11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * linux/fs/namespace.c 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * (C) Copyright Al Viro 2000, 2001 51da177e4SLinus Torvalds * Released under GPL v2. 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Based on code from fs/super.c, copyright Linus Torvalds and others. 81da177e4SLinus Torvalds * Heavily rewritten. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/syscalls.h> 12d10577a8SAl Viro #include <linux/export.h> 1316f7e0feSRandy Dunlap #include <linux/capability.h> 146b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 15771b1371SEric W. Biederman #include <linux/user_namespace.h> 161da177e4SLinus Torvalds #include <linux/namei.h> 171da177e4SLinus Torvalds #include <linux/security.h> 1873cd49ecSMiklos Szeredi #include <linux/idr.h> 19d10577a8SAl Viro #include <linux/acct.h> /* acct_auto_close_mnt */ 2057f150a5SRob Landley #include <linux/init.h> /* init_rootfs */ 21d10577a8SAl Viro #include <linux/fs_struct.h> /* get_fs_root et.al. */ 22d10577a8SAl Viro #include <linux/fsnotify.h> /* fsnotify_vfsmount_delete */ 23d10577a8SAl Viro #include <linux/uaccess.h> 240bb80f24SDavid Howells #include <linux/proc_ns.h> 2520b4fb48SLinus Torvalds #include <linux/magic.h> 2607b20889SRam Pai #include "pnode.h" 27948730b0SAdrian Bunk #include "internal.h" 281da177e4SLinus Torvalds 2913f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 3013f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 3113f14b4dSEric Dumazet 325addc5ddSAl Viro static int event; 3373cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 34719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 3599b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 36f21f6220SAl Viro static int mnt_id_start = 0; 37f21f6220SAl Viro static int mnt_group_start = 1; 385addc5ddSAl Viro 39fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 4084d17192SAl Viro static struct list_head *mountpoint_hashtable __read_mostly; 41e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 4259aa0da8SAl Viro static DECLARE_RWSEM(namespace_sem); 431da177e4SLinus Torvalds 44f87fd4c2SMiklos Szeredi /* /sys/fs */ 4500d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 4600d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 47f87fd4c2SMiklos Szeredi 4899b7db7bSNick Piggin /* 4999b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 5099b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 5199b7db7bSNick Piggin * up the tree. 5299b7db7bSNick Piggin * 5399b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 5499b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 5599b7db7bSNick Piggin */ 5648a066e7SAl Viro __cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock); 5799b7db7bSNick Piggin 581da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 591da177e4SLinus Torvalds { 601da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 611da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 6213f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 6313f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 641da177e4SLinus Torvalds } 651da177e4SLinus Torvalds 6699b7db7bSNick Piggin /* 6799b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 6899b7db7bSNick Piggin * serialize with freeing. 6999b7db7bSNick Piggin */ 70b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 7173cd49ecSMiklos Szeredi { 7273cd49ecSMiklos Szeredi int res; 7373cd49ecSMiklos Szeredi 7473cd49ecSMiklos Szeredi retry: 7573cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 7699b7db7bSNick Piggin spin_lock(&mnt_id_lock); 7715169fe7SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 78f21f6220SAl Viro if (!res) 7915169fe7SAl Viro mnt_id_start = mnt->mnt_id + 1; 8099b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 8173cd49ecSMiklos Szeredi if (res == -EAGAIN) 8273cd49ecSMiklos Szeredi goto retry; 8373cd49ecSMiklos Szeredi 8473cd49ecSMiklos Szeredi return res; 8573cd49ecSMiklos Szeredi } 8673cd49ecSMiklos Szeredi 87b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 8873cd49ecSMiklos Szeredi { 8915169fe7SAl Viro int id = mnt->mnt_id; 9099b7db7bSNick Piggin spin_lock(&mnt_id_lock); 91f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 92f21f6220SAl Viro if (mnt_id_start > id) 93f21f6220SAl Viro mnt_id_start = id; 9499b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 9573cd49ecSMiklos Szeredi } 9673cd49ecSMiklos Szeredi 97719f5d7fSMiklos Szeredi /* 98719f5d7fSMiklos Szeredi * Allocate a new peer group ID 99719f5d7fSMiklos Szeredi * 100719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 101719f5d7fSMiklos Szeredi */ 1024b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 103719f5d7fSMiklos Szeredi { 104f21f6220SAl Viro int res; 105f21f6220SAl Viro 106719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 107719f5d7fSMiklos Szeredi return -ENOMEM; 108719f5d7fSMiklos Szeredi 109f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 110f21f6220SAl Viro mnt_group_start, 11115169fe7SAl Viro &mnt->mnt_group_id); 112f21f6220SAl Viro if (!res) 11315169fe7SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 114f21f6220SAl Viro 115f21f6220SAl Viro return res; 116719f5d7fSMiklos Szeredi } 117719f5d7fSMiklos Szeredi 118719f5d7fSMiklos Szeredi /* 119719f5d7fSMiklos Szeredi * Release a peer group ID 120719f5d7fSMiklos Szeredi */ 1214b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 122719f5d7fSMiklos Szeredi { 12315169fe7SAl Viro int id = mnt->mnt_group_id; 124f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 125f21f6220SAl Viro if (mnt_group_start > id) 126f21f6220SAl Viro mnt_group_start = id; 12715169fe7SAl Viro mnt->mnt_group_id = 0; 128719f5d7fSMiklos Szeredi } 129719f5d7fSMiklos Szeredi 130b3e19d92SNick Piggin /* 131b3e19d92SNick Piggin * vfsmount lock must be held for read 132b3e19d92SNick Piggin */ 13383adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 134b3e19d92SNick Piggin { 135b3e19d92SNick Piggin #ifdef CONFIG_SMP 13668e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 137b3e19d92SNick Piggin #else 138b3e19d92SNick Piggin preempt_disable(); 13968e8a9feSAl Viro mnt->mnt_count += n; 140b3e19d92SNick Piggin preempt_enable(); 141b3e19d92SNick Piggin #endif 142b3e19d92SNick Piggin } 143b3e19d92SNick Piggin 144b3e19d92SNick Piggin /* 145b3e19d92SNick Piggin * vfsmount lock must be held for write 146b3e19d92SNick Piggin */ 14783adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 148b3e19d92SNick Piggin { 149b3e19d92SNick Piggin #ifdef CONFIG_SMP 150f03c6599SAl Viro unsigned int count = 0; 151b3e19d92SNick Piggin int cpu; 152b3e19d92SNick Piggin 153b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 15468e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 155b3e19d92SNick Piggin } 156b3e19d92SNick Piggin 157b3e19d92SNick Piggin return count; 158b3e19d92SNick Piggin #else 15968e8a9feSAl Viro return mnt->mnt_count; 160b3e19d92SNick Piggin #endif 161b3e19d92SNick Piggin } 162b3e19d92SNick Piggin 163b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1641da177e4SLinus Torvalds { 165c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 166c63181e6SAl Viro if (mnt) { 16773cd49ecSMiklos Szeredi int err; 16873cd49ecSMiklos Szeredi 169c63181e6SAl Viro err = mnt_alloc_id(mnt); 17088b38782SLi Zefan if (err) 17188b38782SLi Zefan goto out_free_cache; 17288b38782SLi Zefan 17388b38782SLi Zefan if (name) { 174c63181e6SAl Viro mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 175c63181e6SAl Viro if (!mnt->mnt_devname) 17688b38782SLi Zefan goto out_free_id; 17773cd49ecSMiklos Szeredi } 17873cd49ecSMiklos Szeredi 179b3e19d92SNick Piggin #ifdef CONFIG_SMP 180c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 181c63181e6SAl Viro if (!mnt->mnt_pcp) 182b3e19d92SNick Piggin goto out_free_devname; 183b3e19d92SNick Piggin 184c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 185b3e19d92SNick Piggin #else 186c63181e6SAl Viro mnt->mnt_count = 1; 187c63181e6SAl Viro mnt->mnt_writers = 0; 188b3e19d92SNick Piggin #endif 189b3e19d92SNick Piggin 190c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_hash); 191c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 192c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 193c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 194c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 195c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 196c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 197c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 1982504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 1992504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 2002504c5d6SAndreas Gruenbacher #endif 2011da177e4SLinus Torvalds } 202c63181e6SAl Viro return mnt; 20388b38782SLi Zefan 204d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 205d3ef3d73Snpiggin@suse.de out_free_devname: 206c63181e6SAl Viro kfree(mnt->mnt_devname); 207d3ef3d73Snpiggin@suse.de #endif 20888b38782SLi Zefan out_free_id: 209c63181e6SAl Viro mnt_free_id(mnt); 21088b38782SLi Zefan out_free_cache: 211c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 21288b38782SLi Zefan return NULL; 2131da177e4SLinus Torvalds } 2141da177e4SLinus Torvalds 2158366025eSDave Hansen /* 2168366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2178366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2188366025eSDave Hansen * We must keep track of when those operations start 2198366025eSDave Hansen * (for permission checks) and when they end, so that 2208366025eSDave Hansen * we can determine when writes are able to occur to 2218366025eSDave Hansen * a filesystem. 2228366025eSDave Hansen */ 2233d733633SDave Hansen /* 2243d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2253d733633SDave Hansen * @mnt: the mount to check for its write status 2263d733633SDave Hansen * 2273d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2283d733633SDave Hansen * It does not guarantee that the filesystem will stay 2293d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2303d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2313d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2323d733633SDave Hansen * r/w. 2333d733633SDave Hansen */ 2343d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2353d733633SDave Hansen { 2362e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2372e4b7fcdSDave Hansen return 1; 2382e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2392e4b7fcdSDave Hansen return 1; 2402e4b7fcdSDave Hansen return 0; 2413d733633SDave Hansen } 2423d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2433d733633SDave Hansen 24483adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2453d733633SDave Hansen { 246d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 24768e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 248d3ef3d73Snpiggin@suse.de #else 24968e8a9feSAl Viro mnt->mnt_writers++; 250d3ef3d73Snpiggin@suse.de #endif 2513d733633SDave Hansen } 2523d733633SDave Hansen 25383adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2543d733633SDave Hansen { 255d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 25668e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 257d3ef3d73Snpiggin@suse.de #else 25868e8a9feSAl Viro mnt->mnt_writers--; 259d3ef3d73Snpiggin@suse.de #endif 260d3ef3d73Snpiggin@suse.de } 261d3ef3d73Snpiggin@suse.de 26283adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 263d3ef3d73Snpiggin@suse.de { 264d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 265d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2663d733633SDave Hansen int cpu; 2673d733633SDave Hansen 2683d733633SDave Hansen for_each_possible_cpu(cpu) { 26968e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2703d733633SDave Hansen } 2713d733633SDave Hansen 272d3ef3d73Snpiggin@suse.de return count; 273d3ef3d73Snpiggin@suse.de #else 274d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 275d3ef3d73Snpiggin@suse.de #endif 2763d733633SDave Hansen } 2773d733633SDave Hansen 2784ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2794ed5e82fSMiklos Szeredi { 2804ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2814ed5e82fSMiklos Szeredi return 1; 2824ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2834ed5e82fSMiklos Szeredi smp_rmb(); 2844ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2854ed5e82fSMiklos Szeredi } 2864ed5e82fSMiklos Szeredi 2873d733633SDave Hansen /* 288eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 289eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 290eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 291eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 2923d733633SDave Hansen */ 2938366025eSDave Hansen /** 294eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 29583adc753SAl Viro * @m: the mount on which to take a write 2968366025eSDave Hansen * 297eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 298eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 299eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 300eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 301eb04c282SJan Kara * called. This is effectively a refcount. 3028366025eSDave Hansen */ 303eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3048366025eSDave Hansen { 30583adc753SAl Viro struct mount *mnt = real_mount(m); 3063d733633SDave Hansen int ret = 0; 3073d733633SDave Hansen 308d3ef3d73Snpiggin@suse.de preempt_disable(); 309c6653a83SNick Piggin mnt_inc_writers(mnt); 310d3ef3d73Snpiggin@suse.de /* 311c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 312d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 313d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 314d3ef3d73Snpiggin@suse.de */ 315d3ef3d73Snpiggin@suse.de smp_mb(); 3161e75529eSMiao Xie while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 317d3ef3d73Snpiggin@suse.de cpu_relax(); 318d3ef3d73Snpiggin@suse.de /* 319d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 320d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 321d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 322d3ef3d73Snpiggin@suse.de */ 323d3ef3d73Snpiggin@suse.de smp_rmb(); 3244ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 325c6653a83SNick Piggin mnt_dec_writers(mnt); 3263d733633SDave Hansen ret = -EROFS; 3273d733633SDave Hansen } 328d3ef3d73Snpiggin@suse.de preempt_enable(); 329eb04c282SJan Kara 330eb04c282SJan Kara return ret; 331eb04c282SJan Kara } 332eb04c282SJan Kara 333eb04c282SJan Kara /** 334eb04c282SJan Kara * mnt_want_write - get write access to a mount 335eb04c282SJan Kara * @m: the mount on which to take a write 336eb04c282SJan Kara * 337eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 338eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 339eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 340eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 341eb04c282SJan Kara */ 342eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 343eb04c282SJan Kara { 344eb04c282SJan Kara int ret; 345eb04c282SJan Kara 346eb04c282SJan Kara sb_start_write(m->mnt_sb); 347eb04c282SJan Kara ret = __mnt_want_write(m); 348eb04c282SJan Kara if (ret) 349eb04c282SJan Kara sb_end_write(m->mnt_sb); 3503d733633SDave Hansen return ret; 3518366025eSDave Hansen } 3528366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3538366025eSDave Hansen 3548366025eSDave Hansen /** 35596029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 35696029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 35796029c4eSnpiggin@suse.de * 35896029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 35996029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 36096029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 36196029c4eSnpiggin@suse.de * on it. This allows some optimisation. 36296029c4eSnpiggin@suse.de * 36396029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 36496029c4eSnpiggin@suse.de * drop the reference. 36596029c4eSnpiggin@suse.de */ 36696029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 36796029c4eSnpiggin@suse.de { 36896029c4eSnpiggin@suse.de /* superblock may be r/o */ 36996029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 37096029c4eSnpiggin@suse.de return -EROFS; 37196029c4eSnpiggin@suse.de preempt_disable(); 37283adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 37396029c4eSnpiggin@suse.de preempt_enable(); 37496029c4eSnpiggin@suse.de return 0; 37596029c4eSnpiggin@suse.de } 37696029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 37796029c4eSnpiggin@suse.de 37896029c4eSnpiggin@suse.de /** 379eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 380eb04c282SJan Kara * @file: the file who's mount on which to take a write 381eb04c282SJan Kara * 382eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 383eb04c282SJan Kara * do some optimisations if the file is open for write already 384eb04c282SJan Kara */ 385eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 386eb04c282SJan Kara { 387496ad9aaSAl Viro struct inode *inode = file_inode(file); 388eb04c282SJan Kara 389eb04c282SJan Kara if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) 390eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 391eb04c282SJan Kara else 392eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 393eb04c282SJan Kara } 394eb04c282SJan Kara 395eb04c282SJan Kara /** 39696029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 39796029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 39896029c4eSnpiggin@suse.de * 39996029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 40096029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 40196029c4eSnpiggin@suse.de */ 40296029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 40396029c4eSnpiggin@suse.de { 404eb04c282SJan Kara int ret; 405eb04c282SJan Kara 406eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 407eb04c282SJan Kara ret = __mnt_want_write_file(file); 408eb04c282SJan Kara if (ret) 409eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 410eb04c282SJan Kara return ret; 41196029c4eSnpiggin@suse.de } 41296029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 41396029c4eSnpiggin@suse.de 41496029c4eSnpiggin@suse.de /** 415eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4168366025eSDave Hansen * @mnt: the mount on which to give up write access 4178366025eSDave Hansen * 4188366025eSDave Hansen * Tells the low-level filesystem that we are done 4198366025eSDave Hansen * performing writes to it. Must be matched with 420eb04c282SJan Kara * __mnt_want_write() call above. 4218366025eSDave Hansen */ 422eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4238366025eSDave Hansen { 424d3ef3d73Snpiggin@suse.de preempt_disable(); 42583adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 426d3ef3d73Snpiggin@suse.de preempt_enable(); 4278366025eSDave Hansen } 428eb04c282SJan Kara 429eb04c282SJan Kara /** 430eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 431eb04c282SJan Kara * @mnt: the mount on which to give up write access 432eb04c282SJan Kara * 433eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 434eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 435eb04c282SJan Kara * mnt_want_write() call above. 436eb04c282SJan Kara */ 437eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 438eb04c282SJan Kara { 439eb04c282SJan Kara __mnt_drop_write(mnt); 440eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 441eb04c282SJan Kara } 4428366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4438366025eSDave Hansen 444eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 445eb04c282SJan Kara { 446eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 447eb04c282SJan Kara } 448eb04c282SJan Kara 4492a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4502a79f17eSAl Viro { 4512a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4522a79f17eSAl Viro } 4532a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4542a79f17eSAl Viro 45583adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4568366025eSDave Hansen { 4573d733633SDave Hansen int ret = 0; 4583d733633SDave Hansen 459719ea2fbSAl Viro lock_mount_hash(); 46083adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 461d3ef3d73Snpiggin@suse.de /* 462d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 463d3ef3d73Snpiggin@suse.de * should be visible before we do. 464d3ef3d73Snpiggin@suse.de */ 465d3ef3d73Snpiggin@suse.de smp_mb(); 466d3ef3d73Snpiggin@suse.de 467d3ef3d73Snpiggin@suse.de /* 468d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 469d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 470d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 471d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 472d3ef3d73Snpiggin@suse.de * 473d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 474d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 475d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 476d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 477d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 478d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 479d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 480d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 481d3ef3d73Snpiggin@suse.de * we're counting up here. 482d3ef3d73Snpiggin@suse.de */ 483c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 484d3ef3d73Snpiggin@suse.de ret = -EBUSY; 485d3ef3d73Snpiggin@suse.de else 48683adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 487d3ef3d73Snpiggin@suse.de /* 488d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 489d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 490d3ef3d73Snpiggin@suse.de */ 491d3ef3d73Snpiggin@suse.de smp_wmb(); 49283adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 493719ea2fbSAl Viro unlock_mount_hash(); 4943d733633SDave Hansen return ret; 4953d733633SDave Hansen } 4968366025eSDave Hansen 49783adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 4982e4b7fcdSDave Hansen { 499719ea2fbSAl Viro lock_mount_hash(); 50083adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 501719ea2fbSAl Viro unlock_mount_hash(); 5022e4b7fcdSDave Hansen } 5032e4b7fcdSDave Hansen 5044ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5054ed5e82fSMiklos Szeredi { 5064ed5e82fSMiklos Szeredi struct mount *mnt; 5074ed5e82fSMiklos Szeredi int err = 0; 5084ed5e82fSMiklos Szeredi 5098e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5108e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5118e8b8796SMiklos Szeredi return -EBUSY; 5128e8b8796SMiklos Szeredi 513719ea2fbSAl Viro lock_mount_hash(); 5144ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5154ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5164ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5174ed5e82fSMiklos Szeredi smp_mb(); 5184ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5194ed5e82fSMiklos Szeredi err = -EBUSY; 5204ed5e82fSMiklos Szeredi break; 5214ed5e82fSMiklos Szeredi } 5224ed5e82fSMiklos Szeredi } 5234ed5e82fSMiklos Szeredi } 5248e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5258e8b8796SMiklos Szeredi err = -EBUSY; 5268e8b8796SMiklos Szeredi 5274ed5e82fSMiklos Szeredi if (!err) { 5284ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5294ed5e82fSMiklos Szeredi smp_wmb(); 5304ed5e82fSMiklos Szeredi } 5314ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5324ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5334ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5344ed5e82fSMiklos Szeredi } 535719ea2fbSAl Viro unlock_mount_hash(); 5364ed5e82fSMiklos Szeredi 5374ed5e82fSMiklos Szeredi return err; 5384ed5e82fSMiklos Szeredi } 5394ed5e82fSMiklos Szeredi 540b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5411da177e4SLinus Torvalds { 54252ba1621SAl Viro kfree(mnt->mnt_devname); 54373cd49ecSMiklos Szeredi mnt_free_id(mnt); 544d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 54568e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 546d3ef3d73Snpiggin@suse.de #endif 547b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5481da177e4SLinus Torvalds } 5491da177e4SLinus Torvalds 55048a066e7SAl Viro /* call under rcu_read_lock */ 55148a066e7SAl Viro bool legitimize_mnt(struct vfsmount *bastard, unsigned seq) 55248a066e7SAl Viro { 55348a066e7SAl Viro struct mount *mnt; 55448a066e7SAl Viro if (read_seqretry(&mount_lock, seq)) 55548a066e7SAl Viro return false; 55648a066e7SAl Viro if (bastard == NULL) 55748a066e7SAl Viro return true; 55848a066e7SAl Viro mnt = real_mount(bastard); 55948a066e7SAl Viro mnt_add_count(mnt, 1); 56048a066e7SAl Viro if (likely(!read_seqretry(&mount_lock, seq))) 56148a066e7SAl Viro return true; 56248a066e7SAl Viro if (bastard->mnt_flags & MNT_SYNC_UMOUNT) { 56348a066e7SAl Viro mnt_add_count(mnt, -1); 56448a066e7SAl Viro return false; 56548a066e7SAl Viro } 56648a066e7SAl Viro rcu_read_unlock(); 56748a066e7SAl Viro mntput(bastard); 56848a066e7SAl Viro rcu_read_lock(); 56948a066e7SAl Viro return false; 57048a066e7SAl Viro } 57148a066e7SAl Viro 5721da177e4SLinus Torvalds /* 573474279dcSAl Viro * find the first mount at @dentry on vfsmount @mnt. 57448a066e7SAl Viro * call under rcu_read_lock() 5751da177e4SLinus Torvalds */ 576474279dcSAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 5771da177e4SLinus Torvalds { 5781da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 579474279dcSAl Viro struct mount *p; 5801da177e4SLinus Torvalds 58148a066e7SAl Viro list_for_each_entry_rcu(p, head, mnt_hash) 582474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 583474279dcSAl Viro return p; 584474279dcSAl Viro return NULL; 5851da177e4SLinus Torvalds } 586474279dcSAl Viro 587474279dcSAl Viro /* 588474279dcSAl Viro * find the last mount at @dentry on vfsmount @mnt. 58948a066e7SAl Viro * mount_lock must be held. 590474279dcSAl Viro */ 591474279dcSAl Viro struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry) 592474279dcSAl Viro { 593474279dcSAl Viro struct list_head *head = mount_hashtable + hash(mnt, dentry); 594474279dcSAl Viro struct mount *p; 595474279dcSAl Viro 596474279dcSAl Viro list_for_each_entry_reverse(p, head, mnt_hash) 597474279dcSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) 598474279dcSAl Viro return p; 599474279dcSAl Viro return NULL; 6001da177e4SLinus Torvalds } 6011da177e4SLinus Torvalds 602a05964f3SRam Pai /* 603f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 604f015f126SDavid Howells * 605f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 606f015f126SDavid Howells * following mounts: 607f015f126SDavid Howells * 608f015f126SDavid Howells * mount /dev/sda1 /mnt 609f015f126SDavid Howells * mount /dev/sda2 /mnt 610f015f126SDavid Howells * mount /dev/sda3 /mnt 611f015f126SDavid Howells * 612f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 613f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 614f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 615f015f126SDavid Howells * 616f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 617a05964f3SRam Pai */ 6181c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 619a05964f3SRam Pai { 620c7105365SAl Viro struct mount *child_mnt; 62148a066e7SAl Viro struct vfsmount *m; 62248a066e7SAl Viro unsigned seq; 62399b7db7bSNick Piggin 62448a066e7SAl Viro rcu_read_lock(); 62548a066e7SAl Viro do { 62648a066e7SAl Viro seq = read_seqbegin(&mount_lock); 627474279dcSAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry); 62848a066e7SAl Viro m = child_mnt ? &child_mnt->mnt : NULL; 62948a066e7SAl Viro } while (!legitimize_mnt(m, seq)); 63048a066e7SAl Viro rcu_read_unlock(); 63148a066e7SAl Viro return m; 632a05964f3SRam Pai } 633a05964f3SRam Pai 63484d17192SAl Viro static struct mountpoint *new_mountpoint(struct dentry *dentry) 63584d17192SAl Viro { 63684d17192SAl Viro struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry); 63784d17192SAl Viro struct mountpoint *mp; 638eed81007SMiklos Szeredi int ret; 63984d17192SAl Viro 64084d17192SAl Viro list_for_each_entry(mp, chain, m_hash) { 64184d17192SAl Viro if (mp->m_dentry == dentry) { 64284d17192SAl Viro /* might be worth a WARN_ON() */ 64384d17192SAl Viro if (d_unlinked(dentry)) 64484d17192SAl Viro return ERR_PTR(-ENOENT); 64584d17192SAl Viro mp->m_count++; 64684d17192SAl Viro return mp; 64784d17192SAl Viro } 64884d17192SAl Viro } 64984d17192SAl Viro 65084d17192SAl Viro mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 65184d17192SAl Viro if (!mp) 65284d17192SAl Viro return ERR_PTR(-ENOMEM); 65384d17192SAl Viro 654eed81007SMiklos Szeredi ret = d_set_mounted(dentry); 655eed81007SMiklos Szeredi if (ret) { 65684d17192SAl Viro kfree(mp); 657eed81007SMiklos Szeredi return ERR_PTR(ret); 65884d17192SAl Viro } 659eed81007SMiklos Szeredi 66084d17192SAl Viro mp->m_dentry = dentry; 66184d17192SAl Viro mp->m_count = 1; 66284d17192SAl Viro list_add(&mp->m_hash, chain); 66384d17192SAl Viro return mp; 66484d17192SAl Viro } 66584d17192SAl Viro 66684d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 66784d17192SAl Viro { 66884d17192SAl Viro if (!--mp->m_count) { 66984d17192SAl Viro struct dentry *dentry = mp->m_dentry; 67084d17192SAl Viro spin_lock(&dentry->d_lock); 67184d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 67284d17192SAl Viro spin_unlock(&dentry->d_lock); 67384d17192SAl Viro list_del(&mp->m_hash); 67484d17192SAl Viro kfree(mp); 67584d17192SAl Viro } 67684d17192SAl Viro } 67784d17192SAl Viro 678143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 6791da177e4SLinus Torvalds { 6806b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 6811da177e4SLinus Torvalds } 6821da177e4SLinus Torvalds 68399b7db7bSNick Piggin /* 68499b7db7bSNick Piggin * vfsmount lock must be held for write 68599b7db7bSNick Piggin */ 6866b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 6875addc5ddSAl Viro { 6885addc5ddSAl Viro if (ns) { 6895addc5ddSAl Viro ns->event = ++event; 6905addc5ddSAl Viro wake_up_interruptible(&ns->poll); 6915addc5ddSAl Viro } 6925addc5ddSAl Viro } 6935addc5ddSAl Viro 69499b7db7bSNick Piggin /* 69599b7db7bSNick Piggin * vfsmount lock must be held for write 69699b7db7bSNick Piggin */ 6976b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 6985addc5ddSAl Viro { 6995addc5ddSAl Viro if (ns && ns->event != event) { 7005addc5ddSAl Viro ns->event = event; 7015addc5ddSAl Viro wake_up_interruptible(&ns->poll); 7025addc5ddSAl Viro } 7035addc5ddSAl Viro } 7045addc5ddSAl Viro 70599b7db7bSNick Piggin /* 70699b7db7bSNick Piggin * vfsmount lock must be held for write 70799b7db7bSNick Piggin */ 708419148daSAl Viro static void detach_mnt(struct mount *mnt, struct path *old_path) 7091da177e4SLinus Torvalds { 710a73324daSAl Viro old_path->dentry = mnt->mnt_mountpoint; 7110714a533SAl Viro old_path->mnt = &mnt->mnt_parent->mnt; 7120714a533SAl Viro mnt->mnt_parent = mnt; 713a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 7146b41d536SAl Viro list_del_init(&mnt->mnt_child); 7151b8e5564SAl Viro list_del_init(&mnt->mnt_hash); 71684d17192SAl Viro put_mountpoint(mnt->mnt_mp); 71784d17192SAl Viro mnt->mnt_mp = NULL; 7181da177e4SLinus Torvalds } 7191da177e4SLinus Torvalds 72099b7db7bSNick Piggin /* 72199b7db7bSNick Piggin * vfsmount lock must be held for write 72299b7db7bSNick Piggin */ 72384d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 72484d17192SAl Viro struct mountpoint *mp, 72544d964d6SAl Viro struct mount *child_mnt) 726b90fa9aeSRam Pai { 72784d17192SAl Viro mp->m_count++; 7283a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 72984d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 7303a2393d7SAl Viro child_mnt->mnt_parent = mnt; 73184d17192SAl Viro child_mnt->mnt_mp = mp; 732b90fa9aeSRam Pai } 733b90fa9aeSRam Pai 73499b7db7bSNick Piggin /* 73599b7db7bSNick Piggin * vfsmount lock must be held for write 73699b7db7bSNick Piggin */ 73784d17192SAl Viro static void attach_mnt(struct mount *mnt, 73884d17192SAl Viro struct mount *parent, 73984d17192SAl Viro struct mountpoint *mp) 7401da177e4SLinus Torvalds { 74184d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 7421b8e5564SAl Viro list_add_tail(&mnt->mnt_hash, mount_hashtable + 74384d17192SAl Viro hash(&parent->mnt, mp->m_dentry)); 74484d17192SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 745b90fa9aeSRam Pai } 746b90fa9aeSRam Pai 747b90fa9aeSRam Pai /* 74899b7db7bSNick Piggin * vfsmount lock must be held for write 749b90fa9aeSRam Pai */ 7504b2619a5SAl Viro static void commit_tree(struct mount *mnt) 751b90fa9aeSRam Pai { 7520714a533SAl Viro struct mount *parent = mnt->mnt_parent; 75383adc753SAl Viro struct mount *m; 754b90fa9aeSRam Pai LIST_HEAD(head); 755143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 756b90fa9aeSRam Pai 7570714a533SAl Viro BUG_ON(parent == mnt); 758b90fa9aeSRam Pai 7591a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 760f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 761143c8c91SAl Viro m->mnt_ns = n; 762f03c6599SAl Viro 763b90fa9aeSRam Pai list_splice(&head, n->list.prev); 764b90fa9aeSRam Pai 7651b8e5564SAl Viro list_add_tail(&mnt->mnt_hash, mount_hashtable + 766a73324daSAl Viro hash(&parent->mnt, mnt->mnt_mountpoint)); 7676b41d536SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 7686b3286edSKirill Korotaev touch_mnt_namespace(n); 7691da177e4SLinus Torvalds } 7701da177e4SLinus Torvalds 771909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 7721da177e4SLinus Torvalds { 7736b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 7746b41d536SAl Viro if (next == &p->mnt_mounts) { 7751da177e4SLinus Torvalds while (1) { 776909b0a88SAl Viro if (p == root) 7771da177e4SLinus Torvalds return NULL; 7786b41d536SAl Viro next = p->mnt_child.next; 7796b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 7801da177e4SLinus Torvalds break; 7810714a533SAl Viro p = p->mnt_parent; 7821da177e4SLinus Torvalds } 7831da177e4SLinus Torvalds } 7846b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 7851da177e4SLinus Torvalds } 7861da177e4SLinus Torvalds 787315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 7889676f0c6SRam Pai { 7896b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 7906b41d536SAl Viro while (prev != &p->mnt_mounts) { 7916b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 7926b41d536SAl Viro prev = p->mnt_mounts.prev; 7939676f0c6SRam Pai } 7949676f0c6SRam Pai return p; 7959676f0c6SRam Pai } 7969676f0c6SRam Pai 7979d412a43SAl Viro struct vfsmount * 7989d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 7999d412a43SAl Viro { 800b105e270SAl Viro struct mount *mnt; 8019d412a43SAl Viro struct dentry *root; 8029d412a43SAl Viro 8039d412a43SAl Viro if (!type) 8049d412a43SAl Viro return ERR_PTR(-ENODEV); 8059d412a43SAl Viro 8069d412a43SAl Viro mnt = alloc_vfsmnt(name); 8079d412a43SAl Viro if (!mnt) 8089d412a43SAl Viro return ERR_PTR(-ENOMEM); 8099d412a43SAl Viro 8109d412a43SAl Viro if (flags & MS_KERNMOUNT) 811b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 8129d412a43SAl Viro 8139d412a43SAl Viro root = mount_fs(type, flags, name, data); 8149d412a43SAl Viro if (IS_ERR(root)) { 8159d412a43SAl Viro free_vfsmnt(mnt); 8169d412a43SAl Viro return ERR_CAST(root); 8179d412a43SAl Viro } 8189d412a43SAl Viro 819b105e270SAl Viro mnt->mnt.mnt_root = root; 820b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 821a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8220714a533SAl Viro mnt->mnt_parent = mnt; 823719ea2fbSAl Viro lock_mount_hash(); 82439f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 825719ea2fbSAl Viro unlock_mount_hash(); 826b105e270SAl Viro return &mnt->mnt; 8279d412a43SAl Viro } 8289d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 8299d412a43SAl Viro 83087129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 83136341f64SRam Pai int flag) 8321da177e4SLinus Torvalds { 83387129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 834be34d1a3SDavid Howells struct mount *mnt; 835be34d1a3SDavid Howells int err; 8361da177e4SLinus Torvalds 837be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 838be34d1a3SDavid Howells if (!mnt) 839be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 840be34d1a3SDavid Howells 8417a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 84215169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 843719f5d7fSMiklos Szeredi else 84415169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 845719f5d7fSMiklos Szeredi 84615169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 847be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 848719f5d7fSMiklos Szeredi if (err) 849719f5d7fSMiklos Szeredi goto out_free; 850719f5d7fSMiklos Szeredi } 851719f5d7fSMiklos Szeredi 85287129cc0SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; 853132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 854132c94e3SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY)) 855132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 856132c94e3SEric W. Biederman 8575ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 8585ff9d8a6SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) 8595ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 8605ff9d8a6SEric W. Biederman 8611da177e4SLinus Torvalds atomic_inc(&sb->s_active); 862b105e270SAl Viro mnt->mnt.mnt_sb = sb; 863b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 864a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8650714a533SAl Viro mnt->mnt_parent = mnt; 866719ea2fbSAl Viro lock_mount_hash(); 86739f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 868719ea2fbSAl Viro unlock_mount_hash(); 869b90fa9aeSRam Pai 8707a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 8717a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 8726776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 87332301920SAl Viro mnt->mnt_master = old; 874fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 8758aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 876fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 8776776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 878d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 8796776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 880d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 8815afe0022SRam Pai } 882b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 8830f0afb1dSAl Viro set_mnt_shared(mnt); 8841da177e4SLinus Torvalds 8851da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 8861da177e4SLinus Torvalds * as the original if that was on one */ 88736341f64SRam Pai if (flag & CL_EXPIRE) { 8886776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 8896776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 8901da177e4SLinus Torvalds } 891be34d1a3SDavid Howells 892cb338d06SAl Viro return mnt; 893719f5d7fSMiklos Szeredi 894719f5d7fSMiklos Szeredi out_free: 895719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 896be34d1a3SDavid Howells return ERR_PTR(err); 8971da177e4SLinus Torvalds } 8981da177e4SLinus Torvalds 89948a066e7SAl Viro static void delayed_free(struct rcu_head *head) 90048a066e7SAl Viro { 90148a066e7SAl Viro struct mount *mnt = container_of(head, struct mount, mnt_rcu); 90248a066e7SAl Viro kfree(mnt->mnt_devname); 90348a066e7SAl Viro #ifdef CONFIG_SMP 90448a066e7SAl Viro free_percpu(mnt->mnt_pcp); 90548a066e7SAl Viro #endif 90648a066e7SAl Viro kmem_cache_free(mnt_cache, mnt); 90748a066e7SAl Viro } 90848a066e7SAl Viro 909900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 9107b7b1aceSAl Viro { 911b3e19d92SNick Piggin put_again: 91248a066e7SAl Viro rcu_read_lock(); 913aa9c0e07SAl Viro mnt_add_count(mnt, -1); 91448a066e7SAl Viro if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */ 91548a066e7SAl Viro rcu_read_unlock(); 91699b7db7bSNick Piggin return; 917b3e19d92SNick Piggin } 918719ea2fbSAl Viro lock_mount_hash(); 919b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 92048a066e7SAl Viro rcu_read_unlock(); 921719ea2fbSAl Viro unlock_mount_hash(); 92299b7db7bSNick Piggin return; 92399b7db7bSNick Piggin } 924863d684fSAl Viro if (unlikely(mnt->mnt_pinned)) { 925863d684fSAl Viro mnt_add_count(mnt, mnt->mnt_pinned + 1); 926863d684fSAl Viro mnt->mnt_pinned = 0; 92748a066e7SAl Viro rcu_read_unlock(); 928719ea2fbSAl Viro unlock_mount_hash(); 929900148dcSAl Viro acct_auto_close_mnt(&mnt->mnt); 930b3e19d92SNick Piggin goto put_again; 931b3e19d92SNick Piggin } 93248a066e7SAl Viro if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) { 93348a066e7SAl Viro rcu_read_unlock(); 93448a066e7SAl Viro unlock_mount_hash(); 93548a066e7SAl Viro return; 93648a066e7SAl Viro } 93748a066e7SAl Viro mnt->mnt.mnt_flags |= MNT_DOOMED; 93848a066e7SAl Viro rcu_read_unlock(); 939962830dfSAndi Kleen 94039f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 941719ea2fbSAl Viro unlock_mount_hash(); 942649a795aSAl Viro 943649a795aSAl Viro /* 944649a795aSAl Viro * This probably indicates that somebody messed 945649a795aSAl Viro * up a mnt_want/drop_write() pair. If this 946649a795aSAl Viro * happens, the filesystem was probably unable 947649a795aSAl Viro * to make r/w->r/o transitions. 948649a795aSAl Viro */ 949649a795aSAl Viro /* 950649a795aSAl Viro * The locking used to deal with mnt_count decrement provides barriers, 951649a795aSAl Viro * so mnt_get_writers() below is safe. 952649a795aSAl Viro */ 953649a795aSAl Viro WARN_ON(mnt_get_writers(mnt)); 954649a795aSAl Viro fsnotify_vfsmount_delete(&mnt->mnt); 955649a795aSAl Viro dput(mnt->mnt.mnt_root); 956649a795aSAl Viro deactivate_super(mnt->mnt.mnt_sb); 95748a066e7SAl Viro mnt_free_id(mnt); 95848a066e7SAl Viro call_rcu(&mnt->mnt_rcu, delayed_free); 959b3e19d92SNick Piggin } 960b3e19d92SNick Piggin 961b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 962b3e19d92SNick Piggin { 963b3e19d92SNick Piggin if (mnt) { 964863d684fSAl Viro struct mount *m = real_mount(mnt); 965b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 966863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 967863d684fSAl Viro m->mnt_expiry_mark = 0; 968863d684fSAl Viro mntput_no_expire(m); 969b3e19d92SNick Piggin } 970b3e19d92SNick Piggin } 971b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 972b3e19d92SNick Piggin 973b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 974b3e19d92SNick Piggin { 975b3e19d92SNick Piggin if (mnt) 97683adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 977b3e19d92SNick Piggin return mnt; 978b3e19d92SNick Piggin } 979b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 980b3e19d92SNick Piggin 9817b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 9827b7b1aceSAl Viro { 983719ea2fbSAl Viro lock_mount_hash(); 984863d684fSAl Viro real_mount(mnt)->mnt_pinned++; 985719ea2fbSAl Viro unlock_mount_hash(); 9867b7b1aceSAl Viro } 9877b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 9887b7b1aceSAl Viro 989863d684fSAl Viro void mnt_unpin(struct vfsmount *m) 9907b7b1aceSAl Viro { 991863d684fSAl Viro struct mount *mnt = real_mount(m); 992719ea2fbSAl Viro lock_mount_hash(); 9937b7b1aceSAl Viro if (mnt->mnt_pinned) { 994863d684fSAl Viro mnt_add_count(mnt, 1); 9957b7b1aceSAl Viro mnt->mnt_pinned--; 9967b7b1aceSAl Viro } 997719ea2fbSAl Viro unlock_mount_hash(); 9987b7b1aceSAl Viro } 9997b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 10001da177e4SLinus Torvalds 1001b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 1002b3b304a2SMiklos Szeredi { 1003b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 1004b3b304a2SMiklos Szeredi } 1005b3b304a2SMiklos Szeredi 1006b3b304a2SMiklos Szeredi /* 1007b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 1008b3b304a2SMiklos Szeredi * implement more complex mount option showing. 1009b3b304a2SMiklos Szeredi * 1010b3b304a2SMiklos Szeredi * See also save_mount_options(). 1011b3b304a2SMiklos Szeredi */ 101234c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 1013b3b304a2SMiklos Szeredi { 10142a32cebdSAl Viro const char *options; 10152a32cebdSAl Viro 10162a32cebdSAl Viro rcu_read_lock(); 101734c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 1018b3b304a2SMiklos Szeredi 1019b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 1020b3b304a2SMiklos Szeredi seq_putc(m, ','); 1021b3b304a2SMiklos Szeredi mangle(m, options); 1022b3b304a2SMiklos Szeredi } 10232a32cebdSAl Viro rcu_read_unlock(); 1024b3b304a2SMiklos Szeredi 1025b3b304a2SMiklos Szeredi return 0; 1026b3b304a2SMiklos Szeredi } 1027b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 1028b3b304a2SMiklos Szeredi 1029b3b304a2SMiklos Szeredi /* 1030b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 1031b3b304a2SMiklos Szeredi * called from the fill_super() callback. 1032b3b304a2SMiklos Szeredi * 1033b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 1034b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1035b3b304a2SMiklos Szeredi * remount fails. 1036b3b304a2SMiklos Szeredi * 1037b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1038b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1039b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1040b3b304a2SMiklos Szeredi * any more. 1041b3b304a2SMiklos Szeredi */ 1042b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1043b3b304a2SMiklos Szeredi { 10442a32cebdSAl Viro BUG_ON(sb->s_options); 10452a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1046b3b304a2SMiklos Szeredi } 1047b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1048b3b304a2SMiklos Szeredi 10492a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 10502a32cebdSAl Viro { 10512a32cebdSAl Viro char *old = sb->s_options; 10522a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 10532a32cebdSAl Viro if (old) { 10542a32cebdSAl Viro synchronize_rcu(); 10552a32cebdSAl Viro kfree(old); 10562a32cebdSAl Viro } 10572a32cebdSAl Viro } 10582a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 10592a32cebdSAl Viro 1060a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 10610226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 10621da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 10631da177e4SLinus Torvalds { 10646ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 10651da177e4SLinus Torvalds 1066390c6843SRam Pai down_read(&namespace_sem); 1067a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 10681da177e4SLinus Torvalds } 10691da177e4SLinus Torvalds 10701da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 10711da177e4SLinus Torvalds { 10726ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 1073b0765fb8SPavel Emelianov 1074a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 10751da177e4SLinus Torvalds } 10761da177e4SLinus Torvalds 10771da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 10781da177e4SLinus Torvalds { 1079390c6843SRam Pai up_read(&namespace_sem); 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds 10820226f492SAl Viro static int m_show(struct seq_file *m, void *v) 10839f5596afSAl Viro { 10846ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 10851a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 10860226f492SAl Viro return p->show(m, &r->mnt); 10871da177e4SLinus Torvalds } 10881da177e4SLinus Torvalds 1089a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 10901da177e4SLinus Torvalds .start = m_start, 10911da177e4SLinus Torvalds .next = m_next, 10921da177e4SLinus Torvalds .stop = m_stop, 10930226f492SAl Viro .show = m_show, 1094b4629fe2SChuck Lever }; 1095a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1096b4629fe2SChuck Lever 10971da177e4SLinus Torvalds /** 10981da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 10991da177e4SLinus Torvalds * @mnt: root of mount tree 11001da177e4SLinus Torvalds * 11011da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 11021da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 11031da177e4SLinus Torvalds * busy. 11041da177e4SLinus Torvalds */ 1105909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 11061da177e4SLinus Torvalds { 1107909b0a88SAl Viro struct mount *mnt = real_mount(m); 110836341f64SRam Pai int actual_refs = 0; 110936341f64SRam Pai int minimum_refs = 0; 1110315fc83eSAl Viro struct mount *p; 1111909b0a88SAl Viro BUG_ON(!m); 11121da177e4SLinus Torvalds 1113b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1114719ea2fbSAl Viro lock_mount_hash(); 1115909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 111683adc753SAl Viro actual_refs += mnt_get_count(p); 11171da177e4SLinus Torvalds minimum_refs += 2; 11181da177e4SLinus Torvalds } 1119719ea2fbSAl Viro unlock_mount_hash(); 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds if (actual_refs > minimum_refs) 11221da177e4SLinus Torvalds return 0; 1123e3474a8eSIan Kent 1124e3474a8eSIan Kent return 1; 11251da177e4SLinus Torvalds } 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 11281da177e4SLinus Torvalds 11291da177e4SLinus Torvalds /** 11301da177e4SLinus Torvalds * may_umount - check if a mount point is busy 11311da177e4SLinus Torvalds * @mnt: root of mount 11321da177e4SLinus Torvalds * 11331da177e4SLinus Torvalds * This is called to check if a mount point has any 11341da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 11351da177e4SLinus Torvalds * mount has sub mounts this will return busy 11361da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 11371da177e4SLinus Torvalds * 11381da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 11391da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 11401da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 11411da177e4SLinus Torvalds */ 11421da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 11431da177e4SLinus Torvalds { 1144e3474a8eSIan Kent int ret = 1; 11458ad08d8aSAl Viro down_read(&namespace_sem); 1146719ea2fbSAl Viro lock_mount_hash(); 11471ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1148e3474a8eSIan Kent ret = 0; 1149719ea2fbSAl Viro unlock_mount_hash(); 11508ad08d8aSAl Viro up_read(&namespace_sem); 1151a05964f3SRam Pai return ret; 11521da177e4SLinus Torvalds } 11531da177e4SLinus Torvalds 11541da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 11551da177e4SLinus Torvalds 1156e3197d83SAl Viro static LIST_HEAD(unmounted); /* protected by namespace_sem */ 1157e3197d83SAl Viro 115897216be0SAl Viro static void namespace_unlock(void) 11591da177e4SLinus Torvalds { 1160d5e50f74SAl Viro struct mount *mnt; 116197216be0SAl Viro LIST_HEAD(head); 116297216be0SAl Viro 116397216be0SAl Viro if (likely(list_empty(&unmounted))) { 116497216be0SAl Viro up_write(&namespace_sem); 116597216be0SAl Viro return; 116697216be0SAl Viro } 116797216be0SAl Viro 116897216be0SAl Viro list_splice_init(&unmounted, &head); 116997216be0SAl Viro up_write(&namespace_sem); 117097216be0SAl Viro 117148a066e7SAl Viro synchronize_rcu(); 117248a066e7SAl Viro 117397216be0SAl Viro while (!list_empty(&head)) { 117497216be0SAl Viro mnt = list_first_entry(&head, struct mount, mnt_hash); 11751b8e5564SAl Viro list_del_init(&mnt->mnt_hash); 1176aba809cfSAl Viro if (mnt->mnt_ex_mountpoint.mnt) 1177aba809cfSAl Viro path_put(&mnt->mnt_ex_mountpoint); 1178d5e50f74SAl Viro mntput(&mnt->mnt); 117970fbcdf4SRam Pai } 118070fbcdf4SRam Pai } 118170fbcdf4SRam Pai 118297216be0SAl Viro static inline void namespace_lock(void) 1183e3197d83SAl Viro { 118497216be0SAl Viro down_write(&namespace_sem); 1185e3197d83SAl Viro } 1186e3197d83SAl Viro 118799b7db7bSNick Piggin /* 118848a066e7SAl Viro * mount_lock must be held 118999b7db7bSNick Piggin * namespace_sem must be held for write 119048a066e7SAl Viro * how = 0 => just this tree, don't propagate 119148a066e7SAl Viro * how = 1 => propagate; we know that nobody else has reference to any victims 119248a066e7SAl Viro * how = 2 => lazy umount 119399b7db7bSNick Piggin */ 119448a066e7SAl Viro void umount_tree(struct mount *mnt, int how) 119570fbcdf4SRam Pai { 11967b8a53fdSAl Viro LIST_HEAD(tmp_list); 1197315fc83eSAl Viro struct mount *p; 119870fbcdf4SRam Pai 1199909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) 12001b8e5564SAl Viro list_move(&p->mnt_hash, &tmp_list); 120170fbcdf4SRam Pai 120248a066e7SAl Viro if (how) 12037b8a53fdSAl Viro propagate_umount(&tmp_list); 1204a05964f3SRam Pai 12051b8e5564SAl Viro list_for_each_entry(p, &tmp_list, mnt_hash) { 12066776db3dSAl Viro list_del_init(&p->mnt_expire); 12071a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1208143c8c91SAl Viro __touch_mnt_namespace(p->mnt_ns); 120963d37a84SAl Viro p->mnt_ns = NULL; 121048a066e7SAl Viro if (how < 2) 121148a066e7SAl Viro p->mnt.mnt_flags |= MNT_SYNC_UMOUNT; 12126b41d536SAl Viro list_del_init(&p->mnt_child); 1213676da58dSAl Viro if (mnt_has_parent(p)) { 121484d17192SAl Viro put_mountpoint(p->mnt_mp); 1215aba809cfSAl Viro /* move the reference to mountpoint into ->mnt_ex_mountpoint */ 1216aba809cfSAl Viro p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint; 1217aba809cfSAl Viro p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt; 1218aba809cfSAl Viro p->mnt_mountpoint = p->mnt.mnt_root; 1219aba809cfSAl Viro p->mnt_parent = p; 122084d17192SAl Viro p->mnt_mp = NULL; 12217c4b93d8SAl Viro } 12220f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 12231da177e4SLinus Torvalds } 1224328e6d90SAl Viro list_splice(&tmp_list, &unmounted); 12251da177e4SLinus Torvalds } 12261da177e4SLinus Torvalds 1227b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1228c35038beSAl Viro 12291ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 12301da177e4SLinus Torvalds { 12311ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 12321da177e4SLinus Torvalds int retval; 12331da177e4SLinus Torvalds 12341ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 12351da177e4SLinus Torvalds if (retval) 12361da177e4SLinus Torvalds return retval; 12371da177e4SLinus Torvalds 12381da177e4SLinus Torvalds /* 12391da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 12401da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 12411da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 12421da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 12431da177e4SLinus Torvalds */ 12441da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 12451ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 12461da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 12471da177e4SLinus Torvalds return -EINVAL; 12481da177e4SLinus Torvalds 1249b3e19d92SNick Piggin /* 1250b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1251b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1252b3e19d92SNick Piggin */ 1253719ea2fbSAl Viro lock_mount_hash(); 125483adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1255719ea2fbSAl Viro unlock_mount_hash(); 12561da177e4SLinus Torvalds return -EBUSY; 1257b3e19d92SNick Piggin } 1258719ea2fbSAl Viro unlock_mount_hash(); 12591da177e4SLinus Torvalds 1260863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 12611da177e4SLinus Torvalds return -EAGAIN; 12621da177e4SLinus Torvalds } 12631da177e4SLinus Torvalds 12641da177e4SLinus Torvalds /* 12651da177e4SLinus Torvalds * If we may have to abort operations to get out of this 12661da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 12671da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 12681da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 12691da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 12701da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 12711da177e4SLinus Torvalds * about for the moment. 12721da177e4SLinus Torvalds */ 12731da177e4SLinus Torvalds 127442faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 127542faad99SAl Viro sb->s_op->umount_begin(sb); 127642faad99SAl Viro } 12771da177e4SLinus Torvalds 12781da177e4SLinus Torvalds /* 12791da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 12801da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 12811da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 12821da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 12831da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 12841da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 12851da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 12861da177e4SLinus Torvalds */ 12871ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 12881da177e4SLinus Torvalds /* 12891da177e4SLinus Torvalds * Special case for "unmounting" root ... 12901da177e4SLinus Torvalds * we just try to remount it readonly. 12911da177e4SLinus Torvalds */ 12921da177e4SLinus Torvalds down_write(&sb->s_umount); 12934aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 12941da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 12951da177e4SLinus Torvalds up_write(&sb->s_umount); 12961da177e4SLinus Torvalds return retval; 12971da177e4SLinus Torvalds } 12981da177e4SLinus Torvalds 129997216be0SAl Viro namespace_lock(); 1300719ea2fbSAl Viro lock_mount_hash(); 13015addc5ddSAl Viro event++; 13021da177e4SLinus Torvalds 130348a066e7SAl Viro if (flags & MNT_DETACH) { 130448a066e7SAl Viro if (!list_empty(&mnt->mnt_list)) 130548a066e7SAl Viro umount_tree(mnt, 2); 130648a066e7SAl Viro retval = 0; 130748a066e7SAl Viro } else { 1308b54b9be7SAl Viro shrink_submounts(mnt); 13091da177e4SLinus Torvalds retval = -EBUSY; 131048a066e7SAl Viro if (!propagate_mount_busy(mnt, 2)) { 13111a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1312328e6d90SAl Viro umount_tree(mnt, 1); 13131da177e4SLinus Torvalds retval = 0; 13141da177e4SLinus Torvalds } 131548a066e7SAl Viro } 1316719ea2fbSAl Viro unlock_mount_hash(); 1317e3197d83SAl Viro namespace_unlock(); 13181da177e4SLinus Torvalds return retval; 13191da177e4SLinus Torvalds } 13201da177e4SLinus Torvalds 13211da177e4SLinus Torvalds /* 13229b40bc90SAl Viro * Is the caller allowed to modify his namespace? 13239b40bc90SAl Viro */ 13249b40bc90SAl Viro static inline bool may_mount(void) 13259b40bc90SAl Viro { 13269b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 13279b40bc90SAl Viro } 13289b40bc90SAl Viro 13299b40bc90SAl Viro /* 13301da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 13311da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 13321da177e4SLinus Torvalds * 13331da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 13341da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 13351da177e4SLinus Torvalds */ 13361da177e4SLinus Torvalds 1337bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 13381da177e4SLinus Torvalds { 13392d8f3038SAl Viro struct path path; 1340900148dcSAl Viro struct mount *mnt; 13411da177e4SLinus Torvalds int retval; 1342db1f05bbSMiklos Szeredi int lookup_flags = 0; 13431da177e4SLinus Torvalds 1344db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1345db1f05bbSMiklos Szeredi return -EINVAL; 1346db1f05bbSMiklos Szeredi 13479b40bc90SAl Viro if (!may_mount()) 13489b40bc90SAl Viro return -EPERM; 13499b40bc90SAl Viro 1350db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1351db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1352db1f05bbSMiklos Szeredi 1353197df04cSAl Viro retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path); 13541da177e4SLinus Torvalds if (retval) 13551da177e4SLinus Torvalds goto out; 1356900148dcSAl Viro mnt = real_mount(path.mnt); 13571da177e4SLinus Torvalds retval = -EINVAL; 13582d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 13591da177e4SLinus Torvalds goto dput_and_out; 1360143c8c91SAl Viro if (!check_mnt(mnt)) 13611da177e4SLinus Torvalds goto dput_and_out; 13625ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 13635ff9d8a6SEric W. Biederman goto dput_and_out; 13641da177e4SLinus Torvalds 1365900148dcSAl Viro retval = do_umount(mnt, flags); 13661da177e4SLinus Torvalds dput_and_out: 1367429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 13682d8f3038SAl Viro dput(path.dentry); 1369900148dcSAl Viro mntput_no_expire(mnt); 13701da177e4SLinus Torvalds out: 13711da177e4SLinus Torvalds return retval; 13721da177e4SLinus Torvalds } 13731da177e4SLinus Torvalds 13741da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 13751da177e4SLinus Torvalds 13761da177e4SLinus Torvalds /* 13771da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 13781da177e4SLinus Torvalds */ 1379bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 13801da177e4SLinus Torvalds { 13811da177e4SLinus Torvalds return sys_umount(name, 0); 13821da177e4SLinus Torvalds } 13831da177e4SLinus Torvalds 13841da177e4SLinus Torvalds #endif 13851da177e4SLinus Torvalds 13864ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 13878823c079SEric W. Biederman { 13884ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 13894ce5d2b1SEric W. Biederman struct inode *inode = dentry->d_inode; 13900bb80f24SDavid Howells struct proc_ns *ei; 13918823c079SEric W. Biederman 13928823c079SEric W. Biederman if (!proc_ns_inode(inode)) 13938823c079SEric W. Biederman return false; 13948823c079SEric W. Biederman 13950bb80f24SDavid Howells ei = get_proc_ns(inode); 13968823c079SEric W. Biederman if (ei->ns_ops != &mntns_operations) 13978823c079SEric W. Biederman return false; 13988823c079SEric W. Biederman 13994ce5d2b1SEric W. Biederman return true; 14004ce5d2b1SEric W. Biederman } 14014ce5d2b1SEric W. Biederman 14024ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 14034ce5d2b1SEric W. Biederman { 14044ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 14054ce5d2b1SEric W. Biederman * mount namespace loop? 14064ce5d2b1SEric W. Biederman */ 14074ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 14084ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 14094ce5d2b1SEric W. Biederman return false; 14104ce5d2b1SEric W. Biederman 14114ce5d2b1SEric W. Biederman mnt_ns = get_proc_ns(dentry->d_inode)->ns; 14128823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 14138823c079SEric W. Biederman } 14148823c079SEric W. Biederman 141587129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 141636341f64SRam Pai int flag) 14171da177e4SLinus Torvalds { 141884d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 14191da177e4SLinus Torvalds 14204ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 14214ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 14224ce5d2b1SEric W. Biederman 14234ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1424be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 14259676f0c6SRam Pai 142636341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1427be34d1a3SDavid Howells if (IS_ERR(q)) 1428be34d1a3SDavid Howells return q; 1429be34d1a3SDavid Howells 14305ff9d8a6SEric W. Biederman q->mnt.mnt_flags &= ~MNT_LOCKED; 1431a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 14321da177e4SLinus Torvalds 14331da177e4SLinus Torvalds p = mnt; 14346b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1435315fc83eSAl Viro struct mount *s; 14367ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 14371da177e4SLinus Torvalds continue; 14381da177e4SLinus Torvalds 1439909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 14404ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 14414ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 14424ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 14434ce5d2b1SEric W. Biederman continue; 14444ce5d2b1SEric W. Biederman } 14454ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 14464ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 14479676f0c6SRam Pai s = skip_mnt_tree(s); 14489676f0c6SRam Pai continue; 14499676f0c6SRam Pai } 14500714a533SAl Viro while (p != s->mnt_parent) { 14510714a533SAl Viro p = p->mnt_parent; 14520714a533SAl Viro q = q->mnt_parent; 14531da177e4SLinus Torvalds } 145487129cc0SAl Viro p = s; 145584d17192SAl Viro parent = q; 145687129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1457be34d1a3SDavid Howells if (IS_ERR(q)) 1458be34d1a3SDavid Howells goto out; 1459719ea2fbSAl Viro lock_mount_hash(); 14601a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 146184d17192SAl Viro attach_mnt(q, parent, p->mnt_mp); 1462719ea2fbSAl Viro unlock_mount_hash(); 14631da177e4SLinus Torvalds } 14641da177e4SLinus Torvalds } 14651da177e4SLinus Torvalds return res; 1466be34d1a3SDavid Howells out: 14671da177e4SLinus Torvalds if (res) { 1468719ea2fbSAl Viro lock_mount_hash(); 1469328e6d90SAl Viro umount_tree(res, 0); 1470719ea2fbSAl Viro unlock_mount_hash(); 14711da177e4SLinus Torvalds } 1472be34d1a3SDavid Howells return q; 14731da177e4SLinus Torvalds } 14741da177e4SLinus Torvalds 1475be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1476be34d1a3SDavid Howells 1477589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 14788aec0809SAl Viro { 1479cb338d06SAl Viro struct mount *tree; 148097216be0SAl Viro namespace_lock(); 148187129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 148287129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1483328e6d90SAl Viro namespace_unlock(); 1484be34d1a3SDavid Howells if (IS_ERR(tree)) 148552e220d3SDan Carpenter return ERR_CAST(tree); 1486be34d1a3SDavid Howells return &tree->mnt; 14878aec0809SAl Viro } 14888aec0809SAl Viro 14898aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 14908aec0809SAl Viro { 149197216be0SAl Viro namespace_lock(); 1492719ea2fbSAl Viro lock_mount_hash(); 1493328e6d90SAl Viro umount_tree(real_mount(mnt), 0); 1494719ea2fbSAl Viro unlock_mount_hash(); 14953ab6abeeSAl Viro namespace_unlock(); 14968aec0809SAl Viro } 14978aec0809SAl Viro 14981f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 14991f707137SAl Viro struct vfsmount *root) 15001f707137SAl Viro { 15011a4eeaf2SAl Viro struct mount *mnt; 15021f707137SAl Viro int res = f(root, arg); 15031f707137SAl Viro if (res) 15041f707137SAl Viro return res; 15051a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 15061a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 15071f707137SAl Viro if (res) 15081f707137SAl Viro return res; 15091f707137SAl Viro } 15101f707137SAl Viro return 0; 15111f707137SAl Viro } 15121f707137SAl Viro 15134b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1514719f5d7fSMiklos Szeredi { 1515315fc83eSAl Viro struct mount *p; 1516719f5d7fSMiklos Szeredi 1517909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1518fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 15194b8b21f4SAl Viro mnt_release_group_id(p); 1520719f5d7fSMiklos Szeredi } 1521719f5d7fSMiklos Szeredi } 1522719f5d7fSMiklos Szeredi 15234b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1524719f5d7fSMiklos Szeredi { 1525315fc83eSAl Viro struct mount *p; 1526719f5d7fSMiklos Szeredi 1527909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1528fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 15294b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1530719f5d7fSMiklos Szeredi if (err) { 15314b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1532719f5d7fSMiklos Szeredi return err; 1533719f5d7fSMiklos Szeredi } 1534719f5d7fSMiklos Szeredi } 1535719f5d7fSMiklos Szeredi } 1536719f5d7fSMiklos Szeredi 1537719f5d7fSMiklos Szeredi return 0; 1538719f5d7fSMiklos Szeredi } 1539719f5d7fSMiklos Szeredi 1540b90fa9aeSRam Pai /* 1541b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1542b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 154321444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 154421444403SRam Pai * store the parent mount and mountpoint dentry. 154521444403SRam Pai * (done when source_mnt is moved) 1546b90fa9aeSRam Pai * 1547b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1548b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 15499676f0c6SRam Pai * --------------------------------------------------------------------------- 1550b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 15519676f0c6SRam Pai * |************************************************************************** 15529676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15539676f0c6SRam Pai * | dest | | | | | 15549676f0c6SRam Pai * | | | | | | | 15559676f0c6SRam Pai * | v | | | | | 15569676f0c6SRam Pai * |************************************************************************** 15579676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 15585afe0022SRam Pai * | | | | | | 15599676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 15609676f0c6SRam Pai * *************************************************************************** 1561b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1562b90fa9aeSRam Pai * destination mount. 1563b90fa9aeSRam Pai * 1564b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1565b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1566b90fa9aeSRam Pai * the peer group of the source mount. 1567b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1568b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1569b90fa9aeSRam Pai * mount. 15705afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 15715afe0022SRam Pai * of the destination mount and the cloned mount is made slave 15725afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 15735afe0022SRam Pai * is marked as 'shared and slave'. 15745afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 15755afe0022SRam Pai * source mount. 15765afe0022SRam Pai * 15779676f0c6SRam Pai * --------------------------------------------------------------------------- 157821444403SRam Pai * | MOVE MOUNT OPERATION | 15799676f0c6SRam Pai * |************************************************************************** 15809676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15819676f0c6SRam Pai * | dest | | | | | 15829676f0c6SRam Pai * | | | | | | | 15839676f0c6SRam Pai * | v | | | | | 15849676f0c6SRam Pai * |************************************************************************** 15859676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 15865afe0022SRam Pai * | | | | | | 15879676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 15889676f0c6SRam Pai * *************************************************************************** 15895afe0022SRam Pai * 15905afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 15915afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 159221444403SRam Pai * (+*) the mount is moved to the destination. 15935afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 15945afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 15955afe0022SRam Pai * the mount is marked as 'shared and slave'. 15965afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1597b90fa9aeSRam Pai * 1598b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1599b90fa9aeSRam Pai * applied to each mount in the tree. 1600b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1601b90fa9aeSRam Pai * in allocations. 1602b90fa9aeSRam Pai */ 16030fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 160484d17192SAl Viro struct mount *dest_mnt, 160584d17192SAl Viro struct mountpoint *dest_mp, 160684d17192SAl Viro struct path *parent_path) 1607b90fa9aeSRam Pai { 1608b90fa9aeSRam Pai LIST_HEAD(tree_list); 1609315fc83eSAl Viro struct mount *child, *p; 1610719f5d7fSMiklos Szeredi int err; 1611b90fa9aeSRam Pai 1612fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 16130fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1614719f5d7fSMiklos Szeredi if (err) 1615719f5d7fSMiklos Szeredi goto out; 1616719f5d7fSMiklos Szeredi } 161784d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1618719f5d7fSMiklos Szeredi if (err) 1619719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1620b90fa9aeSRam Pai 1621719ea2fbSAl Viro lock_mount_hash(); 1622df1a1ad2SAl Viro 1623fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 1624909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 16250f0afb1dSAl Viro set_mnt_shared(p); 1626b90fa9aeSRam Pai } 16271a390689SAl Viro if (parent_path) { 16280fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 162984d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1630143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 163121444403SRam Pai } else { 163284d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 16330fb54e50SAl Viro commit_tree(source_mnt); 163421444403SRam Pai } 1635b90fa9aeSRam Pai 16361b8e5564SAl Viro list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 16371b8e5564SAl Viro list_del_init(&child->mnt_hash); 16384b2619a5SAl Viro commit_tree(child); 1639b90fa9aeSRam Pai } 1640719ea2fbSAl Viro unlock_mount_hash(); 164199b7db7bSNick Piggin 1642b90fa9aeSRam Pai return 0; 1643719f5d7fSMiklos Szeredi 1644719f5d7fSMiklos Szeredi out_cleanup_ids: 1645fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) 16460fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1647719f5d7fSMiklos Szeredi out: 1648719f5d7fSMiklos Szeredi return err; 1649b90fa9aeSRam Pai } 1650b90fa9aeSRam Pai 165184d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 1652b12cea91SAl Viro { 1653b12cea91SAl Viro struct vfsmount *mnt; 165484d17192SAl Viro struct dentry *dentry = path->dentry; 1655b12cea91SAl Viro retry: 165684d17192SAl Viro mutex_lock(&dentry->d_inode->i_mutex); 165784d17192SAl Viro if (unlikely(cant_mount(dentry))) { 165884d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 165984d17192SAl Viro return ERR_PTR(-ENOENT); 1660b12cea91SAl Viro } 166197216be0SAl Viro namespace_lock(); 1662b12cea91SAl Viro mnt = lookup_mnt(path); 166384d17192SAl Viro if (likely(!mnt)) { 166484d17192SAl Viro struct mountpoint *mp = new_mountpoint(dentry); 166584d17192SAl Viro if (IS_ERR(mp)) { 166697216be0SAl Viro namespace_unlock(); 166784d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 166884d17192SAl Viro return mp; 166984d17192SAl Viro } 167084d17192SAl Viro return mp; 167184d17192SAl Viro } 167297216be0SAl Viro namespace_unlock(); 1673b12cea91SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 1674b12cea91SAl Viro path_put(path); 1675b12cea91SAl Viro path->mnt = mnt; 167684d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 1677b12cea91SAl Viro goto retry; 1678b12cea91SAl Viro } 1679b12cea91SAl Viro 168084d17192SAl Viro static void unlock_mount(struct mountpoint *where) 1681b12cea91SAl Viro { 168284d17192SAl Viro struct dentry *dentry = where->m_dentry; 168384d17192SAl Viro put_mountpoint(where); 1684328e6d90SAl Viro namespace_unlock(); 168584d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 1686b12cea91SAl Viro } 1687b12cea91SAl Viro 168884d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 16891da177e4SLinus Torvalds { 169095bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 16911da177e4SLinus Torvalds return -EINVAL; 16921da177e4SLinus Torvalds 169384d17192SAl Viro if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != 169495bc5f25SAl Viro S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) 16951da177e4SLinus Torvalds return -ENOTDIR; 16961da177e4SLinus Torvalds 169784d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 16981da177e4SLinus Torvalds } 16991da177e4SLinus Torvalds 17001da177e4SLinus Torvalds /* 17017a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 17027a2e8a8fSValerie Aurora */ 17037a2e8a8fSValerie Aurora 17047a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 17057a2e8a8fSValerie Aurora { 17067c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 17077a2e8a8fSValerie Aurora 17087a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 17097a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 17107a2e8a8fSValerie Aurora return 0; 17117a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 17127a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 17137a2e8a8fSValerie Aurora return 0; 17147a2e8a8fSValerie Aurora return type; 17157a2e8a8fSValerie Aurora } 17167a2e8a8fSValerie Aurora 17177a2e8a8fSValerie Aurora /* 171807b20889SRam Pai * recursively change the type of the mountpoint. 171907b20889SRam Pai */ 17200a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 172107b20889SRam Pai { 1722315fc83eSAl Viro struct mount *m; 17234b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 172407b20889SRam Pai int recurse = flag & MS_REC; 17257a2e8a8fSValerie Aurora int type; 1726719f5d7fSMiklos Szeredi int err = 0; 172707b20889SRam Pai 17282d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 172907b20889SRam Pai return -EINVAL; 173007b20889SRam Pai 17317a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 17327a2e8a8fSValerie Aurora if (!type) 17337a2e8a8fSValerie Aurora return -EINVAL; 17347a2e8a8fSValerie Aurora 173597216be0SAl Viro namespace_lock(); 1736719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1737719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1738719f5d7fSMiklos Szeredi if (err) 1739719f5d7fSMiklos Szeredi goto out_unlock; 1740719f5d7fSMiklos Szeredi } 1741719f5d7fSMiklos Szeredi 1742719ea2fbSAl Viro lock_mount_hash(); 1743909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 17440f0afb1dSAl Viro change_mnt_propagation(m, type); 1745719ea2fbSAl Viro unlock_mount_hash(); 1746719f5d7fSMiklos Szeredi 1747719f5d7fSMiklos Szeredi out_unlock: 174897216be0SAl Viro namespace_unlock(); 1749719f5d7fSMiklos Szeredi return err; 175007b20889SRam Pai } 175107b20889SRam Pai 17525ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 17535ff9d8a6SEric W. Biederman { 17545ff9d8a6SEric W. Biederman struct mount *child; 17555ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 17565ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 17575ff9d8a6SEric W. Biederman continue; 17585ff9d8a6SEric W. Biederman 17595ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 17605ff9d8a6SEric W. Biederman return true; 17615ff9d8a6SEric W. Biederman } 17625ff9d8a6SEric W. Biederman return false; 17635ff9d8a6SEric W. Biederman } 17645ff9d8a6SEric W. Biederman 176507b20889SRam Pai /* 17661da177e4SLinus Torvalds * do loopback mount. 17671da177e4SLinus Torvalds */ 1768808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 17692dafe1c4SEric Sandeen int recurse) 17701da177e4SLinus Torvalds { 17712d92ab3cSAl Viro struct path old_path; 177284d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 177384d17192SAl Viro struct mountpoint *mp; 177457eccb83SAl Viro int err; 17751da177e4SLinus Torvalds if (!old_name || !*old_name) 17761da177e4SLinus Torvalds return -EINVAL; 1777815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 17781da177e4SLinus Torvalds if (err) 17791da177e4SLinus Torvalds return err; 17801da177e4SLinus Torvalds 17818823c079SEric W. Biederman err = -EINVAL; 17824ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 17838823c079SEric W. Biederman goto out; 17848823c079SEric W. Biederman 178584d17192SAl Viro mp = lock_mount(path); 178684d17192SAl Viro err = PTR_ERR(mp); 178784d17192SAl Viro if (IS_ERR(mp)) 17889676f0c6SRam Pai goto out; 17899676f0c6SRam Pai 179087129cc0SAl Viro old = real_mount(old_path.mnt); 179184d17192SAl Viro parent = real_mount(path->mnt); 179287129cc0SAl Viro 1793b12cea91SAl Viro err = -EINVAL; 1794fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 1795b12cea91SAl Viro goto out2; 1796b12cea91SAl Viro 179784d17192SAl Viro if (!check_mnt(parent) || !check_mnt(old)) 1798b12cea91SAl Viro goto out2; 1799ccd48bc7SAl Viro 18005ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 18015ff9d8a6SEric W. Biederman goto out2; 18025ff9d8a6SEric W. Biederman 18031da177e4SLinus Torvalds if (recurse) 18044ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 18051da177e4SLinus Torvalds else 180687129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 18071da177e4SLinus Torvalds 1808be34d1a3SDavid Howells if (IS_ERR(mnt)) { 1809be34d1a3SDavid Howells err = PTR_ERR(mnt); 1810e9c5d8a5SAndrey Vagin goto out2; 1811be34d1a3SDavid Howells } 1812ccd48bc7SAl Viro 18135ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 18145ff9d8a6SEric W. Biederman 181584d17192SAl Viro err = graft_tree(mnt, parent, mp); 18161da177e4SLinus Torvalds if (err) { 1817719ea2fbSAl Viro lock_mount_hash(); 1818328e6d90SAl Viro umount_tree(mnt, 0); 1819719ea2fbSAl Viro unlock_mount_hash(); 18205b83d2c5SRam Pai } 1821b12cea91SAl Viro out2: 182284d17192SAl Viro unlock_mount(mp); 1823ccd48bc7SAl Viro out: 18242d92ab3cSAl Viro path_put(&old_path); 18251da177e4SLinus Torvalds return err; 18261da177e4SLinus Torvalds } 18271da177e4SLinus Torvalds 18282e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 18292e4b7fcdSDave Hansen { 18302e4b7fcdSDave Hansen int error = 0; 18312e4b7fcdSDave Hansen int readonly_request = 0; 18322e4b7fcdSDave Hansen 18332e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 18342e4b7fcdSDave Hansen readonly_request = 1; 18352e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 18362e4b7fcdSDave Hansen return 0; 18372e4b7fcdSDave Hansen 183890563b19SEric W. Biederman if (mnt->mnt_flags & MNT_LOCK_READONLY) 183990563b19SEric W. Biederman return -EPERM; 184090563b19SEric W. Biederman 18412e4b7fcdSDave Hansen if (readonly_request) 184283adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 18432e4b7fcdSDave Hansen else 184483adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 18452e4b7fcdSDave Hansen return error; 18462e4b7fcdSDave Hansen } 18472e4b7fcdSDave Hansen 18481da177e4SLinus Torvalds /* 18491da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 18501da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 18511da177e4SLinus Torvalds * on it - tough luck. 18521da177e4SLinus Torvalds */ 18530a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 18541da177e4SLinus Torvalds void *data) 18551da177e4SLinus Torvalds { 18561da177e4SLinus Torvalds int err; 18572d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 1858143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 18591da177e4SLinus Torvalds 1860143c8c91SAl Viro if (!check_mnt(mnt)) 18611da177e4SLinus Torvalds return -EINVAL; 18621da177e4SLinus Torvalds 18632d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 18641da177e4SLinus Torvalds return -EINVAL; 18651da177e4SLinus Torvalds 1866ff36fe2cSEric Paris err = security_sb_remount(sb, data); 1867ff36fe2cSEric Paris if (err) 1868ff36fe2cSEric Paris return err; 1869ff36fe2cSEric Paris 18701da177e4SLinus Torvalds down_write(&sb->s_umount); 18712e4b7fcdSDave Hansen if (flags & MS_BIND) 18722d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 187357eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 187457eccb83SAl Viro err = -EPERM; 18754aa98cf7SAl Viro else 18761da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 18777b43a79fSAl Viro if (!err) { 1878719ea2fbSAl Viro lock_mount_hash(); 1879143c8c91SAl Viro mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; 1880143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 1881143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1882719ea2fbSAl Viro unlock_mount_hash(); 18830e55a7ccSDan Williams } 18846339dab8SAl Viro up_write(&sb->s_umount); 18851da177e4SLinus Torvalds return err; 18861da177e4SLinus Torvalds } 18871da177e4SLinus Torvalds 1888cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 18899676f0c6SRam Pai { 1890315fc83eSAl Viro struct mount *p; 1891909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 1892fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 18939676f0c6SRam Pai return 1; 18949676f0c6SRam Pai } 18959676f0c6SRam Pai return 0; 18969676f0c6SRam Pai } 18979676f0c6SRam Pai 1898808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 18991da177e4SLinus Torvalds { 19002d92ab3cSAl Viro struct path old_path, parent_path; 1901676da58dSAl Viro struct mount *p; 19020fb54e50SAl Viro struct mount *old; 190384d17192SAl Viro struct mountpoint *mp; 190457eccb83SAl Viro int err; 19051da177e4SLinus Torvalds if (!old_name || !*old_name) 19061da177e4SLinus Torvalds return -EINVAL; 19072d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 19081da177e4SLinus Torvalds if (err) 19091da177e4SLinus Torvalds return err; 19101da177e4SLinus Torvalds 191184d17192SAl Viro mp = lock_mount(path); 191284d17192SAl Viro err = PTR_ERR(mp); 191384d17192SAl Viro if (IS_ERR(mp)) 1914cc53ce53SDavid Howells goto out; 1915cc53ce53SDavid Howells 1916143c8c91SAl Viro old = real_mount(old_path.mnt); 1917fc7be130SAl Viro p = real_mount(path->mnt); 1918143c8c91SAl Viro 19191da177e4SLinus Torvalds err = -EINVAL; 1920fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 19211da177e4SLinus Torvalds goto out1; 19221da177e4SLinus Torvalds 19235ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 19245ff9d8a6SEric W. Biederman goto out1; 19255ff9d8a6SEric W. Biederman 19261da177e4SLinus Torvalds err = -EINVAL; 19272d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 192821444403SRam Pai goto out1; 19291da177e4SLinus Torvalds 1930676da58dSAl Viro if (!mnt_has_parent(old)) 193121444403SRam Pai goto out1; 19321da177e4SLinus Torvalds 19332d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 19342d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 193521444403SRam Pai goto out1; 193621444403SRam Pai /* 193721444403SRam Pai * Don't move a mount residing in a shared parent. 193821444403SRam Pai */ 1939fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 194021444403SRam Pai goto out1; 19419676f0c6SRam Pai /* 19429676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 19439676f0c6SRam Pai * mount which is shared. 19449676f0c6SRam Pai */ 1945fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 19469676f0c6SRam Pai goto out1; 19471da177e4SLinus Torvalds err = -ELOOP; 1948fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 1949676da58dSAl Viro if (p == old) 195021444403SRam Pai goto out1; 19511da177e4SLinus Torvalds 195284d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 19534ac91378SJan Blunck if (err) 195421444403SRam Pai goto out1; 19551da177e4SLinus Torvalds 19561da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 19571da177e4SLinus Torvalds * automatically */ 19586776db3dSAl Viro list_del_init(&old->mnt_expire); 19591da177e4SLinus Torvalds out1: 196084d17192SAl Viro unlock_mount(mp); 19611da177e4SLinus Torvalds out: 19621da177e4SLinus Torvalds if (!err) 19631a390689SAl Viro path_put(&parent_path); 19642d92ab3cSAl Viro path_put(&old_path); 19651da177e4SLinus Torvalds return err; 19661da177e4SLinus Torvalds } 19671da177e4SLinus Torvalds 19689d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 19699d412a43SAl Viro { 19709d412a43SAl Viro int err; 19719d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 19729d412a43SAl Viro if (subtype) { 19739d412a43SAl Viro subtype++; 19749d412a43SAl Viro err = -EINVAL; 19759d412a43SAl Viro if (!subtype[0]) 19769d412a43SAl Viro goto err; 19779d412a43SAl Viro } else 19789d412a43SAl Viro subtype = ""; 19799d412a43SAl Viro 19809d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 19819d412a43SAl Viro err = -ENOMEM; 19829d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 19839d412a43SAl Viro goto err; 19849d412a43SAl Viro return mnt; 19859d412a43SAl Viro 19869d412a43SAl Viro err: 19879d412a43SAl Viro mntput(mnt); 19889d412a43SAl Viro return ERR_PTR(err); 19899d412a43SAl Viro } 19909d412a43SAl Viro 19919d412a43SAl Viro /* 19929d412a43SAl Viro * add a mount into a namespace's mount tree 19939d412a43SAl Viro */ 199495bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 19959d412a43SAl Viro { 199684d17192SAl Viro struct mountpoint *mp; 199784d17192SAl Viro struct mount *parent; 19989d412a43SAl Viro int err; 19999d412a43SAl Viro 200048a066e7SAl Viro mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL | MNT_DOOMED | MNT_SYNC_UMOUNT); 20019d412a43SAl Viro 200284d17192SAl Viro mp = lock_mount(path); 200384d17192SAl Viro if (IS_ERR(mp)) 200484d17192SAl Viro return PTR_ERR(mp); 20059d412a43SAl Viro 200684d17192SAl Viro parent = real_mount(path->mnt); 20079d412a43SAl Viro err = -EINVAL; 200884d17192SAl Viro if (unlikely(!check_mnt(parent))) { 2009156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 2010156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 20119d412a43SAl Viro goto unlock; 2012156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 201384d17192SAl Viro if (!parent->mnt_ns) 2014156cacb1SAl Viro goto unlock; 2015156cacb1SAl Viro } 20169d412a43SAl Viro 20179d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 20189d412a43SAl Viro err = -EBUSY; 201995bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 20209d412a43SAl Viro path->mnt->mnt_root == path->dentry) 20219d412a43SAl Viro goto unlock; 20229d412a43SAl Viro 20239d412a43SAl Viro err = -EINVAL; 202495bc5f25SAl Viro if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) 20259d412a43SAl Viro goto unlock; 20269d412a43SAl Viro 202795bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 202884d17192SAl Viro err = graft_tree(newmnt, parent, mp); 20299d412a43SAl Viro 20309d412a43SAl Viro unlock: 203184d17192SAl Viro unlock_mount(mp); 20329d412a43SAl Viro return err; 20339d412a43SAl Viro } 2034b1e75df4SAl Viro 20351da177e4SLinus Torvalds /* 20361da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 20371da177e4SLinus Torvalds * namespace's tree 20381da177e4SLinus Torvalds */ 20390c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2040808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 20411da177e4SLinus Torvalds { 20420c55cfc4SEric W. Biederman struct file_system_type *type; 20439b40bc90SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 20441da177e4SLinus Torvalds struct vfsmount *mnt; 204515f9a3f3SAl Viro int err; 20461da177e4SLinus Torvalds 20470c55cfc4SEric W. Biederman if (!fstype) 20481da177e4SLinus Torvalds return -EINVAL; 20491da177e4SLinus Torvalds 20500c55cfc4SEric W. Biederman type = get_fs_type(fstype); 20510c55cfc4SEric W. Biederman if (!type) 20520c55cfc4SEric W. Biederman return -ENODEV; 20530c55cfc4SEric W. Biederman 20540c55cfc4SEric W. Biederman if (user_ns != &init_user_ns) { 20550c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_MOUNT)) { 20560c55cfc4SEric W. Biederman put_filesystem(type); 20570c55cfc4SEric W. Biederman return -EPERM; 20580c55cfc4SEric W. Biederman } 20590c55cfc4SEric W. Biederman /* Only in special cases allow devices from mounts 20600c55cfc4SEric W. Biederman * created outside the initial user namespace. 20610c55cfc4SEric W. Biederman */ 20620c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { 20630c55cfc4SEric W. Biederman flags |= MS_NODEV; 20640c55cfc4SEric W. Biederman mnt_flags |= MNT_NODEV; 20650c55cfc4SEric W. Biederman } 20660c55cfc4SEric W. Biederman } 20670c55cfc4SEric W. Biederman 20680c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 20690c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 20700c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 20710c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 20720c55cfc4SEric W. Biederman 20730c55cfc4SEric W. Biederman put_filesystem(type); 20741da177e4SLinus Torvalds if (IS_ERR(mnt)) 20751da177e4SLinus Torvalds return PTR_ERR(mnt); 20761da177e4SLinus Torvalds 207795bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 207815f9a3f3SAl Viro if (err) 207915f9a3f3SAl Viro mntput(mnt); 208015f9a3f3SAl Viro return err; 20811da177e4SLinus Torvalds } 20821da177e4SLinus Torvalds 208319a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 208419a167afSAl Viro { 20856776db3dSAl Viro struct mount *mnt = real_mount(m); 208619a167afSAl Viro int err; 208719a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 208819a167afSAl Viro * expired before we get a chance to add it 208919a167afSAl Viro */ 20906776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 209119a167afSAl Viro 209219a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 209319a167afSAl Viro m->mnt_root == path->dentry) { 2094b1e75df4SAl Viro err = -ELOOP; 2095b1e75df4SAl Viro goto fail; 209619a167afSAl Viro } 209719a167afSAl Viro 209895bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2099b1e75df4SAl Viro if (!err) 2100b1e75df4SAl Viro return 0; 2101b1e75df4SAl Viro fail: 2102b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 21036776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 210497216be0SAl Viro namespace_lock(); 21056776db3dSAl Viro list_del_init(&mnt->mnt_expire); 210697216be0SAl Viro namespace_unlock(); 210719a167afSAl Viro } 2108b1e75df4SAl Viro mntput(m); 2109b1e75df4SAl Viro mntput(m); 211019a167afSAl Viro return err; 211119a167afSAl Viro } 211219a167afSAl Viro 2113ea5b778aSDavid Howells /** 2114ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2115ea5b778aSDavid Howells * @mnt: The mount to list. 2116ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2117ea5b778aSDavid Howells */ 2118ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2119ea5b778aSDavid Howells { 212097216be0SAl Viro namespace_lock(); 2121ea5b778aSDavid Howells 21226776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2123ea5b778aSDavid Howells 212497216be0SAl Viro namespace_unlock(); 2125ea5b778aSDavid Howells } 2126ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2127ea5b778aSDavid Howells 2128ea5b778aSDavid Howells /* 21291da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 21301da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 21311da177e4SLinus Torvalds * here 21321da177e4SLinus Torvalds */ 21331da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 21341da177e4SLinus Torvalds { 2135761d5c38SAl Viro struct mount *mnt, *next; 21361da177e4SLinus Torvalds LIST_HEAD(graveyard); 21371da177e4SLinus Torvalds 21381da177e4SLinus Torvalds if (list_empty(mounts)) 21391da177e4SLinus Torvalds return; 21401da177e4SLinus Torvalds 214197216be0SAl Viro namespace_lock(); 2142719ea2fbSAl Viro lock_mount_hash(); 21431da177e4SLinus Torvalds 21441da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 21451da177e4SLinus Torvalds * following criteria: 21461da177e4SLinus Torvalds * - only referenced by its parent vfsmount 21471da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 21481da177e4SLinus Torvalds * cleared by mntput()) 21491da177e4SLinus Torvalds */ 21506776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2151863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 21521ab59738SAl Viro propagate_mount_busy(mnt, 1)) 21531da177e4SLinus Torvalds continue; 21546776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 21551da177e4SLinus Torvalds } 2156bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 21576776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2158143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2159328e6d90SAl Viro umount_tree(mnt, 1); 2160bcc5c7d2SAl Viro } 2161719ea2fbSAl Viro unlock_mount_hash(); 21623ab6abeeSAl Viro namespace_unlock(); 21631da177e4SLinus Torvalds } 21641da177e4SLinus Torvalds 21651da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 21661da177e4SLinus Torvalds 21671da177e4SLinus Torvalds /* 21685528f911STrond Myklebust * Ripoff of 'select_parent()' 21695528f911STrond Myklebust * 21705528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 21715528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 21725528f911STrond Myklebust */ 2173692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 21745528f911STrond Myklebust { 2175692afc31SAl Viro struct mount *this_parent = parent; 21765528f911STrond Myklebust struct list_head *next; 21775528f911STrond Myklebust int found = 0; 21785528f911STrond Myklebust 21795528f911STrond Myklebust repeat: 21806b41d536SAl Viro next = this_parent->mnt_mounts.next; 21815528f911STrond Myklebust resume: 21826b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 21835528f911STrond Myklebust struct list_head *tmp = next; 21846b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 21855528f911STrond Myklebust 21865528f911STrond Myklebust next = tmp->next; 2187692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 21885528f911STrond Myklebust continue; 21895528f911STrond Myklebust /* 21905528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 21915528f911STrond Myklebust */ 21926b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 21935528f911STrond Myklebust this_parent = mnt; 21945528f911STrond Myklebust goto repeat; 21955528f911STrond Myklebust } 21965528f911STrond Myklebust 21971ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 21986776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 21995528f911STrond Myklebust found++; 22005528f911STrond Myklebust } 22015528f911STrond Myklebust } 22025528f911STrond Myklebust /* 22035528f911STrond Myklebust * All done at this level ... ascend and resume the search 22045528f911STrond Myklebust */ 22055528f911STrond Myklebust if (this_parent != parent) { 22066b41d536SAl Viro next = this_parent->mnt_child.next; 22070714a533SAl Viro this_parent = this_parent->mnt_parent; 22085528f911STrond Myklebust goto resume; 22095528f911STrond Myklebust } 22105528f911STrond Myklebust return found; 22115528f911STrond Myklebust } 22125528f911STrond Myklebust 22135528f911STrond Myklebust /* 22145528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 22155528f911STrond Myklebust * submounts of a specific parent mountpoint 221699b7db7bSNick Piggin * 221748a066e7SAl Viro * mount_lock must be held for write 22185528f911STrond Myklebust */ 2219b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 22205528f911STrond Myklebust { 22215528f911STrond Myklebust LIST_HEAD(graveyard); 2222761d5c38SAl Viro struct mount *m; 22235528f911STrond Myklebust 22245528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2225c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2226bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2227761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 22286776db3dSAl Viro mnt_expire); 2229143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2230328e6d90SAl Viro umount_tree(m, 1); 2231bcc5c7d2SAl Viro } 2232bcc5c7d2SAl Viro } 22335528f911STrond Myklebust } 22345528f911STrond Myklebust 22355528f911STrond Myklebust /* 22361da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 22371da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 22381da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 22391da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 22401da177e4SLinus Torvalds */ 2241b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2242b58fed8bSRam Pai unsigned long n) 22431da177e4SLinus Torvalds { 22441da177e4SLinus Torvalds char *t = to; 22451da177e4SLinus Torvalds const char __user *f = from; 22461da177e4SLinus Torvalds char c; 22471da177e4SLinus Torvalds 22481da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 22491da177e4SLinus Torvalds return n; 22501da177e4SLinus Torvalds 22511da177e4SLinus Torvalds while (n) { 22521da177e4SLinus Torvalds if (__get_user(c, f)) { 22531da177e4SLinus Torvalds memset(t, 0, n); 22541da177e4SLinus Torvalds break; 22551da177e4SLinus Torvalds } 22561da177e4SLinus Torvalds *t++ = c; 22571da177e4SLinus Torvalds f++; 22581da177e4SLinus Torvalds n--; 22591da177e4SLinus Torvalds } 22601da177e4SLinus Torvalds return n; 22611da177e4SLinus Torvalds } 22621da177e4SLinus Torvalds 22631da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 22641da177e4SLinus Torvalds { 22651da177e4SLinus Torvalds int i; 22661da177e4SLinus Torvalds unsigned long page; 22671da177e4SLinus Torvalds unsigned long size; 22681da177e4SLinus Torvalds 22691da177e4SLinus Torvalds *where = 0; 22701da177e4SLinus Torvalds if (!data) 22711da177e4SLinus Torvalds return 0; 22721da177e4SLinus Torvalds 22731da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 22741da177e4SLinus Torvalds return -ENOMEM; 22751da177e4SLinus Torvalds 22761da177e4SLinus Torvalds /* We only care that *some* data at the address the user 22771da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 22781da177e4SLinus Torvalds * the remainder of the page. 22791da177e4SLinus Torvalds */ 22801da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 22811da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 22821da177e4SLinus Torvalds if (size > PAGE_SIZE) 22831da177e4SLinus Torvalds size = PAGE_SIZE; 22841da177e4SLinus Torvalds 22851da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 22861da177e4SLinus Torvalds if (!i) { 22871da177e4SLinus Torvalds free_page(page); 22881da177e4SLinus Torvalds return -EFAULT; 22891da177e4SLinus Torvalds } 22901da177e4SLinus Torvalds if (i != PAGE_SIZE) 22911da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 22921da177e4SLinus Torvalds *where = page; 22931da177e4SLinus Torvalds return 0; 22941da177e4SLinus Torvalds } 22951da177e4SLinus Torvalds 2296eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 2297eca6f534SVegard Nossum { 2298eca6f534SVegard Nossum char *tmp; 2299eca6f534SVegard Nossum 2300eca6f534SVegard Nossum if (!data) { 2301eca6f534SVegard Nossum *where = NULL; 2302eca6f534SVegard Nossum return 0; 2303eca6f534SVegard Nossum } 2304eca6f534SVegard Nossum 2305eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 2306eca6f534SVegard Nossum if (IS_ERR(tmp)) 2307eca6f534SVegard Nossum return PTR_ERR(tmp); 2308eca6f534SVegard Nossum 2309eca6f534SVegard Nossum *where = tmp; 2310eca6f534SVegard Nossum return 0; 2311eca6f534SVegard Nossum } 2312eca6f534SVegard Nossum 23131da177e4SLinus Torvalds /* 23141da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 23151da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 23161da177e4SLinus Torvalds * 23171da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 23181da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 23191da177e4SLinus Torvalds * information (or be NULL). 23201da177e4SLinus Torvalds * 23211da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 23221da177e4SLinus Torvalds * When the flags word was introduced its top half was required 23231da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 23241da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 23251da177e4SLinus Torvalds * and must be discarded. 23261da177e4SLinus Torvalds */ 2327808d4e3cSAl Viro long do_mount(const char *dev_name, const char *dir_name, 2328808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 23291da177e4SLinus Torvalds { 23302d92ab3cSAl Viro struct path path; 23311da177e4SLinus Torvalds int retval = 0; 23321da177e4SLinus Torvalds int mnt_flags = 0; 23331da177e4SLinus Torvalds 23341da177e4SLinus Torvalds /* Discard magic */ 23351da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 23361da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 23371da177e4SLinus Torvalds 23381da177e4SLinus Torvalds /* Basic sanity checks */ 23391da177e4SLinus Torvalds 23401da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 23411da177e4SLinus Torvalds return -EINVAL; 23421da177e4SLinus Torvalds 23431da177e4SLinus Torvalds if (data_page) 23441da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 23451da177e4SLinus Torvalds 2346a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 2347a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2348a27ab9f2STetsuo Handa if (retval) 2349a27ab9f2STetsuo Handa return retval; 2350a27ab9f2STetsuo Handa 2351a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2352a27ab9f2STetsuo Handa type_page, flags, data_page); 23530d5cadb8SAl Viro if (!retval && !may_mount()) 23540d5cadb8SAl Viro retval = -EPERM; 2355a27ab9f2STetsuo Handa if (retval) 2356a27ab9f2STetsuo Handa goto dput_out; 2357a27ab9f2STetsuo Handa 2358613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2359613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 23600a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 23610a1c01c9SMatthew Garrett 23621da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 23631da177e4SLinus Torvalds if (flags & MS_NOSUID) 23641da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 23651da177e4SLinus Torvalds if (flags & MS_NODEV) 23661da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 23671da177e4SLinus Torvalds if (flags & MS_NOEXEC) 23681da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2369fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2370fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2371fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2372fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2373d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2374d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 23752e4b7fcdSDave Hansen if (flags & MS_RDONLY) 23762e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2377fc33a7bbSChristoph Hellwig 23787a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2379d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2380d0adde57SMatthew Garrett MS_STRICTATIME); 23811da177e4SLinus Torvalds 23821da177e4SLinus Torvalds if (flags & MS_REMOUNT) 23832d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 23841da177e4SLinus Torvalds data_page); 23851da177e4SLinus Torvalds else if (flags & MS_BIND) 23862d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 23879676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 23882d92ab3cSAl Viro retval = do_change_type(&path, flags); 23891da177e4SLinus Torvalds else if (flags & MS_MOVE) 23902d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 23911da177e4SLinus Torvalds else 23922d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 23931da177e4SLinus Torvalds dev_name, data_page); 23941da177e4SLinus Torvalds dput_out: 23952d92ab3cSAl Viro path_put(&path); 23961da177e4SLinus Torvalds return retval; 23971da177e4SLinus Torvalds } 23981da177e4SLinus Torvalds 2399771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2400771b1371SEric W. Biederman { 240198f842e6SEric W. Biederman proc_free_inum(ns->proc_inum); 2402771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2403771b1371SEric W. Biederman kfree(ns); 2404771b1371SEric W. Biederman } 2405771b1371SEric W. Biederman 24068823c079SEric W. Biederman /* 24078823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 24088823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 24098823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 24108823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 24118823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 24128823c079SEric W. Biederman */ 24138823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 24148823c079SEric W. Biederman 2415771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2416cf8d2c11STrond Myklebust { 2417cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 241898f842e6SEric W. Biederman int ret; 2419cf8d2c11STrond Myklebust 2420cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2421cf8d2c11STrond Myklebust if (!new_ns) 2422cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 242398f842e6SEric W. Biederman ret = proc_alloc_inum(&new_ns->proc_inum); 242498f842e6SEric W. Biederman if (ret) { 242598f842e6SEric W. Biederman kfree(new_ns); 242698f842e6SEric W. Biederman return ERR_PTR(ret); 242798f842e6SEric W. Biederman } 24288823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2429cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2430cf8d2c11STrond Myklebust new_ns->root = NULL; 2431cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2432cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2433cf8d2c11STrond Myklebust new_ns->event = 0; 2434771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2435cf8d2c11STrond Myklebust return new_ns; 2436cf8d2c11STrond Myklebust } 2437cf8d2c11STrond Myklebust 24389559f689SAl Viro struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 24399559f689SAl Viro struct user_namespace *user_ns, struct fs_struct *new_fs) 24401da177e4SLinus Torvalds { 24416b3286edSKirill Korotaev struct mnt_namespace *new_ns; 24427f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2443315fc83eSAl Viro struct mount *p, *q; 24449559f689SAl Viro struct mount *old; 2445cb338d06SAl Viro struct mount *new; 24467a472ef4SEric W. Biederman int copy_flags; 24471da177e4SLinus Torvalds 24489559f689SAl Viro BUG_ON(!ns); 24499559f689SAl Viro 24509559f689SAl Viro if (likely(!(flags & CLONE_NEWNS))) { 24519559f689SAl Viro get_mnt_ns(ns); 24529559f689SAl Viro return ns; 24539559f689SAl Viro } 24549559f689SAl Viro 24559559f689SAl Viro old = ns->root; 24569559f689SAl Viro 2457771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2458cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2459cf8d2c11STrond Myklebust return new_ns; 24601da177e4SLinus Torvalds 246197216be0SAl Viro namespace_lock(); 24621da177e4SLinus Torvalds /* First pass: copy the tree topology */ 24634ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 24649559f689SAl Viro if (user_ns != ns->user_ns) 2465132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 24667a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2467be34d1a3SDavid Howells if (IS_ERR(new)) { 2468328e6d90SAl Viro namespace_unlock(); 2469771b1371SEric W. Biederman free_mnt_ns(new_ns); 2470be34d1a3SDavid Howells return ERR_CAST(new); 24711da177e4SLinus Torvalds } 2472be08d6d2SAl Viro new_ns->root = new; 24731a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 24741da177e4SLinus Torvalds 24751da177e4SLinus Torvalds /* 24761da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 24771da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 24781da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 24791da177e4SLinus Torvalds */ 2480909b0a88SAl Viro p = old; 2481cb338d06SAl Viro q = new; 24821da177e4SLinus Torvalds while (p) { 2483143c8c91SAl Viro q->mnt_ns = new_ns; 24849559f689SAl Viro if (new_fs) { 24859559f689SAl Viro if (&p->mnt == new_fs->root.mnt) { 24869559f689SAl Viro new_fs->root.mnt = mntget(&q->mnt); 2487315fc83eSAl Viro rootmnt = &p->mnt; 24881da177e4SLinus Torvalds } 24899559f689SAl Viro if (&p->mnt == new_fs->pwd.mnt) { 24909559f689SAl Viro new_fs->pwd.mnt = mntget(&q->mnt); 2491315fc83eSAl Viro pwdmnt = &p->mnt; 24921da177e4SLinus Torvalds } 24931da177e4SLinus Torvalds } 2494909b0a88SAl Viro p = next_mnt(p, old); 2495909b0a88SAl Viro q = next_mnt(q, new); 24964ce5d2b1SEric W. Biederman if (!q) 24974ce5d2b1SEric W. Biederman break; 24984ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 24994ce5d2b1SEric W. Biederman p = next_mnt(p, old); 25001da177e4SLinus Torvalds } 2501328e6d90SAl Viro namespace_unlock(); 25021da177e4SLinus Torvalds 25031da177e4SLinus Torvalds if (rootmnt) 2504f03c6599SAl Viro mntput(rootmnt); 25051da177e4SLinus Torvalds if (pwdmnt) 2506f03c6599SAl Viro mntput(pwdmnt); 25071da177e4SLinus Torvalds 2508741a2951SJANAK DESAI return new_ns; 2509741a2951SJANAK DESAI } 2510741a2951SJANAK DESAI 2511cf8d2c11STrond Myklebust /** 2512cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2513cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2514cf8d2c11STrond Myklebust */ 25151a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2516cf8d2c11STrond Myklebust { 2517771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2518cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 25191a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 25201a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2521be08d6d2SAl Viro new_ns->root = mnt; 2522b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2523c1334495SAl Viro } else { 25241a4eeaf2SAl Viro mntput(m); 2525cf8d2c11STrond Myklebust } 2526cf8d2c11STrond Myklebust return new_ns; 2527cf8d2c11STrond Myklebust } 2528cf8d2c11STrond Myklebust 2529ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2530ea441d11SAl Viro { 2531ea441d11SAl Viro struct mnt_namespace *ns; 2532d31da0f0SAl Viro struct super_block *s; 2533ea441d11SAl Viro struct path path; 2534ea441d11SAl Viro int err; 2535ea441d11SAl Viro 2536ea441d11SAl Viro ns = create_mnt_ns(mnt); 2537ea441d11SAl Viro if (IS_ERR(ns)) 2538ea441d11SAl Viro return ERR_CAST(ns); 2539ea441d11SAl Viro 2540ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2541ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2542ea441d11SAl Viro 2543ea441d11SAl Viro put_mnt_ns(ns); 2544ea441d11SAl Viro 2545ea441d11SAl Viro if (err) 2546ea441d11SAl Viro return ERR_PTR(err); 2547ea441d11SAl Viro 2548ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2549d31da0f0SAl Viro s = path.mnt->mnt_sb; 2550d31da0f0SAl Viro atomic_inc(&s->s_active); 2551ea441d11SAl Viro mntput(path.mnt); 2552ea441d11SAl Viro /* lock the sucker */ 2553d31da0f0SAl Viro down_write(&s->s_umount); 2554ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2555ea441d11SAl Viro return path.dentry; 2556ea441d11SAl Viro } 2557ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2558ea441d11SAl Viro 2559bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2560bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 25611da177e4SLinus Torvalds { 2562eca6f534SVegard Nossum int ret; 2563eca6f534SVegard Nossum char *kernel_type; 256491a27b2aSJeff Layton struct filename *kernel_dir; 2565eca6f534SVegard Nossum char *kernel_dev; 25661da177e4SLinus Torvalds unsigned long data_page; 25671da177e4SLinus Torvalds 2568eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2569eca6f534SVegard Nossum if (ret < 0) 2570eca6f534SVegard Nossum goto out_type; 25711da177e4SLinus Torvalds 2572eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2573eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2574eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2575eca6f534SVegard Nossum goto out_dir; 2576eca6f534SVegard Nossum } 25771da177e4SLinus Torvalds 2578eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2579eca6f534SVegard Nossum if (ret < 0) 2580eca6f534SVegard Nossum goto out_dev; 25811da177e4SLinus Torvalds 2582eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2583eca6f534SVegard Nossum if (ret < 0) 2584eca6f534SVegard Nossum goto out_data; 25851da177e4SLinus Torvalds 258691a27b2aSJeff Layton ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, 2587eca6f534SVegard Nossum (void *) data_page); 2588eca6f534SVegard Nossum 25891da177e4SLinus Torvalds free_page(data_page); 2590eca6f534SVegard Nossum out_data: 2591eca6f534SVegard Nossum kfree(kernel_dev); 2592eca6f534SVegard Nossum out_dev: 2593eca6f534SVegard Nossum putname(kernel_dir); 2594eca6f534SVegard Nossum out_dir: 2595eca6f534SVegard Nossum kfree(kernel_type); 2596eca6f534SVegard Nossum out_type: 2597eca6f534SVegard Nossum return ret; 25981da177e4SLinus Torvalds } 25991da177e4SLinus Torvalds 26001da177e4SLinus Torvalds /* 2601afac7cbaSAl Viro * Return true if path is reachable from root 2602afac7cbaSAl Viro * 260348a066e7SAl Viro * namespace_sem or mount_lock is held 2604afac7cbaSAl Viro */ 2605643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2606afac7cbaSAl Viro const struct path *root) 2607afac7cbaSAl Viro { 2608643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2609a73324daSAl Viro dentry = mnt->mnt_mountpoint; 26100714a533SAl Viro mnt = mnt->mnt_parent; 2611afac7cbaSAl Viro } 2612643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2613afac7cbaSAl Viro } 2614afac7cbaSAl Viro 2615afac7cbaSAl Viro int path_is_under(struct path *path1, struct path *path2) 2616afac7cbaSAl Viro { 2617afac7cbaSAl Viro int res; 261848a066e7SAl Viro read_seqlock_excl(&mount_lock); 2619643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 262048a066e7SAl Viro read_sequnlock_excl(&mount_lock); 2621afac7cbaSAl Viro return res; 2622afac7cbaSAl Viro } 2623afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 2624afac7cbaSAl Viro 2625afac7cbaSAl Viro /* 26261da177e4SLinus Torvalds * pivot_root Semantics: 26271da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 26281da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 26291da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 26301da177e4SLinus Torvalds * 26311da177e4SLinus Torvalds * Restrictions: 26321da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 26331da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 26341da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 26351da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 26361da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 26371da177e4SLinus Torvalds * 26384a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 26394a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 26404a0d11faSNeil Brown * in this situation. 26414a0d11faSNeil Brown * 26421da177e4SLinus Torvalds * Notes: 26431da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 26441da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 26451da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 26461da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 26471da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 26481da177e4SLinus Torvalds * first. 26491da177e4SLinus Torvalds */ 26503480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 26513480b257SHeiko Carstens const char __user *, put_old) 26521da177e4SLinus Torvalds { 26532d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 265484d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 265584d17192SAl Viro struct mountpoint *old_mp, *root_mp; 26561da177e4SLinus Torvalds int error; 26571da177e4SLinus Torvalds 26589b40bc90SAl Viro if (!may_mount()) 26591da177e4SLinus Torvalds return -EPERM; 26601da177e4SLinus Torvalds 26612d8f3038SAl Viro error = user_path_dir(new_root, &new); 26621da177e4SLinus Torvalds if (error) 26631da177e4SLinus Torvalds goto out0; 26641da177e4SLinus Torvalds 26652d8f3038SAl Viro error = user_path_dir(put_old, &old); 26661da177e4SLinus Torvalds if (error) 26671da177e4SLinus Torvalds goto out1; 26681da177e4SLinus Torvalds 26692d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 2670b12cea91SAl Viro if (error) 2671b12cea91SAl Viro goto out2; 26721da177e4SLinus Torvalds 2673f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 267484d17192SAl Viro old_mp = lock_mount(&old); 267584d17192SAl Viro error = PTR_ERR(old_mp); 267684d17192SAl Viro if (IS_ERR(old_mp)) 2677b12cea91SAl Viro goto out3; 2678b12cea91SAl Viro 26791da177e4SLinus Torvalds error = -EINVAL; 2680419148daSAl Viro new_mnt = real_mount(new.mnt); 2681419148daSAl Viro root_mnt = real_mount(root.mnt); 268284d17192SAl Viro old_mnt = real_mount(old.mnt); 268384d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 2684fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 2685fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 2686b12cea91SAl Viro goto out4; 2687143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 2688b12cea91SAl Viro goto out4; 26895ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 26905ff9d8a6SEric W. Biederman goto out4; 26911da177e4SLinus Torvalds error = -ENOENT; 2692f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 2693b12cea91SAl Viro goto out4; 26941da177e4SLinus Torvalds error = -EBUSY; 269584d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 2696b12cea91SAl Viro goto out4; /* loop, on the same file system */ 26971da177e4SLinus Torvalds error = -EINVAL; 26988c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 2699b12cea91SAl Viro goto out4; /* not a mountpoint */ 2700676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 2701b12cea91SAl Viro goto out4; /* not attached */ 270284d17192SAl Viro root_mp = root_mnt->mnt_mp; 27032d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 2704b12cea91SAl Viro goto out4; /* not a mountpoint */ 2705676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 2706b12cea91SAl Viro goto out4; /* not attached */ 27074ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 270884d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 2709b12cea91SAl Viro goto out4; 271084d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 2711719ea2fbSAl Viro lock_mount_hash(); 2712419148daSAl Viro detach_mnt(new_mnt, &parent_path); 2713419148daSAl Viro detach_mnt(root_mnt, &root_parent); 27145ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 27155ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 27165ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 27175ff9d8a6SEric W. Biederman } 27184ac91378SJan Blunck /* mount old root on put_old */ 271984d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 27204ac91378SJan Blunck /* mount new_root on / */ 272184d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 27226b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 2723719ea2fbSAl Viro unlock_mount_hash(); 27242d8f3038SAl Viro chroot_fs_refs(&root, &new); 272584d17192SAl Viro put_mountpoint(root_mp); 27261da177e4SLinus Torvalds error = 0; 2727b12cea91SAl Viro out4: 272884d17192SAl Viro unlock_mount(old_mp); 2729b12cea91SAl Viro if (!error) { 27301a390689SAl Viro path_put(&root_parent); 27311a390689SAl Viro path_put(&parent_path); 2732b12cea91SAl Viro } 2733b12cea91SAl Viro out3: 27348c3ee42eSAl Viro path_put(&root); 2735b12cea91SAl Viro out2: 27362d8f3038SAl Viro path_put(&old); 27371da177e4SLinus Torvalds out1: 27382d8f3038SAl Viro path_put(&new); 27391da177e4SLinus Torvalds out0: 27401da177e4SLinus Torvalds return error; 27411da177e4SLinus Torvalds } 27421da177e4SLinus Torvalds 27431da177e4SLinus Torvalds static void __init init_mount_tree(void) 27441da177e4SLinus Torvalds { 27451da177e4SLinus Torvalds struct vfsmount *mnt; 27466b3286edSKirill Korotaev struct mnt_namespace *ns; 2747ac748a09SJan Blunck struct path root; 27480c55cfc4SEric W. Biederman struct file_system_type *type; 27491da177e4SLinus Torvalds 27500c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 27510c55cfc4SEric W. Biederman if (!type) 27520c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 27530c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 27540c55cfc4SEric W. Biederman put_filesystem(type); 27551da177e4SLinus Torvalds if (IS_ERR(mnt)) 27561da177e4SLinus Torvalds panic("Can't create rootfs"); 2757b3e19d92SNick Piggin 27583b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 27593b22edc5STrond Myklebust if (IS_ERR(ns)) 27601da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 27611da177e4SLinus Torvalds 27626b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 27636b3286edSKirill Korotaev get_mnt_ns(ns); 27641da177e4SLinus Torvalds 2765be08d6d2SAl Viro root.mnt = mnt; 2766be08d6d2SAl Viro root.dentry = mnt->mnt_root; 2767ac748a09SJan Blunck 2768ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2769ac748a09SJan Blunck set_fs_root(current->fs, &root); 27701da177e4SLinus Torvalds } 27711da177e4SLinus Torvalds 277274bf17cfSDenis Cheng void __init mnt_init(void) 27731da177e4SLinus Torvalds { 277413f14b4dSEric Dumazet unsigned u; 277515a67dd8SRandy Dunlap int err; 27761da177e4SLinus Torvalds 27777d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 277820c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 27791da177e4SLinus Torvalds 2780b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 278184d17192SAl Viro mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 27821da177e4SLinus Torvalds 278384d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 27841da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 27851da177e4SLinus Torvalds 278680cdc6daSMandeep Singh Baines printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); 27871da177e4SLinus Torvalds 278813f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 278913f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 279084d17192SAl Viro for (u = 0; u < HASH_SIZE; u++) 279184d17192SAl Viro INIT_LIST_HEAD(&mountpoint_hashtable[u]); 27921da177e4SLinus Torvalds 27934b93dc9bSTejun Heo kernfs_init(); 27944b93dc9bSTejun Heo 279515a67dd8SRandy Dunlap err = sysfs_init(); 279615a67dd8SRandy Dunlap if (err) 279715a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 27988e24eea7SHarvey Harrison __func__, err); 279900d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 280000d26666SGreg Kroah-Hartman if (!fs_kobj) 28018e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 28021da177e4SLinus Torvalds init_rootfs(); 28031da177e4SLinus Torvalds init_mount_tree(); 28041da177e4SLinus Torvalds } 28051da177e4SLinus Torvalds 2806616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 28071da177e4SLinus Torvalds { 2808d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 2809616511d0STrond Myklebust return; 28107b00ed6fSAl Viro drop_collected_mounts(&ns->root->mnt); 2811771b1371SEric W. Biederman free_mnt_ns(ns); 28121da177e4SLinus Torvalds } 28139d412a43SAl Viro 28149d412a43SAl Viro struct vfsmount *kern_mount_data(struct file_system_type *type, void *data) 28159d412a43SAl Viro { 2816423e0ab0STim Chen struct vfsmount *mnt; 2817423e0ab0STim Chen mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data); 2818423e0ab0STim Chen if (!IS_ERR(mnt)) { 2819423e0ab0STim Chen /* 2820423e0ab0STim Chen * it is a longterm mount, don't release mnt until 2821423e0ab0STim Chen * we unmount before file sys is unregistered 2822423e0ab0STim Chen */ 2823f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL; 2824423e0ab0STim Chen } 2825423e0ab0STim Chen return mnt; 28269d412a43SAl Viro } 28279d412a43SAl Viro EXPORT_SYMBOL_GPL(kern_mount_data); 2828423e0ab0STim Chen 2829423e0ab0STim Chen void kern_unmount(struct vfsmount *mnt) 2830423e0ab0STim Chen { 2831423e0ab0STim Chen /* release long term mount so mount point can be released */ 2832423e0ab0STim Chen if (!IS_ERR_OR_NULL(mnt)) { 2833f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 283448a066e7SAl Viro synchronize_rcu(); /* yecchhh... */ 2835423e0ab0STim Chen mntput(mnt); 2836423e0ab0STim Chen } 2837423e0ab0STim Chen } 2838423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 283902125a82SAl Viro 284002125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 284102125a82SAl Viro { 2842143c8c91SAl Viro return check_mnt(real_mount(mnt)); 284302125a82SAl Viro } 28448823c079SEric W. Biederman 28453151527eSEric W. Biederman bool current_chrooted(void) 28463151527eSEric W. Biederman { 28473151527eSEric W. Biederman /* Does the current process have a non-standard root */ 28483151527eSEric W. Biederman struct path ns_root; 28493151527eSEric W. Biederman struct path fs_root; 28503151527eSEric W. Biederman bool chrooted; 28513151527eSEric W. Biederman 28523151527eSEric W. Biederman /* Find the namespace root */ 28533151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 28543151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 28553151527eSEric W. Biederman path_get(&ns_root); 28563151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 28573151527eSEric W. Biederman ; 28583151527eSEric W. Biederman 28593151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 28603151527eSEric W. Biederman 28613151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 28623151527eSEric W. Biederman 28633151527eSEric W. Biederman path_put(&fs_root); 28643151527eSEric W. Biederman path_put(&ns_root); 28653151527eSEric W. Biederman 28663151527eSEric W. Biederman return chrooted; 28673151527eSEric W. Biederman } 28683151527eSEric W. Biederman 2869e51db735SEric W. Biederman bool fs_fully_visible(struct file_system_type *type) 287087a8ebd6SEric W. Biederman { 287187a8ebd6SEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 287287a8ebd6SEric W. Biederman struct mount *mnt; 2873e51db735SEric W. Biederman bool visible = false; 287487a8ebd6SEric W. Biederman 2875e51db735SEric W. Biederman if (unlikely(!ns)) 2876e51db735SEric W. Biederman return false; 2877e51db735SEric W. Biederman 287844bb4385SAl Viro down_read(&namespace_sem); 287987a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 2880e51db735SEric W. Biederman struct mount *child; 2881e51db735SEric W. Biederman if (mnt->mnt.mnt_sb->s_type != type) 2882e51db735SEric W. Biederman continue; 2883e51db735SEric W. Biederman 2884e51db735SEric W. Biederman /* This mount is not fully visible if there are any child mounts 2885e51db735SEric W. Biederman * that cover anything except for empty directories. 2886e51db735SEric W. Biederman */ 2887e51db735SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 2888e51db735SEric W. Biederman struct inode *inode = child->mnt_mountpoint->d_inode; 2889e51db735SEric W. Biederman if (!S_ISDIR(inode->i_mode)) 2890e51db735SEric W. Biederman goto next; 2891e51db735SEric W. Biederman if (inode->i_nlink != 2) 2892e51db735SEric W. Biederman goto next; 289387a8ebd6SEric W. Biederman } 2894e51db735SEric W. Biederman visible = true; 2895e51db735SEric W. Biederman goto found; 2896e51db735SEric W. Biederman next: ; 289787a8ebd6SEric W. Biederman } 2898e51db735SEric W. Biederman found: 289944bb4385SAl Viro up_read(&namespace_sem); 2900e51db735SEric W. Biederman return visible; 290187a8ebd6SEric W. Biederman } 290287a8ebd6SEric W. Biederman 29038823c079SEric W. Biederman static void *mntns_get(struct task_struct *task) 29048823c079SEric W. Biederman { 29058823c079SEric W. Biederman struct mnt_namespace *ns = NULL; 29068823c079SEric W. Biederman struct nsproxy *nsproxy; 29078823c079SEric W. Biederman 29088823c079SEric W. Biederman rcu_read_lock(); 29098823c079SEric W. Biederman nsproxy = task_nsproxy(task); 29108823c079SEric W. Biederman if (nsproxy) { 29118823c079SEric W. Biederman ns = nsproxy->mnt_ns; 29128823c079SEric W. Biederman get_mnt_ns(ns); 29138823c079SEric W. Biederman } 29148823c079SEric W. Biederman rcu_read_unlock(); 29158823c079SEric W. Biederman 29168823c079SEric W. Biederman return ns; 29178823c079SEric W. Biederman } 29188823c079SEric W. Biederman 29198823c079SEric W. Biederman static void mntns_put(void *ns) 29208823c079SEric W. Biederman { 29218823c079SEric W. Biederman put_mnt_ns(ns); 29228823c079SEric W. Biederman } 29238823c079SEric W. Biederman 29248823c079SEric W. Biederman static int mntns_install(struct nsproxy *nsproxy, void *ns) 29258823c079SEric W. Biederman { 29268823c079SEric W. Biederman struct fs_struct *fs = current->fs; 29278823c079SEric W. Biederman struct mnt_namespace *mnt_ns = ns; 29288823c079SEric W. Biederman struct path root; 29298823c079SEric W. Biederman 29300c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 2931c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_CHROOT) || 2932c7b96acfSEric W. Biederman !ns_capable(current_user_ns(), CAP_SYS_ADMIN)) 2933ae11e0f1SZhao Hongjiang return -EPERM; 29348823c079SEric W. Biederman 29358823c079SEric W. Biederman if (fs->users != 1) 29368823c079SEric W. Biederman return -EINVAL; 29378823c079SEric W. Biederman 29388823c079SEric W. Biederman get_mnt_ns(mnt_ns); 29398823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 29408823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 29418823c079SEric W. Biederman 29428823c079SEric W. Biederman /* Find the root */ 29438823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 29448823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 29458823c079SEric W. Biederman path_get(&root); 29468823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 29478823c079SEric W. Biederman ; 29488823c079SEric W. Biederman 29498823c079SEric W. Biederman /* Update the pwd and root */ 29508823c079SEric W. Biederman set_fs_pwd(fs, &root); 29518823c079SEric W. Biederman set_fs_root(fs, &root); 29528823c079SEric W. Biederman 29538823c079SEric W. Biederman path_put(&root); 29548823c079SEric W. Biederman return 0; 29558823c079SEric W. Biederman } 29568823c079SEric W. Biederman 295798f842e6SEric W. Biederman static unsigned int mntns_inum(void *ns) 295898f842e6SEric W. Biederman { 295998f842e6SEric W. Biederman struct mnt_namespace *mnt_ns = ns; 296098f842e6SEric W. Biederman return mnt_ns->proc_inum; 296198f842e6SEric W. Biederman } 296298f842e6SEric W. Biederman 29638823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 29648823c079SEric W. Biederman .name = "mnt", 29658823c079SEric W. Biederman .type = CLONE_NEWNS, 29668823c079SEric W. Biederman .get = mntns_get, 29678823c079SEric W. Biederman .put = mntns_put, 29688823c079SEric W. Biederman .install = mntns_install, 296998f842e6SEric W. Biederman .inum = mntns_inum, 29708823c079SEric W. Biederman }; 2971