 478d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 479d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 480d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 481d3ef3d73Snpiggin@suse.de * we're counting up here. 482d3ef3d73Snpiggin@suse.de */ 483c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 484d3ef3d73Snpiggin@suse.de ret = -EBUSY; 485d3ef3d73Snpiggin@suse.de else 48683adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 487d3ef3d73Snpiggin@suse.de /* 488d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 489d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 490d3ef3d73Snpiggin@suse.de */ 491d3ef3d73Snpiggin@suse.de smp_wmb(); 49283adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 493962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 4943d733633SDave Hansen return ret; 4953d733633SDave Hansen } 4968366025eSDave Hansen 49783adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 4982e4b7fcdSDave Hansen { 499962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 50083adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 501962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 5022e4b7fcdSDave Hansen } 5032e4b7fcdSDave Hansen 5044ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5054ed5e82fSMiklos Szeredi { 5064ed5e82fSMiklos Szeredi struct mount *mnt; 5074ed5e82fSMiklos Szeredi int err = 0; 5084ed5e82fSMiklos Szeredi 5098e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5108e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5118e8b8796SMiklos Szeredi return -EBUSY; 5128e8b8796SMiklos Szeredi 513962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 5144ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5154ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5164ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5174ed5e82fSMiklos Szeredi smp_mb(); 5184ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5194ed5e82fSMiklos Szeredi err = -EBUSY; 5204ed5e82fSMiklos Szeredi break; 5214ed5e82fSMiklos Szeredi } 5224ed5e82fSMiklos Szeredi } 5234ed5e82fSMiklos Szeredi } 5248e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5258e8b8796SMiklos Szeredi err = -EBUSY; 5268e8b8796SMiklos Szeredi 5274ed5e82fSMiklos Szeredi if (!err) { 5284ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5294ed5e82fSMiklos Szeredi smp_wmb(); 5304ed5e82fSMiklos Szeredi } 5314ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5324ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5334ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5344ed5e82fSMiklos Szeredi } 535962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 5364ed5e82fSMiklos Szeredi 5374ed5e82fSMiklos Szeredi return err; 5384ed5e82fSMiklos Szeredi } 5394ed5e82fSMiklos Szeredi 540b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5411da177e4SLinus Torvalds { 54252ba1621SAl Viro kfree(mnt->mnt_devname); 54373cd49ecSMiklos Szeredi mnt_free_id(mnt); 544d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 54568e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 546d3ef3d73Snpiggin@suse.de #endif 547b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5481da177e4SLinus Torvalds } 5491da177e4SLinus Torvalds 5501da177e4SLinus Torvalds /* 551a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 552a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 55399b7db7bSNick Piggin * vfsmount_lock must be held for read or write. 5541da177e4SLinus Torvalds */ 555c7105365SAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 556a05964f3SRam Pai int dir) 5571da177e4SLinus Torvalds { 5581da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 5591da177e4SLinus Torvalds struct list_head *tmp = head; 560c7105365SAl Viro struct mount *p, *found = NULL; 5611da177e4SLinus Torvalds 5621da177e4SLinus Torvalds for (;;) { 563a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 5641da177e4SLinus Torvalds p = NULL; 5651da177e4SLinus Torvalds if (tmp == head) 5661da177e4SLinus Torvalds break; 5671b8e5564SAl Viro p = list_entry(tmp, struct mount, mnt_hash); 568a73324daSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) { 569a05964f3SRam Pai found = p; 5701da177e4SLinus Torvalds break; 5711da177e4SLinus Torvalds } 5721da177e4SLinus Torvalds } 5731da177e4SLinus Torvalds return found; 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds 576a05964f3SRam Pai /* 577f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 578f015f126SDavid Howells * 579f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 580f015f126SDavid Howells * following mounts: 581f015f126SDavid Howells * 582f015f126SDavid Howells * mount /dev/sda1 /mnt 583f015f126SDavid Howells * mount /dev/sda2 /mnt 584f015f126SDavid Howells * mount /dev/sda3 /mnt 585f015f126SDavid Howells * 586f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 587f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 588f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 589f015f126SDavid Howells * 590f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 591a05964f3SRam Pai */ 5921c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 593a05964f3SRam Pai { 594c7105365SAl Viro struct mount *child_mnt; 59599b7db7bSNick Piggin 596962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 597c7105365SAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry, 1); 598c7105365SAl Viro if (child_mnt) { 599c7105365SAl Viro mnt_add_count(child_mnt, 1); 600962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 601c7105365SAl Viro return &child_mnt->mnt; 602c7105365SAl Viro } else { 603962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 604c7105365SAl Viro return NULL; 605c7105365SAl Viro } 606a05964f3SRam Pai } 607a05964f3SRam Pai 60884d17192SAl Viro static struct mountpoint *new_mountpoint(struct dentry *dentry) 60984d17192SAl Viro { 61084d17192SAl Viro struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry); 61184d17192SAl Viro struct mountpoint *mp; 612eed81007SMiklos Szeredi int ret; 61384d17192SAl Viro 61484d17192SAl Viro list_for_each_entry(mp, chain, m_hash) { 61584d17192SAl Viro if (mp->m_dentry == dentry) { 61684d17192SAl Viro /* might be worth a WARN_ON() */ 61784d17192SAl Viro if (d_unlinked(dentry)) 61884d17192SAl Viro return ERR_PTR(-ENOENT); 61984d17192SAl Viro mp->m_count++; 62084d17192SAl Viro return mp; 62184d17192SAl Viro } 62284d17192SAl Viro } 62384d17192SAl Viro 62484d17192SAl Viro mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 62584d17192SAl Viro if (!mp) 62684d17192SAl Viro return ERR_PTR(-ENOMEM); 62784d17192SAl Viro 628eed81007SMiklos Szeredi ret = d_set_mounted(dentry); 629eed81007SMiklos Szeredi if (ret) { 63084d17192SAl Viro kfree(mp); 631eed81007SMiklos Szeredi return ERR_PTR(ret); 63284d17192SAl Viro } 633eed81007SMiklos Szeredi 63484d17192SAl Viro mp->m_dentry = dentry; 63584d17192SAl Viro mp->m_count = 1; 63684d17192SAl Viro list_add(&mp->m_hash, chain); 63784d17192SAl Viro return mp; 63884d17192SAl Viro } 63984d17192SAl Viro 64084d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 64184d17192SAl Viro { 64284d17192SAl Viro if (!--mp->m_count) { 64384d17192SAl Viro struct dentry *dentry = mp->m_dentry; 64484d17192SAl Viro spin_lock(&dentry->d_lock); 64584d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 64684d17192SAl Viro spin_unlock(&dentry->d_lock); 64784d17192SAl Viro list_del(&mp->m_hash); 64884d17192SAl Viro kfree(mp); 64984d17192SAl Viro } 65084d17192SAl Viro } 65184d17192SAl Viro 652143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 6531da177e4SLinus Torvalds { 6546b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds 65799b7db7bSNick Piggin /* 65899b7db7bSNick Piggin * vfsmount lock must be held for write 65999b7db7bSNick Piggin */ 6606b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 6615addc5ddSAl Viro { 6625addc5ddSAl Viro if (ns) { 6635addc5ddSAl Viro ns->event = ++event; 6645addc5ddSAl Viro wake_up_interruptible(&ns->poll); 6655addc5ddSAl Viro } 6665addc5ddSAl Viro } 6675addc5ddSAl Viro 66899b7db7bSNick Piggin /* 66999b7db7bSNick Piggin * vfsmount lock must be held for write 67099b7db7bSNick Piggin */ 6716b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 6725addc5ddSAl Viro { 6735addc5ddSAl Viro if (ns && ns->event != event) { 6745addc5ddSAl Viro ns->event = event; 6755addc5ddSAl Viro wake_up_interruptible(&ns->poll); 6765addc5ddSAl Viro } 6775addc5ddSAl Viro } 6785addc5ddSAl Viro 67999b7db7bSNick Piggin /* 68099b7db7bSNick Piggin * vfsmount lock must be held for write 68199b7db7bSNick Piggin */ 682419148daSAl Viro static void detach_mnt(struct mount *mnt, struct path *old_path) 6831da177e4SLinus Torvalds { 684a73324daSAl Viro old_path->dentry = mnt->mnt_mountpoint; 6850714a533SAl Viro old_path->mnt = &mnt->mnt_parent->mnt; 6860714a533SAl Viro mnt->mnt_parent = mnt; 687a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 6886b41d536SAl Viro list_del_init(&mnt->mnt_child); 6891b8e5564SAl Viro list_del_init(&mnt->mnt_hash); 69084d17192SAl Viro put_mountpoint(mnt->mnt_mp); 69184d17192SAl Viro mnt->mnt_mp = NULL; 6921da177e4SLinus Torvalds } 6931da177e4SLinus Torvalds 69499b7db7bSNick Piggin /* 69599b7db7bSNick Piggin * vfsmount lock must be held for write 69699b7db7bSNick Piggin */ 69784d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 69884d17192SAl Viro struct mountpoint *mp, 69944d964d6SAl Viro struct mount *child_mnt) 700b90fa9aeSRam Pai { 70184d17192SAl Viro mp->m_count++; 7023a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 70384d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 7043a2393d7SAl Viro child_mnt->mnt_parent = mnt; 70584d17192SAl Viro child_mnt->mnt_mp = mp; 706b90fa9aeSRam Pai } 707b90fa9aeSRam Pai 70899b7db7bSNick Piggin /* 70999b7db7bSNick Piggin * vfsmount lock must be held for write 71099b7db7bSNick Piggin */ 71184d17192SAl Viro static void attach_mnt(struct mount *mnt, 71284d17192SAl Viro struct mount *parent, 71384d17192SAl Viro struct mountpoint *mp) 7141da177e4SLinus Torvalds { 71584d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 7161b8e5564SAl Viro list_add_tail(&mnt->mnt_hash, mount_hashtable + 71784d17192SAl Viro hash(&parent->mnt, mp->m_dentry)); 71884d17192SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 719b90fa9aeSRam Pai } 720b90fa9aeSRam Pai 721b90fa9aeSRam Pai /* 72299b7db7bSNick Piggin * vfsmount lock must be held for write 723b90fa9aeSRam Pai */ 7244b2619a5SAl Viro static void commit_tree(struct mount *mnt) 725b90fa9aeSRam Pai { 7260714a533SAl Viro struct mount *parent = mnt->mnt_parent; 72783adc753SAl Viro struct mount *m; 728b90fa9aeSRam Pai LIST_HEAD(head); 729143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 730b90fa9aeSRam Pai 7310714a533SAl Viro BUG_ON(parent == mnt); 732b90fa9aeSRam Pai 7331a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 734f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 735143c8c91SAl Viro m->mnt_ns = n; 736f03c6599SAl Viro 737b90fa9aeSRam Pai list_splice(&head, n->list.prev); 738b90fa9aeSRam Pai 7391b8e5564SAl Viro list_add_tail(&mnt->mnt_hash, mount_hashtable + 740a73324daSAl Viro hash(&parent->mnt, mnt->mnt_mountpoint)); 7416b41d536SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 7426b3286edSKirill Korotaev touch_mnt_namespace(n); 7431da177e4SLinus Torvalds } 7441da177e4SLinus Torvalds 745909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 7461da177e4SLinus Torvalds { 7476b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 7486b41d536SAl Viro if (next == &p->mnt_mounts) { 7491da177e4SLinus Torvalds while (1) { 750909b0a88SAl Viro if (p == root) 7511da177e4SLinus Torvalds return NULL; 7526b41d536SAl Viro next = p->mnt_child.next; 7536b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 7541da177e4SLinus Torvalds break; 7550714a533SAl Viro p = p->mnt_parent; 7561da177e4SLinus Torvalds } 7571da177e4SLinus Torvalds } 7586b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 7591da177e4SLinus Torvalds } 7601da177e4SLinus Torvalds 761315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 7629676f0c6SRam Pai { 7636b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 7646b41d536SAl Viro while (prev != &p->mnt_mounts) { 7656b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 7666b41d536SAl Viro prev = p->mnt_mounts.prev; 7679676f0c6SRam Pai } 7689676f0c6SRam Pai return p; 7699676f0c6SRam Pai } 7709676f0c6SRam Pai 7719d412a43SAl Viro struct vfsmount * 7729d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 7739d412a43SAl Viro { 774b105e270SAl Viro struct mount *mnt; 7759d412a43SAl Viro struct dentry *root; 7769d412a43SAl Viro 7779d412a43SAl Viro if (!type) 7789d412a43SAl Viro return ERR_PTR(-ENODEV); 7799d412a43SAl Viro 7809d412a43SAl Viro mnt = alloc_vfsmnt(name); 7819d412a43SAl Viro if (!mnt) 7829d412a43SAl Viro return ERR_PTR(-ENOMEM); 7839d412a43SAl Viro 7849d412a43SAl Viro if (flags & MS_KERNMOUNT) 785b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 7869d412a43SAl Viro 7879d412a43SAl Viro root = mount_fs(type, flags, name, data); 7889d412a43SAl Viro if (IS_ERR(root)) { 7899d412a43SAl Viro free_vfsmnt(mnt); 7909d412a43SAl Viro return ERR_CAST(root); 7919d412a43SAl Viro } 7929d412a43SAl Viro 793b105e270SAl Viro mnt->mnt.mnt_root = root; 794b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 795a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 7960714a533SAl Viro mnt->mnt_parent = mnt; 797962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 79839f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 799962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 800b105e270SAl Viro return &mnt->mnt; 8019d412a43SAl Viro } 8029d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 8039d412a43SAl Viro 80487129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 80536341f64SRam Pai int flag) 8061da177e4SLinus Torvalds { 80787129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 808be34d1a3SDavid Howells struct mount *mnt; 809be34d1a3SDavid Howells int err; 8101da177e4SLinus Torvalds 811be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 812be34d1a3SDavid Howells if (!mnt) 813be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 814be34d1a3SDavid Howells 8157a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 81615169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 817719f5d7fSMiklos Szeredi else 81815169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 819719f5d7fSMiklos Szeredi 82015169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 821be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 822719f5d7fSMiklos Szeredi if (err) 823719f5d7fSMiklos Szeredi goto out_free; 824719f5d7fSMiklos Szeredi } 825719f5d7fSMiklos Szeredi 82687129cc0SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; 827132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 828132c94e3SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY)) 829132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 830132c94e3SEric W. Biederman 8315ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 8325ff9d8a6SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) 8335ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 8345ff9d8a6SEric W. Biederman 8351da177e4SLinus Torvalds atomic_inc(&sb->s_active); 836b105e270SAl Viro mnt->mnt.mnt_sb = sb; 837b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 838a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8390714a533SAl Viro mnt->mnt_parent = mnt; 840962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 84139f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 842962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 843b90fa9aeSRam Pai 8447a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 8457a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 8466776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 84732301920SAl Viro mnt->mnt_master = old; 848fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 8498aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 850fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 8516776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 852d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 8536776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 854d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 8555afe0022SRam Pai } 856b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 8570f0afb1dSAl Viro set_mnt_shared(mnt); 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 8601da177e4SLinus Torvalds * as the original if that was on one */ 86136341f64SRam Pai if (flag & CL_EXPIRE) { 8626776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 8636776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 8641da177e4SLinus Torvalds } 865be34d1a3SDavid Howells 866cb338d06SAl Viro return mnt; 867719f5d7fSMiklos Szeredi 868719f5d7fSMiklos Szeredi out_free: 869719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 870be34d1a3SDavid Howells return ERR_PTR(err); 8711da177e4SLinus Torvalds } 8721da177e4SLinus Torvalds 873900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 8747b7b1aceSAl Viro { 875b3e19d92SNick Piggin put_again: 876f03c6599SAl Viro #ifdef CONFIG_SMP 877962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 878f7a99c5bSAl Viro if (likely(mnt->mnt_ns)) { 879f7a99c5bSAl Viro /* shouldn't be the last one */ 880aa9c0e07SAl Viro mnt_add_count(mnt, -1); 881962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 88299b7db7bSNick Piggin return; 883b3e19d92SNick Piggin } 884962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 885b3e19d92SNick Piggin 886962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 887aa9c0e07SAl Viro mnt_add_count(mnt, -1); 888b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 889962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 89099b7db7bSNick Piggin return; 89199b7db7bSNick Piggin } 892b3e19d92SNick Piggin #else 893aa9c0e07SAl Viro mnt_add_count(mnt, -1); 894b3e19d92SNick Piggin if (likely(mnt_get_count(mnt))) 895b3e19d92SNick Piggin return; 896962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 897f03c6599SAl Viro #endif 898863d684fSAl Viro if (unlikely(mnt->mnt_pinned)) { 899863d684fSAl Viro mnt_add_count(mnt, mnt->mnt_pinned + 1); 900863d684fSAl Viro mnt->mnt_pinned = 0; 901962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 902900148dcSAl Viro acct_auto_close_mnt(&mnt->mnt); 903b3e19d92SNick Piggin goto put_again; 904b3e19d92SNick Piggin } 905962830dfSAndi Kleen 90639f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 907962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 908649a795aSAl Viro 909649a795aSAl Viro /* 910649a795aSAl Viro * This probably indicates that somebody messed 911649a795aSAl Viro * up a mnt_want/drop_write() pair. If this 912649a795aSAl Viro * happens, the filesystem was probably unable 913649a795aSAl Viro * to make r/w->r/o transitions. 914649a795aSAl Viro */ 915649a795aSAl Viro /* 916649a795aSAl Viro * The locking used to deal with mnt_count decrement provides barriers, 917649a795aSAl Viro * so mnt_get_writers() below is safe. 918649a795aSAl Viro */ 919649a795aSAl Viro WARN_ON(mnt_get_writers(mnt)); 920649a795aSAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 921649a795aSAl Viro dput(mnt->mnt.mnt_root); 922649a795aSAl Viro deactivate_super(mnt->mnt.mnt_sb); 923649a795aSAl Viro free_vfsmnt(mnt); 924b3e19d92SNick Piggin } 925b3e19d92SNick Piggin 926b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 927b3e19d92SNick Piggin { 928b3e19d92SNick Piggin if (mnt) { 929863d684fSAl Viro struct mount *m = real_mount(mnt); 930b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 931863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 932863d684fSAl Viro m->mnt_expiry_mark = 0; 933863d684fSAl Viro mntput_no_expire(m); 934b3e19d92SNick Piggin } 935b3e19d92SNick Piggin } 936b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 937b3e19d92SNick Piggin 938b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 939b3e19d92SNick Piggin { 940b3e19d92SNick Piggin if (mnt) 94183adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 942b3e19d92SNick Piggin return mnt; 943b3e19d92SNick Piggin } 944b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 945b3e19d92SNick Piggin 9467b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 9477b7b1aceSAl Viro { 948962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 949863d684fSAl Viro real_mount(mnt)->mnt_pinned++; 950962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 9517b7b1aceSAl Viro } 9527b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 9537b7b1aceSAl Viro 954863d684fSAl Viro void mnt_unpin(struct vfsmount *m) 9557b7b1aceSAl Viro { 956863d684fSAl Viro struct mount *mnt = real_mount(m); 957962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 9587b7b1aceSAl Viro if (mnt->mnt_pinned) { 959863d684fSAl Viro mnt_add_count(mnt, 1); 9607b7b1aceSAl Viro mnt->mnt_pinned--; 9617b7b1aceSAl Viro } 962962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 9637b7b1aceSAl Viro } 9647b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 9651da177e4SLinus Torvalds 966b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 967b3b304a2SMiklos Szeredi { 968b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 969b3b304a2SMiklos Szeredi } 970b3b304a2SMiklos Szeredi 971b3b304a2SMiklos Szeredi /* 972b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 973b3b304a2SMiklos Szeredi * implement more complex mount option showing. 974b3b304a2SMiklos Szeredi * 975b3b304a2SMiklos Szeredi * See also save_mount_options(). 976b3b304a2SMiklos Szeredi */ 97734c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 978b3b304a2SMiklos Szeredi { 9792a32cebdSAl Viro const char *options; 9802a32cebdSAl Viro 9812a32cebdSAl Viro rcu_read_lock(); 98234c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 983b3b304a2SMiklos Szeredi 984b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 985b3b304a2SMiklos Szeredi seq_putc(m, ','); 986b3b304a2SMiklos Szeredi mangle(m, options); 987b3b304a2SMiklos Szeredi } 9882a32cebdSAl Viro rcu_read_unlock(); 989b3b304a2SMiklos Szeredi 990b3b304a2SMiklos Szeredi return 0; 991b3b304a2SMiklos Szeredi } 992b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 993b3b304a2SMiklos Szeredi 994b3b304a2SMiklos Szeredi /* 995b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 996b3b304a2SMiklos Szeredi * called from the fill_super() callback. 997b3b304a2SMiklos Szeredi * 998b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 999b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1000b3b304a2SMiklos Szeredi * remount fails. 1001b3b304a2SMiklos Szeredi * 1002b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1003b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1004b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1005b3b304a2SMiklos Szeredi * any more. 1006b3b304a2SMiklos Szeredi */ 1007b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1008b3b304a2SMiklos Szeredi { 10092a32cebdSAl Viro BUG_ON(sb->s_options); 10102a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1011b3b304a2SMiklos Szeredi } 1012b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1013b3b304a2SMiklos Szeredi 10142a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 10152a32cebdSAl Viro { 10162a32cebdSAl Viro char *old = sb->s_options; 10172a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 10182a32cebdSAl Viro if (old) { 10192a32cebdSAl Viro synchronize_rcu(); 10202a32cebdSAl Viro kfree(old); 10212a32cebdSAl Viro } 10222a32cebdSAl Viro } 10232a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 10242a32cebdSAl Viro 1025a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 10260226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 10271da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 10281da177e4SLinus Torvalds { 10296ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 10301da177e4SLinus Torvalds 1031390c6843SRam Pai down_read(&namespace_sem); 1032a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 10331da177e4SLinus Torvalds } 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 10361da177e4SLinus Torvalds { 10376ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 1038b0765fb8SPavel Emelianov 1039a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 10401da177e4SLinus Torvalds } 10411da177e4SLinus Torvalds 10421da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 10431da177e4SLinus Torvalds { 1044390c6843SRam Pai up_read(&namespace_sem); 10451da177e4SLinus Torvalds } 10461da177e4SLinus Torvalds 10470226f492SAl Viro static int m_show(struct seq_file *m, void *v) 10489f5596afSAl Viro { 10496ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 10501a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 10510226f492SAl Viro return p->show(m, &r->mnt); 10521da177e4SLinus Torvalds } 10531da177e4SLinus Torvalds 1054a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 10551da177e4SLinus Torvalds .start = m_start, 10561da177e4SLinus Torvalds .next = m_next, 10571da177e4SLinus Torvalds .stop = m_stop, 10580226f492SAl Viro .show = m_show, 1059b4629fe2SChuck Lever }; 1060a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1061b4629fe2SChuck Lever 10621da177e4SLinus Torvalds /** 10631da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 10641da177e4SLinus Torvalds * @mnt: root of mount tree 10651da177e4SLinus Torvalds * 10661da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 10671da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 10681da177e4SLinus Torvalds * busy. 10691da177e4SLinus Torvalds */ 1070909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 10711da177e4SLinus Torvalds { 1072909b0a88SAl Viro struct mount *mnt = real_mount(m); 107336341f64SRam Pai int actual_refs = 0; 107436341f64SRam Pai int minimum_refs = 0; 1075315fc83eSAl Viro struct mount *p; 1076909b0a88SAl Viro BUG_ON(!m); 10771da177e4SLinus Torvalds 1078b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1079962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1080909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 108183adc753SAl Viro actual_refs += mnt_get_count(p); 10821da177e4SLinus Torvalds minimum_refs += 2; 10831da177e4SLinus Torvalds } 1084962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 10851da177e4SLinus Torvalds 10861da177e4SLinus Torvalds if (actual_refs > minimum_refs) 10871da177e4SLinus Torvalds return 0; 1088e3474a8eSIan Kent 1089e3474a8eSIan Kent return 1; 10901da177e4SLinus Torvalds } 10911da177e4SLinus Torvalds 10921da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 10931da177e4SLinus Torvalds 10941da177e4SLinus Torvalds /** 10951da177e4SLinus Torvalds * may_umount - check if a mount point is busy 10961da177e4SLinus Torvalds * @mnt: root of mount 10971da177e4SLinus Torvalds * 10981da177e4SLinus Torvalds * This is called to check if a mount point has any 10991da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 11001da177e4SLinus Torvalds * mount has sub mounts this will return busy 11011da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 11021da177e4SLinus Torvalds * 11031da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 11041da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 11051da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 11061da177e4SLinus Torvalds */ 11071da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 11081da177e4SLinus Torvalds { 1109e3474a8eSIan Kent int ret = 1; 11108ad08d8aSAl Viro down_read(&namespace_sem); 1111962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 11121ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1113e3474a8eSIan Kent ret = 0; 1114962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 11158ad08d8aSAl Viro up_read(&namespace_sem); 1116a05964f3SRam Pai return ret; 11171da177e4SLinus Torvalds } 11181da177e4SLinus Torvalds 11191da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 11201da177e4SLinus Torvalds 1121e3197d83SAl Viro static LIST_HEAD(unmounted); /* protected by namespace_sem */ 1122e3197d83SAl Viro 112397216be0SAl Viro static void namespace_unlock(void) 11241da177e4SLinus Torvalds { 1125d5e50f74SAl Viro struct mount *mnt; 112697216be0SAl Viro LIST_HEAD(head); 112797216be0SAl Viro 112897216be0SAl Viro if (likely(list_empty(&unmounted))) { 112997216be0SAl Viro up_write(&namespace_sem); 113097216be0SAl Viro return; 113197216be0SAl Viro } 113297216be0SAl Viro 113397216be0SAl Viro list_splice_init(&unmounted, &head); 113497216be0SAl Viro up_write(&namespace_sem); 113597216be0SAl Viro 113697216be0SAl Viro while (!list_empty(&head)) { 113797216be0SAl Viro mnt = list_first_entry(&head, struct mount, mnt_hash); 11381b8e5564SAl Viro list_del_init(&mnt->mnt_hash); 1139676da58dSAl Viro if (mnt_has_parent(mnt)) { 114070fbcdf4SRam Pai struct dentry *dentry; 1141863d684fSAl Viro struct mount *m; 114299b7db7bSNick Piggin 1143962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1144a73324daSAl Viro dentry = mnt->mnt_mountpoint; 1145863d684fSAl Viro m = mnt->mnt_parent; 1146a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 11470714a533SAl Viro mnt->mnt_parent = mnt; 11487c4b93d8SAl Viro m->mnt_ghosts--; 1149962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 115070fbcdf4SRam Pai dput(dentry); 1151863d684fSAl Viro mntput(&m->mnt); 11521da177e4SLinus Torvalds } 1153d5e50f74SAl Viro mntput(&mnt->mnt); 115470fbcdf4SRam Pai } 115570fbcdf4SRam Pai } 115670fbcdf4SRam Pai 115797216be0SAl Viro static inline void namespace_lock(void) 1158e3197d83SAl Viro { 115997216be0SAl Viro down_write(&namespace_sem); 1160e3197d83SAl Viro } 1161e3197d83SAl Viro 116299b7db7bSNick Piggin /* 116399b7db7bSNick Piggin * vfsmount lock must be held for write 116499b7db7bSNick Piggin * namespace_sem must be held for write 116599b7db7bSNick Piggin */ 1166328e6d90SAl Viro void umount_tree(struct mount *mnt, int propagate) 116770fbcdf4SRam Pai { 11687b8a53fdSAl Viro LIST_HEAD(tmp_list); 1169315fc83eSAl Viro struct mount *p; 117070fbcdf4SRam Pai 1171909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) 11721b8e5564SAl Viro list_move(&p->mnt_hash, &tmp_list); 117370fbcdf4SRam Pai 1174a05964f3SRam Pai if (propagate) 11757b8a53fdSAl Viro propagate_umount(&tmp_list); 1176a05964f3SRam Pai 11771b8e5564SAl Viro list_for_each_entry(p, &tmp_list, mnt_hash) { 11786776db3dSAl Viro list_del_init(&p->mnt_expire); 11791a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1180143c8c91SAl Viro __touch_mnt_namespace(p->mnt_ns); 118163d37a84SAl Viro p->mnt_ns = NULL; 11826b41d536SAl Viro list_del_init(&p->mnt_child); 1183676da58dSAl Viro if (mnt_has_parent(p)) { 1184863d684fSAl Viro p->mnt_parent->mnt_ghosts++; 118584d17192SAl Viro put_mountpoint(p->mnt_mp); 118684d17192SAl Viro p->mnt_mp = NULL; 11877c4b93d8SAl Viro } 11880f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 11891da177e4SLinus Torvalds } 1190328e6d90SAl Viro list_splice(&tmp_list, &unmounted); 11911da177e4SLinus Torvalds } 11921da177e4SLinus Torvalds 1193b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1194c35038beSAl Viro 11951ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 11961da177e4SLinus Torvalds { 11971ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 11981da177e4SLinus Torvalds int retval; 11991da177e4SLinus Torvalds 12001ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 12011da177e4SLinus Torvalds if (retval) 12021da177e4SLinus Torvalds return retval; 12031da177e4SLinus Torvalds 12041da177e4SLinus Torvalds /* 12051da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 12061da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 12071da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 12081da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 12091da177e4SLinus Torvalds */ 12101da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 12111ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 12121da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 12131da177e4SLinus Torvalds return -EINVAL; 12141da177e4SLinus Torvalds 1215b3e19d92SNick Piggin /* 1216b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1217b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1218b3e19d92SNick Piggin */ 1219962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 122083adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1221962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 12221da177e4SLinus Torvalds return -EBUSY; 1223b3e19d92SNick Piggin } 1224962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 12251da177e4SLinus Torvalds 1226863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 12271da177e4SLinus Torvalds return -EAGAIN; 12281da177e4SLinus Torvalds } 12291da177e4SLinus Torvalds 12301da177e4SLinus Torvalds /* 12311da177e4SLinus Torvalds * If we may have to abort operations to get out of this 12321da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 12331da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 12341da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 12351da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 12361da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 12371da177e4SLinus Torvalds * about for the moment. 12381da177e4SLinus Torvalds */ 12391da177e4SLinus Torvalds 124042faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 124142faad99SAl Viro sb->s_op->umount_begin(sb); 124242faad99SAl Viro } 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds /* 12451da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 12461da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 12471da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 12481da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 12491da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 12501da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 12511da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 12521da177e4SLinus Torvalds */ 12531ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 12541da177e4SLinus Torvalds /* 12551da177e4SLinus Torvalds * Special case for "unmounting" root ... 12561da177e4SLinus Torvalds * we just try to remount it readonly. 12571da177e4SLinus Torvalds */ 12581da177e4SLinus Torvalds down_write(&sb->s_umount); 12594aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 12601da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 12611da177e4SLinus Torvalds up_write(&sb->s_umount); 12621da177e4SLinus Torvalds return retval; 12631da177e4SLinus Torvalds } 12641da177e4SLinus Torvalds 126597216be0SAl Viro namespace_lock(); 1266962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 12675addc5ddSAl Viro event++; 12681da177e4SLinus Torvalds 1269c35038beSAl Viro if (!(flags & MNT_DETACH)) 1270b54b9be7SAl Viro shrink_submounts(mnt); 1271c35038beSAl Viro 12721da177e4SLinus Torvalds retval = -EBUSY; 1273a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 12741a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1275328e6d90SAl Viro umount_tree(mnt, 1); 12761da177e4SLinus Torvalds retval = 0; 12771da177e4SLinus Torvalds } 1278962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 1279e3197d83SAl Viro namespace_unlock(); 12801da177e4SLinus Torvalds return retval; 12811da177e4SLinus Torvalds } 12821da177e4SLinus Torvalds 12831da177e4SLinus Torvalds /* 12849b40bc90SAl Viro * Is the caller allowed to modify his namespace? 12859b40bc90SAl Viro */ 12869b40bc90SAl Viro static inline bool may_mount(void) 12879b40bc90SAl Viro { 12889b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 12899b40bc90SAl Viro } 12909b40bc90SAl Viro 12919b40bc90SAl Viro /* 12921da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 12931da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 12941da177e4SLinus Torvalds * 12951da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 12961da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 12971da177e4SLinus Torvalds */ 12981da177e4SLinus Torvalds 1299bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 13001da177e4SLinus Torvalds { 13012d8f3038SAl Viro struct path path; 1302900148dcSAl Viro struct mount *mnt; 13031da177e4SLinus Torvalds int retval; 1304db1f05bbSMiklos Szeredi int lookup_flags = 0; 13051da177e4SLinus Torvalds 1306db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1307db1f05bbSMiklos Szeredi return -EINVAL; 1308db1f05bbSMiklos Szeredi 13099b40bc90SAl Viro if (!may_mount()) 13109b40bc90SAl Viro return -EPERM; 13119b40bc90SAl Viro 1312db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1313db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1314db1f05bbSMiklos Szeredi 1315197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 13161da177e4SLinus Torvalds if (retval) 13171da177e4SLinus Torvalds goto out; 1318900148dcSAl Viro mnt = real_mount(path.mnt); 13191da177e4SLinus Torvalds retval = -EINVAL; 13202d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 13211da177e4SLinus Torvalds goto dput_and_out; 1322143c8c91SAl Viro if (!check_mnt(mnt)) 13231da177e4SLinus Torvalds goto dput_and_out; 13245ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 13255ff9d8a6SEric W. Biederman goto dput_and_out; 13261da177e4SLinus Torvalds 1327900148dcSAl Viro retval = do_umount(mnt, flags); 13281da177e4SLinus Torvalds dput_and_out: 1329429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 13302d8f3038SAl Viro dput(path.dentry); 1331900148dcSAl Viro mntput_no_expire(mnt); 13321da177e4SLinus Torvalds out: 13331da177e4SLinus Torvalds return retval; 13341da177e4SLinus Torvalds } 13351da177e4SLinus Torvalds 13361da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 13371da177e4SLinus Torvalds 13381da177e4SLinus Torvalds /* 13391da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 13401da177e4SLinus Torvalds */ 1341bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 13421da177e4SLinus Torvalds { 13431da177e4SLinus Torvalds return sys_umount(name, 0); 13441da177e4SLinus Torvalds } 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds #endif 13471da177e4SLinus Torvalds 13484ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 13498823c079SEric W. Biederman { 13504ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 13514ce5d2b1SEric W. Biederman struct inode *inode = dentry->d_inode; 13520bb80f24SDavid Howells struct proc_ns *ei; 13538823c079SEric W. Biederman 13548823c079SEric W. Biederman if (!proc_ns_inode(inode)) 13558823c079SEric W. Biederman return false; 13568823c079SEric W. Biederman 13570bb80f24SDavid Howells ei = get_proc_ns(inode); 13588823c079SEric W. Biederman if (ei->ns_ops != &mntns_operations) 13598823c079SEric W. Biederman return false; 13608823c079SEric W. Biederman 13614ce5d2b1SEric W. Biederman return true; 13624ce5d2b1SEric W. Biederman } 13634ce5d2b1SEric W. Biederman 13644ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 13654ce5d2b1SEric W. Biederman { 13664ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 13674ce5d2b1SEric W. Biederman * mount namespace loop? 13684ce5d2b1SEric W. Biederman */ 13694ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 13704ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 13714ce5d2b1SEric W. Biederman return false; 13724ce5d2b1SEric W. Biederman 13734ce5d2b1SEric W. Biederman mnt_ns = get_proc_ns(dentry->d_inode)->ns; 13748823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 13758823c079SEric W. Biederman } 13768823c079SEric W. Biederman 137787129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 137836341f64SRam Pai int flag) 13791da177e4SLinus Torvalds { 138084d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 13811da177e4SLinus Torvalds 13824ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 13834ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 13844ce5d2b1SEric W. Biederman 13854ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1386be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 13879676f0c6SRam Pai 138836341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1389be34d1a3SDavid Howells if (IS_ERR(q)) 1390be34d1a3SDavid Howells return q; 1391be34d1a3SDavid Howells 13925ff9d8a6SEric W. Biederman q->mnt.mnt_flags &= ~MNT_LOCKED; 1393a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 13941da177e4SLinus Torvalds 13951da177e4SLinus Torvalds p = mnt; 13966b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1397315fc83eSAl Viro struct mount *s; 13987ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 13991da177e4SLinus Torvalds continue; 14001da177e4SLinus Torvalds 1401909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 14024ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 14034ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 14044ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 14054ce5d2b1SEric W. Biederman continue; 14064ce5d2b1SEric W. Biederman } 14074ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 14084ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 14099676f0c6SRam Pai s = skip_mnt_tree(s); 14109676f0c6SRam Pai continue; 14119676f0c6SRam Pai } 14120714a533SAl Viro while (p != s->mnt_parent) { 14130714a533SAl Viro p = p->mnt_parent; 14140714a533SAl Viro q = q->mnt_parent; 14151da177e4SLinus Torvalds } 141687129cc0SAl Viro p = s; 141784d17192SAl Viro parent = q; 141887129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1419be34d1a3SDavid Howells if (IS_ERR(q)) 1420be34d1a3SDavid Howells goto out; 1421962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 14221a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 142384d17192SAl Viro attach_mnt(q, parent, p->mnt_mp); 1424962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 14251da177e4SLinus Torvalds } 14261da177e4SLinus Torvalds } 14271da177e4SLinus Torvalds return res; 1428be34d1a3SDavid Howells out: 14291da177e4SLinus Torvalds if (res) { 1430962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1431328e6d90SAl Viro umount_tree(res, 0); 1432962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 14331da177e4SLinus Torvalds } 1434be34d1a3SDavid Howells return q; 14351da177e4SLinus Torvalds } 14361da177e4SLinus Torvalds 1437be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1438be34d1a3SDavid Howells 1439589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 14408aec0809SAl Viro { 1441cb338d06SAl Viro struct mount *tree; 144297216be0SAl Viro namespace_lock(); 144387129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 144487129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1445328e6d90SAl Viro namespace_unlock(); 1446be34d1a3SDavid Howells if (IS_ERR(tree)) 144752e220d3SDan Carpenter return ERR_CAST(tree); 1448be34d1a3SDavid Howells return &tree->mnt; 14498aec0809SAl Viro } 14508aec0809SAl Viro 14518aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 14528aec0809SAl Viro { 145397216be0SAl Viro namespace_lock(); 1454962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1455328e6d90SAl Viro umount_tree(real_mount(mnt), 0); 1456962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 14573ab6abeeSAl Viro namespace_unlock(); 14588aec0809SAl Viro } 14598aec0809SAl Viro 14601f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 14611f707137SAl Viro struct vfsmount *root) 14621f707137SAl Viro { 14631a4eeaf2SAl Viro struct mount *mnt; 14641f707137SAl Viro int res = f(root, arg); 14651f707137SAl Viro if (res) 14661f707137SAl Viro return res; 14671a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 14681a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 14691f707137SAl Viro if (res) 14701f707137SAl Viro return res; 14711f707137SAl Viro } 14721f707137SAl Viro return 0; 14731f707137SAl Viro } 14741f707137SAl Viro 14754b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1476719f5d7fSMiklos Szeredi { 1477315fc83eSAl Viro struct mount *p; 1478719f5d7fSMiklos Szeredi 1479909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1480fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 14814b8b21f4SAl Viro mnt_release_group_id(p); 1482719f5d7fSMiklos Szeredi } 1483719f5d7fSMiklos Szeredi } 1484719f5d7fSMiklos Szeredi 14854b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1486719f5d7fSMiklos Szeredi { 1487315fc83eSAl Viro struct mount *p; 1488719f5d7fSMiklos Szeredi 1489909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1490fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 14914b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1492719f5d7fSMiklos Szeredi if (err) { 14934b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1494719f5d7fSMiklos Szeredi return err; 1495719f5d7fSMiklos Szeredi } 1496719f5d7fSMiklos Szeredi } 1497719f5d7fSMiklos Szeredi } 1498719f5d7fSMiklos Szeredi 1499719f5d7fSMiklos Szeredi return 0; 1500719f5d7fSMiklos Szeredi } 1501719f5d7fSMiklos Szeredi 1502b90fa9aeSRam Pai /* 1503b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1504b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 150521444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 150621444403SRam Pai * store the parent mount and mountpoint dentry. 150721444403SRam Pai * (done when source_mnt is moved) 1508b90fa9aeSRam Pai * 1509b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1510b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 15119676f0c6SRam Pai * --------------------------------------------------------------------------- 1512b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 15139676f0c6SRam Pai * |************************************************************************** 15149676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15159676f0c6SRam Pai * | dest | | | | | 15169676f0c6SRam Pai * | | | | | | | 15179676f0c6SRam Pai * | v | | | | | 15189676f0c6SRam Pai * |************************************************************************** 15199676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 15205afe0022SRam Pai * | | | | | | 15219676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 15229676f0c6SRam Pai * *************************************************************************** 1523b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1524b90fa9aeSRam Pai * destination mount. 1525b90fa9aeSRam Pai * 1526b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1527b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1528b90fa9aeSRam Pai * the peer group of the source mount. 1529b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1530b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1531b90fa9aeSRam Pai * mount. 15325afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 15335afe0022SRam Pai * of the destination mount and the cloned mount is made slave 15345afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 15355afe0022SRam Pai * is marked as 'shared and slave'. 15365afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 15375afe0022SRam Pai * source mount. 15385afe0022SRam Pai * 15399676f0c6SRam Pai * --------------------------------------------------------------------------- 154021444403SRam Pai * | MOVE MOUNT OPERATION | 15419676f0c6SRam Pai * |************************************************************************** 15429676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15439676f0c6SRam Pai * | dest | | | | | 15449676f0c6SRam Pai * | | | | | | | 15459676f0c6SRam Pai * | v | | | | | 15469676f0c6SRam Pai * |************************************************************************** 15479676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 15485afe0022SRam Pai * | | | | | | 15499676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 15509676f0c6SRam Pai * *************************************************************************** 15515afe0022SRam Pai * 15525afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 15535afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 155421444403SRam Pai * (+*) the mount is moved to the destination. 15555afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 15565afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 15575afe0022SRam Pai * the mount is marked as 'shared and slave'. 15585afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1559b90fa9aeSRam Pai * 1560b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1561b90fa9aeSRam Pai * applied to each mount in the tree. 1562b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1563b90fa9aeSRam Pai * in allocations. 1564b90fa9aeSRam Pai */ 15650fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 156684d17192SAl Viro struct mount *dest_mnt, 156784d17192SAl Viro struct mountpoint *dest_mp, 156884d17192SAl Viro struct path *parent_path) 1569b90fa9aeSRam Pai { 1570b90fa9aeSRam Pai LIST_HEAD(tree_list); 1571315fc83eSAl Viro struct mount *child, *p; 1572719f5d7fSMiklos Szeredi int err; 1573b90fa9aeSRam Pai 1574fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 15750fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1576719f5d7fSMiklos Szeredi if (err) 1577719f5d7fSMiklos Szeredi goto out; 1578719f5d7fSMiklos Szeredi } 157984d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1580719f5d7fSMiklos Szeredi if (err) 1581719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1582b90fa9aeSRam Pai 1583962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1584df1a1ad2SAl Viro 1585fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 1586909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 15870f0afb1dSAl Viro set_mnt_shared(p); 1588b90fa9aeSRam Pai } 15891a390689SAl Viro if (parent_path) { 15900fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 159184d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1592143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 159321444403SRam Pai } else { 159484d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 15950fb54e50SAl Viro commit_tree(source_mnt); 159621444403SRam Pai } 1597b90fa9aeSRam Pai 15981b8e5564SAl Viro list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 15991b8e5564SAl Viro list_del_init(&child->mnt_hash); 16004b2619a5SAl Viro commit_tree(child); 1601b90fa9aeSRam Pai } 1602962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 160399b7db7bSNick Piggin 1604b90fa9aeSRam Pai return 0; 1605719f5d7fSMiklos Szeredi 1606719f5d7fSMiklos Szeredi out_cleanup_ids: 1607fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) 16080fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1609719f5d7fSMiklos Szeredi out: 1610719f5d7fSMiklos Szeredi return err; 1611b90fa9aeSRam Pai } 1612b90fa9aeSRam Pai 161384d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 1614b12cea91SAl Viro { 1615b12cea91SAl Viro struct vfsmount *mnt; 161684d17192SAl Viro struct dentry *dentry = path->dentry; 1617b12cea91SAl Viro retry: 161884d17192SAl Viro mutex_lock(&dentry->d_inode->i_mutex); 161984d17192SAl Viro if (unlikely(cant_mount(dentry))) { 162084d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 162184d17192SAl Viro return ERR_PTR(-ENOENT); 1622b12cea91SAl Viro } 162397216be0SAl Viro namespace_lock(); 1624b12cea91SAl Viro mnt = lookup_mnt(path); 162584d17192SAl Viro if (likely(!mnt)) { 162684d17192SAl Viro struct mountpoint *mp = new_mountpoint(dentry); 162784d17192SAl Viro if (IS_ERR(mp)) { 162897216be0SAl Viro namespace_unlock(); 162984d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 163084d17192SAl Viro return mp; 163184d17192SAl Viro } 163284d17192SAl Viro return mp; 163384d17192SAl Viro } 163497216be0SAl Viro namespace_unlock(); 1635b12cea91SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 1636b12cea91SAl Viro path_put(path); 1637b12cea91SAl Viro path->mnt = mnt; 163884d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 1639b12cea91SAl Viro goto retry; 1640b12cea91SAl Viro } 1641b12cea91SAl Viro 164284d17192SAl Viro static void unlock_mount(struct mountpoint *where) 1643b12cea91SAl Viro { 164484d17192SAl Viro struct dentry *dentry = where->m_dentry; 164584d17192SAl Viro put_mountpoint(where); 1646328e6d90SAl Viro namespace_unlock(); 164784d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 1648b12cea91SAl Viro } 1649b12cea91SAl Viro 165084d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 16511da177e4SLinus Torvalds { 165295bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 16531da177e4SLinus Torvalds return -EINVAL; 16541da177e4SLinus Torvalds 165584d17192SAl Viro if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != 165695bc5f25SAl Viro S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) 16571da177e4SLinus Torvalds return -ENOTDIR; 16581da177e4SLinus Torvalds 165984d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 16601da177e4SLinus Torvalds } 16611da177e4SLinus Torvalds 16621da177e4SLinus Torvalds /* 16637a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 16647a2e8a8fSValerie Aurora */ 16657a2e8a8fSValerie Aurora 16667a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 16677a2e8a8fSValerie Aurora { 16687c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 16697a2e8a8fSValerie Aurora 16707a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 16717a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 16727a2e8a8fSValerie Aurora return 0; 16737a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 16747a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 16757a2e8a8fSValerie Aurora return 0; 16767a2e8a8fSValerie Aurora return type; 16777a2e8a8fSValerie Aurora } 16787a2e8a8fSValerie Aurora 16797a2e8a8fSValerie Aurora /* 168007b20889SRam Pai * recursively change the type of the mountpoint. 168107b20889SRam Pai */ 16820a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 168307b20889SRam Pai { 1684315fc83eSAl Viro struct mount *m; 16854b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 168607b20889SRam Pai int recurse = flag & MS_REC; 16877a2e8a8fSValerie Aurora int type; 1688719f5d7fSMiklos Szeredi int err = 0; 168907b20889SRam Pai 16902d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 169107b20889SRam Pai return -EINVAL; 169207b20889SRam Pai 16937a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 16947a2e8a8fSValerie Aurora if (!type) 16957a2e8a8fSValerie Aurora return -EINVAL; 16967a2e8a8fSValerie Aurora 169797216be0SAl Viro namespace_lock(); 1698719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1699719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1700719f5d7fSMiklos Szeredi if (err) 1701719f5d7fSMiklos Szeredi goto out_unlock; 1702719f5d7fSMiklos Szeredi } 1703719f5d7fSMiklos Szeredi 1704962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1705909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 17060f0afb1dSAl Viro change_mnt_propagation(m, type); 1707962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 1708719f5d7fSMiklos Szeredi 1709719f5d7fSMiklos Szeredi out_unlock: 171097216be0SAl Viro namespace_unlock(); 1711719f5d7fSMiklos Szeredi return err; 171207b20889SRam Pai } 171307b20889SRam Pai 17145ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 17155ff9d8a6SEric W. Biederman { 17165ff9d8a6SEric W. Biederman struct mount *child; 17175ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 17185ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 17195ff9d8a6SEric W. Biederman continue; 17205ff9d8a6SEric W. Biederman 17215ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 17225ff9d8a6SEric W. Biederman return true; 17235ff9d8a6SEric W. Biederman } 17245ff9d8a6SEric W. Biederman return false; 17255ff9d8a6SEric W. Biederman } 17265ff9d8a6SEric W. Biederman 172707b20889SRam Pai /* 17281da177e4SLinus Torvalds * do loopback mount. 17291da177e4SLinus Torvalds */ 1730808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 17312dafe1c4SEric Sandeen int recurse) 17321da177e4SLinus Torvalds { 17332d92ab3cSAl Viro struct path old_path; 173484d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 173584d17192SAl Viro struct mountpoint *mp; 173657eccb83SAl Viro int err; 17371da177e4SLinus Torvalds if (!old_name || !*old_name) 17381da177e4SLinus Torvalds return -EINVAL; 1739815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 17401da177e4SLinus Torvalds if (err) 17411da177e4SLinus Torvalds return err; 17421da177e4SLinus Torvalds 17438823c079SEric W. Biederman err = -EINVAL; 17444ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 17458823c079SEric W. Biederman goto out; 17468823c079SEric W. Biederman 174784d17192SAl Viro mp = lock_mount(path); 174884d17192SAl Viro err = PTR_ERR(mp); 174984d17192SAl Viro if (IS_ERR(mp)) 17509676f0c6SRam Pai goto out; 17519676f0c6SRam Pai 175287129cc0SAl Viro old = real_mount(old_path.mnt); 175384d17192SAl Viro parent = real_mount(path->mnt); 175487129cc0SAl Viro 1755b12cea91SAl Viro err = -EINVAL; 1756fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 1757b12cea91SAl Viro goto out2; 1758b12cea91SAl Viro 175984d17192SAl Viro if (!check_mnt(parent) || !check_mnt(old)) 1760b12cea91SAl Viro goto out2; 1761ccd48bc7SAl Viro 17625ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 17635ff9d8a6SEric W. Biederman goto out2; 17645ff9d8a6SEric W. Biederman 17651da177e4SLinus Torvalds if (recurse) 17664ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 17671da177e4SLinus Torvalds else 176887129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 17691da177e4SLinus Torvalds 1770be34d1a3SDavid Howells if (IS_ERR(mnt)) { 1771be34d1a3SDavid Howells err = PTR_ERR(mnt); 1772e9c5d8a5SAndrey Vagin goto out2; 1773be34d1a3SDavid Howells } 1774ccd48bc7SAl Viro 17755ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 17765ff9d8a6SEric W. Biederman 177784d17192SAl Viro err = graft_tree(mnt, parent, mp); 17781da177e4SLinus Torvalds if (err) { 1779962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1780328e6d90SAl Viro umount_tree(mnt, 0); 1781962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 17825b83d2c5SRam Pai } 1783b12cea91SAl Viro out2: 178484d17192SAl Viro unlock_mount(mp); 1785ccd48bc7SAl Viro out: 17862d92ab3cSAl Viro path_put(&old_path); 17871da177e4SLinus Torvalds return err; 17881da177e4SLinus Torvalds } 17891da177e4SLinus Torvalds 17902e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 17912e4b7fcdSDave Hansen { 17922e4b7fcdSDave Hansen int error = 0; 17932e4b7fcdSDave Hansen int readonly_request = 0; 17942e4b7fcdSDave Hansen 17952e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 17962e4b7fcdSDave Hansen readonly_request = 1; 17972e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 17982e4b7fcdSDave Hansen return 0; 17992e4b7fcdSDave Hansen 180090563b19SEric W. Biederman if (mnt->mnt_flags & MNT_LOCK_READONLY) 180190563b19SEric W. Biederman return -EPERM; 180290563b19SEric W. Biederman 18032e4b7fcdSDave Hansen if (readonly_request) 180483adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 18052e4b7fcdSDave Hansen else 180683adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 18072e4b7fcdSDave Hansen return error; 18082e4b7fcdSDave Hansen } 18092e4b7fcdSDave Hansen 18101da177e4SLinus Torvalds /* 18111da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 18121da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 18131da177e4SLinus Torvalds * on it - tough luck. 18141da177e4SLinus Torvalds */ 18150a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 18161da177e4SLinus Torvalds void *data) 18171da177e4SLinus Torvalds { 18181da177e4SLinus Torvalds int err; 18192d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 1820143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 18211da177e4SLinus Torvalds 1822143c8c91SAl Viro if (!check_mnt(mnt)) 18231da177e4SLinus Torvalds return -EINVAL; 18241da177e4SLinus Torvalds 18252d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 18261da177e4SLinus Torvalds return -EINVAL; 18271da177e4SLinus Torvalds 1828ff36fe2cSEric Paris err = security_sb_remount(sb, data); 1829ff36fe2cSEric Paris if (err) 1830ff36fe2cSEric Paris return err; 1831ff36fe2cSEric Paris 18321da177e4SLinus Torvalds down_write(&sb->s_umount); 18332e4b7fcdSDave Hansen if (flags & MS_BIND) 18342d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 183557eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 183657eccb83SAl Viro err = -EPERM; 18374aa98cf7SAl Viro else 18381da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 18397b43a79fSAl Viro if (!err) { 1840962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1841143c8c91SAl Viro mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; 1842143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 1843143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1844962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 18450e55a7ccSDan Williams } 18466339dab8SAl Viro up_write(&sb->s_umount); 18471da177e4SLinus Torvalds return err; 18481da177e4SLinus Torvalds } 18491da177e4SLinus Torvalds 1850cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 18519676f0c6SRam Pai { 1852315fc83eSAl Viro struct mount *p; 1853909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 1854fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 18559676f0c6SRam Pai return 1; 18569676f0c6SRam Pai } 18579676f0c6SRam Pai return 0; 18589676f0c6SRam Pai } 18599676f0c6SRam Pai 1860808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 18611da177e4SLinus Torvalds { 18622d92ab3cSAl Viro struct path old_path, parent_path; 1863676da58dSAl Viro struct mount *p; 18640fb54e50SAl Viro struct mount *old; 186584d17192SAl Viro struct mountpoint *mp; 186657eccb83SAl Viro int err; 18671da177e4SLinus Torvalds if (!old_name || !*old_name) 18681da177e4SLinus Torvalds return -EINVAL; 18692d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 18701da177e4SLinus Torvalds if (err) 18711da177e4SLinus Torvalds return err; 18721da177e4SLinus Torvalds 187384d17192SAl Viro mp = lock_mount(path); 187484d17192SAl Viro err = PTR_ERR(mp); 187584d17192SAl Viro if (IS_ERR(mp)) 1876cc53ce53SDavid Howells goto out; 1877cc53ce53SDavid Howells 1878143c8c91SAl Viro old = real_mount(old_path.mnt); 1879fc7be130SAl Viro p = real_mount(path->mnt); 1880143c8c91SAl Viro 18811da177e4SLinus Torvalds err = -EINVAL; 1882fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 18831da177e4SLinus Torvalds goto out1; 18841da177e4SLinus Torvalds 18855ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 18865ff9d8a6SEric W. Biederman goto out1; 18875ff9d8a6SEric W. Biederman 18881da177e4SLinus Torvalds err = -EINVAL; 18892d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 189021444403SRam Pai goto out1; 18911da177e4SLinus Torvalds 1892676da58dSAl Viro if (!mnt_has_parent(old)) 189321444403SRam Pai goto out1; 18941da177e4SLinus Torvalds 18952d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 18962d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 189721444403SRam Pai goto out1; 189821444403SRam Pai /* 189921444403SRam Pai * Don't move a mount residing in a shared parent. 190021444403SRam Pai */ 1901fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 190221444403SRam Pai goto out1; 19039676f0c6SRam Pai /* 19049676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 19059676f0c6SRam Pai * mount which is shared. 19069676f0c6SRam Pai */ 1907fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 19089676f0c6SRam Pai goto out1; 19091da177e4SLinus Torvalds err = -ELOOP; 1910fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 1911676da58dSAl Viro if (p == old) 191221444403SRam Pai goto out1; 19131da177e4SLinus Torvalds 191484d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 19154ac91378SJan Blunck if (err) 191621444403SRam Pai goto out1; 19171da177e4SLinus Torvalds 19181da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 19191da177e4SLinus Torvalds * automatically */ 19206776db3dSAl Viro list_del_init(&old->mnt_expire); 19211da177e4SLinus Torvalds out1: 192284d17192SAl Viro unlock_mount(mp); 19231da177e4SLinus Torvalds out: 19241da177e4SLinus Torvalds if (!err) 19251a390689SAl Viro path_put(&parent_path); 19262d92ab3cSAl Viro path_put(&old_path); 19271da177e4SLinus Torvalds return err; 19281da177e4SLinus Torvalds } 19291da177e4SLinus Torvalds 19309d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 19319d412a43SAl Viro { 19329d412a43SAl Viro int err; 19339d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 19349d412a43SAl Viro if (subtype) { 19359d412a43SAl Viro subtype++; 19369d412a43SAl Viro err = -EINVAL; 19379d412a43SAl Viro if (!subtype[0]) 19389d412a43SAl Viro goto err; 19399d412a43SAl Viro } else 19409d412a43SAl Viro subtype = ""; 19419d412a43SAl Viro 19429d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 19439d412a43SAl Viro err = -ENOMEM; 19449d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 19459d412a43SAl Viro goto err; 19469d412a43SAl Viro return mnt; 19479d412a43SAl Viro 19489d412a43SAl Viro err: 19499d412a43SAl Viro mntput(mnt); 19509d412a43SAl Viro return ERR_PTR(err); 19519d412a43SAl Viro } 19529d412a43SAl Viro 19539d412a43SAl Viro /* 19549d412a43SAl Viro * add a mount into a namespace's mount tree 19559d412a43SAl Viro */ 195695bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 19579d412a43SAl Viro { 195884d17192SAl Viro struct mountpoint *mp; 195984d17192SAl Viro struct mount *parent; 19609d412a43SAl Viro int err; 19619d412a43SAl Viro 19629d412a43SAl Viro mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); 19639d412a43SAl Viro 196484d17192SAl Viro mp = lock_mount(path); 196584d17192SAl Viro if (IS_ERR(mp)) 196684d17192SAl Viro return PTR_ERR(mp); 19679d412a43SAl Viro 196884d17192SAl Viro parent = real_mount(path->mnt); 19699d412a43SAl Viro err = -EINVAL; 197084d17192SAl Viro if (unlikely(!check_mnt(parent))) { 1971156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 1972156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 19739d412a43SAl Viro goto unlock; 1974156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 197584d17192SAl Viro if (!parent->mnt_ns) 1976156cacb1SAl Viro goto unlock; 1977156cacb1SAl Viro } 19789d412a43SAl Viro 19799d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 19809d412a43SAl Viro err = -EBUSY; 198195bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 19829d412a43SAl Viro path->mnt->mnt_root == path->dentry) 19839d412a43SAl Viro goto unlock; 19849d412a43SAl Viro 19859d412a43SAl Viro err = -EINVAL; 198695bc5f25SAl Viro if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) 19879d412a43SAl Viro goto unlock; 19889d412a43SAl Viro 198995bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 199084d17192SAl Viro err = graft_tree(newmnt, parent, mp); 19919d412a43SAl Viro 19929d412a43SAl Viro unlock: 199384d17192SAl Viro unlock_mount(mp); 19949d412a43SAl Viro return err; 19959d412a43SAl Viro } 1996b1e75df4SAl Viro 19971da177e4SLinus Torvalds /* 19981da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 19991da177e4SLinus Torvalds * namespace's tree 20001da177e4SLinus Torvalds */ 20010c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2002808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 20031da177e4SLinus Torvalds { 20040c55cfc4SEric W. Biederman struct file_system_type *type; 20059b40bc90SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 20061da177e4SLinus Torvalds struct vfsmount *mnt; 200715f9a3f3SAl Viro int err; 20081da177e4SLinus Torvalds 20090c55cfc4SEric W. Biederman if (!fstype) 20101da177e4SLinus Torvalds return -EINVAL; 20111da177e4SLinus Torvalds 20120c55cfc4SEric W. Biederman type = get_fs_type(fstype); 20130c55cfc4SEric W. Biederman if (!type) 20140c55cfc4SEric W. Biederman return -ENODEV; 20150c55cfc4SEric W. Biederman 20160c55cfc4SEric W. Biederman if (user_ns != &init_user_ns) { 20170c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_MOUNT)) { 20180c55cfc4SEric W. Biederman put_filesystem(type); 20190c55cfc4SEric W. Biederman return -EPERM; 20200c55cfc4SEric W. Biederman } 20210c55cfc4SEric W. Biederman /* Only in special cases allow devices from mounts 20220c55cfc4SEric W. Biederman * created outside the initial user namespace. 20230c55cfc4SEric W. Biederman */ 20240c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { 20250c55cfc4SEric W. Biederman flags |= MS_NODEV; 20260c55cfc4SEric W. Biederman mnt_flags |= MNT_NODEV; 20270c55cfc4SEric W. Biederman } 20280c55cfc4SEric W. Biederman } 20290c55cfc4SEric W. Biederman 20300c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 20310c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 20320c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 20330c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 20340c55cfc4SEric W. Biederman 20350c55cfc4SEric W. Biederman put_filesystem(type); 20361da177e4SLinus Torvalds if (IS_ERR(mnt)) 20371da177e4SLinus Torvalds return PTR_ERR(mnt); 20381da177e4SLinus Torvalds 203995bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 204015f9a3f3SAl Viro if (err) 204115f9a3f3SAl Viro mntput(mnt); 204215f9a3f3SAl Viro return err; 20431da177e4SLinus Torvalds } 20441da177e4SLinus Torvalds 204519a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 204619a167afSAl Viro { 20476776db3dSAl Viro struct mount *mnt = real_mount(m); 204819a167afSAl Viro int err; 204919a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 205019a167afSAl Viro * expired before we get a chance to add it 205119a167afSAl Viro */ 20526776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 205319a167afSAl Viro 205419a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 205519a167afSAl Viro m->mnt_root == path->dentry) { 2056b1e75df4SAl Viro err = -ELOOP; 2057b1e75df4SAl Viro goto fail; 205819a167afSAl Viro } 205919a167afSAl Viro 206095bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2061b1e75df4SAl Viro if (!err) 2062b1e75df4SAl Viro return 0; 2063b1e75df4SAl Viro fail: 2064b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 20656776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 206697216be0SAl Viro namespace_lock(); 20676776db3dSAl Viro list_del_init(&mnt->mnt_expire); 206897216be0SAl Viro namespace_unlock(); 206919a167afSAl Viro } 2070b1e75df4SAl Viro mntput(m); 2071b1e75df4SAl Viro mntput(m); 207219a167afSAl Viro return err; 207319a167afSAl Viro } 207419a167afSAl Viro 2075ea5b778aSDavid Howells /** 2076ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2077ea5b778aSDavid Howells * @mnt: The mount to list. 2078ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2079ea5b778aSDavid Howells */ 2080ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2081ea5b778aSDavid Howells { 208297216be0SAl Viro namespace_lock(); 2083ea5b778aSDavid Howells 20846776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2085ea5b778aSDavid Howells 208697216be0SAl Viro namespace_unlock(); 2087ea5b778aSDavid Howells } 2088ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2089ea5b778aSDavid Howells 2090ea5b778aSDavid Howells /* 20911da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 20921da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 20931da177e4SLinus Torvalds * here 20941da177e4SLinus Torvalds */ 20951da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 20961da177e4SLinus Torvalds { 2097761d5c38SAl Viro struct mount *mnt, *next; 20981da177e4SLinus Torvalds LIST_HEAD(graveyard); 20991da177e4SLinus Torvalds 21001da177e4SLinus Torvalds if (list_empty(mounts)) 21011da177e4SLinus Torvalds return; 21021da177e4SLinus Torvalds 210397216be0SAl Viro namespace_lock(); 2104962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 21051da177e4SLinus Torvalds 21061da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 21071da177e4SLinus Torvalds * following criteria: 21081da177e4SLinus Torvalds * - only referenced by its parent vfsmount 21091da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 21101da177e4SLinus Torvalds * cleared by mntput()) 21111da177e4SLinus Torvalds */ 21126776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2113863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 21141ab59738SAl Viro propagate_mount_busy(mnt, 1)) 21151da177e4SLinus Torvalds continue; 21166776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 21171da177e4SLinus Torvalds } 2118bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 21196776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2120143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2121328e6d90SAl Viro umount_tree(mnt, 1); 2122bcc5c7d2SAl Viro } 2123962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 21243ab6abeeSAl Viro namespace_unlock(); 21251da177e4SLinus Torvalds } 21261da177e4SLinus Torvalds 21271da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 21281da177e4SLinus Torvalds 21291da177e4SLinus Torvalds /* 21305528f911STrond Myklebust * Ripoff of 'select_parent()' 21315528f911STrond Myklebust * 21325528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 21335528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 21345528f911STrond Myklebust */ 2135692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 21365528f911STrond Myklebust { 2137692afc31SAl Viro struct mount *this_parent = parent; 21385528f911STrond Myklebust struct list_head *next; 21395528f911STrond Myklebust int found = 0; 21405528f911STrond Myklebust 21415528f911STrond Myklebust repeat: 21426b41d536SAl Viro next = this_parent->mnt_mounts.next; 21435528f911STrond Myklebust resume: 21446b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 21455528f911STrond Myklebust struct list_head *tmp = next; 21466b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 21475528f911STrond Myklebust 21485528f911STrond Myklebust next = tmp->next; 2149692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 21505528f911STrond Myklebust continue; 21515528f911STrond Myklebust /* 21525528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 21535528f911STrond Myklebust */ 21546b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 21555528f911STrond Myklebust this_parent = mnt; 21565528f911STrond Myklebust goto repeat; 21575528f911STrond Myklebust } 21585528f911STrond Myklebust 21591ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 21606776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 21615528f911STrond Myklebust found++; 21625528f911STrond Myklebust } 21635528f911STrond Myklebust } 21645528f911STrond Myklebust /* 21655528f911STrond Myklebust * All done at this level ... ascend and resume the search 21665528f911STrond Myklebust */ 21675528f911STrond Myklebust if (this_parent != parent) { 21686b41d536SAl Viro next = this_parent->mnt_child.next; 21690714a533SAl Viro this_parent = this_parent->mnt_parent; 21705528f911STrond Myklebust goto resume; 21715528f911STrond Myklebust } 21725528f911STrond Myklebust return found; 21735528f911STrond Myklebust } 21745528f911STrond Myklebust 21755528f911STrond Myklebust /* 21765528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 21775528f911STrond Myklebust * submounts of a specific parent mountpoint 217899b7db7bSNick Piggin * 217999b7db7bSNick Piggin * vfsmount_lock must be held for write 21805528f911STrond Myklebust */ 2181b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 21825528f911STrond Myklebust { 21835528f911STrond Myklebust LIST_HEAD(graveyard); 2184761d5c38SAl Viro struct mount *m; 21855528f911STrond Myklebust 21865528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2187c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2188bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2189761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 21906776db3dSAl Viro mnt_expire); 2191143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2192328e6d90SAl Viro umount_tree(m, 1); 2193bcc5c7d2SAl Viro } 2194bcc5c7d2SAl Viro } 21955528f911STrond Myklebust } 21965528f911STrond Myklebust 21975528f911STrond Myklebust /* 21981da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 21991da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 22001da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 22011da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 22021da177e4SLinus Torvalds */ 2203b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2204b58fed8bSRam Pai unsigned long n) 22051da177e4SLinus Torvalds { 22061da177e4SLinus Torvalds char *t = to; 22071da177e4SLinus Torvalds const char __user *f = from; 22081da177e4SLinus Torvalds char c; 22091da177e4SLinus Torvalds 22101da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 22111da177e4SLinus Torvalds return n; 22121da177e4SLinus Torvalds 22131da177e4SLinus Torvalds while (n) { 22141da177e4SLinus Torvalds if (__get_user(c, f)) { 22151da177e4SLinus Torvalds memset(t, 0, n); 22161da177e4SLinus Torvalds break; 22171da177e4SLinus Torvalds } 22181da177e4SLinus Torvalds *t++ = c; 22191da177e4SLinus Torvalds f++; 22201da177e4SLinus Torvalds n--; 22211da177e4SLinus Torvalds } 22221da177e4SLinus Torvalds return n; 22231da177e4SLinus Torvalds } 22241da177e4SLinus Torvalds 22251da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 22261da177e4SLinus Torvalds { 22271da177e4SLinus Torvalds int i; 22281da177e4SLinus Torvalds unsigned long page; 22291da177e4SLinus Torvalds unsigned long size; 22301da177e4SLinus Torvalds 22311da177e4SLinus Torvalds *where = 0; 22321da177e4SLinus Torvalds if (!data) 22331da177e4SLinus Torvalds return 0; 22341da177e4SLinus Torvalds 22351da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 22361da177e4SLinus Torvalds return -ENOMEM; 22371da177e4SLinus Torvalds 22381da177e4SLinus Torvalds /* We only care that *some* data at the address the user 22391da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 22401da177e4SLinus Torvalds * the remainder of the page. 22411da177e4SLinus Torvalds */ 22421da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 22431da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 22441da177e4SLinus Torvalds if (size > PAGE_SIZE) 22451da177e4SLinus Torvalds size = PAGE_SIZE; 22461da177e4SLinus Torvalds 22471da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 22481da177e4SLinus Torvalds if (!i) { 22491da177e4SLinus Torvalds free_page(page); 22501da177e4SLinus Torvalds return -EFAULT; 22511da177e4SLinus Torvalds } 22521da177e4SLinus Torvalds if (i != PAGE_SIZE) 22531da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 22541da177e4SLinus Torvalds *where = page; 22551da177e4SLinus Torvalds return 0; 22561da177e4SLinus Torvalds } 22571da177e4SLinus Torvalds 2258eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 2259eca6f534SVegard Nossum { 2260eca6f534SVegard Nossum char *tmp; 2261eca6f534SVegard Nossum 2262eca6f534SVegard Nossum if (!data) { 2263eca6f534SVegard Nossum *where = NULL; 2264eca6f534SVegard Nossum return 0; 2265eca6f534SVegard Nossum } 2266eca6f534SVegard Nossum 2267eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 2268eca6f534SVegard Nossum if (IS_ERR(tmp)) 2269eca6f534SVegard Nossum return PTR_ERR(tmp); 2270eca6f534SVegard Nossum 2271eca6f534SVegard Nossum *where = tmp; 2272eca6f534SVegard Nossum return 0; 2273eca6f534SVegard Nossum } 2274eca6f534SVegard Nossum 22751da177e4SLinus Torvalds /* 22761da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 22771da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 22781da177e4SLinus Torvalds * 22791da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 22801da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 22811da177e4SLinus Torvalds * information (or be NULL). 22821da177e4SLinus Torvalds * 22831da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 22841da177e4SLinus Torvalds * When the flags word was introduced its top half was required 22851da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 22861da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 22871da177e4SLinus Torvalds * and must be discarded. 22881da177e4SLinus Torvalds */ 2289808d4e3cSAl Viro long do_mount(const char *dev_name, const char *dir_name, 2290808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 22911da177e4SLinus Torvalds { 22922d92ab3cSAl Viro struct path path; 22931da177e4SLinus Torvalds int retval = 0; 22941da177e4SLinus Torvalds int mnt_flags = 0; 22951da177e4SLinus Torvalds 22961da177e4SLinus Torvalds /* Discard magic */ 22971da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 22981da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 22991da177e4SLinus Torvalds 23001da177e4SLinus Torvalds /* Basic sanity checks */ 23011da177e4SLinus Torvalds 23021da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 23031da177e4SLinus Torvalds return -EINVAL; 23041da177e4SLinus Torvalds 23051da177e4SLinus Torvalds if (data_page) 23061da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 23071da177e4SLinus Torvalds 2308a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 2309a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2310a27ab9f2STetsuo Handa if (retval) 2311a27ab9f2STetsuo Handa return retval; 2312a27ab9f2STetsuo Handa 2313a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2314a27ab9f2STetsuo Handa type_page, flags, data_page); 23150d5cadb8SAl Viro if (!retval && !may_mount()) 23160d5cadb8SAl Viro retval = -EPERM; 2317a27ab9f2STetsuo Handa if (retval) 2318a27ab9f2STetsuo Handa goto dput_out; 2319a27ab9f2STetsuo Handa 2320613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2321613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 23220a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 23230a1c01c9SMatthew Garrett 23241da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 23251da177e4SLinus Torvalds if (flags & MS_NOSUID) 23261da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 23271da177e4SLinus Torvalds if (flags & MS_NODEV) 23281da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 23291da177e4SLinus Torvalds if (flags & MS_NOEXEC) 23301da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2331fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2332fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2333fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2334fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2335d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2336d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 23372e4b7fcdSDave Hansen if (flags & MS_RDONLY) 23382e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2339fc33a7bbSChristoph Hellwig 23407a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2341d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2342d0adde57SMatthew Garrett MS_STRICTATIME); 23431da177e4SLinus Torvalds 23441da177e4SLinus Torvalds if (flags & MS_REMOUNT) 23452d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 23461da177e4SLinus Torvalds data_page); 23471da177e4SLinus Torvalds else if (flags & MS_BIND) 23482d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 23499676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 23502d92ab3cSAl Viro retval = do_change_type(&path, flags); 23511da177e4SLinus Torvalds else if (flags & MS_MOVE) 23522d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 23531da177e4SLinus Torvalds else 23542d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 23551da177e4SLinus Torvalds dev_name, data_page); 23561da177e4SLinus Torvalds dput_out: 23572d92ab3cSAl Viro path_put(&path); 23581da177e4SLinus Torvalds return retval; 23591da177e4SLinus Torvalds } 23601da177e4SLinus Torvalds 2361771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2362771b1371SEric W. Biederman { 236398f842e6SEric W. Biederman proc_free_inum(ns->proc_inum); 2364771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2365771b1371SEric W. Biederman kfree(ns); 2366771b1371SEric W. Biederman } 2367771b1371SEric W. Biederman 23688823c079SEric W. Biederman /* 23698823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 23708823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 23718823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 23728823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 23738823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 23748823c079SEric W. Biederman */ 23758823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 23768823c079SEric W. Biederman 2377771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2378cf8d2c11STrond Myklebust { 2379cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 238098f842e6SEric W. Biederman int ret; 2381cf8d2c11STrond Myklebust 2382cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2383cf8d2c11STrond Myklebust if (!new_ns) 2384cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 238598f842e6SEric W. Biederman ret = proc_alloc_inum(&new_ns->proc_inum); 238698f842e6SEric W. Biederman if (ret) { 238798f842e6SEric W. Biederman kfree(new_ns); 238898f842e6SEric W. Biederman return ERR_PTR(ret); 238998f842e6SEric W. Biederman } 23908823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2391cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2392cf8d2c11STrond Myklebust new_ns->root = NULL; 2393cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2394cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2395cf8d2c11STrond Myklebust new_ns->event = 0; 2396771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2397cf8d2c11STrond Myklebust return new_ns; 2398cf8d2c11STrond Myklebust } 2399cf8d2c11STrond Myklebust 24009559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 24019559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 24021da177e4SLinus Torvalds { 24036b3286edSKirill Korotaev struct mnt_namespace *new_ns; 24047f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2405315fc83eSAl Viro struct mount *p, *q; 24069559f689SAl Viro struct mount *old; 2407cb338d06SAl Viro struct mount *new; 24087a472ef4SEric W. Biederman int copy_flags; 24091da177e4SLinus Torvalds 24109559f689SAl Viro BUG_ON(!ns); 24119559f689SAl Viro 24129559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 24139559f689SAl Viro get_mnt_ns(ns); 24149559f689SAl Viro return ns; 24159559f689SAl Viro } 24169559f689SAl Viro 24179559f689SAl Viro old = ns->root; 24189559f689SAl Viro 2419771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2420cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2421cf8d2c11STrond Myklebust return new_ns; 24221da177e4SLinus Torvalds 242397216be0SAl Viro namespace_lock(); 24241da177e4SLinus Torvalds /* First pass: copy the tree topology */ 24254ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 24269559f689SAl Viro if (user_ns != ns->user_ns) 2427132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 24287a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2429be34d1a3SDavid Howells if (IS_ERR(new)) { 2430328e6d90SAl Viro namespace_unlock(); 2431771b1371SEric W. Biederman free_mnt_ns(new_ns); 2432be34d1a3SDavid Howells return ERR_CAST(new); 24331da177e4SLinus Torvalds } 2434be08d6d2SAl Viro new_ns->root = new; 24351a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 24361da177e4SLinus Torvalds 24371da177e4SLinus Torvalds /* 24381da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 24391da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 24401da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 24411da177e4SLinus Torvalds */ 2442909b0a88SAl Viro p = old; 2443cb338d06SAl Viro q = new; 24441da177e4SLinus Torvalds while (p) { 2445143c8c91SAl Viro q->mnt_ns = new_ns; 24469559f689SAl Viro if (new_fs) { 24479559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 24489559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2449315fc83eSAl Viro rootmnt = &p->mnt; 24501da177e4SLinus Torvalds } 24519559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 24529559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2453315fc83eSAl Viro pwdmnt = &p->mnt; 24541da177e4SLinus Torvalds } 24551da177e4SLinus Torvalds } 2456909b0a88SAl Viro p = next_mnt(p, old); 2457909b0a88SAl Viro q = next_mnt(q, new); 24584ce5d2b1SEric W. Biederman if (!q) 24594ce5d2b1SEric W. Biederman break; 24604ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 24614ce5d2b1SEric W. Biederman p = next_mnt(p, old); 24621da177e4SLinus Torvalds } 2463328e6d90SAl Viro namespace_unlock(); 24641da177e4SLinus Torvalds 24651da177e4SLinus Torvalds if (rootmnt) 2466f03c6599SAl Viro mntput(rootmnt); 24671da177e4SLinus Torvalds if (pwdmnt) 2468f03c6599SAl Viro mntput(pwdmnt); 24691da177e4SLinus Torvalds 2470741a2951SJANAK DESAI return new_ns; 2471741a2951SJANAK DESAI } 2472741a2951SJANAK DESAI 2473cf8d2c11STrond Myklebust /** 2474cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2475cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2476cf8d2c11STrond Myklebust */ 24771a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2478cf8d2c11STrond Myklebust { 2479771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2480cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 24811a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 24821a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2483be08d6d2SAl Viro new_ns->root = mnt; 2484b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2485c1334495SAl Viro } else { 24861a4eeaf2SAl Viro mntput(m); 2487cf8d2c11STrond Myklebust } 2488cf8d2c11STrond Myklebust return new_ns; 2489cf8d2c11STrond Myklebust } 2490cf8d2c11STrond Myklebust 2491ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2492ea441d11SAl Viro { 2493ea441d11SAl Viro struct mnt_namespace *ns; 2494d31da0f0SAl Viro struct super_block *s; 2495ea441d11SAl Viro struct path path; 2496ea441d11SAl Viro int err; 2497ea441d11SAl Viro 2498ea441d11SAl Viro ns = create_mnt_ns(mnt); 2499ea441d11SAl Viro if (IS_ERR(ns)) 2500ea441d11SAl Viro return ERR_CAST(ns); 2501ea441d11SAl Viro 2502ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2503ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2504ea441d11SAl Viro 2505ea441d11SAl Viro put_mnt_ns(ns); 2506ea441d11SAl Viro 2507ea441d11SAl Viro if (err) 2508ea441d11SAl Viro return ERR_PTR(err); 2509ea441d11SAl Viro 2510ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2511d31da0f0SAl Viro s = path.mnt->mnt_sb; 2512d31da0f0SAl Viro atomic_inc(&s->s_active); 2513ea441d11SAl Viro mntput(path.mnt); 2514ea441d11SAl Viro /* lock the sucker */ 2515d31da0f0SAl Viro down_write(&s->s_umount); 2516ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2517ea441d11SAl Viro return path.dentry; 2518ea441d11SAl Viro } 2519ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2520ea441d11SAl Viro 2521bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2522bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 25231da177e4SLinus Torvalds { 2524eca6f534SVegard Nossum int ret; 2525eca6f534SVegard Nossum char *kernel_type; 252691a27b2aSJeff Layton struct filename *kernel_dir; 2527eca6f534SVegard Nossum char *kernel_dev; 25281da177e4SLinus Torvalds unsigned long data_page; 25291da177e4SLinus Torvalds 2530eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2531eca6f534SVegard Nossum if (ret < 0) 2532eca6f534SVegard Nossum goto out_type; 25331da177e4SLinus Torvalds 2534eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2535eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2536eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2537eca6f534SVegard Nossum goto out_dir; 2538eca6f534SVegard Nossum } 25391da177e4SLinus Torvalds 2540eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2541eca6f534SVegard Nossum if (ret < 0) 2542eca6f534SVegard Nossum goto out_dev; 25431da177e4SLinus Torvalds 2544eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2545eca6f534SVegard Nossum if (ret < 0) 2546eca6f534SVegard Nossum goto out_data; 25471da177e4SLinus Torvalds 254891a27b2aSJeff Layton ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, 2549eca6f534SVegard Nossum (void *) data_page); 2550eca6f534SVegard Nossum 25511da177e4SLinus Torvalds free_page(data_page); 2552eca6f534SVegard Nossum out_data: 2553eca6f534SVegard Nossum kfree(kernel_dev); 2554eca6f534SVegard Nossum out_dev: 2555eca6f534SVegard Nossum putname(kernel_dir); 2556eca6f534SVegard Nossum out_dir: 2557eca6f534SVegard Nossum kfree(kernel_type); 2558eca6f534SVegard Nossum out_type: 2559eca6f534SVegard Nossum return ret; 25601da177e4SLinus Torvalds } 25611da177e4SLinus Torvalds 25621da177e4SLinus Torvalds /* 2563afac7cbaSAl Viro * Return true if path is reachable from root 2564afac7cbaSAl Viro * 2565afac7cbaSAl Viro * namespace_sem or vfsmount_lock is held 2566afac7cbaSAl Viro */ 2567643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2568afac7cbaSAl Viro const struct path *root) 2569afac7cbaSAl Viro { 2570643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2571a73324daSAl Viro dentry = mnt->mnt_mountpoint; 25720714a533SAl Viro mnt = mnt->mnt_parent; 2573afac7cbaSAl Viro } 2574643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2575afac7cbaSAl Viro } 2576afac7cbaSAl Viro 2577afac7cbaSAl Viro int path_is_under(struct path *path1, struct path *path2) 2578afac7cbaSAl Viro { 2579afac7cbaSAl Viro int res; 2580962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 2581643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 2582962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 2583afac7cbaSAl Viro return res; 2584afac7cbaSAl Viro } 2585afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 2586afac7cbaSAl Viro 2587afac7cbaSAl Viro /* 25881da177e4SLinus Torvalds * pivot_root Semantics: 25891da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 25901da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 25911da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 25921da177e4SLinus Torvalds * 25931da177e4SLinus Torvalds * Restrictions: 25941da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 25951da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 25961da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 25971da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 25981da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 25991da177e4SLinus Torvalds * 26004a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 26014a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 26024a0d11faSNeil Brown * in this situation. 26034a0d11faSNeil Brown * 26041da177e4SLinus Torvalds * Notes: 26051da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 26061da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 26071da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 26081da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 26091da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 26101da177e4SLinus Torvalds * first. 26111da177e4SLinus Torvalds */ 26123480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 26133480b257SHeiko Carstens const char __user *, put_old) 26141da177e4SLinus Torvalds { 26152d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 261684d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 261784d17192SAl Viro struct mountpoint *old_mp, *root_mp; 26181da177e4SLinus Torvalds int error; 26191da177e4SLinus Torvalds 26209b40bc90SAl Viro if (!may_mount()) 26211da177e4SLinus Torvalds return -EPERM; 26221da177e4SLinus Torvalds 26232d8f3038SAl Viro error = user_path_dir(new_root, &new); 26241da177e4SLinus Torvalds if (error) 26251da177e4SLinus Torvalds goto out0; 26261da177e4SLinus Torvalds 26272d8f3038SAl Viro error = user_path_dir(put_old, &old); 26281da177e4SLinus Torvalds if (error) 26291da177e4SLinus Torvalds goto out1; 26301da177e4SLinus Torvalds 26312d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 2632b12cea91SAl Viro if (error) 2633b12cea91SAl Viro goto out2; 26341da177e4SLinus Torvalds 2635f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 263684d17192SAl Viro old_mp = lock_mount(&old); 263784d17192SAl Viro error = PTR_ERR(old_mp); 263884d17192SAl Viro if (IS_ERR(old_mp)) 2639b12cea91SAl Viro goto out3; 2640b12cea91SAl Viro 26411da177e4SLinus Torvalds error = -EINVAL; 2642419148daSAl Viro new_mnt = real_mount(new.mnt); 2643419148daSAl Viro root_mnt = real_mount(root.mnt); 264484d17192SAl Viro old_mnt = real_mount(old.mnt); 264584d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 2646fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 2647fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 2648b12cea91SAl Viro goto out4; 2649143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 2650b12cea91SAl Viro goto out4; 26515ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 26525ff9d8a6SEric W. Biederman goto out4; 26531da177e4SLinus Torvalds error = -ENOENT; 2654f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 2655b12cea91SAl Viro goto out4; 26561da177e4SLinus Torvalds error = -EBUSY; 265784d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 2658b12cea91SAl Viro goto out4; /* loop, on the same file system */ 26591da177e4SLinus Torvalds error = -EINVAL; 26608c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 2661b12cea91SAl Viro goto out4; /* not a mountpoint */ 2662676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 2663b12cea91SAl Viro goto out4; /* not attached */ 266484d17192SAl Viro root_mp = root_mnt->mnt_mp; 26652d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 2666b12cea91SAl Viro goto out4; /* not a mountpoint */ 2667676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 2668b12cea91SAl Viro goto out4; /* not attached */ 26694ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 267084d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 2671b12cea91SAl Viro goto out4; 267284d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 2673962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 2674419148daSAl Viro detach_mnt(new_mnt, &parent_path); 2675419148daSAl Viro detach_mnt(root_mnt, &root_parent); 26765ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 26775ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 26785ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 26795ff9d8a6SEric W. Biederman } 26804ac91378SJan Blunck /* mount old root on put_old */ 268184d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 26824ac91378SJan Blunck /* mount new_root on / */ 268384d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 26846b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 2685962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 26862d8f3038SAl Viro chroot_fs_refs(&root, &new); 268784d17192SAl Viro put_mountpoint(root_mp); 26881da177e4SLinus Torvalds error = 0; 2689b12cea91SAl Viro out4: 269084d17192SAl Viro unlock_mount(old_mp); 2691b12cea91SAl Viro if (!error) { 26921a390689SAl Viro path_put(&root_parent); 26931a390689SAl Viro path_put(&parent_path); 2694b12cea91SAl Viro } 2695b12cea91SAl Viro out3: 26968c3ee42eSAl Viro path_put(&root); 2697b12cea91SAl Viro out2: 26982d8f3038SAl Viro path_put(&old); 26991da177e4SLinus Torvalds out1: 27002d8f3038SAl Viro path_put(&new); 27011da177e4SLinus Torvalds out0: 27021da177e4SLinus Torvalds return error; 27031da177e4SLinus Torvalds } 27041da177e4SLinus Torvalds 27051da177e4SLinus Torvalds static void __init init_mount_tree(void) 27061da177e4SLinus Torvalds { 27071da177e4SLinus Torvalds struct vfsmount *mnt; 27086b3286edSKirill Korotaev struct mnt_namespace *ns; 2709ac748a09SJan Blunck struct path root; 27100c55cfc4SEric W. Biederman struct file_system_type *type; 27111da177e4SLinus Torvalds 27120c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 27130c55cfc4SEric W. Biederman if (!type) 27140c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 27150c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 27160c55cfc4SEric W. Biederman put_filesystem(type); 27171da177e4SLinus Torvalds if (IS_ERR(mnt)) 27181da177e4SLinus Torvalds panic("Can't create rootfs"); 2719b3e19d92SNick Piggin 27203b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 27213b22edc5STrond Myklebust if (IS_ERR(ns)) 27221da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 27231da177e4SLinus Torvalds 27246b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 27256b3286edSKirill Korotaev get_mnt_ns(ns); 27261da177e4SLinus Torvalds 2727be08d6d2SAl Viro root.mnt = mnt; 2728be08d6d2SAl Viro root.dentry = mnt->mnt_root; 2729ac748a09SJan Blunck 2730ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2731ac748a09SJan Blunck set_fs_root(current->fs, &root); 27321da177e4SLinus Torvalds } 27331da177e4SLinus Torvalds 273474bf17cfSDenis Cheng void __init mnt_init(void) 27351da177e4SLinus Torvalds { 273613f14b4dSEric Dumazet unsigned u; 273715a67dd8SRandy Dunlap int err; 27381da177e4SLinus Torvalds 27397d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 274020c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 27411da177e4SLinus Torvalds 2742b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 274384d17192SAl Viro mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 27441da177e4SLinus Torvalds 274584d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 27461da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 27471da177e4SLinus Torvalds 274880cdc6daSMandeep Singh Baines printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); 27491da177e4SLinus Torvalds 275013f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 275113f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 275284d17192SAl Viro for (u = 0; u < HASH_SIZE; u++) 275384d17192SAl Viro INIT_LIST_HEAD(&mountpoint_hashtable[u]); 27541da177e4SLinus Torvalds 2755962830dfSAndi Kleen br_lock_init(&vfsmount_lock); 275699b7db7bSNick Piggin 275715a67dd8SRandy Dunlap err = sysfs_init(); 275815a67dd8SRandy Dunlap if (err) 275915a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 27608e24eea7SHarvey Harrison __func__, err); 276100d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 276200d26666SGreg Kroah-Hartman if (!fs_kobj) 27638e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 27641da177e4SLinus Torvalds init_rootfs(); 27651da177e4SLinus Torvalds init_mount_tree(); 27661da177e4SLinus Torvalds } 27671da177e4SLinus Torvalds 2768616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 27691da177e4SLinus Torvalds { 2770d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 2771616511d0STrond Myklebust return; 27727b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 2773771b1371SEric W. Biederman free_mnt_ns(ns); 27741da177e4SLinus Torvalds } 27759d412a43SAl Viro 27769d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 27779d412a43SAl Viro { 2778423e0ab0STim Chen struct vfsmount *mnt; 2779423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 2780423e0ab0STim Chen if (!IS_ERR(mnt)) { 2781423e0ab0STim Chen /* 2782423e0ab0STim Chen * it is a longterm mount, don't release mnt until 2783423e0ab0STim Chen * we unmount before file sys is unregistered 2784423e0ab0STim Chen */ 2785f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 2786423e0ab0STim Chen } 2787423e0ab0STim Chen return mnt; 27889d412a43SAl Viro } 27899d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 2790423e0ab0STim Chen 2791423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 2792423e0ab0STim Chen { 2793423e0ab0STim Chen /* release long term mount so mount point can be released */ 2794423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 2795f7a99c5bSAl Viro br_write_lock(&vfsmount_lock); 2796f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 2797f7a99c5bSAl Viro br_write_unlock(&vfsmount_lock); 2798423e0ab0STim Chen mntput(mnt); 2799423e0ab0STim Chen } 2800423e0ab0STim Chen } 2801423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 280202125a82SAl Viro 280302125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 280402125a82SAl Viro { 2805143c8c91SAl Viro return check_mnt(real_mount(mnt)); 280602125a82SAl Viro } 28078823c079SEric W. Biederman 28083151527eSEric W. Biederman bool current_chrooted(void) 28093151527eSEric W. Biederman { 28103151527eSEric W. Biederman /* Does the current process have a non-standard root */ 28113151527eSEric W. Biederman struct path ns_root; 28123151527eSEric W. Biederman struct path fs_root; 28133151527eSEric W. Biederman bool chrooted; 28143151527eSEric W. Biederman 28153151527eSEric W. Biederman /* Find the namespace root */ 28163151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 28173151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 28183151527eSEric W. Biederman path_get(&ns_root); 28193151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 28203151527eSEric W. Biederman ; 28213151527eSEric W. Biederman 28223151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 28233151527eSEric W. Biederman 28243151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 28253151527eSEric W. Biederman 28263151527eSEric W. Biederman path_put(&fs_root); 28273151527eSEric W. Biederman path_put(&ns_root); 28283151527eSEric W. Biederman 28293151527eSEric W. Biederman return chrooted; 28303151527eSEric W. Biederman } 28313151527eSEric W. Biederman 2832e51db735SEric W. Biederman bool fs_fully_visible(struct file_system_type *type) 283387a8ebd6SEric W. Biederman { 283487a8ebd6SEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 283587a8ebd6SEric W. Biederman struct mount *mnt; 2836e51db735SEric W. Biederman bool visible = false; 283787a8ebd6SEric W. Biederman 2838e51db735SEric W. Biederman if (unlikely(!ns)) 2839e51db735SEric W. Biederman return false; 2840e51db735SEric W. Biederman 284144bb4385SAl Viro down_read(&namespace_sem); 284287a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 2843e51db735SEric W. Biederman struct mount *child; 2844e51db735SEric W. Biederman if (mnt->mnt.mnt_sb->s_type != type) 2845e51db735SEric W. Biederman continue; 2846e51db735SEric W. Biederman 2847e51db735SEric W. Biederman /* This mount is not fully visible if there are any child mounts 2848e51db735SEric W. Biederman * that cover anything except for empty directories. 2849e51db735SEric W. Biederman */ 2850e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 2851e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 2852e51db735SEric W. Biederman if (!S_ISDIR(inode->i_mode)) 2853e51db735SEric W. Biederman goto next; 2854e51db735SEric W. Biederman if (inode->i_nlink != 2) 2855e51db735SEric W. Biederman goto next; 285687a8ebd6SEric W. Biederman } 2857e51db735SEric W. Biederman visible = true; 2858e51db735SEric W. Biederman goto found; 2859e51db735SEric W. Biederman next: ; 286087a8ebd6SEric W. Biederman } 2861e51db735SEric W. Biederman found: 286244bb4385SAl Viro up_read(&namespace_sem); 2863e51db735SEric W. Biederman return visible; 286487a8ebd6SEric W. Biederman } 286587a8ebd6SEric W. Biederman 28668823c079SEric W. Biederman static void *mntns_get(struct task_struct *task) 28678823c079SEric W. Biederman { 28688823c079SEric W. Biederman struct mnt_namespace *ns = NULL; 28698823c079SEric W. Biederman struct nsproxy *nsproxy; 28708823c079SEric W. Biederman 28718823c079SEric W. Biederman rcu_read_lock(); 28728823c079SEric W. Biederman nsproxy = task_nsproxy(task); 28738823c079SEric W. Biederman if (nsproxy) { 28748823c079SEric W. Biederman ns = nsproxy->mnt_ns; 28758823c079SEric W. Biederman get_mnt_ns(ns); 28768823c079SEric W. Biederman } 28778823c079SEric W. Biederman rcu_read_unlock(); 28788823c079SEric W. Biederman 28798823c079SEric W. Biederman return ns; 28808823c079SEric W. Biederman } 28818823c079SEric W. Biederman 28828823c079SEric W. Biederman static void mntns_put(void *ns) 28838823c079SEric W. Biederman { 28848823c079SEric W. Biederman put_mnt_ns(ns); 28858823c079SEric W. Biederman } 28868823c079SEric W. Biederman 28878823c079SEric W. Biederman static int mntns_install(struct nsproxy *nsproxy, void *ns) 28888823c079SEric W. Biederman { 28898823c079SEric W. Biederman struct fs_struct *fs = current->fs; 28908823c079SEric W. Biederman struct mnt_namespace *mnt_ns = ns; 28918823c079SEric W. Biederman struct path root; 28928823c079SEric W. Biederman 28930c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 2894c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 2895c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 2896ae11e0f1SZhao Hongjiang return -EPERM; 28978823c079SEric W. Biederman 28988823c079SEric W. Biederman if (fs->users != 1) 28998823c079SEric W. Biederman return -EINVAL; 29008823c079SEric W. Biederman 29018823c079SEric W. Biederman get_mnt_ns(mnt_ns); 29028823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 29038823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 29048823c079SEric W. Biederman 29058823c079SEric W. Biederman /* Find the root */ 29068823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 29078823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 29088823c079SEric W. Biederman path_get(&root); 29098823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 29108823c079SEric W. Biederman ; 29118823c079SEric W. Biederman 29128823c079SEric W. Biederman /* Update the pwd and root */ 29138823c079SEric W. Biederman set_fs_pwd(fs, &root); 29148823c079SEric W. Biederman set_fs_root(fs, &root); 29158823c079SEric W. Biederman 29168823c079SEric W. Biederman path_put(&root); 29178823c079SEric W. Biederman return 0; 29188823c079SEric W. Biederman } 29198823c079SEric W. Biederman 292098f842e6SEric W. Biederman static unsigned int mntns_inum(void *ns) 292198f842e6SEric W. Biederman { 292298f842e6SEric W. Biederman struct mnt_namespace *mnt_ns = ns; 292398f842e6SEric W. Biederman return mnt_ns->proc_inum; 292498f842e6SEric W. Biederman } 292598f842e6SEric W. Biederman 29268823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 29278823c079SEric W. Biederman .name = "mnt", 29288823c079SEric W. Biederman .type = CLONE_NEWNS, 29298823c079SEric W. Biederman .get = mntns_get, 29308823c079SEric W. Biederman .put = mntns_put, 29318823c079SEric W. Biederman .install = mntns_install, 293298f842e6SEric W. Biederman .inum = mntns_inum, 29338823c079SEric W. Biederman }; 2934