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> 151da177e4SLinus Torvalds #include <linux/namei.h> 161da177e4SLinus Torvalds #include <linux/security.h> 1773cd49ecSMiklos Szeredi #include <linux/idr.h> 18d10577a8SAl Viro #include <linux/acct.h> /* acct_auto_close_mnt */ 19d10577a8SAl Viro #include <linux/ramfs.h> /* init_rootfs */ 20d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 21d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 22d10577a8SAl Viro #include <linux/uaccess.h> 238823c079SEric W. Biederman #include <linux/proc_fs.h> 2407b20889SRam Pai #include "pnode.h" 25948730b0SAdrian Bunk #include "internal.h" 261da177e4SLinus Torvalds 2713f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 2813f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 2913f14b4dSEric Dumazet 305addc5ddSAl Viro static int event; 3173cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 32719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 3399b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 34f21f6220SAl Viro static int mnt_id_start = 0; 35f21f6220SAl Viro static int mnt_group_start = 1; 365addc5ddSAl Viro 37fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 38e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 39390c6843SRam Pai static struct rw_semaphore namespace_sem; 401da177e4SLinus Torvalds 41f87fd4c2SMiklos Szeredi /* /sys/fs */ 4200d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 4300d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 44f87fd4c2SMiklos Szeredi 4599b7db7bSNick Piggin /* 4699b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 4799b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 4899b7db7bSNick Piggin * up the tree. 4999b7db7bSNick Piggin * 5099b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 5199b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 5299b7db7bSNick Piggin */ 5399b7db7bSNick Piggin DEFINE_BRLOCK(vfsmount_lock); 5499b7db7bSNick Piggin 551da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 561da177e4SLinus Torvalds { 571da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 581da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 5913f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 6013f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 611da177e4SLinus Torvalds } 621da177e4SLinus Torvalds 633d733633SDave Hansen #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 643d733633SDave Hansen 6599b7db7bSNick Piggin /* 6699b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 6799b7db7bSNick Piggin * serialize with freeing. 6899b7db7bSNick Piggin */ 69b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 7073cd49ecSMiklos Szeredi { 7173cd49ecSMiklos Szeredi int res; 7273cd49ecSMiklos Szeredi 7373cd49ecSMiklos Szeredi retry: 7473cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 7599b7db7bSNick Piggin spin_lock(&mnt_id_lock); 7615169fe7SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 77f21f6220SAl Viro if (!res) 7815169fe7SAl Viro mnt_id_start = mnt->mnt_id + 1; 7999b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 8073cd49ecSMiklos Szeredi if (res == -EAGAIN) 8173cd49ecSMiklos Szeredi goto retry; 8273cd49ecSMiklos Szeredi 8373cd49ecSMiklos Szeredi return res; 8473cd49ecSMiklos Szeredi } 8573cd49ecSMiklos Szeredi 86b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 8773cd49ecSMiklos Szeredi { 8815169fe7SAl Viro int id = mnt->mnt_id; 8999b7db7bSNick Piggin spin_lock(&mnt_id_lock); 90f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 91f21f6220SAl Viro if (mnt_id_start > id) 92f21f6220SAl Viro mnt_id_start = id; 9399b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 9473cd49ecSMiklos Szeredi } 9573cd49ecSMiklos Szeredi 96719f5d7fSMiklos Szeredi /* 97719f5d7fSMiklos Szeredi * Allocate a new peer group ID 98719f5d7fSMiklos Szeredi * 99719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 100719f5d7fSMiklos Szeredi */ 1014b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 102719f5d7fSMiklos Szeredi { 103f21f6220SAl Viro int res; 104f21f6220SAl Viro 105719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 106719f5d7fSMiklos Szeredi return -ENOMEM; 107719f5d7fSMiklos Szeredi 108f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 109f21f6220SAl Viro mnt_group_start, 11015169fe7SAl Viro &mnt->mnt_group_id); 111f21f6220SAl Viro if (!res) 11215169fe7SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 113f21f6220SAl Viro 114f21f6220SAl Viro return res; 115719f5d7fSMiklos Szeredi } 116719f5d7fSMiklos Szeredi 117719f5d7fSMiklos Szeredi /* 118719f5d7fSMiklos Szeredi * Release a peer group ID 119719f5d7fSMiklos Szeredi */ 1204b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 121719f5d7fSMiklos Szeredi { 12215169fe7SAl Viro int id = mnt->mnt_group_id; 123f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 124f21f6220SAl Viro if (mnt_group_start > id) 125f21f6220SAl Viro mnt_group_start = id; 12615169fe7SAl Viro mnt->mnt_group_id = 0; 127719f5d7fSMiklos Szeredi } 128719f5d7fSMiklos Szeredi 129b3e19d92SNick Piggin /* 130b3e19d92SNick Piggin * vfsmount lock must be held for read 131b3e19d92SNick Piggin */ 13283adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 133b3e19d92SNick Piggin { 134b3e19d92SNick Piggin #ifdef CONFIG_SMP 13568e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 136b3e19d92SNick Piggin #else 137b3e19d92SNick Piggin preempt_disable(); 13868e8a9feSAl Viro mnt->mnt_count += n; 139b3e19d92SNick Piggin preempt_enable(); 140b3e19d92SNick Piggin #endif 141b3e19d92SNick Piggin } 142b3e19d92SNick Piggin 143b3e19d92SNick Piggin /* 144b3e19d92SNick Piggin * vfsmount lock must be held for write 145b3e19d92SNick Piggin */ 14683adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 147b3e19d92SNick Piggin { 148b3e19d92SNick Piggin #ifdef CONFIG_SMP 149f03c6599SAl Viro unsigned int count = 0; 150b3e19d92SNick Piggin int cpu; 151b3e19d92SNick Piggin 152b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 15368e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 154b3e19d92SNick Piggin } 155b3e19d92SNick Piggin 156b3e19d92SNick Piggin return count; 157b3e19d92SNick Piggin #else 15868e8a9feSAl Viro return mnt->mnt_count; 159b3e19d92SNick Piggin #endif 160b3e19d92SNick Piggin } 161b3e19d92SNick Piggin 162b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1631da177e4SLinus Torvalds { 164c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 165c63181e6SAl Viro if (mnt) { 16673cd49ecSMiklos Szeredi int err; 16773cd49ecSMiklos Szeredi 168c63181e6SAl Viro err = mnt_alloc_id(mnt); 16988b38782SLi Zefan if (err) 17088b38782SLi Zefan goto out_free_cache; 17188b38782SLi Zefan 17288b38782SLi Zefan if (name) { 173c63181e6SAl Viro mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 174c63181e6SAl Viro if (!mnt->mnt_devname) 17588b38782SLi Zefan goto out_free_id; 17673cd49ecSMiklos Szeredi } 17773cd49ecSMiklos Szeredi 178b3e19d92SNick Piggin #ifdef CONFIG_SMP 179c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 180c63181e6SAl Viro if (!mnt->mnt_pcp) 181b3e19d92SNick Piggin goto out_free_devname; 182b3e19d92SNick Piggin 183c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 184b3e19d92SNick Piggin #else 185c63181e6SAl Viro mnt->mnt_count = 1; 186c63181e6SAl Viro mnt->mnt_writers = 0; 187b3e19d92SNick Piggin #endif 188b3e19d92SNick Piggin 189c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_hash); 190c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 191c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 192c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 193c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 194c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 195c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 196c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 1972504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 1982504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 1992504c5d6SAndreas Gruenbacher #endif 2001da177e4SLinus Torvalds } 201c63181e6SAl Viro return mnt; 20288b38782SLi Zefan 203d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 204d3ef3d73Snpiggin@suse.de out_free_devname: 205c63181e6SAl Viro kfree(mnt->mnt_devname); 206d3ef3d73Snpiggin@suse.de #endif 20788b38782SLi Zefan out_free_id: 208c63181e6SAl Viro mnt_free_id(mnt); 20988b38782SLi Zefan out_free_cache: 210c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 21188b38782SLi Zefan return NULL; 2121da177e4SLinus Torvalds } 2131da177e4SLinus Torvalds 2148366025eSDave Hansen /* 2158366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2168366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2178366025eSDave Hansen * We must keep track of when those operations start 2188366025eSDave Hansen * (for permission checks) and when they end, so that 2198366025eSDave Hansen * we can determine when writes are able to occur to 2208366025eSDave Hansen * a filesystem. 2218366025eSDave Hansen */ 2223d733633SDave Hansen /* 2233d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2243d733633SDave Hansen * @mnt: the mount to check for its write status 2253d733633SDave Hansen * 2263d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2273d733633SDave Hansen * It does not guarantee that the filesystem will stay 2283d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2293d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2303d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2313d733633SDave Hansen * r/w. 2323d733633SDave Hansen */ 2333d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2343d733633SDave Hansen { 2352e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2362e4b7fcdSDave Hansen return 1; 2372e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2382e4b7fcdSDave Hansen return 1; 2392e4b7fcdSDave Hansen return 0; 2403d733633SDave Hansen } 2413d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2423d733633SDave Hansen 24383adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2443d733633SDave Hansen { 245d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 24668e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 247d3ef3d73Snpiggin@suse.de #else 24868e8a9feSAl Viro mnt->mnt_writers++; 249d3ef3d73Snpiggin@suse.de #endif 2503d733633SDave Hansen } 2513d733633SDave Hansen 25283adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2533d733633SDave Hansen { 254d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 25568e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 256d3ef3d73Snpiggin@suse.de #else 25768e8a9feSAl Viro mnt->mnt_writers--; 258d3ef3d73Snpiggin@suse.de #endif 259d3ef3d73Snpiggin@suse.de } 260d3ef3d73Snpiggin@suse.de 26183adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 262d3ef3d73Snpiggin@suse.de { 263d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 264d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2653d733633SDave Hansen int cpu; 2663d733633SDave Hansen 2673d733633SDave Hansen for_each_possible_cpu(cpu) { 26868e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2693d733633SDave Hansen } 2703d733633SDave Hansen 271d3ef3d73Snpiggin@suse.de return count; 272d3ef3d73Snpiggin@suse.de #else 273d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 274d3ef3d73Snpiggin@suse.de #endif 2753d733633SDave Hansen } 2763d733633SDave Hansen 2774ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2784ed5e82fSMiklos Szeredi { 2794ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2804ed5e82fSMiklos Szeredi return 1; 2814ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2824ed5e82fSMiklos Szeredi smp_rmb(); 2834ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2844ed5e82fSMiklos Szeredi } 2854ed5e82fSMiklos Szeredi 2863d733633SDave Hansen /* 287eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 288eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 289eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 290eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 2913d733633SDave Hansen */ 2928366025eSDave Hansen /** 293eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 29483adc753SAl Viro * @m: the mount on which to take a write 2958366025eSDave Hansen * 296eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 297eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 298eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 299eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 300eb04c282SJan Kara * called. This is effectively a refcount. 3018366025eSDave Hansen */ 302eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3038366025eSDave Hansen { 30483adc753SAl Viro struct mount *mnt = real_mount(m); 3053d733633SDave Hansen int ret = 0; 3063d733633SDave Hansen 307d3ef3d73Snpiggin@suse.de preempt_disable(); 308c6653a83SNick Piggin mnt_inc_writers(mnt); 309d3ef3d73Snpiggin@suse.de /* 310c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 311d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 312d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 313d3ef3d73Snpiggin@suse.de */ 314d3ef3d73Snpiggin@suse.de smp_mb(); 31583adc753SAl Viro while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 316d3ef3d73Snpiggin@suse.de cpu_relax(); 317d3ef3d73Snpiggin@suse.de /* 318d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 319d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 320d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 321d3ef3d73Snpiggin@suse.de */ 322d3ef3d73Snpiggin@suse.de smp_rmb(); 3234ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 324c6653a83SNick Piggin mnt_dec_writers(mnt); 3253d733633SDave Hansen ret = -EROFS; 3263d733633SDave Hansen } 327d3ef3d73Snpiggin@suse.de preempt_enable(); 328eb04c282SJan Kara 329eb04c282SJan Kara return ret; 330eb04c282SJan Kara } 331eb04c282SJan Kara 332eb04c282SJan Kara /** 333eb04c282SJan Kara * mnt_want_write - get write access to a mount 334eb04c282SJan Kara * @m: the mount on which to take a write 335eb04c282SJan Kara * 336eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 337eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 338eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 339eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 340eb04c282SJan Kara */ 341eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 342eb04c282SJan Kara { 343eb04c282SJan Kara int ret; 344eb04c282SJan Kara 345eb04c282SJan Kara sb_start_write(m->mnt_sb); 346eb04c282SJan Kara ret = __mnt_want_write(m); 347eb04c282SJan Kara if (ret) 348eb04c282SJan Kara sb_end_write(m->mnt_sb); 3493d733633SDave Hansen return ret; 3508366025eSDave Hansen } 3518366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3528366025eSDave Hansen 3538366025eSDave Hansen /** 35496029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 35596029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 35696029c4eSnpiggin@suse.de * 35796029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 35896029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 35996029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 36096029c4eSnpiggin@suse.de * on it. This allows some optimisation. 36196029c4eSnpiggin@suse.de * 36296029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 36396029c4eSnpiggin@suse.de * drop the reference. 36496029c4eSnpiggin@suse.de */ 36596029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 36696029c4eSnpiggin@suse.de { 36796029c4eSnpiggin@suse.de /* superblock may be r/o */ 36896029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 36996029c4eSnpiggin@suse.de return -EROFS; 37096029c4eSnpiggin@suse.de preempt_disable(); 37183adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 37296029c4eSnpiggin@suse.de preempt_enable(); 37396029c4eSnpiggin@suse.de return 0; 37496029c4eSnpiggin@suse.de } 37596029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 37696029c4eSnpiggin@suse.de 37796029c4eSnpiggin@suse.de /** 378eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 379eb04c282SJan Kara * @file: the file who's mount on which to take a write 380eb04c282SJan Kara * 381eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 382eb04c282SJan Kara * do some optimisations if the file is open for write already 383eb04c282SJan Kara */ 384eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 385eb04c282SJan Kara { 386eb04c282SJan Kara struct inode *inode = file->f_dentry->d_inode; 387eb04c282SJan Kara 388eb04c282SJan Kara if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) 389eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 390eb04c282SJan Kara else 391eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 392eb04c282SJan Kara } 393eb04c282SJan Kara 394eb04c282SJan Kara /** 39596029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 39696029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 39796029c4eSnpiggin@suse.de * 39896029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 39996029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 40096029c4eSnpiggin@suse.de */ 40196029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 40296029c4eSnpiggin@suse.de { 403eb04c282SJan Kara int ret; 404eb04c282SJan Kara 405eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 406eb04c282SJan Kara ret = __mnt_want_write_file(file); 407eb04c282SJan Kara if (ret) 408eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 409eb04c282SJan Kara return ret; 41096029c4eSnpiggin@suse.de } 41196029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 41296029c4eSnpiggin@suse.de 41396029c4eSnpiggin@suse.de /** 414eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4158366025eSDave Hansen * @mnt: the mount on which to give up write access 4168366025eSDave Hansen * 4178366025eSDave Hansen * Tells the low-level filesystem that we are done 4188366025eSDave Hansen * performing writes to it. Must be matched with 419eb04c282SJan Kara * __mnt_want_write() call above. 4208366025eSDave Hansen */ 421eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4228366025eSDave Hansen { 423d3ef3d73Snpiggin@suse.de preempt_disable(); 42483adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 425d3ef3d73Snpiggin@suse.de preempt_enable(); 4268366025eSDave Hansen } 427eb04c282SJan Kara 428eb04c282SJan Kara /** 429eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 430eb04c282SJan Kara * @mnt: the mount on which to give up write access 431eb04c282SJan Kara * 432eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 433eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 434eb04c282SJan Kara * mnt_want_write() call above. 435eb04c282SJan Kara */ 436eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 437eb04c282SJan Kara { 438eb04c282SJan Kara __mnt_drop_write(mnt); 439eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 440eb04c282SJan Kara } 4418366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4428366025eSDave Hansen 443eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 444eb04c282SJan Kara { 445eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 446eb04c282SJan Kara } 447eb04c282SJan Kara 4482a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4492a79f17eSAl Viro { 4502a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4512a79f17eSAl Viro } 4522a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4532a79f17eSAl Viro 45483adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4558366025eSDave Hansen { 4563d733633SDave Hansen int ret = 0; 4573d733633SDave Hansen 458962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 45983adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 460d3ef3d73Snpiggin@suse.de /* 461d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 462d3ef3d73Snpiggin@suse.de * should be visible before we do. 463d3ef3d73Snpiggin@suse.de */ 464d3ef3d73Snpiggin@suse.de smp_mb(); 465d3ef3d73Snpiggin@suse.de 466d3ef3d73Snpiggin@suse.de /* 467d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 468d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 469d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 470d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 471d3ef3d73Snpiggin@suse.de * 472d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 473d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 474d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 475d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 476d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 477d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 478d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 479d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 480d3ef3d73Snpiggin@suse.de * we're counting up here. 481d3ef3d73Snpiggin@suse.de */ 482c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 483d3ef3d73Snpiggin@suse.de ret = -EBUSY; 484d3ef3d73Snpiggin@suse.de else 48583adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 486d3ef3d73Snpiggin@suse.de /* 487d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 488d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 489d3ef3d73Snpiggin@suse.de */ 490d3ef3d73Snpiggin@suse.de smp_wmb(); 49183adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 492962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 4933d733633SDave Hansen return ret; 4943d733633SDave Hansen } 4958366025eSDave Hansen 49683adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 4972e4b7fcdSDave Hansen { 498962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 49983adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 500962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 5012e4b7fcdSDave Hansen } 5022e4b7fcdSDave Hansen 5034ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5044ed5e82fSMiklos Szeredi { 5054ed5e82fSMiklos Szeredi struct mount *mnt; 5064ed5e82fSMiklos Szeredi int err = 0; 5074ed5e82fSMiklos Szeredi 5088e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5098e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5108e8b8796SMiklos Szeredi return -EBUSY; 5118e8b8796SMiklos Szeredi 512962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 5134ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5144ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5154ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5164ed5e82fSMiklos Szeredi smp_mb(); 5174ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5184ed5e82fSMiklos Szeredi err = -EBUSY; 5194ed5e82fSMiklos Szeredi break; 5204ed5e82fSMiklos Szeredi } 5214ed5e82fSMiklos Szeredi } 5224ed5e82fSMiklos Szeredi } 5238e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5248e8b8796SMiklos Szeredi err = -EBUSY; 5258e8b8796SMiklos Szeredi 5264ed5e82fSMiklos Szeredi if (!err) { 5274ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5284ed5e82fSMiklos Szeredi smp_wmb(); 5294ed5e82fSMiklos Szeredi } 5304ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5314ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5324ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5334ed5e82fSMiklos Szeredi } 534962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 5354ed5e82fSMiklos Szeredi 5364ed5e82fSMiklos Szeredi return err; 5374ed5e82fSMiklos Szeredi } 5384ed5e82fSMiklos Szeredi 539b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5401da177e4SLinus Torvalds { 54152ba1621SAl Viro kfree(mnt->mnt_devname); 54273cd49ecSMiklos Szeredi mnt_free_id(mnt); 543d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 54468e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 545d3ef3d73Snpiggin@suse.de #endif 546b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5471da177e4SLinus Torvalds } 5481da177e4SLinus Torvalds 5491da177e4SLinus Torvalds /* 550a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 551a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 55299b7db7bSNick Piggin * vfsmount_lock must be held for read or write. 5531da177e4SLinus Torvalds */ 554c7105365SAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 555a05964f3SRam Pai int dir) 5561da177e4SLinus Torvalds { 5571da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 5581da177e4SLinus Torvalds struct list_head *tmp = head; 559c7105365SAl Viro struct mount *p, *found = NULL; 5601da177e4SLinus Torvalds 5611da177e4SLinus Torvalds for (;;) { 562a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 5631da177e4SLinus Torvalds p = NULL; 5641da177e4SLinus Torvalds if (tmp == head) 5651da177e4SLinus Torvalds break; 5661b8e5564SAl Viro p = list_entry(tmp, struct mount, mnt_hash); 567a73324daSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) { 568a05964f3SRam Pai found = p; 5691da177e4SLinus Torvalds break; 5701da177e4SLinus Torvalds } 5711da177e4SLinus Torvalds } 5721da177e4SLinus Torvalds return found; 5731da177e4SLinus Torvalds } 5741da177e4SLinus Torvalds 575a05964f3SRam Pai /* 576f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 577f015f126SDavid Howells * 578f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 579f015f126SDavid Howells * following mounts: 580f015f126SDavid Howells * 581f015f126SDavid Howells * mount /dev/sda1 /mnt 582f015f126SDavid Howells * mount /dev/sda2 /mnt 583f015f126SDavid Howells * mount /dev/sda3 /mnt 584f015f126SDavid Howells * 585f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 586f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 587f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 588f015f126SDavid Howells * 589f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 590a05964f3SRam Pai */ 5911c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 592a05964f3SRam Pai { 593c7105365SAl Viro struct mount *child_mnt; 59499b7db7bSNick Piggin 595962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 596c7105365SAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry, 1); 597c7105365SAl Viro if (child_mnt) { 598c7105365SAl Viro mnt_add_count(child_mnt, 1); 599962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 600c7105365SAl Viro return &child_mnt->mnt; 601c7105365SAl Viro } else { 602962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 603c7105365SAl Viro return NULL; 604c7105365SAl Viro } 605a05964f3SRam Pai } 606a05964f3SRam Pai 607143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 6081da177e4SLinus Torvalds { 6096b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 6101da177e4SLinus Torvalds } 6111da177e4SLinus Torvalds 61299b7db7bSNick Piggin /* 61399b7db7bSNick Piggin * vfsmount lock must be held for write 61499b7db7bSNick Piggin */ 6156b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 6165addc5ddSAl Viro { 6175addc5ddSAl Viro if (ns) { 6185addc5ddSAl Viro ns->event = ++event; 6195addc5ddSAl Viro wake_up_interruptible(&ns->poll); 6205addc5ddSAl Viro } 6215addc5ddSAl Viro } 6225addc5ddSAl Viro 62399b7db7bSNick Piggin /* 62499b7db7bSNick Piggin * vfsmount lock must be held for write 62599b7db7bSNick Piggin */ 6266b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 6275addc5ddSAl Viro { 6285addc5ddSAl Viro if (ns && ns->event != event) { 6295addc5ddSAl Viro ns->event = event; 6305addc5ddSAl Viro wake_up_interruptible(&ns->poll); 6315addc5ddSAl Viro } 6325addc5ddSAl Viro } 6335addc5ddSAl Viro 63499b7db7bSNick Piggin /* 6355f57cbccSNick Piggin * Clear dentry's mounted state if it has no remaining mounts. 6365f57cbccSNick Piggin * vfsmount_lock must be held for write. 6375f57cbccSNick Piggin */ 638aa0a4cf0SAl Viro static void dentry_reset_mounted(struct dentry *dentry) 6395f57cbccSNick Piggin { 6405f57cbccSNick Piggin unsigned u; 6415f57cbccSNick Piggin 6425f57cbccSNick Piggin for (u = 0; u < HASH_SIZE; u++) { 643d5e50f74SAl Viro struct mount *p; 6445f57cbccSNick Piggin 6451b8e5564SAl Viro list_for_each_entry(p, &mount_hashtable[u], mnt_hash) { 646a73324daSAl Viro if (p->mnt_mountpoint == dentry) 6475f57cbccSNick Piggin return; 6485f57cbccSNick Piggin } 6495f57cbccSNick Piggin } 6505f57cbccSNick Piggin spin_lock(&dentry->d_lock); 6515f57cbccSNick Piggin dentry->d_flags &= ~DCACHE_MOUNTED; 6525f57cbccSNick Piggin spin_unlock(&dentry->d_lock); 6535f57cbccSNick Piggin } 6545f57cbccSNick Piggin 6555f57cbccSNick Piggin /* 65699b7db7bSNick Piggin * vfsmount lock must be held for write 65799b7db7bSNick Piggin */ 658419148daSAl Viro static void detach_mnt(struct mount *mnt, struct path *old_path) 6591da177e4SLinus Torvalds { 660a73324daSAl Viro old_path->dentry = mnt->mnt_mountpoint; 6610714a533SAl Viro old_path->mnt = &mnt->mnt_parent->mnt; 6620714a533SAl Viro mnt->mnt_parent = mnt; 663a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 6646b41d536SAl Viro list_del_init(&mnt->mnt_child); 6651b8e5564SAl Viro list_del_init(&mnt->mnt_hash); 666aa0a4cf0SAl Viro dentry_reset_mounted(old_path->dentry); 6671da177e4SLinus Torvalds } 6681da177e4SLinus Torvalds 66999b7db7bSNick Piggin /* 67099b7db7bSNick Piggin * vfsmount lock must be held for write 67199b7db7bSNick Piggin */ 67214cf1fa8SAl Viro void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry, 67344d964d6SAl Viro struct mount *child_mnt) 674b90fa9aeSRam Pai { 6753a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 676a73324daSAl Viro child_mnt->mnt_mountpoint = dget(dentry); 6773a2393d7SAl Viro child_mnt->mnt_parent = mnt; 6785f57cbccSNick Piggin spin_lock(&dentry->d_lock); 6795f57cbccSNick Piggin dentry->d_flags |= DCACHE_MOUNTED; 6805f57cbccSNick Piggin spin_unlock(&dentry->d_lock); 681b90fa9aeSRam Pai } 682b90fa9aeSRam Pai 68399b7db7bSNick Piggin /* 68499b7db7bSNick Piggin * vfsmount lock must be held for write 68599b7db7bSNick Piggin */ 686419148daSAl Viro static void attach_mnt(struct mount *mnt, struct path *path) 6871da177e4SLinus Torvalds { 68814cf1fa8SAl Viro mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt); 6891b8e5564SAl Viro list_add_tail(&mnt->mnt_hash, mount_hashtable + 6901a390689SAl Viro hash(path->mnt, path->dentry)); 6916b41d536SAl Viro list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts); 692b90fa9aeSRam Pai } 693b90fa9aeSRam Pai 694b90fa9aeSRam Pai /* 69599b7db7bSNick Piggin * vfsmount lock must be held for write 696b90fa9aeSRam Pai */ 6974b2619a5SAl Viro static void commit_tree(struct mount *mnt) 698b90fa9aeSRam Pai { 6990714a533SAl Viro struct mount *parent = mnt->mnt_parent; 70083adc753SAl Viro struct mount *m; 701b90fa9aeSRam Pai LIST_HEAD(head); 702143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 703b90fa9aeSRam Pai 7040714a533SAl Viro BUG_ON(parent == mnt); 705b90fa9aeSRam Pai 7061a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 707f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 708143c8c91SAl Viro m->mnt_ns = n; 709f03c6599SAl Viro 710b90fa9aeSRam Pai list_splice(&head, n->list.prev); 711b90fa9aeSRam Pai 7121b8e5564SAl Viro list_add_tail(&mnt->mnt_hash, mount_hashtable + 713a73324daSAl Viro hash(&parent->mnt, mnt->mnt_mountpoint)); 7146b41d536SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 7156b3286edSKirill Korotaev touch_mnt_namespace(n); 7161da177e4SLinus Torvalds } 7171da177e4SLinus Torvalds 718909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 7191da177e4SLinus Torvalds { 7206b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 7216b41d536SAl Viro if (next == &p->mnt_mounts) { 7221da177e4SLinus Torvalds while (1) { 723909b0a88SAl Viro if (p == root) 7241da177e4SLinus Torvalds return NULL; 7256b41d536SAl Viro next = p->mnt_child.next; 7266b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 7271da177e4SLinus Torvalds break; 7280714a533SAl Viro p = p->mnt_parent; 7291da177e4SLinus Torvalds } 7301da177e4SLinus Torvalds } 7316b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 7321da177e4SLinus Torvalds } 7331da177e4SLinus Torvalds 734315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 7359676f0c6SRam Pai { 7366b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 7376b41d536SAl Viro while (prev != &p->mnt_mounts) { 7386b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 7396b41d536SAl Viro prev = p->mnt_mounts.prev; 7409676f0c6SRam Pai } 7419676f0c6SRam Pai return p; 7429676f0c6SRam Pai } 7439676f0c6SRam Pai 7449d412a43SAl Viro struct vfsmount * 7459d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 7469d412a43SAl Viro { 747b105e270SAl Viro struct mount *mnt; 7489d412a43SAl Viro struct dentry *root; 7499d412a43SAl Viro 7509d412a43SAl Viro if (!type) 7519d412a43SAl Viro return ERR_PTR(-ENODEV); 7529d412a43SAl Viro 7539d412a43SAl Viro mnt = alloc_vfsmnt(name); 7549d412a43SAl Viro if (!mnt) 7559d412a43SAl Viro return ERR_PTR(-ENOMEM); 7569d412a43SAl Viro 7579d412a43SAl Viro if (flags & MS_KERNMOUNT) 758b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 7599d412a43SAl Viro 7609d412a43SAl Viro root = mount_fs(type, flags, name, data); 7619d412a43SAl Viro if (IS_ERR(root)) { 7629d412a43SAl Viro free_vfsmnt(mnt); 7639d412a43SAl Viro return ERR_CAST(root); 7649d412a43SAl Viro } 7659d412a43SAl Viro 766b105e270SAl Viro mnt->mnt.mnt_root = root; 767b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 768a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 7690714a533SAl Viro mnt->mnt_parent = mnt; 770962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 77139f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 772962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 773b105e270SAl Viro return &mnt->mnt; 7749d412a43SAl Viro } 7759d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 7769d412a43SAl Viro 77787129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 77836341f64SRam Pai int flag) 7791da177e4SLinus Torvalds { 78087129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 781be34d1a3SDavid Howells struct mount *mnt; 782be34d1a3SDavid Howells int err; 7831da177e4SLinus Torvalds 784be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 785be34d1a3SDavid Howells if (!mnt) 786be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 787be34d1a3SDavid Howells 788719f5d7fSMiklos Szeredi if (flag & (CL_SLAVE | CL_PRIVATE)) 78915169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 790719f5d7fSMiklos Szeredi else 79115169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 792719f5d7fSMiklos Szeredi 79315169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 794be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 795719f5d7fSMiklos Szeredi if (err) 796719f5d7fSMiklos Szeredi goto out_free; 797719f5d7fSMiklos Szeredi } 798719f5d7fSMiklos Szeredi 79987129cc0SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; 8001da177e4SLinus Torvalds atomic_inc(&sb->s_active); 801b105e270SAl Viro mnt->mnt.mnt_sb = sb; 802b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 803a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8040714a533SAl Viro mnt->mnt_parent = mnt; 805962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 80639f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 807962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 808b90fa9aeSRam Pai 8095afe0022SRam Pai if (flag & CL_SLAVE) { 8106776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 81132301920SAl Viro mnt->mnt_master = old; 812fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 8138aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 814fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 8156776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 816d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 8176776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 818d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 8195afe0022SRam Pai } 820b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 8210f0afb1dSAl Viro set_mnt_shared(mnt); 8221da177e4SLinus Torvalds 8231da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 8241da177e4SLinus Torvalds * as the original if that was on one */ 82536341f64SRam Pai if (flag & CL_EXPIRE) { 8266776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 8276776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 8281da177e4SLinus Torvalds } 829be34d1a3SDavid Howells 830cb338d06SAl Viro return mnt; 831719f5d7fSMiklos Szeredi 832719f5d7fSMiklos Szeredi out_free: 833719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 834be34d1a3SDavid Howells return ERR_PTR(err); 8351da177e4SLinus Torvalds } 8361da177e4SLinus Torvalds 83783adc753SAl Viro static inline void mntfree(struct mount *mnt) 8381da177e4SLinus Torvalds { 83983adc753SAl Viro struct vfsmount *m = &mnt->mnt; 84083adc753SAl Viro struct super_block *sb = m->mnt_sb; 841b3e19d92SNick Piggin 8423d733633SDave Hansen /* 8433d733633SDave Hansen * This probably indicates that somebody messed 8443d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 8453d733633SDave Hansen * happens, the filesystem was probably unable 8463d733633SDave Hansen * to make r/w->r/o transitions. 8473d733633SDave Hansen */ 848d3ef3d73Snpiggin@suse.de /* 849b3e19d92SNick Piggin * The locking used to deal with mnt_count decrement provides barriers, 850b3e19d92SNick Piggin * so mnt_get_writers() below is safe. 851d3ef3d73Snpiggin@suse.de */ 852c6653a83SNick Piggin WARN_ON(mnt_get_writers(mnt)); 85383adc753SAl Viro fsnotify_vfsmount_delete(m); 85483adc753SAl Viro dput(m->mnt_root); 85583adc753SAl Viro free_vfsmnt(mnt); 8561da177e4SLinus Torvalds deactivate_super(sb); 8571da177e4SLinus Torvalds } 8581da177e4SLinus Torvalds 859900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 8607b7b1aceSAl Viro { 861b3e19d92SNick Piggin put_again: 862f03c6599SAl Viro #ifdef CONFIG_SMP 863962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 864f7a99c5bSAl Viro if (likely(mnt->mnt_ns)) { 865f7a99c5bSAl Viro /* shouldn't be the last one */ 866aa9c0e07SAl Viro mnt_add_count(mnt, -1); 867962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 86899b7db7bSNick Piggin return; 869b3e19d92SNick Piggin } 870962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 871b3e19d92SNick Piggin 872962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 873aa9c0e07SAl Viro mnt_add_count(mnt, -1); 874b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 875962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 87699b7db7bSNick Piggin return; 87799b7db7bSNick Piggin } 878b3e19d92SNick Piggin #else 879aa9c0e07SAl Viro mnt_add_count(mnt, -1); 880b3e19d92SNick Piggin if (likely(mnt_get_count(mnt))) 881b3e19d92SNick Piggin return; 882962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 883f03c6599SAl Viro #endif 884863d684fSAl Viro if (unlikely(mnt->mnt_pinned)) { 885863d684fSAl Viro mnt_add_count(mnt, mnt->mnt_pinned + 1); 886863d684fSAl Viro mnt->mnt_pinned = 0; 887962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 888900148dcSAl Viro acct_auto_close_mnt(&mnt->mnt); 889b3e19d92SNick Piggin goto put_again; 890b3e19d92SNick Piggin } 891962830dfSAndi Kleen 89239f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 893962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 894b3e19d92SNick Piggin mntfree(mnt); 895b3e19d92SNick Piggin } 896b3e19d92SNick Piggin 897b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 898b3e19d92SNick Piggin { 899b3e19d92SNick Piggin if (mnt) { 900863d684fSAl Viro struct mount *m = real_mount(mnt); 901b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 902863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 903863d684fSAl Viro m->mnt_expiry_mark = 0; 904863d684fSAl Viro mntput_no_expire(m); 905b3e19d92SNick Piggin } 906b3e19d92SNick Piggin } 907b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 908b3e19d92SNick Piggin 909b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 910b3e19d92SNick Piggin { 911b3e19d92SNick Piggin if (mnt) 91283adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 913b3e19d92SNick Piggin return mnt; 914b3e19d92SNick Piggin } 915b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 916b3e19d92SNick Piggin 9177b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 9187b7b1aceSAl Viro { 919962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 920863d684fSAl Viro real_mount(mnt)->mnt_pinned++; 921962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 9227b7b1aceSAl Viro } 9237b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 9247b7b1aceSAl Viro 925863d684fSAl Viro void mnt_unpin(struct vfsmount *m) 9267b7b1aceSAl Viro { 927863d684fSAl Viro struct mount *mnt = real_mount(m); 928962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 9297b7b1aceSAl Viro if (mnt->mnt_pinned) { 930863d684fSAl Viro mnt_add_count(mnt, 1); 9317b7b1aceSAl Viro mnt->mnt_pinned--; 9327b7b1aceSAl Viro } 933962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 9347b7b1aceSAl Viro } 9357b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 9361da177e4SLinus Torvalds 937b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 938b3b304a2SMiklos Szeredi { 939b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 940b3b304a2SMiklos Szeredi } 941b3b304a2SMiklos Szeredi 942b3b304a2SMiklos Szeredi /* 943b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 944b3b304a2SMiklos Szeredi * implement more complex mount option showing. 945b3b304a2SMiklos Szeredi * 946b3b304a2SMiklos Szeredi * See also save_mount_options(). 947b3b304a2SMiklos Szeredi */ 94834c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 949b3b304a2SMiklos Szeredi { 9502a32cebdSAl Viro const char *options; 9512a32cebdSAl Viro 9522a32cebdSAl Viro rcu_read_lock(); 95334c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 954b3b304a2SMiklos Szeredi 955b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 956b3b304a2SMiklos Szeredi seq_putc(m, ','); 957b3b304a2SMiklos Szeredi mangle(m, options); 958b3b304a2SMiklos Szeredi } 9592a32cebdSAl Viro rcu_read_unlock(); 960b3b304a2SMiklos Szeredi 961b3b304a2SMiklos Szeredi return 0; 962b3b304a2SMiklos Szeredi } 963b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 964b3b304a2SMiklos Szeredi 965b3b304a2SMiklos Szeredi /* 966b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 967b3b304a2SMiklos Szeredi * called from the fill_super() callback. 968b3b304a2SMiklos Szeredi * 969b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 970b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 971b3b304a2SMiklos Szeredi * remount fails. 972b3b304a2SMiklos Szeredi * 973b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 974b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 975b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 976b3b304a2SMiklos Szeredi * any more. 977b3b304a2SMiklos Szeredi */ 978b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 979b3b304a2SMiklos Szeredi { 9802a32cebdSAl Viro BUG_ON(sb->s_options); 9812a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 982b3b304a2SMiklos Szeredi } 983b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 984b3b304a2SMiklos Szeredi 9852a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 9862a32cebdSAl Viro { 9872a32cebdSAl Viro char *old = sb->s_options; 9882a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 9892a32cebdSAl Viro if (old) { 9902a32cebdSAl Viro synchronize_rcu(); 9912a32cebdSAl Viro kfree(old); 9922a32cebdSAl Viro } 9932a32cebdSAl Viro } 9942a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 9952a32cebdSAl Viro 996a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 9970226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 9981da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 9991da177e4SLinus Torvalds { 10006ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 10011da177e4SLinus Torvalds 1002390c6843SRam Pai down_read(&namespace_sem); 1003a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 10041da177e4SLinus Torvalds } 10051da177e4SLinus Torvalds 10061da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 10071da177e4SLinus Torvalds { 10086ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 1009b0765fb8SPavel Emelianov 1010a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 10111da177e4SLinus Torvalds } 10121da177e4SLinus Torvalds 10131da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 10141da177e4SLinus Torvalds { 1015390c6843SRam Pai up_read(&namespace_sem); 10161da177e4SLinus Torvalds } 10171da177e4SLinus Torvalds 10180226f492SAl Viro static int m_show(struct seq_file *m, void *v) 10199f5596afSAl Viro { 10206ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 10211a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 10220226f492SAl Viro return p->show(m, &r->mnt); 10231da177e4SLinus Torvalds } 10241da177e4SLinus Torvalds 1025a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 10261da177e4SLinus Torvalds .start = m_start, 10271da177e4SLinus Torvalds .next = m_next, 10281da177e4SLinus Torvalds .stop = m_stop, 10290226f492SAl Viro .show = m_show, 1030b4629fe2SChuck Lever }; 1031a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1032b4629fe2SChuck Lever 10331da177e4SLinus Torvalds /** 10341da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 10351da177e4SLinus Torvalds * @mnt: root of mount tree 10361da177e4SLinus Torvalds * 10371da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 10381da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 10391da177e4SLinus Torvalds * busy. 10401da177e4SLinus Torvalds */ 1041909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 10421da177e4SLinus Torvalds { 1043909b0a88SAl Viro struct mount *mnt = real_mount(m); 104436341f64SRam Pai int actual_refs = 0; 104536341f64SRam Pai int minimum_refs = 0; 1046315fc83eSAl Viro struct mount *p; 1047909b0a88SAl Viro BUG_ON(!m); 10481da177e4SLinus Torvalds 1049b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1050962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1051909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 105283adc753SAl Viro actual_refs += mnt_get_count(p); 10531da177e4SLinus Torvalds minimum_refs += 2; 10541da177e4SLinus Torvalds } 1055962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 10561da177e4SLinus Torvalds 10571da177e4SLinus Torvalds if (actual_refs > minimum_refs) 10581da177e4SLinus Torvalds return 0; 1059e3474a8eSIan Kent 1060e3474a8eSIan Kent return 1; 10611da177e4SLinus Torvalds } 10621da177e4SLinus Torvalds 10631da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 10641da177e4SLinus Torvalds 10651da177e4SLinus Torvalds /** 10661da177e4SLinus Torvalds * may_umount - check if a mount point is busy 10671da177e4SLinus Torvalds * @mnt: root of mount 10681da177e4SLinus Torvalds * 10691da177e4SLinus Torvalds * This is called to check if a mount point has any 10701da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 10711da177e4SLinus Torvalds * mount has sub mounts this will return busy 10721da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 10731da177e4SLinus Torvalds * 10741da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 10751da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 10761da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 10771da177e4SLinus Torvalds */ 10781da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 10791da177e4SLinus Torvalds { 1080e3474a8eSIan Kent int ret = 1; 10818ad08d8aSAl Viro down_read(&namespace_sem); 1082962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 10831ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1084e3474a8eSIan Kent ret = 0; 1085962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 10868ad08d8aSAl Viro up_read(&namespace_sem); 1087a05964f3SRam Pai return ret; 10881da177e4SLinus Torvalds } 10891da177e4SLinus Torvalds 10901da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 10911da177e4SLinus Torvalds 1092b90fa9aeSRam Pai void release_mounts(struct list_head *head) 10931da177e4SLinus Torvalds { 1094d5e50f74SAl Viro struct mount *mnt; 109570fbcdf4SRam Pai while (!list_empty(head)) { 10961b8e5564SAl Viro mnt = list_first_entry(head, struct mount, mnt_hash); 10971b8e5564SAl Viro list_del_init(&mnt->mnt_hash); 1098676da58dSAl Viro if (mnt_has_parent(mnt)) { 109970fbcdf4SRam Pai struct dentry *dentry; 1100863d684fSAl Viro struct mount *m; 110199b7db7bSNick Piggin 1102962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1103a73324daSAl Viro dentry = mnt->mnt_mountpoint; 1104863d684fSAl Viro m = mnt->mnt_parent; 1105a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 11060714a533SAl Viro mnt->mnt_parent = mnt; 11077c4b93d8SAl Viro m->mnt_ghosts--; 1108962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 110970fbcdf4SRam Pai dput(dentry); 1110863d684fSAl Viro mntput(&m->mnt); 11111da177e4SLinus Torvalds } 1112d5e50f74SAl Viro mntput(&mnt->mnt); 111370fbcdf4SRam Pai } 111470fbcdf4SRam Pai } 111570fbcdf4SRam Pai 111699b7db7bSNick Piggin /* 111799b7db7bSNick Piggin * vfsmount lock must be held for write 111899b7db7bSNick Piggin * namespace_sem must be held for write 111999b7db7bSNick Piggin */ 1120761d5c38SAl Viro void umount_tree(struct mount *mnt, int propagate, struct list_head *kill) 112170fbcdf4SRam Pai { 11227b8a53fdSAl Viro LIST_HEAD(tmp_list); 1123315fc83eSAl Viro struct mount *p; 112470fbcdf4SRam Pai 1125909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) 11261b8e5564SAl Viro list_move(&p->mnt_hash, &tmp_list); 112770fbcdf4SRam Pai 1128a05964f3SRam Pai if (propagate) 11297b8a53fdSAl Viro propagate_umount(&tmp_list); 1130a05964f3SRam Pai 11311b8e5564SAl Viro list_for_each_entry(p, &tmp_list, mnt_hash) { 11326776db3dSAl Viro list_del_init(&p->mnt_expire); 11331a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1134143c8c91SAl Viro __touch_mnt_namespace(p->mnt_ns); 113563d37a84SAl Viro p->mnt_ns = NULL; 11366b41d536SAl Viro list_del_init(&p->mnt_child); 1137676da58dSAl Viro if (mnt_has_parent(p)) { 1138863d684fSAl Viro p->mnt_parent->mnt_ghosts++; 1139a73324daSAl Viro dentry_reset_mounted(p->mnt_mountpoint); 11407c4b93d8SAl Viro } 11410f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 11421da177e4SLinus Torvalds } 11437b8a53fdSAl Viro list_splice(&tmp_list, kill); 11441da177e4SLinus Torvalds } 11451da177e4SLinus Torvalds 1146692afc31SAl Viro static void shrink_submounts(struct mount *mnt, struct list_head *umounts); 1147c35038beSAl Viro 11481ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 11491da177e4SLinus Torvalds { 11501ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 11511da177e4SLinus Torvalds int retval; 115270fbcdf4SRam Pai LIST_HEAD(umount_list); 11531da177e4SLinus Torvalds 11541ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 11551da177e4SLinus Torvalds if (retval) 11561da177e4SLinus Torvalds return retval; 11571da177e4SLinus Torvalds 11581da177e4SLinus Torvalds /* 11591da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 11601da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 11611da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 11621da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 11631da177e4SLinus Torvalds */ 11641da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 11651ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 11661da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 11671da177e4SLinus Torvalds return -EINVAL; 11681da177e4SLinus Torvalds 1169b3e19d92SNick Piggin /* 1170b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1171b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1172b3e19d92SNick Piggin */ 1173962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 117483adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1175962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 11761da177e4SLinus Torvalds return -EBUSY; 1177b3e19d92SNick Piggin } 1178962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 11791da177e4SLinus Torvalds 1180863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 11811da177e4SLinus Torvalds return -EAGAIN; 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds 11841da177e4SLinus Torvalds /* 11851da177e4SLinus Torvalds * If we may have to abort operations to get out of this 11861da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 11871da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 11881da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 11891da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 11901da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 11911da177e4SLinus Torvalds * about for the moment. 11921da177e4SLinus Torvalds */ 11931da177e4SLinus Torvalds 119442faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 119542faad99SAl Viro sb->s_op->umount_begin(sb); 119642faad99SAl Viro } 11971da177e4SLinus Torvalds 11981da177e4SLinus Torvalds /* 11991da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 12001da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 12011da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 12021da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 12031da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 12041da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 12051da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 12061da177e4SLinus Torvalds */ 12071ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 12081da177e4SLinus Torvalds /* 12091da177e4SLinus Torvalds * Special case for "unmounting" root ... 12101da177e4SLinus Torvalds * we just try to remount it readonly. 12111da177e4SLinus Torvalds */ 12121da177e4SLinus Torvalds down_write(&sb->s_umount); 12134aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 12141da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 12151da177e4SLinus Torvalds up_write(&sb->s_umount); 12161da177e4SLinus Torvalds return retval; 12171da177e4SLinus Torvalds } 12181da177e4SLinus Torvalds 1219390c6843SRam Pai down_write(&namespace_sem); 1220962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 12215addc5ddSAl Viro event++; 12221da177e4SLinus Torvalds 1223c35038beSAl Viro if (!(flags & MNT_DETACH)) 12241ab59738SAl Viro shrink_submounts(mnt, &umount_list); 1225c35038beSAl Viro 12261da177e4SLinus Torvalds retval = -EBUSY; 1227a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 12281a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 12291ab59738SAl Viro umount_tree(mnt, 1, &umount_list); 12301da177e4SLinus Torvalds retval = 0; 12311da177e4SLinus Torvalds } 1232962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 1233390c6843SRam Pai up_write(&namespace_sem); 123470fbcdf4SRam Pai release_mounts(&umount_list); 12351da177e4SLinus Torvalds return retval; 12361da177e4SLinus Torvalds } 12371da177e4SLinus Torvalds 12381da177e4SLinus Torvalds /* 12391da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 12401da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 12411da177e4SLinus Torvalds * 12421da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 12431da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 12441da177e4SLinus Torvalds */ 12451da177e4SLinus Torvalds 1246bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 12471da177e4SLinus Torvalds { 12482d8f3038SAl Viro struct path path; 1249900148dcSAl Viro struct mount *mnt; 12501da177e4SLinus Torvalds int retval; 1251db1f05bbSMiklos Szeredi int lookup_flags = 0; 12521da177e4SLinus Torvalds 1253db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1254db1f05bbSMiklos Szeredi return -EINVAL; 1255db1f05bbSMiklos Szeredi 1256db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1257db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1258db1f05bbSMiklos Szeredi 1259db1f05bbSMiklos Szeredi retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); 12601da177e4SLinus Torvalds if (retval) 12611da177e4SLinus Torvalds goto out; 1262900148dcSAl Viro mnt = real_mount(path.mnt); 12631da177e4SLinus Torvalds retval = -EINVAL; 12642d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 12651da177e4SLinus Torvalds goto dput_and_out; 1266143c8c91SAl Viro if (!check_mnt(mnt)) 12671da177e4SLinus Torvalds goto dput_and_out; 12681da177e4SLinus Torvalds 12691da177e4SLinus Torvalds retval = -EPERM; 12701da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 12711da177e4SLinus Torvalds goto dput_and_out; 12721da177e4SLinus Torvalds 1273900148dcSAl Viro retval = do_umount(mnt, flags); 12741da177e4SLinus Torvalds dput_and_out: 1275429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 12762d8f3038SAl Viro dput(path.dentry); 1277900148dcSAl Viro mntput_no_expire(mnt); 12781da177e4SLinus Torvalds out: 12791da177e4SLinus Torvalds return retval; 12801da177e4SLinus Torvalds } 12811da177e4SLinus Torvalds 12821da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds /* 12851da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 12861da177e4SLinus Torvalds */ 1287bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 12881da177e4SLinus Torvalds { 12891da177e4SLinus Torvalds return sys_umount(name, 0); 12901da177e4SLinus Torvalds } 12911da177e4SLinus Torvalds 12921da177e4SLinus Torvalds #endif 12931da177e4SLinus Torvalds 12942d92ab3cSAl Viro static int mount_is_safe(struct path *path) 12951da177e4SLinus Torvalds { 12961da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 12971da177e4SLinus Torvalds return 0; 12981da177e4SLinus Torvalds return -EPERM; 12991da177e4SLinus Torvalds #ifdef notyet 13002d92ab3cSAl Viro if (S_ISLNK(path->dentry->d_inode->i_mode)) 13011da177e4SLinus Torvalds return -EPERM; 13022d92ab3cSAl Viro if (path->dentry->d_inode->i_mode & S_ISVTX) { 1303da9592edSDavid Howells if (current_uid() != path->dentry->d_inode->i_uid) 13041da177e4SLinus Torvalds return -EPERM; 13051da177e4SLinus Torvalds } 13062d92ab3cSAl Viro if (inode_permission(path->dentry->d_inode, MAY_WRITE)) 13071da177e4SLinus Torvalds return -EPERM; 13081da177e4SLinus Torvalds return 0; 13091da177e4SLinus Torvalds #endif 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds 13128823c079SEric W. Biederman static bool mnt_ns_loop(struct path *path) 13138823c079SEric W. Biederman { 13148823c079SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 13158823c079SEric W. Biederman * mount namespace loop? 13168823c079SEric W. Biederman */ 13178823c079SEric W. Biederman struct inode *inode = path->dentry->d_inode; 13188823c079SEric W. Biederman struct proc_inode *ei; 13198823c079SEric W. Biederman struct mnt_namespace *mnt_ns; 13208823c079SEric W. Biederman 13218823c079SEric W. Biederman if (!proc_ns_inode(inode)) 13228823c079SEric W. Biederman return false; 13238823c079SEric W. Biederman 13248823c079SEric W. Biederman ei = PROC_I(inode); 13258823c079SEric W. Biederman if (ei->ns_ops != &mntns_operations) 13268823c079SEric W. Biederman return false; 13278823c079SEric W. Biederman 13288823c079SEric W. Biederman mnt_ns = ei->ns; 13298823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 13308823c079SEric W. Biederman } 13318823c079SEric W. Biederman 133287129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 133336341f64SRam Pai int flag) 13341da177e4SLinus Torvalds { 1335a73324daSAl Viro struct mount *res, *p, *q, *r; 13361a390689SAl Viro struct path path; 13371da177e4SLinus Torvalds 1338fc7be130SAl Viro if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 1339be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 13409676f0c6SRam Pai 134136341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1342be34d1a3SDavid Howells if (IS_ERR(q)) 1343be34d1a3SDavid Howells return q; 1344be34d1a3SDavid Howells 1345a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 13461da177e4SLinus Torvalds 13471da177e4SLinus Torvalds p = mnt; 13486b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1349315fc83eSAl Viro struct mount *s; 13507ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 13511da177e4SLinus Torvalds continue; 13521da177e4SLinus Torvalds 1353909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 1354fc7be130SAl Viro if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 13559676f0c6SRam Pai s = skip_mnt_tree(s); 13569676f0c6SRam Pai continue; 13579676f0c6SRam Pai } 13580714a533SAl Viro while (p != s->mnt_parent) { 13590714a533SAl Viro p = p->mnt_parent; 13600714a533SAl Viro q = q->mnt_parent; 13611da177e4SLinus Torvalds } 136287129cc0SAl Viro p = s; 1363cb338d06SAl Viro path.mnt = &q->mnt; 1364a73324daSAl Viro path.dentry = p->mnt_mountpoint; 136587129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1366be34d1a3SDavid Howells if (IS_ERR(q)) 1367be34d1a3SDavid Howells goto out; 1368962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 13691a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 1370cb338d06SAl Viro attach_mnt(q, &path); 1371962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 13721da177e4SLinus Torvalds } 13731da177e4SLinus Torvalds } 13741da177e4SLinus Torvalds return res; 1375be34d1a3SDavid Howells out: 13761da177e4SLinus Torvalds if (res) { 137770fbcdf4SRam Pai LIST_HEAD(umount_list); 1378962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1379761d5c38SAl Viro umount_tree(res, 0, &umount_list); 1380962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 138170fbcdf4SRam Pai release_mounts(&umount_list); 13821da177e4SLinus Torvalds } 1383be34d1a3SDavid Howells return q; 13841da177e4SLinus Torvalds } 13851da177e4SLinus Torvalds 1386be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1387be34d1a3SDavid Howells 1388589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 13898aec0809SAl Viro { 1390cb338d06SAl Viro struct mount *tree; 13911a60a280SAl Viro down_write(&namespace_sem); 139287129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 139387129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 13941a60a280SAl Viro up_write(&namespace_sem); 1395be34d1a3SDavid Howells if (IS_ERR(tree)) 1396be34d1a3SDavid Howells return NULL; 1397be34d1a3SDavid Howells return &tree->mnt; 13988aec0809SAl Viro } 13998aec0809SAl Viro 14008aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 14018aec0809SAl Viro { 14028aec0809SAl Viro LIST_HEAD(umount_list); 14031a60a280SAl Viro down_write(&namespace_sem); 1404962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1405761d5c38SAl Viro umount_tree(real_mount(mnt), 0, &umount_list); 1406962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 14071a60a280SAl Viro up_write(&namespace_sem); 14088aec0809SAl Viro release_mounts(&umount_list); 14098aec0809SAl Viro } 14108aec0809SAl Viro 14111f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 14121f707137SAl Viro struct vfsmount *root) 14131f707137SAl Viro { 14141a4eeaf2SAl Viro struct mount *mnt; 14151f707137SAl Viro int res = f(root, arg); 14161f707137SAl Viro if (res) 14171f707137SAl Viro return res; 14181a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 14191a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 14201f707137SAl Viro if (res) 14211f707137SAl Viro return res; 14221f707137SAl Viro } 14231f707137SAl Viro return 0; 14241f707137SAl Viro } 14251f707137SAl Viro 14264b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1427719f5d7fSMiklos Szeredi { 1428315fc83eSAl Viro struct mount *p; 1429719f5d7fSMiklos Szeredi 1430909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1431fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 14324b8b21f4SAl Viro mnt_release_group_id(p); 1433719f5d7fSMiklos Szeredi } 1434719f5d7fSMiklos Szeredi } 1435719f5d7fSMiklos Szeredi 14364b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1437719f5d7fSMiklos Szeredi { 1438315fc83eSAl Viro struct mount *p; 1439719f5d7fSMiklos Szeredi 1440909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1441fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 14424b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1443719f5d7fSMiklos Szeredi if (err) { 14444b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1445719f5d7fSMiklos Szeredi return err; 1446719f5d7fSMiklos Szeredi } 1447719f5d7fSMiklos Szeredi } 1448719f5d7fSMiklos Szeredi } 1449719f5d7fSMiklos Szeredi 1450719f5d7fSMiklos Szeredi return 0; 1451719f5d7fSMiklos Szeredi } 1452719f5d7fSMiklos Szeredi 1453b90fa9aeSRam Pai /* 1454b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1455b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 145621444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 145721444403SRam Pai * store the parent mount and mountpoint dentry. 145821444403SRam Pai * (done when source_mnt is moved) 1459b90fa9aeSRam Pai * 1460b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1461b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 14629676f0c6SRam Pai * --------------------------------------------------------------------------- 1463b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 14649676f0c6SRam Pai * |************************************************************************** 14659676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 14669676f0c6SRam Pai * | dest | | | | | 14679676f0c6SRam Pai * | | | | | | | 14689676f0c6SRam Pai * | v | | | | | 14699676f0c6SRam Pai * |************************************************************************** 14709676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 14715afe0022SRam Pai * | | | | | | 14729676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 14739676f0c6SRam Pai * *************************************************************************** 1474b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1475b90fa9aeSRam Pai * destination mount. 1476b90fa9aeSRam Pai * 1477b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1478b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1479b90fa9aeSRam Pai * the peer group of the source mount. 1480b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1481b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1482b90fa9aeSRam Pai * mount. 14835afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 14845afe0022SRam Pai * of the destination mount and the cloned mount is made slave 14855afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 14865afe0022SRam Pai * is marked as 'shared and slave'. 14875afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 14885afe0022SRam Pai * source mount. 14895afe0022SRam Pai * 14909676f0c6SRam Pai * --------------------------------------------------------------------------- 149121444403SRam Pai * | MOVE MOUNT OPERATION | 14929676f0c6SRam Pai * |************************************************************************** 14939676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 14949676f0c6SRam Pai * | dest | | | | | 14959676f0c6SRam Pai * | | | | | | | 14969676f0c6SRam Pai * | v | | | | | 14979676f0c6SRam Pai * |************************************************************************** 14989676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 14995afe0022SRam Pai * | | | | | | 15009676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 15019676f0c6SRam Pai * *************************************************************************** 15025afe0022SRam Pai * 15035afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 15045afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 150521444403SRam Pai * (+*) the mount is moved to the destination. 15065afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 15075afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 15085afe0022SRam Pai * the mount is marked as 'shared and slave'. 15095afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1510b90fa9aeSRam Pai * 1511b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1512b90fa9aeSRam Pai * applied to each mount in the tree. 1513b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1514b90fa9aeSRam Pai * in allocations. 1515b90fa9aeSRam Pai */ 15160fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 15171a390689SAl Viro struct path *path, struct path *parent_path) 1518b90fa9aeSRam Pai { 1519b90fa9aeSRam Pai LIST_HEAD(tree_list); 1520a8d56d8eSAl Viro struct mount *dest_mnt = real_mount(path->mnt); 15211a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1522315fc83eSAl Viro struct mount *child, *p; 1523719f5d7fSMiklos Szeredi int err; 1524b90fa9aeSRam Pai 1525fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 15260fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1527719f5d7fSMiklos Szeredi if (err) 1528719f5d7fSMiklos Szeredi goto out; 1529719f5d7fSMiklos Szeredi } 1530a8d56d8eSAl Viro err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1531719f5d7fSMiklos Szeredi if (err) 1532719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1533b90fa9aeSRam Pai 1534962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1535df1a1ad2SAl Viro 1536fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 1537909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 15380f0afb1dSAl Viro set_mnt_shared(p); 1539b90fa9aeSRam Pai } 15401a390689SAl Viro if (parent_path) { 15410fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 15420fb54e50SAl Viro attach_mnt(source_mnt, path); 1543143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 154421444403SRam Pai } else { 154514cf1fa8SAl Viro mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 15460fb54e50SAl Viro commit_tree(source_mnt); 154721444403SRam Pai } 1548b90fa9aeSRam Pai 15491b8e5564SAl Viro list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 15501b8e5564SAl Viro list_del_init(&child->mnt_hash); 15514b2619a5SAl Viro commit_tree(child); 1552b90fa9aeSRam Pai } 1553962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 155499b7db7bSNick Piggin 1555b90fa9aeSRam Pai return 0; 1556719f5d7fSMiklos Szeredi 1557719f5d7fSMiklos Szeredi out_cleanup_ids: 1558fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) 15590fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1560719f5d7fSMiklos Szeredi out: 1561719f5d7fSMiklos Szeredi return err; 1562b90fa9aeSRam Pai } 1563b90fa9aeSRam Pai 1564b12cea91SAl Viro static int lock_mount(struct path *path) 1565b12cea91SAl Viro { 1566b12cea91SAl Viro struct vfsmount *mnt; 1567b12cea91SAl Viro retry: 1568b12cea91SAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 1569b12cea91SAl Viro if (unlikely(cant_mount(path->dentry))) { 1570b12cea91SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 1571b12cea91SAl Viro return -ENOENT; 1572b12cea91SAl Viro } 1573b12cea91SAl Viro down_write(&namespace_sem); 1574b12cea91SAl Viro mnt = lookup_mnt(path); 1575b12cea91SAl Viro if (likely(!mnt)) 1576b12cea91SAl Viro return 0; 1577b12cea91SAl Viro up_write(&namespace_sem); 1578b12cea91SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 1579b12cea91SAl Viro path_put(path); 1580b12cea91SAl Viro path->mnt = mnt; 1581b12cea91SAl Viro path->dentry = dget(mnt->mnt_root); 1582b12cea91SAl Viro goto retry; 1583b12cea91SAl Viro } 1584b12cea91SAl Viro 1585b12cea91SAl Viro static void unlock_mount(struct path *path) 1586b12cea91SAl Viro { 1587b12cea91SAl Viro up_write(&namespace_sem); 1588b12cea91SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 1589b12cea91SAl Viro } 1590b12cea91SAl Viro 159195bc5f25SAl Viro static int graft_tree(struct mount *mnt, struct path *path) 15921da177e4SLinus Torvalds { 159395bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 15941da177e4SLinus Torvalds return -EINVAL; 15951da177e4SLinus Torvalds 15968c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 159795bc5f25SAl Viro S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) 15981da177e4SLinus Torvalds return -ENOTDIR; 15991da177e4SLinus Torvalds 1600b12cea91SAl Viro if (d_unlinked(path->dentry)) 1601b12cea91SAl Viro return -ENOENT; 16021da177e4SLinus Torvalds 160395bc5f25SAl Viro return attach_recursive_mnt(mnt, path, NULL); 16041da177e4SLinus Torvalds } 16051da177e4SLinus Torvalds 16061da177e4SLinus Torvalds /* 16077a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 16087a2e8a8fSValerie Aurora */ 16097a2e8a8fSValerie Aurora 16107a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 16117a2e8a8fSValerie Aurora { 16127c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 16137a2e8a8fSValerie Aurora 16147a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 16157a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 16167a2e8a8fSValerie Aurora return 0; 16177a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 16187a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 16197a2e8a8fSValerie Aurora return 0; 16207a2e8a8fSValerie Aurora return type; 16217a2e8a8fSValerie Aurora } 16227a2e8a8fSValerie Aurora 16237a2e8a8fSValerie Aurora /* 162407b20889SRam Pai * recursively change the type of the mountpoint. 162507b20889SRam Pai */ 16260a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 162707b20889SRam Pai { 1628315fc83eSAl Viro struct mount *m; 16294b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 163007b20889SRam Pai int recurse = flag & MS_REC; 16317a2e8a8fSValerie Aurora int type; 1632719f5d7fSMiklos Szeredi int err = 0; 163307b20889SRam Pai 1634ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1635ee6f9582SMiklos Szeredi return -EPERM; 1636ee6f9582SMiklos Szeredi 16372d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 163807b20889SRam Pai return -EINVAL; 163907b20889SRam Pai 16407a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 16417a2e8a8fSValerie Aurora if (!type) 16427a2e8a8fSValerie Aurora return -EINVAL; 16437a2e8a8fSValerie Aurora 164407b20889SRam Pai down_write(&namespace_sem); 1645719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1646719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1647719f5d7fSMiklos Szeredi if (err) 1648719f5d7fSMiklos Szeredi goto out_unlock; 1649719f5d7fSMiklos Szeredi } 1650719f5d7fSMiklos Szeredi 1651962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1652909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 16530f0afb1dSAl Viro change_mnt_propagation(m, type); 1654962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 1655719f5d7fSMiklos Szeredi 1656719f5d7fSMiklos Szeredi out_unlock: 165707b20889SRam Pai up_write(&namespace_sem); 1658719f5d7fSMiklos Szeredi return err; 165907b20889SRam Pai } 166007b20889SRam Pai 166107b20889SRam Pai /* 16621da177e4SLinus Torvalds * do loopback mount. 16631da177e4SLinus Torvalds */ 1664808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 16652dafe1c4SEric Sandeen int recurse) 16661da177e4SLinus Torvalds { 1667b12cea91SAl Viro LIST_HEAD(umount_list); 16682d92ab3cSAl Viro struct path old_path; 166987129cc0SAl Viro struct mount *mnt = NULL, *old; 16702d92ab3cSAl Viro int err = mount_is_safe(path); 16711da177e4SLinus Torvalds if (err) 16721da177e4SLinus Torvalds return err; 16731da177e4SLinus Torvalds if (!old_name || !*old_name) 16741da177e4SLinus Torvalds return -EINVAL; 1675815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 16761da177e4SLinus Torvalds if (err) 16771da177e4SLinus Torvalds return err; 16781da177e4SLinus Torvalds 16798823c079SEric W. Biederman err = -EINVAL; 16808823c079SEric W. Biederman if (mnt_ns_loop(&old_path)) 16818823c079SEric W. Biederman goto out; 16828823c079SEric W. Biederman 1683b12cea91SAl Viro err = lock_mount(path); 1684b12cea91SAl Viro if (err) 16859676f0c6SRam Pai goto out; 16869676f0c6SRam Pai 168787129cc0SAl Viro old = real_mount(old_path.mnt); 168887129cc0SAl Viro 1689b12cea91SAl Viro err = -EINVAL; 1690fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 1691b12cea91SAl Viro goto out2; 1692b12cea91SAl Viro 1693143c8c91SAl Viro if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old)) 1694b12cea91SAl Viro goto out2; 1695ccd48bc7SAl Viro 16961da177e4SLinus Torvalds if (recurse) 169787129cc0SAl Viro mnt = copy_tree(old, old_path.dentry, 0); 16981da177e4SLinus Torvalds else 169987129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 17001da177e4SLinus Torvalds 1701be34d1a3SDavid Howells if (IS_ERR(mnt)) { 1702be34d1a3SDavid Howells err = PTR_ERR(mnt); 1703be34d1a3SDavid Howells goto out; 1704be34d1a3SDavid Howells } 1705ccd48bc7SAl Viro 170695bc5f25SAl Viro err = graft_tree(mnt, path); 17071da177e4SLinus Torvalds if (err) { 1708962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1709761d5c38SAl Viro umount_tree(mnt, 0, &umount_list); 1710962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 17115b83d2c5SRam Pai } 1712b12cea91SAl Viro out2: 1713b12cea91SAl Viro unlock_mount(path); 1714b12cea91SAl Viro release_mounts(&umount_list); 1715ccd48bc7SAl Viro out: 17162d92ab3cSAl Viro path_put(&old_path); 17171da177e4SLinus Torvalds return err; 17181da177e4SLinus Torvalds } 17191da177e4SLinus Torvalds 17202e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 17212e4b7fcdSDave Hansen { 17222e4b7fcdSDave Hansen int error = 0; 17232e4b7fcdSDave Hansen int readonly_request = 0; 17242e4b7fcdSDave Hansen 17252e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 17262e4b7fcdSDave Hansen readonly_request = 1; 17272e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 17282e4b7fcdSDave Hansen return 0; 17292e4b7fcdSDave Hansen 17302e4b7fcdSDave Hansen if (readonly_request) 173183adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 17322e4b7fcdSDave Hansen else 173383adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 17342e4b7fcdSDave Hansen return error; 17352e4b7fcdSDave Hansen } 17362e4b7fcdSDave Hansen 17371da177e4SLinus Torvalds /* 17381da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 17391da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 17401da177e4SLinus Torvalds * on it - tough luck. 17411da177e4SLinus Torvalds */ 17420a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 17431da177e4SLinus Torvalds void *data) 17441da177e4SLinus Torvalds { 17451da177e4SLinus Torvalds int err; 17462d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 1747143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 17481da177e4SLinus Torvalds 17491da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 17501da177e4SLinus Torvalds return -EPERM; 17511da177e4SLinus Torvalds 1752143c8c91SAl Viro if (!check_mnt(mnt)) 17531da177e4SLinus Torvalds return -EINVAL; 17541da177e4SLinus Torvalds 17552d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 17561da177e4SLinus Torvalds return -EINVAL; 17571da177e4SLinus Torvalds 1758ff36fe2cSEric Paris err = security_sb_remount(sb, data); 1759ff36fe2cSEric Paris if (err) 1760ff36fe2cSEric Paris return err; 1761ff36fe2cSEric Paris 17621da177e4SLinus Torvalds down_write(&sb->s_umount); 17632e4b7fcdSDave Hansen if (flags & MS_BIND) 17642d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 17654aa98cf7SAl Viro else 17661da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 17677b43a79fSAl Viro if (!err) { 1768962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1769143c8c91SAl Viro mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; 1770143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 1771962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 17727b43a79fSAl Viro } 17731da177e4SLinus Torvalds up_write(&sb->s_umount); 17740e55a7ccSDan Williams if (!err) { 1775962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1776143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1777962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 17780e55a7ccSDan Williams } 17791da177e4SLinus Torvalds return err; 17801da177e4SLinus Torvalds } 17811da177e4SLinus Torvalds 1782cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 17839676f0c6SRam Pai { 1784315fc83eSAl Viro struct mount *p; 1785909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 1786fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 17879676f0c6SRam Pai return 1; 17889676f0c6SRam Pai } 17899676f0c6SRam Pai return 0; 17909676f0c6SRam Pai } 17919676f0c6SRam Pai 1792808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 17931da177e4SLinus Torvalds { 17942d92ab3cSAl Viro struct path old_path, parent_path; 1795676da58dSAl Viro struct mount *p; 17960fb54e50SAl Viro struct mount *old; 17971da177e4SLinus Torvalds int err = 0; 17981da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 17991da177e4SLinus Torvalds return -EPERM; 18001da177e4SLinus Torvalds if (!old_name || !*old_name) 18011da177e4SLinus Torvalds return -EINVAL; 18022d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 18031da177e4SLinus Torvalds if (err) 18041da177e4SLinus Torvalds return err; 18051da177e4SLinus Torvalds 1806b12cea91SAl Viro err = lock_mount(path); 1807cc53ce53SDavid Howells if (err < 0) 1808cc53ce53SDavid Howells goto out; 1809cc53ce53SDavid Howells 1810143c8c91SAl Viro old = real_mount(old_path.mnt); 1811fc7be130SAl Viro p = real_mount(path->mnt); 1812143c8c91SAl Viro 18131da177e4SLinus Torvalds err = -EINVAL; 1814fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 18151da177e4SLinus Torvalds goto out1; 18161da177e4SLinus Torvalds 1817f3da392eSAlexey Dobriyan if (d_unlinked(path->dentry)) 181821444403SRam Pai goto out1; 18191da177e4SLinus Torvalds 18201da177e4SLinus Torvalds err = -EINVAL; 18212d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 182221444403SRam Pai goto out1; 18231da177e4SLinus Torvalds 1824676da58dSAl Viro if (!mnt_has_parent(old)) 182521444403SRam Pai goto out1; 18261da177e4SLinus Torvalds 18272d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 18282d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 182921444403SRam Pai goto out1; 183021444403SRam Pai /* 183121444403SRam Pai * Don't move a mount residing in a shared parent. 183221444403SRam Pai */ 1833fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 183421444403SRam Pai goto out1; 18359676f0c6SRam Pai /* 18369676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 18379676f0c6SRam Pai * mount which is shared. 18389676f0c6SRam Pai */ 1839fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 18409676f0c6SRam Pai goto out1; 18411da177e4SLinus Torvalds err = -ELOOP; 1842fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 1843676da58dSAl Viro if (p == old) 184421444403SRam Pai goto out1; 18451da177e4SLinus Torvalds 18460fb54e50SAl Viro err = attach_recursive_mnt(old, path, &parent_path); 18474ac91378SJan Blunck if (err) 184821444403SRam Pai goto out1; 18491da177e4SLinus Torvalds 18501da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 18511da177e4SLinus Torvalds * automatically */ 18526776db3dSAl Viro list_del_init(&old->mnt_expire); 18531da177e4SLinus Torvalds out1: 1854b12cea91SAl Viro unlock_mount(path); 18551da177e4SLinus Torvalds out: 18561da177e4SLinus Torvalds if (!err) 18571a390689SAl Viro path_put(&parent_path); 18582d92ab3cSAl Viro path_put(&old_path); 18591da177e4SLinus Torvalds return err; 18601da177e4SLinus Torvalds } 18611da177e4SLinus Torvalds 18629d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 18639d412a43SAl Viro { 18649d412a43SAl Viro int err; 18659d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 18669d412a43SAl Viro if (subtype) { 18679d412a43SAl Viro subtype++; 18689d412a43SAl Viro err = -EINVAL; 18699d412a43SAl Viro if (!subtype[0]) 18709d412a43SAl Viro goto err; 18719d412a43SAl Viro } else 18729d412a43SAl Viro subtype = ""; 18739d412a43SAl Viro 18749d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 18759d412a43SAl Viro err = -ENOMEM; 18769d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 18779d412a43SAl Viro goto err; 18789d412a43SAl Viro return mnt; 18799d412a43SAl Viro 18809d412a43SAl Viro err: 18819d412a43SAl Viro mntput(mnt); 18829d412a43SAl Viro return ERR_PTR(err); 18839d412a43SAl Viro } 18849d412a43SAl Viro 188579e801a9SAl Viro static struct vfsmount * 18869d412a43SAl Viro do_kern_mount(const char *fstype, int flags, const char *name, void *data) 18879d412a43SAl Viro { 18889d412a43SAl Viro struct file_system_type *type = get_fs_type(fstype); 18899d412a43SAl Viro struct vfsmount *mnt; 18909d412a43SAl Viro if (!type) 18919d412a43SAl Viro return ERR_PTR(-ENODEV); 18929d412a43SAl Viro mnt = vfs_kern_mount(type, flags, name, data); 18939d412a43SAl Viro if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 18949d412a43SAl Viro !mnt->mnt_sb->s_subtype) 18959d412a43SAl Viro mnt = fs_set_subtype(mnt, fstype); 18969d412a43SAl Viro put_filesystem(type); 18979d412a43SAl Viro return mnt; 18989d412a43SAl Viro } 18999d412a43SAl Viro 19009d412a43SAl Viro /* 19019d412a43SAl Viro * add a mount into a namespace's mount tree 19029d412a43SAl Viro */ 190395bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 19049d412a43SAl Viro { 19059d412a43SAl Viro int err; 19069d412a43SAl Viro 19079d412a43SAl Viro mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); 19089d412a43SAl Viro 1909b12cea91SAl Viro err = lock_mount(path); 1910b12cea91SAl Viro if (err) 1911b12cea91SAl Viro return err; 19129d412a43SAl Viro 19139d412a43SAl Viro err = -EINVAL; 1914156cacb1SAl Viro if (unlikely(!check_mnt(real_mount(path->mnt)))) { 1915156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 1916156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 19179d412a43SAl Viro goto unlock; 1918156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 1919156cacb1SAl Viro if (!real_mount(path->mnt)->mnt_ns) 1920156cacb1SAl Viro goto unlock; 1921156cacb1SAl Viro } 19229d412a43SAl Viro 19239d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 19249d412a43SAl Viro err = -EBUSY; 192595bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 19269d412a43SAl Viro path->mnt->mnt_root == path->dentry) 19279d412a43SAl Viro goto unlock; 19289d412a43SAl Viro 19299d412a43SAl Viro err = -EINVAL; 193095bc5f25SAl Viro if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) 19319d412a43SAl Viro goto unlock; 19329d412a43SAl Viro 193395bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 19349d412a43SAl Viro err = graft_tree(newmnt, path); 19359d412a43SAl Viro 19369d412a43SAl Viro unlock: 1937b12cea91SAl Viro unlock_mount(path); 19389d412a43SAl Viro return err; 19399d412a43SAl Viro } 1940b1e75df4SAl Viro 19411da177e4SLinus Torvalds /* 19421da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 19431da177e4SLinus Torvalds * namespace's tree 19441da177e4SLinus Torvalds */ 1945808d4e3cSAl Viro static int do_new_mount(struct path *path, const char *type, int flags, 1946808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 19471da177e4SLinus Torvalds { 19481da177e4SLinus Torvalds struct vfsmount *mnt; 194915f9a3f3SAl Viro int err; 19501da177e4SLinus Torvalds 1951eca6f534SVegard Nossum if (!type) 19521da177e4SLinus Torvalds return -EINVAL; 19531da177e4SLinus Torvalds 19541da177e4SLinus Torvalds /* we need capabilities... */ 19551da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 19561da177e4SLinus Torvalds return -EPERM; 19571da177e4SLinus Torvalds 19581da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 19591da177e4SLinus Torvalds if (IS_ERR(mnt)) 19601da177e4SLinus Torvalds return PTR_ERR(mnt); 19611da177e4SLinus Torvalds 196295bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 196315f9a3f3SAl Viro if (err) 196415f9a3f3SAl Viro mntput(mnt); 196515f9a3f3SAl Viro return err; 19661da177e4SLinus Torvalds } 19671da177e4SLinus Torvalds 196819a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 196919a167afSAl Viro { 19706776db3dSAl Viro struct mount *mnt = real_mount(m); 197119a167afSAl Viro int err; 197219a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 197319a167afSAl Viro * expired before we get a chance to add it 197419a167afSAl Viro */ 19756776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 197619a167afSAl Viro 197719a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 197819a167afSAl Viro m->mnt_root == path->dentry) { 1979b1e75df4SAl Viro err = -ELOOP; 1980b1e75df4SAl Viro goto fail; 198119a167afSAl Viro } 198219a167afSAl Viro 198395bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 1984b1e75df4SAl Viro if (!err) 1985b1e75df4SAl Viro return 0; 1986b1e75df4SAl Viro fail: 1987b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 19886776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 1989b1e75df4SAl Viro down_write(&namespace_sem); 1990962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 19916776db3dSAl Viro list_del_init(&mnt->mnt_expire); 1992962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 1993b1e75df4SAl Viro up_write(&namespace_sem); 199419a167afSAl Viro } 1995b1e75df4SAl Viro mntput(m); 1996b1e75df4SAl Viro mntput(m); 199719a167afSAl Viro return err; 199819a167afSAl Viro } 199919a167afSAl Viro 2000ea5b778aSDavid Howells /** 2001ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2002ea5b778aSDavid Howells * @mnt: The mount to list. 2003ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2004ea5b778aSDavid Howells */ 2005ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2006ea5b778aSDavid Howells { 2007ea5b778aSDavid Howells down_write(&namespace_sem); 2008962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 2009ea5b778aSDavid Howells 20106776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2011ea5b778aSDavid Howells 2012962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 2013ea5b778aSDavid Howells up_write(&namespace_sem); 2014ea5b778aSDavid Howells } 2015ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2016ea5b778aSDavid Howells 2017ea5b778aSDavid Howells /* 20181da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 20191da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 20201da177e4SLinus Torvalds * here 20211da177e4SLinus Torvalds */ 20221da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 20231da177e4SLinus Torvalds { 2024761d5c38SAl Viro struct mount *mnt, *next; 20251da177e4SLinus Torvalds LIST_HEAD(graveyard); 2026bcc5c7d2SAl Viro LIST_HEAD(umounts); 20271da177e4SLinus Torvalds 20281da177e4SLinus Torvalds if (list_empty(mounts)) 20291da177e4SLinus Torvalds return; 20301da177e4SLinus Torvalds 2031bcc5c7d2SAl Viro down_write(&namespace_sem); 2032962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 20331da177e4SLinus Torvalds 20341da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 20351da177e4SLinus Torvalds * following criteria: 20361da177e4SLinus Torvalds * - only referenced by its parent vfsmount 20371da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 20381da177e4SLinus Torvalds * cleared by mntput()) 20391da177e4SLinus Torvalds */ 20406776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2041863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 20421ab59738SAl Viro propagate_mount_busy(mnt, 1)) 20431da177e4SLinus Torvalds continue; 20446776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 20451da177e4SLinus Torvalds } 2046bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 20476776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2048143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2049bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 2050bcc5c7d2SAl Viro } 2051962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 2052bcc5c7d2SAl Viro up_write(&namespace_sem); 2053bcc5c7d2SAl Viro 2054bcc5c7d2SAl Viro release_mounts(&umounts); 20551da177e4SLinus Torvalds } 20561da177e4SLinus Torvalds 20571da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 20581da177e4SLinus Torvalds 20591da177e4SLinus Torvalds /* 20605528f911STrond Myklebust * Ripoff of 'select_parent()' 20615528f911STrond Myklebust * 20625528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 20635528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 20645528f911STrond Myklebust */ 2065692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 20665528f911STrond Myklebust { 2067692afc31SAl Viro struct mount *this_parent = parent; 20685528f911STrond Myklebust struct list_head *next; 20695528f911STrond Myklebust int found = 0; 20705528f911STrond Myklebust 20715528f911STrond Myklebust repeat: 20726b41d536SAl Viro next = this_parent->mnt_mounts.next; 20735528f911STrond Myklebust resume: 20746b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 20755528f911STrond Myklebust struct list_head *tmp = next; 20766b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 20775528f911STrond Myklebust 20785528f911STrond Myklebust next = tmp->next; 2079692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 20805528f911STrond Myklebust continue; 20815528f911STrond Myklebust /* 20825528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 20835528f911STrond Myklebust */ 20846b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 20855528f911STrond Myklebust this_parent = mnt; 20865528f911STrond Myklebust goto repeat; 20875528f911STrond Myklebust } 20885528f911STrond Myklebust 20891ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 20906776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 20915528f911STrond Myklebust found++; 20925528f911STrond Myklebust } 20935528f911STrond Myklebust } 20945528f911STrond Myklebust /* 20955528f911STrond Myklebust * All done at this level ... ascend and resume the search 20965528f911STrond Myklebust */ 20975528f911STrond Myklebust if (this_parent != parent) { 20986b41d536SAl Viro next = this_parent->mnt_child.next; 20990714a533SAl Viro this_parent = this_parent->mnt_parent; 21005528f911STrond Myklebust goto resume; 21015528f911STrond Myklebust } 21025528f911STrond Myklebust return found; 21035528f911STrond Myklebust } 21045528f911STrond Myklebust 21055528f911STrond Myklebust /* 21065528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 21075528f911STrond Myklebust * submounts of a specific parent mountpoint 210899b7db7bSNick Piggin * 210999b7db7bSNick Piggin * vfsmount_lock must be held for write 21105528f911STrond Myklebust */ 2111692afc31SAl Viro static void shrink_submounts(struct mount *mnt, struct list_head *umounts) 21125528f911STrond Myklebust { 21135528f911STrond Myklebust LIST_HEAD(graveyard); 2114761d5c38SAl Viro struct mount *m; 21155528f911STrond Myklebust 21165528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2117c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2118bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2119761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 21206776db3dSAl Viro mnt_expire); 2121143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2122afef80b3SEric W. Biederman umount_tree(m, 1, umounts); 2123bcc5c7d2SAl Viro } 2124bcc5c7d2SAl Viro } 21255528f911STrond Myklebust } 21265528f911STrond Myklebust 21275528f911STrond Myklebust /* 21281da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 21291da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 21301da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 21311da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 21321da177e4SLinus Torvalds */ 2133b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2134b58fed8bSRam Pai unsigned long n) 21351da177e4SLinus Torvalds { 21361da177e4SLinus Torvalds char *t = to; 21371da177e4SLinus Torvalds const char __user *f = from; 21381da177e4SLinus Torvalds char c; 21391da177e4SLinus Torvalds 21401da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 21411da177e4SLinus Torvalds return n; 21421da177e4SLinus Torvalds 21431da177e4SLinus Torvalds while (n) { 21441da177e4SLinus Torvalds if (__get_user(c, f)) { 21451da177e4SLinus Torvalds memset(t, 0, n); 21461da177e4SLinus Torvalds break; 21471da177e4SLinus Torvalds } 21481da177e4SLinus Torvalds *t++ = c; 21491da177e4SLinus Torvalds f++; 21501da177e4SLinus Torvalds n--; 21511da177e4SLinus Torvalds } 21521da177e4SLinus Torvalds return n; 21531da177e4SLinus Torvalds } 21541da177e4SLinus Torvalds 21551da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 21561da177e4SLinus Torvalds { 21571da177e4SLinus Torvalds int i; 21581da177e4SLinus Torvalds unsigned long page; 21591da177e4SLinus Torvalds unsigned long size; 21601da177e4SLinus Torvalds 21611da177e4SLinus Torvalds *where = 0; 21621da177e4SLinus Torvalds if (!data) 21631da177e4SLinus Torvalds return 0; 21641da177e4SLinus Torvalds 21651da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 21661da177e4SLinus Torvalds return -ENOMEM; 21671da177e4SLinus Torvalds 21681da177e4SLinus Torvalds /* We only care that *some* data at the address the user 21691da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 21701da177e4SLinus Torvalds * the remainder of the page. 21711da177e4SLinus Torvalds */ 21721da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 21731da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 21741da177e4SLinus Torvalds if (size > PAGE_SIZE) 21751da177e4SLinus Torvalds size = PAGE_SIZE; 21761da177e4SLinus Torvalds 21771da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 21781da177e4SLinus Torvalds if (!i) { 21791da177e4SLinus Torvalds free_page(page); 21801da177e4SLinus Torvalds return -EFAULT; 21811da177e4SLinus Torvalds } 21821da177e4SLinus Torvalds if (i != PAGE_SIZE) 21831da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 21841da177e4SLinus Torvalds *where = page; 21851da177e4SLinus Torvalds return 0; 21861da177e4SLinus Torvalds } 21871da177e4SLinus Torvalds 2188eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 2189eca6f534SVegard Nossum { 2190eca6f534SVegard Nossum char *tmp; 2191eca6f534SVegard Nossum 2192eca6f534SVegard Nossum if (!data) { 2193eca6f534SVegard Nossum *where = NULL; 2194eca6f534SVegard Nossum return 0; 2195eca6f534SVegard Nossum } 2196eca6f534SVegard Nossum 2197eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 2198eca6f534SVegard Nossum if (IS_ERR(tmp)) 2199eca6f534SVegard Nossum return PTR_ERR(tmp); 2200eca6f534SVegard Nossum 2201eca6f534SVegard Nossum *where = tmp; 2202eca6f534SVegard Nossum return 0; 2203eca6f534SVegard Nossum } 2204eca6f534SVegard Nossum 22051da177e4SLinus Torvalds /* 22061da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 22071da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 22081da177e4SLinus Torvalds * 22091da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 22101da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 22111da177e4SLinus Torvalds * information (or be NULL). 22121da177e4SLinus Torvalds * 22131da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 22141da177e4SLinus Torvalds * When the flags word was introduced its top half was required 22151da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 22161da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 22171da177e4SLinus Torvalds * and must be discarded. 22181da177e4SLinus Torvalds */ 2219808d4e3cSAl Viro long do_mount(const char *dev_name, const char *dir_name, 2220808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 22211da177e4SLinus Torvalds { 22222d92ab3cSAl Viro struct path path; 22231da177e4SLinus Torvalds int retval = 0; 22241da177e4SLinus Torvalds int mnt_flags = 0; 22251da177e4SLinus Torvalds 22261da177e4SLinus Torvalds /* Discard magic */ 22271da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 22281da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 22291da177e4SLinus Torvalds 22301da177e4SLinus Torvalds /* Basic sanity checks */ 22311da177e4SLinus Torvalds 22321da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 22331da177e4SLinus Torvalds return -EINVAL; 22341da177e4SLinus Torvalds 22351da177e4SLinus Torvalds if (data_page) 22361da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 22371da177e4SLinus Torvalds 2238a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 2239a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2240a27ab9f2STetsuo Handa if (retval) 2241a27ab9f2STetsuo Handa return retval; 2242a27ab9f2STetsuo Handa 2243a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2244a27ab9f2STetsuo Handa type_page, flags, data_page); 2245a27ab9f2STetsuo Handa if (retval) 2246a27ab9f2STetsuo Handa goto dput_out; 2247a27ab9f2STetsuo Handa 2248613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2249613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 22500a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 22510a1c01c9SMatthew Garrett 22521da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 22531da177e4SLinus Torvalds if (flags & MS_NOSUID) 22541da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 22551da177e4SLinus Torvalds if (flags & MS_NODEV) 22561da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 22571da177e4SLinus Torvalds if (flags & MS_NOEXEC) 22581da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2259fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2260fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2261fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2262fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2263d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2264d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 22652e4b7fcdSDave Hansen if (flags & MS_RDONLY) 22662e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2267fc33a7bbSChristoph Hellwig 22687a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2269d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2270d0adde57SMatthew Garrett MS_STRICTATIME); 22711da177e4SLinus Torvalds 22721da177e4SLinus Torvalds if (flags & MS_REMOUNT) 22732d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 22741da177e4SLinus Torvalds data_page); 22751da177e4SLinus Torvalds else if (flags & MS_BIND) 22762d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 22779676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 22782d92ab3cSAl Viro retval = do_change_type(&path, flags); 22791da177e4SLinus Torvalds else if (flags & MS_MOVE) 22802d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 22811da177e4SLinus Torvalds else 22822d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 22831da177e4SLinus Torvalds dev_name, data_page); 22841da177e4SLinus Torvalds dput_out: 22852d92ab3cSAl Viro path_put(&path); 22861da177e4SLinus Torvalds return retval; 22871da177e4SLinus Torvalds } 22881da177e4SLinus Torvalds 22898823c079SEric W. Biederman /* 22908823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 22918823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 22928823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 22938823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 22948823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 22958823c079SEric W. Biederman */ 22968823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 22978823c079SEric W. Biederman 2298cf8d2c11STrond Myklebust static struct mnt_namespace *alloc_mnt_ns(void) 2299cf8d2c11STrond Myklebust { 2300cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2301cf8d2c11STrond Myklebust 2302cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2303cf8d2c11STrond Myklebust if (!new_ns) 2304cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 23058823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2306cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2307cf8d2c11STrond Myklebust new_ns->root = NULL; 2308cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2309cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2310cf8d2c11STrond Myklebust new_ns->event = 0; 2311cf8d2c11STrond Myklebust return new_ns; 2312cf8d2c11STrond Myklebust } 2313cf8d2c11STrond Myklebust 2314741a2951SJANAK DESAI /* 2315741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 2316741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 2317741a2951SJANAK DESAI */ 2318e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 23196b3286edSKirill Korotaev struct fs_struct *fs) 23201da177e4SLinus Torvalds { 23216b3286edSKirill Korotaev struct mnt_namespace *new_ns; 23227f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2323315fc83eSAl Viro struct mount *p, *q; 2324be08d6d2SAl Viro struct mount *old = mnt_ns->root; 2325cb338d06SAl Viro struct mount *new; 23261da177e4SLinus Torvalds 2327cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2328cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2329cf8d2c11STrond Myklebust return new_ns; 23301da177e4SLinus Torvalds 2331390c6843SRam Pai down_write(&namespace_sem); 23321da177e4SLinus Torvalds /* First pass: copy the tree topology */ 2333909b0a88SAl Viro new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE); 2334be34d1a3SDavid Howells if (IS_ERR(new)) { 2335390c6843SRam Pai up_write(&namespace_sem); 23361da177e4SLinus Torvalds kfree(new_ns); 2337be34d1a3SDavid Howells return ERR_CAST(new); 23381da177e4SLinus Torvalds } 2339be08d6d2SAl Viro new_ns->root = new; 2340962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 23411a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 2342962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 23431da177e4SLinus Torvalds 23441da177e4SLinus Torvalds /* 23451da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 23461da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 23471da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 23481da177e4SLinus Torvalds */ 2349909b0a88SAl Viro p = old; 2350cb338d06SAl Viro q = new; 23511da177e4SLinus Torvalds while (p) { 2352143c8c91SAl Viro q->mnt_ns = new_ns; 23531da177e4SLinus Torvalds if (fs) { 2354315fc83eSAl Viro if (&p->mnt == fs->root.mnt) { 2355315fc83eSAl Viro fs->root.mnt = mntget(&q->mnt); 2356315fc83eSAl Viro rootmnt = &p->mnt; 23571da177e4SLinus Torvalds } 2358315fc83eSAl Viro if (&p->mnt == fs->pwd.mnt) { 2359315fc83eSAl Viro fs->pwd.mnt = mntget(&q->mnt); 2360315fc83eSAl Viro pwdmnt = &p->mnt; 23611da177e4SLinus Torvalds } 23621da177e4SLinus Torvalds } 2363909b0a88SAl Viro p = next_mnt(p, old); 2364909b0a88SAl Viro q = next_mnt(q, new); 23651da177e4SLinus Torvalds } 2366390c6843SRam Pai up_write(&namespace_sem); 23671da177e4SLinus Torvalds 23681da177e4SLinus Torvalds if (rootmnt) 2369f03c6599SAl Viro mntput(rootmnt); 23701da177e4SLinus Torvalds if (pwdmnt) 2371f03c6599SAl Viro mntput(pwdmnt); 23721da177e4SLinus Torvalds 2373741a2951SJANAK DESAI return new_ns; 2374741a2951SJANAK DESAI } 2375741a2951SJANAK DESAI 2376213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2377e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 2378741a2951SJANAK DESAI { 23796b3286edSKirill Korotaev struct mnt_namespace *new_ns; 2380741a2951SJANAK DESAI 2381e3222c4eSBadari Pulavarty BUG_ON(!ns); 23826b3286edSKirill Korotaev get_mnt_ns(ns); 2383741a2951SJANAK DESAI 2384741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 2385e3222c4eSBadari Pulavarty return ns; 2386741a2951SJANAK DESAI 2387e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 2388741a2951SJANAK DESAI 23896b3286edSKirill Korotaev put_mnt_ns(ns); 2390e3222c4eSBadari Pulavarty return new_ns; 23911da177e4SLinus Torvalds } 23921da177e4SLinus Torvalds 2393cf8d2c11STrond Myklebust /** 2394cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2395cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2396cf8d2c11STrond Myklebust */ 23971a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2398cf8d2c11STrond Myklebust { 23991a4eeaf2SAl Viro struct mnt_namespace *new_ns = alloc_mnt_ns(); 2400cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 24011a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 24021a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2403be08d6d2SAl Viro new_ns->root = mnt; 24041a4eeaf2SAl Viro list_add(&new_ns->list, &mnt->mnt_list); 2405c1334495SAl Viro } else { 24061a4eeaf2SAl Viro mntput(m); 2407cf8d2c11STrond Myklebust } 2408cf8d2c11STrond Myklebust return new_ns; 2409cf8d2c11STrond Myklebust } 2410cf8d2c11STrond Myklebust 2411ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2412ea441d11SAl Viro { 2413ea441d11SAl Viro struct mnt_namespace *ns; 2414d31da0f0SAl Viro struct super_block *s; 2415ea441d11SAl Viro struct path path; 2416ea441d11SAl Viro int err; 2417ea441d11SAl Viro 2418ea441d11SAl Viro ns = create_mnt_ns(mnt); 2419ea441d11SAl Viro if (IS_ERR(ns)) 2420ea441d11SAl Viro return ERR_CAST(ns); 2421ea441d11SAl Viro 2422ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2423ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2424ea441d11SAl Viro 2425ea441d11SAl Viro put_mnt_ns(ns); 2426ea441d11SAl Viro 2427ea441d11SAl Viro if (err) 2428ea441d11SAl Viro return ERR_PTR(err); 2429ea441d11SAl Viro 2430ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2431d31da0f0SAl Viro s = path.mnt->mnt_sb; 2432d31da0f0SAl Viro atomic_inc(&s->s_active); 2433ea441d11SAl Viro mntput(path.mnt); 2434ea441d11SAl Viro /* lock the sucker */ 2435d31da0f0SAl Viro down_write(&s->s_umount); 2436ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2437ea441d11SAl Viro return path.dentry; 2438ea441d11SAl Viro } 2439ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2440ea441d11SAl Viro 2441bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2442bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 24431da177e4SLinus Torvalds { 2444eca6f534SVegard Nossum int ret; 2445eca6f534SVegard Nossum char *kernel_type; 244691a27b2aSJeff Layton struct filename *kernel_dir; 2447eca6f534SVegard Nossum char *kernel_dev; 24481da177e4SLinus Torvalds unsigned long data_page; 24491da177e4SLinus Torvalds 2450eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2451eca6f534SVegard Nossum if (ret < 0) 2452eca6f534SVegard Nossum goto out_type; 24531da177e4SLinus Torvalds 2454eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2455eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2456eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2457eca6f534SVegard Nossum goto out_dir; 2458eca6f534SVegard Nossum } 24591da177e4SLinus Torvalds 2460eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2461eca6f534SVegard Nossum if (ret < 0) 2462eca6f534SVegard Nossum goto out_dev; 24631da177e4SLinus Torvalds 2464eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2465eca6f534SVegard Nossum if (ret < 0) 2466eca6f534SVegard Nossum goto out_data; 24671da177e4SLinus Torvalds 246891a27b2aSJeff Layton ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, 2469eca6f534SVegard Nossum (void *) data_page); 2470eca6f534SVegard Nossum 24711da177e4SLinus Torvalds free_page(data_page); 2472eca6f534SVegard Nossum out_data: 2473eca6f534SVegard Nossum kfree(kernel_dev); 2474eca6f534SVegard Nossum out_dev: 2475eca6f534SVegard Nossum putname(kernel_dir); 2476eca6f534SVegard Nossum out_dir: 2477eca6f534SVegard Nossum kfree(kernel_type); 2478eca6f534SVegard Nossum out_type: 2479eca6f534SVegard Nossum return ret; 24801da177e4SLinus Torvalds } 24811da177e4SLinus Torvalds 24821da177e4SLinus Torvalds /* 2483afac7cbaSAl Viro * Return true if path is reachable from root 2484afac7cbaSAl Viro * 2485afac7cbaSAl Viro * namespace_sem or vfsmount_lock is held 2486afac7cbaSAl Viro */ 2487643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2488afac7cbaSAl Viro const struct path *root) 2489afac7cbaSAl Viro { 2490643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2491a73324daSAl Viro dentry = mnt->mnt_mountpoint; 24920714a533SAl Viro mnt = mnt->mnt_parent; 2493afac7cbaSAl Viro } 2494643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2495afac7cbaSAl Viro } 2496afac7cbaSAl Viro 2497afac7cbaSAl Viro int path_is_under(struct path *path1, struct path *path2) 2498afac7cbaSAl Viro { 2499afac7cbaSAl Viro int res; 2500962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 2501643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 2502962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 2503afac7cbaSAl Viro return res; 2504afac7cbaSAl Viro } 2505afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 2506afac7cbaSAl Viro 2507afac7cbaSAl Viro /* 25081da177e4SLinus Torvalds * pivot_root Semantics: 25091da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 25101da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 25111da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 25121da177e4SLinus Torvalds * 25131da177e4SLinus Torvalds * Restrictions: 25141da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 25151da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 25161da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 25171da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 25181da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 25191da177e4SLinus Torvalds * 25204a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 25214a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 25224a0d11faSNeil Brown * in this situation. 25234a0d11faSNeil Brown * 25241da177e4SLinus Torvalds * Notes: 25251da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 25261da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 25271da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 25281da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 25291da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 25301da177e4SLinus Torvalds * first. 25311da177e4SLinus Torvalds */ 25323480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 25333480b257SHeiko Carstens const char __user *, put_old) 25341da177e4SLinus Torvalds { 25352d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 2536419148daSAl Viro struct mount *new_mnt, *root_mnt; 25371da177e4SLinus Torvalds int error; 25381da177e4SLinus Torvalds 25391da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 25401da177e4SLinus Torvalds return -EPERM; 25411da177e4SLinus Torvalds 25422d8f3038SAl Viro error = user_path_dir(new_root, &new); 25431da177e4SLinus Torvalds if (error) 25441da177e4SLinus Torvalds goto out0; 25451da177e4SLinus Torvalds 25462d8f3038SAl Viro error = user_path_dir(put_old, &old); 25471da177e4SLinus Torvalds if (error) 25481da177e4SLinus Torvalds goto out1; 25491da177e4SLinus Torvalds 25502d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 2551b12cea91SAl Viro if (error) 2552b12cea91SAl Viro goto out2; 25531da177e4SLinus Torvalds 2554f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 2555b12cea91SAl Viro error = lock_mount(&old); 2556b12cea91SAl Viro if (error) 2557b12cea91SAl Viro goto out3; 2558b12cea91SAl Viro 25591da177e4SLinus Torvalds error = -EINVAL; 2560419148daSAl Viro new_mnt = real_mount(new.mnt); 2561419148daSAl Viro root_mnt = real_mount(root.mnt); 2562fc7be130SAl Viro if (IS_MNT_SHARED(real_mount(old.mnt)) || 2563fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 2564fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 2565b12cea91SAl Viro goto out4; 2566143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 2567b12cea91SAl Viro goto out4; 25681da177e4SLinus Torvalds error = -ENOENT; 2569f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 2570b12cea91SAl Viro goto out4; 2571f3da392eSAlexey Dobriyan if (d_unlinked(old.dentry)) 2572b12cea91SAl Viro goto out4; 25731da177e4SLinus Torvalds error = -EBUSY; 25742d8f3038SAl Viro if (new.mnt == root.mnt || 25752d8f3038SAl Viro old.mnt == root.mnt) 2576b12cea91SAl Viro goto out4; /* loop, on the same file system */ 25771da177e4SLinus Torvalds error = -EINVAL; 25788c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 2579b12cea91SAl Viro goto out4; /* not a mountpoint */ 2580676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 2581b12cea91SAl Viro goto out4; /* not attached */ 25822d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 2583b12cea91SAl Viro goto out4; /* not a mountpoint */ 2584676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 2585b12cea91SAl Viro goto out4; /* not attached */ 25864ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 2587643822b4SAl Viro if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new)) 2588b12cea91SAl Viro goto out4; 2589962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 2590419148daSAl Viro detach_mnt(new_mnt, &parent_path); 2591419148daSAl Viro detach_mnt(root_mnt, &root_parent); 25924ac91378SJan Blunck /* mount old root on put_old */ 2593419148daSAl Viro attach_mnt(root_mnt, &old); 25944ac91378SJan Blunck /* mount new_root on / */ 2595419148daSAl Viro attach_mnt(new_mnt, &root_parent); 25966b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 2597962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 25982d8f3038SAl Viro chroot_fs_refs(&root, &new); 25991da177e4SLinus Torvalds error = 0; 2600b12cea91SAl Viro out4: 2601b12cea91SAl Viro unlock_mount(&old); 2602b12cea91SAl Viro if (!error) { 26031a390689SAl Viro path_put(&root_parent); 26041a390689SAl Viro path_put(&parent_path); 2605b12cea91SAl Viro } 2606b12cea91SAl Viro out3: 26078c3ee42eSAl Viro path_put(&root); 2608b12cea91SAl Viro out2: 26092d8f3038SAl Viro path_put(&old); 26101da177e4SLinus Torvalds out1: 26112d8f3038SAl Viro path_put(&new); 26121da177e4SLinus Torvalds out0: 26131da177e4SLinus Torvalds return error; 26141da177e4SLinus Torvalds } 26151da177e4SLinus Torvalds 26161da177e4SLinus Torvalds static void __init init_mount_tree(void) 26171da177e4SLinus Torvalds { 26181da177e4SLinus Torvalds struct vfsmount *mnt; 26196b3286edSKirill Korotaev struct mnt_namespace *ns; 2620ac748a09SJan Blunck struct path root; 26211da177e4SLinus Torvalds 26221da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 26231da177e4SLinus Torvalds if (IS_ERR(mnt)) 26241da177e4SLinus Torvalds panic("Can't create rootfs"); 2625b3e19d92SNick Piggin 26263b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 26273b22edc5STrond Myklebust if (IS_ERR(ns)) 26281da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 26291da177e4SLinus Torvalds 26306b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 26316b3286edSKirill Korotaev get_mnt_ns(ns); 26321da177e4SLinus Torvalds 2633be08d6d2SAl Viro root.mnt = mnt; 2634be08d6d2SAl Viro root.dentry = mnt->mnt_root; 2635ac748a09SJan Blunck 2636ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2637ac748a09SJan Blunck set_fs_root(current->fs, &root); 26381da177e4SLinus Torvalds } 26391da177e4SLinus Torvalds 264074bf17cfSDenis Cheng void __init mnt_init(void) 26411da177e4SLinus Torvalds { 264213f14b4dSEric Dumazet unsigned u; 264315a67dd8SRandy Dunlap int err; 26441da177e4SLinus Torvalds 2645390c6843SRam Pai init_rwsem(&namespace_sem); 2646390c6843SRam Pai 26477d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 264820c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 26491da177e4SLinus Torvalds 2650b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 26511da177e4SLinus Torvalds 26521da177e4SLinus Torvalds if (!mount_hashtable) 26531da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 26541da177e4SLinus Torvalds 265580cdc6daSMandeep Singh Baines printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); 26561da177e4SLinus Torvalds 265713f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 265813f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 26591da177e4SLinus Torvalds 2660962830dfSAndi Kleen br_lock_init(&vfsmount_lock); 266199b7db7bSNick Piggin 266215a67dd8SRandy Dunlap err = sysfs_init(); 266315a67dd8SRandy Dunlap if (err) 266415a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 26658e24eea7SHarvey Harrison __func__, err); 266600d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 266700d26666SGreg Kroah-Hartman if (!fs_kobj) 26688e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 26691da177e4SLinus Torvalds init_rootfs(); 26701da177e4SLinus Torvalds init_mount_tree(); 26711da177e4SLinus Torvalds } 26721da177e4SLinus Torvalds 2673616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 26741da177e4SLinus Torvalds { 267570fbcdf4SRam Pai LIST_HEAD(umount_list); 2676616511d0STrond Myklebust 2677d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 2678616511d0STrond Myklebust return; 2679390c6843SRam Pai down_write(&namespace_sem); 2680962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 2681be08d6d2SAl Viro umount_tree(ns->root, 0, &umount_list); 2682962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 2683390c6843SRam Pai up_write(&namespace_sem); 268470fbcdf4SRam Pai release_mounts(&umount_list); 26856b3286edSKirill Korotaev kfree(ns); 26861da177e4SLinus Torvalds } 26879d412a43SAl Viro 26889d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 26899d412a43SAl Viro { 2690423e0ab0STim Chen struct vfsmount *mnt; 2691423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 2692423e0ab0STim Chen if (!IS_ERR(mnt)) { 2693423e0ab0STim Chen /* 2694423e0ab0STim Chen * it is a longterm mount, don't release mnt until 2695423e0ab0STim Chen * we unmount before file sys is unregistered 2696423e0ab0STim Chen */ 2697f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 2698423e0ab0STim Chen } 2699423e0ab0STim Chen return mnt; 27009d412a43SAl Viro } 27019d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 2702423e0ab0STim Chen 2703423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 2704423e0ab0STim Chen { 2705423e0ab0STim Chen /* release long term mount so mount point can be released */ 2706423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 2707f7a99c5bSAl Viro br_write_lock(&vfsmount_lock); 2708f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 2709f7a99c5bSAl Viro br_write_unlock(&vfsmount_lock); 2710423e0ab0STim Chen mntput(mnt); 2711423e0ab0STim Chen } 2712423e0ab0STim Chen } 2713423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 271402125a82SAl Viro 271502125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 271602125a82SAl Viro { 2717143c8c91SAl Viro return check_mnt(real_mount(mnt)); 271802125a82SAl Viro } 27198823c079SEric W. Biederman 27208823c079SEric W. Biederman static void *mntns_get(struct task_struct *task) 27218823c079SEric W. Biederman { 27228823c079SEric W. Biederman struct mnt_namespace *ns = NULL; 27238823c079SEric W. Biederman struct nsproxy *nsproxy; 27248823c079SEric W. Biederman 27258823c079SEric W. Biederman rcu_read_lock(); 27268823c079SEric W. Biederman nsproxy = task_nsproxy(task); 27278823c079SEric W. Biederman if (nsproxy) { 27288823c079SEric W. Biederman ns = nsproxy->mnt_ns; 27298823c079SEric W. Biederman get_mnt_ns(ns); 27308823c079SEric W. Biederman } 27318823c079SEric W. Biederman rcu_read_unlock(); 27328823c079SEric W. Biederman 27338823c079SEric W. Biederman return ns; 27348823c079SEric W. Biederman } 27358823c079SEric W. Biederman 27368823c079SEric W. Biederman static void mntns_put(void *ns) 27378823c079SEric W. Biederman { 27388823c079SEric W. Biederman put_mnt_ns(ns); 27398823c079SEric W. Biederman } 27408823c079SEric W. Biederman 27418823c079SEric W. Biederman static int mntns_install(struct nsproxy *nsproxy, void *ns) 27428823c079SEric W. Biederman { 27438823c079SEric W. Biederman struct fs_struct *fs = current->fs; 27448823c079SEric W. Biederman struct mnt_namespace *mnt_ns = ns; 27458823c079SEric W. Biederman struct path root; 27468823c079SEric W. Biederman 27478823c079SEric W. Biederman if (!capable(CAP_SYS_ADMIN) || !capable(CAP_SYS_CHROOT)) 27488823c079SEric W. Biederman return -EINVAL; 27498823c079SEric W. Biederman 27508823c079SEric W. Biederman if (fs->users != 1) 27518823c079SEric W. Biederman return -EINVAL; 27528823c079SEric W. Biederman 27538823c079SEric W. Biederman get_mnt_ns(mnt_ns); 27548823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 27558823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 27568823c079SEric W. Biederman 27578823c079SEric W. Biederman /* Find the root */ 27588823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 27598823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 27608823c079SEric W. Biederman path_get(&root); 27618823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 27628823c079SEric W. Biederman ; 27638823c079SEric W. Biederman 27648823c079SEric W. Biederman /* Update the pwd and root */ 27658823c079SEric W. Biederman set_fs_pwd(fs, &root); 27668823c079SEric W. Biederman set_fs_root(fs, &root); 27678823c079SEric W. Biederman 27688823c079SEric W. Biederman path_put(&root); 27698823c079SEric W. Biederman return 0; 27708823c079SEric W. Biederman } 27718823c079SEric W. Biederman 27728823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 27738823c079SEric W. Biederman .name = "mnt", 27748823c079SEric W. Biederman .type = CLONE_NEWNS, 27758823c079SEric W. Biederman .get = mntns_get, 27768823c079SEric W. Biederman .put = mntns_put, 27778823c079SEric W. Biederman .install = mntns_install, 27788823c079SEric W. Biederman }; 2779