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 */ 20d10577a8SAl Viro #include <linux/ramfs.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; 42390c6843SRam Pai static struct rw_semaphore namespace_sem; 431da177e4SLinus Torvalds 44f87fd4c2SMiklos Szeredi /* /sys/fs */ 4500d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 4600d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 47f87fd4c2SMiklos Szeredi 4899b7db7bSNick Piggin /* 4999b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 5099b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 5199b7db7bSNick Piggin * up the tree. 5299b7db7bSNick Piggin * 5399b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 5499b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 5599b7db7bSNick Piggin */ 5699b7db7bSNick Piggin DEFINE_BRLOCK(vfsmount_lock); 5799b7db7bSNick Piggin 581da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 591da177e4SLinus Torvalds { 601da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 611da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 6213f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 6313f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 641da177e4SLinus Torvalds } 651da177e4SLinus Torvalds 663d733633SDave Hansen #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 673d733633SDave Hansen 6899b7db7bSNick Piggin /* 6999b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 7099b7db7bSNick Piggin * serialize with freeing. 7199b7db7bSNick Piggin */ 72b105e270SAl Viro static int mnt_alloc_id(struct mount *mnt) 7373cd49ecSMiklos Szeredi { 7473cd49ecSMiklos Szeredi int res; 7573cd49ecSMiklos Szeredi 7673cd49ecSMiklos Szeredi retry: 7773cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 7899b7db7bSNick Piggin spin_lock(&mnt_id_lock); 7915169fe7SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 80f21f6220SAl Viro if (!res) 8115169fe7SAl Viro mnt_id_start = mnt->mnt_id + 1; 8299b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 8373cd49ecSMiklos Szeredi if (res == -EAGAIN) 8473cd49ecSMiklos Szeredi goto retry; 8573cd49ecSMiklos Szeredi 8673cd49ecSMiklos Szeredi return res; 8773cd49ecSMiklos Szeredi } 8873cd49ecSMiklos Szeredi 89b105e270SAl Viro static void mnt_free_id(struct mount *mnt) 9073cd49ecSMiklos Szeredi { 9115169fe7SAl Viro int id = mnt->mnt_id; 9299b7db7bSNick Piggin spin_lock(&mnt_id_lock); 93f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 94f21f6220SAl Viro if (mnt_id_start > id) 95f21f6220SAl Viro mnt_id_start = id; 9699b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 9773cd49ecSMiklos Szeredi } 9873cd49ecSMiklos Szeredi 99719f5d7fSMiklos Szeredi /* 100719f5d7fSMiklos Szeredi * Allocate a new peer group ID 101719f5d7fSMiklos Szeredi * 102719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 103719f5d7fSMiklos Szeredi */ 1044b8b21f4SAl Viro static int mnt_alloc_group_id(struct mount *mnt) 105719f5d7fSMiklos Szeredi { 106f21f6220SAl Viro int res; 107f21f6220SAl Viro 108719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 109719f5d7fSMiklos Szeredi return -ENOMEM; 110719f5d7fSMiklos Szeredi 111f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 112f21f6220SAl Viro mnt_group_start, 11315169fe7SAl Viro &mnt->mnt_group_id); 114f21f6220SAl Viro if (!res) 11515169fe7SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 116f21f6220SAl Viro 117f21f6220SAl Viro return res; 118719f5d7fSMiklos Szeredi } 119719f5d7fSMiklos Szeredi 120719f5d7fSMiklos Szeredi /* 121719f5d7fSMiklos Szeredi * Release a peer group ID 122719f5d7fSMiklos Szeredi */ 1234b8b21f4SAl Viro void mnt_release_group_id(struct mount *mnt) 124719f5d7fSMiklos Szeredi { 12515169fe7SAl Viro int id = mnt->mnt_group_id; 126f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 127f21f6220SAl Viro if (mnt_group_start > id) 128f21f6220SAl Viro mnt_group_start = id; 12915169fe7SAl Viro mnt->mnt_group_id = 0; 130719f5d7fSMiklos Szeredi } 131719f5d7fSMiklos Szeredi 132b3e19d92SNick Piggin /* 133b3e19d92SNick Piggin * vfsmount lock must be held for read 134b3e19d92SNick Piggin */ 13583adc753SAl Viro static inline void mnt_add_count(struct mount *mnt, int n) 136b3e19d92SNick Piggin { 137b3e19d92SNick Piggin #ifdef CONFIG_SMP 13868e8a9feSAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, n); 139b3e19d92SNick Piggin #else 140b3e19d92SNick Piggin preempt_disable(); 14168e8a9feSAl Viro mnt->mnt_count += n; 142b3e19d92SNick Piggin preempt_enable(); 143b3e19d92SNick Piggin #endif 144b3e19d92SNick Piggin } 145b3e19d92SNick Piggin 146b3e19d92SNick Piggin /* 147b3e19d92SNick Piggin * vfsmount lock must be held for write 148b3e19d92SNick Piggin */ 14983adc753SAl Viro unsigned int mnt_get_count(struct mount *mnt) 150b3e19d92SNick Piggin { 151b3e19d92SNick Piggin #ifdef CONFIG_SMP 152f03c6599SAl Viro unsigned int count = 0; 153b3e19d92SNick Piggin int cpu; 154b3e19d92SNick Piggin 155b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 15668e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 157b3e19d92SNick Piggin } 158b3e19d92SNick Piggin 159b3e19d92SNick Piggin return count; 160b3e19d92SNick Piggin #else 16168e8a9feSAl Viro return mnt->mnt_count; 162b3e19d92SNick Piggin #endif 163b3e19d92SNick Piggin } 164b3e19d92SNick Piggin 165b105e270SAl Viro static struct mount *alloc_vfsmnt(const char *name) 1661da177e4SLinus Torvalds { 167c63181e6SAl Viro struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 168c63181e6SAl Viro if (mnt) { 16973cd49ecSMiklos Szeredi int err; 17073cd49ecSMiklos Szeredi 171c63181e6SAl Viro err = mnt_alloc_id(mnt); 17288b38782SLi Zefan if (err) 17388b38782SLi Zefan goto out_free_cache; 17488b38782SLi Zefan 17588b38782SLi Zefan if (name) { 176c63181e6SAl Viro mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 177c63181e6SAl Viro if (!mnt->mnt_devname) 17888b38782SLi Zefan goto out_free_id; 17973cd49ecSMiklos Szeredi } 18073cd49ecSMiklos Szeredi 181b3e19d92SNick Piggin #ifdef CONFIG_SMP 182c63181e6SAl Viro mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 183c63181e6SAl Viro if (!mnt->mnt_pcp) 184b3e19d92SNick Piggin goto out_free_devname; 185b3e19d92SNick Piggin 186c63181e6SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 1); 187b3e19d92SNick Piggin #else 188c63181e6SAl Viro mnt->mnt_count = 1; 189c63181e6SAl Viro mnt->mnt_writers = 0; 190b3e19d92SNick Piggin #endif 191b3e19d92SNick Piggin 192c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_hash); 193c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_child); 194c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_mounts); 195c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_list); 196c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_expire); 197c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_share); 198c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave_list); 199c63181e6SAl Viro INIT_LIST_HEAD(&mnt->mnt_slave); 2002504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 2012504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 2022504c5d6SAndreas Gruenbacher #endif 2031da177e4SLinus Torvalds } 204c63181e6SAl Viro return mnt; 20588b38782SLi Zefan 206d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 207d3ef3d73Snpiggin@suse.de out_free_devname: 208c63181e6SAl Viro kfree(mnt->mnt_devname); 209d3ef3d73Snpiggin@suse.de #endif 21088b38782SLi Zefan out_free_id: 211c63181e6SAl Viro mnt_free_id(mnt); 21288b38782SLi Zefan out_free_cache: 213c63181e6SAl Viro kmem_cache_free(mnt_cache, mnt); 21488b38782SLi Zefan return NULL; 2151da177e4SLinus Torvalds } 2161da177e4SLinus Torvalds 2178366025eSDave Hansen /* 2188366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2198366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2208366025eSDave Hansen * We must keep track of when those operations start 2218366025eSDave Hansen * (for permission checks) and when they end, so that 2228366025eSDave Hansen * we can determine when writes are able to occur to 2238366025eSDave Hansen * a filesystem. 2248366025eSDave Hansen */ 2253d733633SDave Hansen /* 2263d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2273d733633SDave Hansen * @mnt: the mount to check for its write status 2283d733633SDave Hansen * 2293d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2303d733633SDave Hansen * It does not guarantee that the filesystem will stay 2313d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2323d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2333d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2343d733633SDave Hansen * r/w. 2353d733633SDave Hansen */ 2363d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2373d733633SDave Hansen { 2382e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2392e4b7fcdSDave Hansen return 1; 2402e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2412e4b7fcdSDave Hansen return 1; 2422e4b7fcdSDave Hansen return 0; 2433d733633SDave Hansen } 2443d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2453d733633SDave Hansen 24683adc753SAl Viro static inline void mnt_inc_writers(struct mount *mnt) 2473d733633SDave Hansen { 248d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 24968e8a9feSAl Viro this_cpu_inc(mnt->mnt_pcp->mnt_writers); 250d3ef3d73Snpiggin@suse.de #else 25168e8a9feSAl Viro mnt->mnt_writers++; 252d3ef3d73Snpiggin@suse.de #endif 2533d733633SDave Hansen } 2543d733633SDave Hansen 25583adc753SAl Viro static inline void mnt_dec_writers(struct mount *mnt) 2563d733633SDave Hansen { 257d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 25868e8a9feSAl Viro this_cpu_dec(mnt->mnt_pcp->mnt_writers); 259d3ef3d73Snpiggin@suse.de #else 26068e8a9feSAl Viro mnt->mnt_writers--; 261d3ef3d73Snpiggin@suse.de #endif 262d3ef3d73Snpiggin@suse.de } 263d3ef3d73Snpiggin@suse.de 26483adc753SAl Viro static unsigned int mnt_get_writers(struct mount *mnt) 265d3ef3d73Snpiggin@suse.de { 266d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 267d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2683d733633SDave Hansen int cpu; 2693d733633SDave Hansen 2703d733633SDave Hansen for_each_possible_cpu(cpu) { 27168e8a9feSAl Viro count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 2723d733633SDave Hansen } 2733d733633SDave Hansen 274d3ef3d73Snpiggin@suse.de return count; 275d3ef3d73Snpiggin@suse.de #else 276d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 277d3ef3d73Snpiggin@suse.de #endif 2783d733633SDave Hansen } 2793d733633SDave Hansen 2804ed5e82fSMiklos Szeredi static int mnt_is_readonly(struct vfsmount *mnt) 2814ed5e82fSMiklos Szeredi { 2824ed5e82fSMiklos Szeredi if (mnt->mnt_sb->s_readonly_remount) 2834ed5e82fSMiklos Szeredi return 1; 2844ed5e82fSMiklos Szeredi /* Order wrt setting s_flags/s_readonly_remount in do_remount() */ 2854ed5e82fSMiklos Szeredi smp_rmb(); 2864ed5e82fSMiklos Szeredi return __mnt_is_readonly(mnt); 2874ed5e82fSMiklos Szeredi } 2884ed5e82fSMiklos Szeredi 2893d733633SDave Hansen /* 290eb04c282SJan Kara * Most r/o & frozen checks on a fs are for operations that take discrete 291eb04c282SJan Kara * amounts of time, like a write() or unlink(). We must keep track of when 292eb04c282SJan Kara * those operations start (for permission checks) and when they end, so that we 293eb04c282SJan Kara * can determine when writes are able to occur to a filesystem. 2943d733633SDave Hansen */ 2958366025eSDave Hansen /** 296eb04c282SJan Kara * __mnt_want_write - get write access to a mount without freeze protection 29783adc753SAl Viro * @m: the mount on which to take a write 2988366025eSDave Hansen * 299eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 300eb04c282SJan Kara * it, and makes sure that writes are allowed (mnt it read-write) before 301eb04c282SJan Kara * returning success. This operation does not protect against filesystem being 302eb04c282SJan Kara * frozen. When the write operation is finished, __mnt_drop_write() must be 303eb04c282SJan Kara * called. This is effectively a refcount. 3048366025eSDave Hansen */ 305eb04c282SJan Kara int __mnt_want_write(struct vfsmount *m) 3068366025eSDave Hansen { 30783adc753SAl Viro struct mount *mnt = real_mount(m); 3083d733633SDave Hansen int ret = 0; 3093d733633SDave Hansen 310d3ef3d73Snpiggin@suse.de preempt_disable(); 311c6653a83SNick Piggin mnt_inc_writers(mnt); 312d3ef3d73Snpiggin@suse.de /* 313c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 314d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 315d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 316d3ef3d73Snpiggin@suse.de */ 317d3ef3d73Snpiggin@suse.de smp_mb(); 3181e75529eSMiao Xie while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD) 319d3ef3d73Snpiggin@suse.de cpu_relax(); 320d3ef3d73Snpiggin@suse.de /* 321d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 322d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 323d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 324d3ef3d73Snpiggin@suse.de */ 325d3ef3d73Snpiggin@suse.de smp_rmb(); 3264ed5e82fSMiklos Szeredi if (mnt_is_readonly(m)) { 327c6653a83SNick Piggin mnt_dec_writers(mnt); 3283d733633SDave Hansen ret = -EROFS; 3293d733633SDave Hansen } 330d3ef3d73Snpiggin@suse.de preempt_enable(); 331eb04c282SJan Kara 332eb04c282SJan Kara return ret; 333eb04c282SJan Kara } 334eb04c282SJan Kara 335eb04c282SJan Kara /** 336eb04c282SJan Kara * mnt_want_write - get write access to a mount 337eb04c282SJan Kara * @m: the mount on which to take a write 338eb04c282SJan Kara * 339eb04c282SJan Kara * This tells the low-level filesystem that a write is about to be performed to 340eb04c282SJan Kara * it, and makes sure that writes are allowed (mount is read-write, filesystem 341eb04c282SJan Kara * is not frozen) before returning success. When the write operation is 342eb04c282SJan Kara * finished, mnt_drop_write() must be called. This is effectively a refcount. 343eb04c282SJan Kara */ 344eb04c282SJan Kara int mnt_want_write(struct vfsmount *m) 345eb04c282SJan Kara { 346eb04c282SJan Kara int ret; 347eb04c282SJan Kara 348eb04c282SJan Kara sb_start_write(m->mnt_sb); 349eb04c282SJan Kara ret = __mnt_want_write(m); 350eb04c282SJan Kara if (ret) 351eb04c282SJan Kara sb_end_write(m->mnt_sb); 3523d733633SDave Hansen return ret; 3538366025eSDave Hansen } 3548366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3558366025eSDave Hansen 3568366025eSDave Hansen /** 35796029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 35896029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 35996029c4eSnpiggin@suse.de * 36096029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 36196029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 36296029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 36396029c4eSnpiggin@suse.de * on it. This allows some optimisation. 36496029c4eSnpiggin@suse.de * 36596029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 36696029c4eSnpiggin@suse.de * drop the reference. 36796029c4eSnpiggin@suse.de */ 36896029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 36996029c4eSnpiggin@suse.de { 37096029c4eSnpiggin@suse.de /* superblock may be r/o */ 37196029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 37296029c4eSnpiggin@suse.de return -EROFS; 37396029c4eSnpiggin@suse.de preempt_disable(); 37483adc753SAl Viro mnt_inc_writers(real_mount(mnt)); 37596029c4eSnpiggin@suse.de preempt_enable(); 37696029c4eSnpiggin@suse.de return 0; 37796029c4eSnpiggin@suse.de } 37896029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 37996029c4eSnpiggin@suse.de 38096029c4eSnpiggin@suse.de /** 381eb04c282SJan Kara * __mnt_want_write_file - get write access to a file's mount 382eb04c282SJan Kara * @file: the file who's mount on which to take a write 383eb04c282SJan Kara * 384eb04c282SJan Kara * This is like __mnt_want_write, but it takes a file and can 385eb04c282SJan Kara * do some optimisations if the file is open for write already 386eb04c282SJan Kara */ 387eb04c282SJan Kara int __mnt_want_write_file(struct file *file) 388eb04c282SJan Kara { 389496ad9aaSAl Viro struct inode *inode = file_inode(file); 390eb04c282SJan Kara 391eb04c282SJan Kara if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) 392eb04c282SJan Kara return __mnt_want_write(file->f_path.mnt); 393eb04c282SJan Kara else 394eb04c282SJan Kara return mnt_clone_write(file->f_path.mnt); 395eb04c282SJan Kara } 396eb04c282SJan Kara 397eb04c282SJan Kara /** 39896029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 39996029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 40096029c4eSnpiggin@suse.de * 40196029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 40296029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 40396029c4eSnpiggin@suse.de */ 40496029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 40596029c4eSnpiggin@suse.de { 406eb04c282SJan Kara int ret; 407eb04c282SJan Kara 408eb04c282SJan Kara sb_start_write(file->f_path.mnt->mnt_sb); 409eb04c282SJan Kara ret = __mnt_want_write_file(file); 410eb04c282SJan Kara if (ret) 411eb04c282SJan Kara sb_end_write(file->f_path.mnt->mnt_sb); 412eb04c282SJan Kara return ret; 41396029c4eSnpiggin@suse.de } 41496029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 41596029c4eSnpiggin@suse.de 41696029c4eSnpiggin@suse.de /** 417eb04c282SJan Kara * __mnt_drop_write - give up write access to a mount 4188366025eSDave Hansen * @mnt: the mount on which to give up write access 4198366025eSDave Hansen * 4208366025eSDave Hansen * Tells the low-level filesystem that we are done 4218366025eSDave Hansen * performing writes to it. Must be matched with 422eb04c282SJan Kara * __mnt_want_write() call above. 4238366025eSDave Hansen */ 424eb04c282SJan Kara void __mnt_drop_write(struct vfsmount *mnt) 4258366025eSDave Hansen { 426d3ef3d73Snpiggin@suse.de preempt_disable(); 42783adc753SAl Viro mnt_dec_writers(real_mount(mnt)); 428d3ef3d73Snpiggin@suse.de preempt_enable(); 4298366025eSDave Hansen } 430eb04c282SJan Kara 431eb04c282SJan Kara /** 432eb04c282SJan Kara * mnt_drop_write - give up write access to a mount 433eb04c282SJan Kara * @mnt: the mount on which to give up write access 434eb04c282SJan Kara * 435eb04c282SJan Kara * Tells the low-level filesystem that we are done performing writes to it and 436eb04c282SJan Kara * also allows filesystem to be frozen again. Must be matched with 437eb04c282SJan Kara * mnt_want_write() call above. 438eb04c282SJan Kara */ 439eb04c282SJan Kara void mnt_drop_write(struct vfsmount *mnt) 440eb04c282SJan Kara { 441eb04c282SJan Kara __mnt_drop_write(mnt); 442eb04c282SJan Kara sb_end_write(mnt->mnt_sb); 443eb04c282SJan Kara } 4448366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4458366025eSDave Hansen 446eb04c282SJan Kara void __mnt_drop_write_file(struct file *file) 447eb04c282SJan Kara { 448eb04c282SJan Kara __mnt_drop_write(file->f_path.mnt); 449eb04c282SJan Kara } 450eb04c282SJan Kara 4512a79f17eSAl Viro void mnt_drop_write_file(struct file *file) 4522a79f17eSAl Viro { 4532a79f17eSAl Viro mnt_drop_write(file->f_path.mnt); 4542a79f17eSAl Viro } 4552a79f17eSAl Viro EXPORT_SYMBOL(mnt_drop_write_file); 4562a79f17eSAl Viro 45783adc753SAl Viro static int mnt_make_readonly(struct mount *mnt) 4588366025eSDave Hansen { 4593d733633SDave Hansen int ret = 0; 4603d733633SDave Hansen 461962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 46283adc753SAl Viro mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 463d3ef3d73Snpiggin@suse.de /* 464d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 465d3ef3d73Snpiggin@suse.de * should be visible before we do. 466d3ef3d73Snpiggin@suse.de */ 467d3ef3d73Snpiggin@suse.de smp_mb(); 468d3ef3d73Snpiggin@suse.de 469d3ef3d73Snpiggin@suse.de /* 470d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 471d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 472d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 473d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 474d3ef3d73Snpiggin@suse.de * 475d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 476d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 477d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 478d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 479d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 480d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 481d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 482d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 483d3ef3d73Snpiggin@suse.de * we're counting up here. 484d3ef3d73Snpiggin@suse.de */ 485c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 486d3ef3d73Snpiggin@suse.de ret = -EBUSY; 487d3ef3d73Snpiggin@suse.de else 48883adc753SAl Viro mnt->mnt.mnt_flags |= MNT_READONLY; 489d3ef3d73Snpiggin@suse.de /* 490d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 491d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 492d3ef3d73Snpiggin@suse.de */ 493d3ef3d73Snpiggin@suse.de smp_wmb(); 49483adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 495962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 4963d733633SDave Hansen return ret; 4973d733633SDave Hansen } 4988366025eSDave Hansen 49983adc753SAl Viro static void __mnt_unmake_readonly(struct mount *mnt) 5002e4b7fcdSDave Hansen { 501962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 50283adc753SAl Viro mnt->mnt.mnt_flags &= ~MNT_READONLY; 503962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 5042e4b7fcdSDave Hansen } 5052e4b7fcdSDave Hansen 5064ed5e82fSMiklos Szeredi int sb_prepare_remount_readonly(struct super_block *sb) 5074ed5e82fSMiklos Szeredi { 5084ed5e82fSMiklos Szeredi struct mount *mnt; 5094ed5e82fSMiklos Szeredi int err = 0; 5104ed5e82fSMiklos Szeredi 5118e8b8796SMiklos Szeredi /* Racy optimization. Recheck the counter under MNT_WRITE_HOLD */ 5128e8b8796SMiklos Szeredi if (atomic_long_read(&sb->s_remove_count)) 5138e8b8796SMiklos Szeredi return -EBUSY; 5148e8b8796SMiklos Szeredi 515962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 5164ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5174ed5e82fSMiklos Szeredi if (!(mnt->mnt.mnt_flags & MNT_READONLY)) { 5184ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags |= MNT_WRITE_HOLD; 5194ed5e82fSMiklos Szeredi smp_mb(); 5204ed5e82fSMiklos Szeredi if (mnt_get_writers(mnt) > 0) { 5214ed5e82fSMiklos Szeredi err = -EBUSY; 5224ed5e82fSMiklos Szeredi break; 5234ed5e82fSMiklos Szeredi } 5244ed5e82fSMiklos Szeredi } 5254ed5e82fSMiklos Szeredi } 5268e8b8796SMiklos Szeredi if (!err && atomic_long_read(&sb->s_remove_count)) 5278e8b8796SMiklos Szeredi err = -EBUSY; 5288e8b8796SMiklos Szeredi 5294ed5e82fSMiklos Szeredi if (!err) { 5304ed5e82fSMiklos Szeredi sb->s_readonly_remount = 1; 5314ed5e82fSMiklos Szeredi smp_wmb(); 5324ed5e82fSMiklos Szeredi } 5334ed5e82fSMiklos Szeredi list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) { 5344ed5e82fSMiklos Szeredi if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD) 5354ed5e82fSMiklos Szeredi mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD; 5364ed5e82fSMiklos Szeredi } 537962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 5384ed5e82fSMiklos Szeredi 5394ed5e82fSMiklos Szeredi return err; 5404ed5e82fSMiklos Szeredi } 5414ed5e82fSMiklos Szeredi 542b105e270SAl Viro static void free_vfsmnt(struct mount *mnt) 5431da177e4SLinus Torvalds { 54452ba1621SAl Viro kfree(mnt->mnt_devname); 54573cd49ecSMiklos Szeredi mnt_free_id(mnt); 546d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 54768e8a9feSAl Viro free_percpu(mnt->mnt_pcp); 548d3ef3d73Snpiggin@suse.de #endif 549b105e270SAl Viro kmem_cache_free(mnt_cache, mnt); 5501da177e4SLinus Torvalds } 5511da177e4SLinus Torvalds 5521da177e4SLinus Torvalds /* 553a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 554a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 55599b7db7bSNick Piggin * vfsmount_lock must be held for read or write. 5561da177e4SLinus Torvalds */ 557c7105365SAl Viro struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 558a05964f3SRam Pai int dir) 5591da177e4SLinus Torvalds { 5601da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 5611da177e4SLinus Torvalds struct list_head *tmp = head; 562c7105365SAl Viro struct mount *p, *found = NULL; 5631da177e4SLinus Torvalds 5641da177e4SLinus Torvalds for (;;) { 565a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 5661da177e4SLinus Torvalds p = NULL; 5671da177e4SLinus Torvalds if (tmp == head) 5681da177e4SLinus Torvalds break; 5691b8e5564SAl Viro p = list_entry(tmp, struct mount, mnt_hash); 570a73324daSAl Viro if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) { 571a05964f3SRam Pai found = p; 5721da177e4SLinus Torvalds break; 5731da177e4SLinus Torvalds } 5741da177e4SLinus Torvalds } 5751da177e4SLinus Torvalds return found; 5761da177e4SLinus Torvalds } 5771da177e4SLinus Torvalds 578a05964f3SRam Pai /* 579f015f126SDavid Howells * lookup_mnt - Return the first child mount mounted at path 580f015f126SDavid Howells * 581f015f126SDavid Howells * "First" means first mounted chronologically. If you create the 582f015f126SDavid Howells * following mounts: 583f015f126SDavid Howells * 584f015f126SDavid Howells * mount /dev/sda1 /mnt 585f015f126SDavid Howells * mount /dev/sda2 /mnt 586f015f126SDavid Howells * mount /dev/sda3 /mnt 587f015f126SDavid Howells * 588f015f126SDavid Howells * Then lookup_mnt() on the base /mnt dentry in the root mount will 589f015f126SDavid Howells * return successively the root dentry and vfsmount of /dev/sda1, then 590f015f126SDavid Howells * /dev/sda2, then /dev/sda3, then NULL. 591f015f126SDavid Howells * 592f015f126SDavid Howells * lookup_mnt takes a reference to the found vfsmount. 593a05964f3SRam Pai */ 5941c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 595a05964f3SRam Pai { 596c7105365SAl Viro struct mount *child_mnt; 59799b7db7bSNick Piggin 598962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 599c7105365SAl Viro child_mnt = __lookup_mnt(path->mnt, path->dentry, 1); 600c7105365SAl Viro if (child_mnt) { 601c7105365SAl Viro mnt_add_count(child_mnt, 1); 602962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 603c7105365SAl Viro return &child_mnt->mnt; 604c7105365SAl Viro } else { 605962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 606c7105365SAl Viro return NULL; 607c7105365SAl Viro } 608a05964f3SRam Pai } 609a05964f3SRam Pai 61084d17192SAl Viro static struct mountpoint *new_mountpoint(struct dentry *dentry) 61184d17192SAl Viro { 61284d17192SAl Viro struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry); 61384d17192SAl Viro struct mountpoint *mp; 61484d17192SAl Viro 61584d17192SAl Viro list_for_each_entry(mp, chain, m_hash) { 61684d17192SAl Viro if (mp->m_dentry == dentry) { 61784d17192SAl Viro /* might be worth a WARN_ON() */ 61884d17192SAl Viro if (d_unlinked(dentry)) 61984d17192SAl Viro return ERR_PTR(-ENOENT); 62084d17192SAl Viro mp->m_count++; 62184d17192SAl Viro return mp; 62284d17192SAl Viro } 62384d17192SAl Viro } 62484d17192SAl Viro 62584d17192SAl Viro mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL); 62684d17192SAl Viro if (!mp) 62784d17192SAl Viro return ERR_PTR(-ENOMEM); 62884d17192SAl Viro 62984d17192SAl Viro spin_lock(&dentry->d_lock); 63084d17192SAl Viro if (d_unlinked(dentry)) { 63184d17192SAl Viro spin_unlock(&dentry->d_lock); 63284d17192SAl Viro kfree(mp); 63384d17192SAl Viro return ERR_PTR(-ENOENT); 63484d17192SAl Viro } 63584d17192SAl Viro dentry->d_flags |= DCACHE_MOUNTED; 63684d17192SAl Viro spin_unlock(&dentry->d_lock); 63784d17192SAl Viro mp->m_dentry = dentry; 63884d17192SAl Viro mp->m_count = 1; 63984d17192SAl Viro list_add(&mp->m_hash, chain); 64084d17192SAl Viro return mp; 64184d17192SAl Viro } 64284d17192SAl Viro 64384d17192SAl Viro static void put_mountpoint(struct mountpoint *mp) 64484d17192SAl Viro { 64584d17192SAl Viro if (!--mp->m_count) { 64684d17192SAl Viro struct dentry *dentry = mp->m_dentry; 64784d17192SAl Viro spin_lock(&dentry->d_lock); 64884d17192SAl Viro dentry->d_flags &= ~DCACHE_MOUNTED; 64984d17192SAl Viro spin_unlock(&dentry->d_lock); 65084d17192SAl Viro list_del(&mp->m_hash); 65184d17192SAl Viro kfree(mp); 65284d17192SAl Viro } 65384d17192SAl Viro } 65484d17192SAl Viro 655143c8c91SAl Viro static inline int check_mnt(struct mount *mnt) 6561da177e4SLinus Torvalds { 6576b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds 66099b7db7bSNick Piggin /* 66199b7db7bSNick Piggin * vfsmount lock must be held for write 66299b7db7bSNick Piggin */ 6636b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 6645addc5ddSAl Viro { 6655addc5ddSAl Viro if (ns) { 6665addc5ddSAl Viro ns->event = ++event; 6675addc5ddSAl Viro wake_up_interruptible(&ns->poll); 6685addc5ddSAl Viro } 6695addc5ddSAl Viro } 6705addc5ddSAl Viro 67199b7db7bSNick Piggin /* 67299b7db7bSNick Piggin * vfsmount lock must be held for write 67399b7db7bSNick Piggin */ 6746b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 6755addc5ddSAl Viro { 6765addc5ddSAl Viro if (ns && ns->event != event) { 6775addc5ddSAl Viro ns->event = event; 6785addc5ddSAl Viro wake_up_interruptible(&ns->poll); 6795addc5ddSAl Viro } 6805addc5ddSAl Viro } 6815addc5ddSAl Viro 68299b7db7bSNick Piggin /* 68399b7db7bSNick Piggin * vfsmount lock must be held for write 68499b7db7bSNick Piggin */ 685419148daSAl Viro static void detach_mnt(struct mount *mnt, struct path *old_path) 6861da177e4SLinus Torvalds { 687a73324daSAl Viro old_path->dentry = mnt->mnt_mountpoint; 6880714a533SAl Viro old_path->mnt = &mnt->mnt_parent->mnt; 6890714a533SAl Viro mnt->mnt_parent = mnt; 690a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 6916b41d536SAl Viro list_del_init(&mnt->mnt_child); 6921b8e5564SAl Viro list_del_init(&mnt->mnt_hash); 69384d17192SAl Viro put_mountpoint(mnt->mnt_mp); 69484d17192SAl Viro mnt->mnt_mp = NULL; 6951da177e4SLinus Torvalds } 6961da177e4SLinus Torvalds 69799b7db7bSNick Piggin /* 69899b7db7bSNick Piggin * vfsmount lock must be held for write 69999b7db7bSNick Piggin */ 70084d17192SAl Viro void mnt_set_mountpoint(struct mount *mnt, 70184d17192SAl Viro struct mountpoint *mp, 70244d964d6SAl Viro struct mount *child_mnt) 703b90fa9aeSRam Pai { 70484d17192SAl Viro mp->m_count++; 7053a2393d7SAl Viro mnt_add_count(mnt, 1); /* essentially, that's mntget */ 70684d17192SAl Viro child_mnt->mnt_mountpoint = dget(mp->m_dentry); 7073a2393d7SAl Viro child_mnt->mnt_parent = mnt; 70884d17192SAl Viro child_mnt->mnt_mp = mp; 709b90fa9aeSRam Pai } 710b90fa9aeSRam Pai 71199b7db7bSNick Piggin /* 71299b7db7bSNick Piggin * vfsmount lock must be held for write 71399b7db7bSNick Piggin */ 71484d17192SAl Viro static void attach_mnt(struct mount *mnt, 71584d17192SAl Viro struct mount *parent, 71684d17192SAl Viro struct mountpoint *mp) 7171da177e4SLinus Torvalds { 71884d17192SAl Viro mnt_set_mountpoint(parent, mp, mnt); 7191b8e5564SAl Viro list_add_tail(&mnt->mnt_hash, mount_hashtable + 72084d17192SAl Viro hash(&parent->mnt, mp->m_dentry)); 72184d17192SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 722b90fa9aeSRam Pai } 723b90fa9aeSRam Pai 724b90fa9aeSRam Pai /* 72599b7db7bSNick Piggin * vfsmount lock must be held for write 726b90fa9aeSRam Pai */ 7274b2619a5SAl Viro static void commit_tree(struct mount *mnt) 728b90fa9aeSRam Pai { 7290714a533SAl Viro struct mount *parent = mnt->mnt_parent; 73083adc753SAl Viro struct mount *m; 731b90fa9aeSRam Pai LIST_HEAD(head); 732143c8c91SAl Viro struct mnt_namespace *n = parent->mnt_ns; 733b90fa9aeSRam Pai 7340714a533SAl Viro BUG_ON(parent == mnt); 735b90fa9aeSRam Pai 7361a4eeaf2SAl Viro list_add_tail(&head, &mnt->mnt_list); 737f7a99c5bSAl Viro list_for_each_entry(m, &head, mnt_list) 738143c8c91SAl Viro m->mnt_ns = n; 739f03c6599SAl Viro 740b90fa9aeSRam Pai list_splice(&head, n->list.prev); 741b90fa9aeSRam Pai 7421b8e5564SAl Viro list_add_tail(&mnt->mnt_hash, mount_hashtable + 743a73324daSAl Viro hash(&parent->mnt, mnt->mnt_mountpoint)); 7446b41d536SAl Viro list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 7456b3286edSKirill Korotaev touch_mnt_namespace(n); 7461da177e4SLinus Torvalds } 7471da177e4SLinus Torvalds 748909b0a88SAl Viro static struct mount *next_mnt(struct mount *p, struct mount *root) 7491da177e4SLinus Torvalds { 7506b41d536SAl Viro struct list_head *next = p->mnt_mounts.next; 7516b41d536SAl Viro if (next == &p->mnt_mounts) { 7521da177e4SLinus Torvalds while (1) { 753909b0a88SAl Viro if (p == root) 7541da177e4SLinus Torvalds return NULL; 7556b41d536SAl Viro next = p->mnt_child.next; 7566b41d536SAl Viro if (next != &p->mnt_parent->mnt_mounts) 7571da177e4SLinus Torvalds break; 7580714a533SAl Viro p = p->mnt_parent; 7591da177e4SLinus Torvalds } 7601da177e4SLinus Torvalds } 7616b41d536SAl Viro return list_entry(next, struct mount, mnt_child); 7621da177e4SLinus Torvalds } 7631da177e4SLinus Torvalds 764315fc83eSAl Viro static struct mount *skip_mnt_tree(struct mount *p) 7659676f0c6SRam Pai { 7666b41d536SAl Viro struct list_head *prev = p->mnt_mounts.prev; 7676b41d536SAl Viro while (prev != &p->mnt_mounts) { 7686b41d536SAl Viro p = list_entry(prev, struct mount, mnt_child); 7696b41d536SAl Viro prev = p->mnt_mounts.prev; 7709676f0c6SRam Pai } 7719676f0c6SRam Pai return p; 7729676f0c6SRam Pai } 7739676f0c6SRam Pai 7749d412a43SAl Viro struct vfsmount * 7759d412a43SAl Viro vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data) 7769d412a43SAl Viro { 777b105e270SAl Viro struct mount *mnt; 7789d412a43SAl Viro struct dentry *root; 7799d412a43SAl Viro 7809d412a43SAl Viro if (!type) 7819d412a43SAl Viro return ERR_PTR(-ENODEV); 7829d412a43SAl Viro 7839d412a43SAl Viro mnt = alloc_vfsmnt(name); 7849d412a43SAl Viro if (!mnt) 7859d412a43SAl Viro return ERR_PTR(-ENOMEM); 7869d412a43SAl Viro 7879d412a43SAl Viro if (flags & MS_KERNMOUNT) 788b105e270SAl Viro mnt->mnt.mnt_flags = MNT_INTERNAL; 7899d412a43SAl Viro 7909d412a43SAl Viro root = mount_fs(type, flags, name, data); 7919d412a43SAl Viro if (IS_ERR(root)) { 7929d412a43SAl Viro free_vfsmnt(mnt); 7939d412a43SAl Viro return ERR_CAST(root); 7949d412a43SAl Viro } 7959d412a43SAl Viro 796b105e270SAl Viro mnt->mnt.mnt_root = root; 797b105e270SAl Viro mnt->mnt.mnt_sb = root->d_sb; 798a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 7990714a533SAl Viro mnt->mnt_parent = mnt; 800962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 80139f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts); 802962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 803b105e270SAl Viro return &mnt->mnt; 8049d412a43SAl Viro } 8059d412a43SAl Viro EXPORT_SYMBOL_GPL(vfs_kern_mount); 8069d412a43SAl Viro 80787129cc0SAl Viro static struct mount *clone_mnt(struct mount *old, struct dentry *root, 80836341f64SRam Pai int flag) 8091da177e4SLinus Torvalds { 81087129cc0SAl Viro struct super_block *sb = old->mnt.mnt_sb; 811be34d1a3SDavid Howells struct mount *mnt; 812be34d1a3SDavid Howells int err; 8131da177e4SLinus Torvalds 814be34d1a3SDavid Howells mnt = alloc_vfsmnt(old->mnt_devname); 815be34d1a3SDavid Howells if (!mnt) 816be34d1a3SDavid Howells return ERR_PTR(-ENOMEM); 817be34d1a3SDavid Howells 8187a472ef4SEric W. Biederman if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE)) 81915169fe7SAl Viro mnt->mnt_group_id = 0; /* not a peer of original */ 820719f5d7fSMiklos Szeredi else 82115169fe7SAl Viro mnt->mnt_group_id = old->mnt_group_id; 822719f5d7fSMiklos Szeredi 82315169fe7SAl Viro if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 824be34d1a3SDavid Howells err = mnt_alloc_group_id(mnt); 825719f5d7fSMiklos Szeredi if (err) 826719f5d7fSMiklos Szeredi goto out_free; 827719f5d7fSMiklos Szeredi } 828719f5d7fSMiklos Szeredi 82987129cc0SAl Viro mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD; 830132c94e3SEric W. Biederman /* Don't allow unprivileged users to change mount flags */ 831132c94e3SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY)) 832132c94e3SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCK_READONLY; 833132c94e3SEric W. Biederman 8345ff9d8a6SEric W. Biederman /* Don't allow unprivileged users to reveal what is under a mount */ 8355ff9d8a6SEric W. Biederman if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire)) 8365ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags |= MNT_LOCKED; 8375ff9d8a6SEric W. Biederman 8381da177e4SLinus Torvalds atomic_inc(&sb->s_active); 839b105e270SAl Viro mnt->mnt.mnt_sb = sb; 840b105e270SAl Viro mnt->mnt.mnt_root = dget(root); 841a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 8420714a533SAl Viro mnt->mnt_parent = mnt; 843962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 84439f7c4dbSMiklos Szeredi list_add_tail(&mnt->mnt_instance, &sb->s_mounts); 845962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 846b90fa9aeSRam Pai 8477a472ef4SEric W. Biederman if ((flag & CL_SLAVE) || 8487a472ef4SEric W. Biederman ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) { 8496776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave_list); 85032301920SAl Viro mnt->mnt_master = old; 851fc7be130SAl Viro CLEAR_MNT_SHARED(mnt); 8528aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 853fc7be130SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 8546776db3dSAl Viro list_add(&mnt->mnt_share, &old->mnt_share); 855d10e8defSAl Viro if (IS_MNT_SLAVE(old)) 8566776db3dSAl Viro list_add(&mnt->mnt_slave, &old->mnt_slave); 857d10e8defSAl Viro mnt->mnt_master = old->mnt_master; 8585afe0022SRam Pai } 859b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 8600f0afb1dSAl Viro set_mnt_shared(mnt); 8611da177e4SLinus Torvalds 8621da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 8631da177e4SLinus Torvalds * as the original if that was on one */ 86436341f64SRam Pai if (flag & CL_EXPIRE) { 8656776db3dSAl Viro if (!list_empty(&old->mnt_expire)) 8666776db3dSAl Viro list_add(&mnt->mnt_expire, &old->mnt_expire); 8671da177e4SLinus Torvalds } 868be34d1a3SDavid Howells 869cb338d06SAl Viro return mnt; 870719f5d7fSMiklos Szeredi 871719f5d7fSMiklos Szeredi out_free: 872719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 873be34d1a3SDavid Howells return ERR_PTR(err); 8741da177e4SLinus Torvalds } 8751da177e4SLinus Torvalds 87683adc753SAl Viro static inline void mntfree(struct mount *mnt) 8771da177e4SLinus Torvalds { 87883adc753SAl Viro struct vfsmount *m = &mnt->mnt; 87983adc753SAl Viro struct super_block *sb = m->mnt_sb; 880b3e19d92SNick Piggin 8813d733633SDave Hansen /* 8823d733633SDave Hansen * This probably indicates that somebody messed 8833d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 8843d733633SDave Hansen * happens, the filesystem was probably unable 8853d733633SDave Hansen * to make r/w->r/o transitions. 8863d733633SDave Hansen */ 887d3ef3d73Snpiggin@suse.de /* 888b3e19d92SNick Piggin * The locking used to deal with mnt_count decrement provides barriers, 889b3e19d92SNick Piggin * so mnt_get_writers() below is safe. 890d3ef3d73Snpiggin@suse.de */ 891c6653a83SNick Piggin WARN_ON(mnt_get_writers(mnt)); 89283adc753SAl Viro fsnotify_vfsmount_delete(m); 89383adc753SAl Viro dput(m->mnt_root); 89483adc753SAl Viro free_vfsmnt(mnt); 8951da177e4SLinus Torvalds deactivate_super(sb); 8961da177e4SLinus Torvalds } 8971da177e4SLinus Torvalds 898900148dcSAl Viro static void mntput_no_expire(struct mount *mnt) 8997b7b1aceSAl Viro { 900b3e19d92SNick Piggin put_again: 901f03c6599SAl Viro #ifdef CONFIG_SMP 902962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 903f7a99c5bSAl Viro if (likely(mnt->mnt_ns)) { 904f7a99c5bSAl Viro /* shouldn't be the last one */ 905aa9c0e07SAl Viro mnt_add_count(mnt, -1); 906962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 90799b7db7bSNick Piggin return; 908b3e19d92SNick Piggin } 909962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 910b3e19d92SNick Piggin 911962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 912aa9c0e07SAl Viro mnt_add_count(mnt, -1); 913b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 914962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 91599b7db7bSNick Piggin return; 91699b7db7bSNick Piggin } 917b3e19d92SNick Piggin #else 918aa9c0e07SAl Viro mnt_add_count(mnt, -1); 919b3e19d92SNick Piggin if (likely(mnt_get_count(mnt))) 920b3e19d92SNick Piggin return; 921962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 922f03c6599SAl Viro #endif 923863d684fSAl Viro if (unlikely(mnt->mnt_pinned)) { 924863d684fSAl Viro mnt_add_count(mnt, mnt->mnt_pinned + 1); 925863d684fSAl Viro mnt->mnt_pinned = 0; 926962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 927900148dcSAl Viro acct_auto_close_mnt(&mnt->mnt); 928b3e19d92SNick Piggin goto put_again; 929b3e19d92SNick Piggin } 930962830dfSAndi Kleen 93139f7c4dbSMiklos Szeredi list_del(&mnt->mnt_instance); 932962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 933b3e19d92SNick Piggin mntfree(mnt); 934b3e19d92SNick Piggin } 935b3e19d92SNick Piggin 936b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 937b3e19d92SNick Piggin { 938b3e19d92SNick Piggin if (mnt) { 939863d684fSAl Viro struct mount *m = real_mount(mnt); 940b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 941863d684fSAl Viro if (unlikely(m->mnt_expiry_mark)) 942863d684fSAl Viro m->mnt_expiry_mark = 0; 943863d684fSAl Viro mntput_no_expire(m); 944b3e19d92SNick Piggin } 945b3e19d92SNick Piggin } 946b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 947b3e19d92SNick Piggin 948b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 949b3e19d92SNick Piggin { 950b3e19d92SNick Piggin if (mnt) 95183adc753SAl Viro mnt_add_count(real_mount(mnt), 1); 952b3e19d92SNick Piggin return mnt; 953b3e19d92SNick Piggin } 954b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 955b3e19d92SNick Piggin 9567b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 9577b7b1aceSAl Viro { 958962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 959863d684fSAl Viro real_mount(mnt)->mnt_pinned++; 960962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 9617b7b1aceSAl Viro } 9627b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 9637b7b1aceSAl Viro 964863d684fSAl Viro void mnt_unpin(struct vfsmount *m) 9657b7b1aceSAl Viro { 966863d684fSAl Viro struct mount *mnt = real_mount(m); 967962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 9687b7b1aceSAl Viro if (mnt->mnt_pinned) { 969863d684fSAl Viro mnt_add_count(mnt, 1); 9707b7b1aceSAl Viro mnt->mnt_pinned--; 9717b7b1aceSAl Viro } 972962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 9737b7b1aceSAl Viro } 9747b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 9751da177e4SLinus Torvalds 976b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 977b3b304a2SMiklos Szeredi { 978b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 979b3b304a2SMiklos Szeredi } 980b3b304a2SMiklos Szeredi 981b3b304a2SMiklos Szeredi /* 982b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 983b3b304a2SMiklos Szeredi * implement more complex mount option showing. 984b3b304a2SMiklos Szeredi * 985b3b304a2SMiklos Szeredi * See also save_mount_options(). 986b3b304a2SMiklos Szeredi */ 98734c80b1dSAl Viro int generic_show_options(struct seq_file *m, struct dentry *root) 988b3b304a2SMiklos Szeredi { 9892a32cebdSAl Viro const char *options; 9902a32cebdSAl Viro 9912a32cebdSAl Viro rcu_read_lock(); 99234c80b1dSAl Viro options = rcu_dereference(root->d_sb->s_options); 993b3b304a2SMiklos Szeredi 994b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 995b3b304a2SMiklos Szeredi seq_putc(m, ','); 996b3b304a2SMiklos Szeredi mangle(m, options); 997b3b304a2SMiklos Szeredi } 9982a32cebdSAl Viro rcu_read_unlock(); 999b3b304a2SMiklos Szeredi 1000b3b304a2SMiklos Szeredi return 0; 1001b3b304a2SMiklos Szeredi } 1002b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 1003b3b304a2SMiklos Szeredi 1004b3b304a2SMiklos Szeredi /* 1005b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 1006b3b304a2SMiklos Szeredi * called from the fill_super() callback. 1007b3b304a2SMiklos Szeredi * 1008b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 1009b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 1010b3b304a2SMiklos Szeredi * remount fails. 1011b3b304a2SMiklos Szeredi * 1012b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 1013b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 1014b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 1015b3b304a2SMiklos Szeredi * any more. 1016b3b304a2SMiklos Szeredi */ 1017b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 1018b3b304a2SMiklos Szeredi { 10192a32cebdSAl Viro BUG_ON(sb->s_options); 10202a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 1021b3b304a2SMiklos Szeredi } 1022b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 1023b3b304a2SMiklos Szeredi 10242a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 10252a32cebdSAl Viro { 10262a32cebdSAl Viro char *old = sb->s_options; 10272a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 10282a32cebdSAl Viro if (old) { 10292a32cebdSAl Viro synchronize_rcu(); 10302a32cebdSAl Viro kfree(old); 10312a32cebdSAl Viro } 10322a32cebdSAl Viro } 10332a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 10342a32cebdSAl Viro 1035a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 10360226f492SAl Viro /* iterator; we want it to have access to namespace_sem, thus here... */ 10371da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 10381da177e4SLinus Torvalds { 10396ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 10401da177e4SLinus Torvalds 1041390c6843SRam Pai down_read(&namespace_sem); 1042a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 10431da177e4SLinus Torvalds } 10441da177e4SLinus Torvalds 10451da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 10461da177e4SLinus Torvalds { 10476ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 1048b0765fb8SPavel Emelianov 1049a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 10501da177e4SLinus Torvalds } 10511da177e4SLinus Torvalds 10521da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 10531da177e4SLinus Torvalds { 1054390c6843SRam Pai up_read(&namespace_sem); 10551da177e4SLinus Torvalds } 10561da177e4SLinus Torvalds 10570226f492SAl Viro static int m_show(struct seq_file *m, void *v) 10589f5596afSAl Viro { 10596ce6e24eSAl Viro struct proc_mounts *p = proc_mounts(m); 10601a4eeaf2SAl Viro struct mount *r = list_entry(v, struct mount, mnt_list); 10610226f492SAl Viro return p->show(m, &r->mnt); 10621da177e4SLinus Torvalds } 10631da177e4SLinus Torvalds 1064a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 10651da177e4SLinus Torvalds .start = m_start, 10661da177e4SLinus Torvalds .next = m_next, 10671da177e4SLinus Torvalds .stop = m_stop, 10680226f492SAl Viro .show = m_show, 1069b4629fe2SChuck Lever }; 1070a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1071b4629fe2SChuck Lever 10721da177e4SLinus Torvalds /** 10731da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 10741da177e4SLinus Torvalds * @mnt: root of mount tree 10751da177e4SLinus Torvalds * 10761da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 10771da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 10781da177e4SLinus Torvalds * busy. 10791da177e4SLinus Torvalds */ 1080909b0a88SAl Viro int may_umount_tree(struct vfsmount *m) 10811da177e4SLinus Torvalds { 1082909b0a88SAl Viro struct mount *mnt = real_mount(m); 108336341f64SRam Pai int actual_refs = 0; 108436341f64SRam Pai int minimum_refs = 0; 1085315fc83eSAl Viro struct mount *p; 1086909b0a88SAl Viro BUG_ON(!m); 10871da177e4SLinus Torvalds 1088b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1089962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1090909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 109183adc753SAl Viro actual_refs += mnt_get_count(p); 10921da177e4SLinus Torvalds minimum_refs += 2; 10931da177e4SLinus Torvalds } 1094962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 10951da177e4SLinus Torvalds 10961da177e4SLinus Torvalds if (actual_refs > minimum_refs) 10971da177e4SLinus Torvalds return 0; 1098e3474a8eSIan Kent 1099e3474a8eSIan Kent return 1; 11001da177e4SLinus Torvalds } 11011da177e4SLinus Torvalds 11021da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 11031da177e4SLinus Torvalds 11041da177e4SLinus Torvalds /** 11051da177e4SLinus Torvalds * may_umount - check if a mount point is busy 11061da177e4SLinus Torvalds * @mnt: root of mount 11071da177e4SLinus Torvalds * 11081da177e4SLinus Torvalds * This is called to check if a mount point has any 11091da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 11101da177e4SLinus Torvalds * mount has sub mounts this will return busy 11111da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 11121da177e4SLinus Torvalds * 11131da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 11141da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 11151da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 11161da177e4SLinus Torvalds */ 11171da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 11181da177e4SLinus Torvalds { 1119e3474a8eSIan Kent int ret = 1; 11208ad08d8aSAl Viro down_read(&namespace_sem); 1121962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 11221ab59738SAl Viro if (propagate_mount_busy(real_mount(mnt), 2)) 1123e3474a8eSIan Kent ret = 0; 1124962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 11258ad08d8aSAl Viro up_read(&namespace_sem); 1126a05964f3SRam Pai return ret; 11271da177e4SLinus Torvalds } 11281da177e4SLinus Torvalds 11291da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 11301da177e4SLinus Torvalds 1131e3197d83SAl Viro static LIST_HEAD(unmounted); /* protected by namespace_sem */ 1132e3197d83SAl Viro 113397216be0SAl Viro static void namespace_unlock(void) 11341da177e4SLinus Torvalds { 1135d5e50f74SAl Viro struct mount *mnt; 113697216be0SAl Viro LIST_HEAD(head); 113797216be0SAl Viro 113897216be0SAl Viro if (likely(list_empty(&unmounted))) { 113997216be0SAl Viro up_write(&namespace_sem); 114097216be0SAl Viro return; 114197216be0SAl Viro } 114297216be0SAl Viro 114397216be0SAl Viro list_splice_init(&unmounted, &head); 114497216be0SAl Viro up_write(&namespace_sem); 114597216be0SAl Viro 114697216be0SAl Viro while (!list_empty(&head)) { 114797216be0SAl Viro mnt = list_first_entry(&head, struct mount, mnt_hash); 11481b8e5564SAl Viro list_del_init(&mnt->mnt_hash); 1149676da58dSAl Viro if (mnt_has_parent(mnt)) { 115070fbcdf4SRam Pai struct dentry *dentry; 1151863d684fSAl Viro struct mount *m; 115299b7db7bSNick Piggin 1153962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1154a73324daSAl Viro dentry = mnt->mnt_mountpoint; 1155863d684fSAl Viro m = mnt->mnt_parent; 1156a73324daSAl Viro mnt->mnt_mountpoint = mnt->mnt.mnt_root; 11570714a533SAl Viro mnt->mnt_parent = mnt; 11587c4b93d8SAl Viro m->mnt_ghosts--; 1159962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 116070fbcdf4SRam Pai dput(dentry); 1161863d684fSAl Viro mntput(&m->mnt); 11621da177e4SLinus Torvalds } 1163d5e50f74SAl Viro mntput(&mnt->mnt); 116470fbcdf4SRam Pai } 116570fbcdf4SRam Pai } 116670fbcdf4SRam Pai 116797216be0SAl Viro static inline void namespace_lock(void) 1168e3197d83SAl Viro { 116997216be0SAl Viro down_write(&namespace_sem); 1170e3197d83SAl Viro } 1171e3197d83SAl Viro 117299b7db7bSNick Piggin /* 117399b7db7bSNick Piggin * vfsmount lock must be held for write 117499b7db7bSNick Piggin * namespace_sem must be held for write 117599b7db7bSNick Piggin */ 1176328e6d90SAl Viro void umount_tree(struct mount *mnt, int propagate) 117770fbcdf4SRam Pai { 11787b8a53fdSAl Viro LIST_HEAD(tmp_list); 1179315fc83eSAl Viro struct mount *p; 118070fbcdf4SRam Pai 1181909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) 11821b8e5564SAl Viro list_move(&p->mnt_hash, &tmp_list); 118370fbcdf4SRam Pai 1184a05964f3SRam Pai if (propagate) 11857b8a53fdSAl Viro propagate_umount(&tmp_list); 1186a05964f3SRam Pai 11871b8e5564SAl Viro list_for_each_entry(p, &tmp_list, mnt_hash) { 11886776db3dSAl Viro list_del_init(&p->mnt_expire); 11891a4eeaf2SAl Viro list_del_init(&p->mnt_list); 1190143c8c91SAl Viro __touch_mnt_namespace(p->mnt_ns); 119163d37a84SAl Viro p->mnt_ns = NULL; 11926b41d536SAl Viro list_del_init(&p->mnt_child); 1193676da58dSAl Viro if (mnt_has_parent(p)) { 1194863d684fSAl Viro p->mnt_parent->mnt_ghosts++; 119584d17192SAl Viro put_mountpoint(p->mnt_mp); 119684d17192SAl Viro p->mnt_mp = NULL; 11977c4b93d8SAl Viro } 11980f0afb1dSAl Viro change_mnt_propagation(p, MS_PRIVATE); 11991da177e4SLinus Torvalds } 1200328e6d90SAl Viro list_splice(&tmp_list, &unmounted); 12011da177e4SLinus Torvalds } 12021da177e4SLinus Torvalds 1203b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt); 1204c35038beSAl Viro 12051ab59738SAl Viro static int do_umount(struct mount *mnt, int flags) 12061da177e4SLinus Torvalds { 12071ab59738SAl Viro struct super_block *sb = mnt->mnt.mnt_sb; 12081da177e4SLinus Torvalds int retval; 12091da177e4SLinus Torvalds 12101ab59738SAl Viro retval = security_sb_umount(&mnt->mnt, flags); 12111da177e4SLinus Torvalds if (retval) 12121da177e4SLinus Torvalds return retval; 12131da177e4SLinus Torvalds 12141da177e4SLinus Torvalds /* 12151da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 12161da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 12171da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 12181da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 12191da177e4SLinus Torvalds */ 12201da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 12211ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt || 12221da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 12231da177e4SLinus Torvalds return -EINVAL; 12241da177e4SLinus Torvalds 1225b3e19d92SNick Piggin /* 1226b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1227b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1228b3e19d92SNick Piggin */ 1229962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 123083adc753SAl Viro if (mnt_get_count(mnt) != 2) { 1231962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 12321da177e4SLinus Torvalds return -EBUSY; 1233b3e19d92SNick Piggin } 1234962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 12351da177e4SLinus Torvalds 1236863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1)) 12371da177e4SLinus Torvalds return -EAGAIN; 12381da177e4SLinus Torvalds } 12391da177e4SLinus Torvalds 12401da177e4SLinus Torvalds /* 12411da177e4SLinus Torvalds * If we may have to abort operations to get out of this 12421da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 12431da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 12441da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 12451da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 12461da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 12471da177e4SLinus Torvalds * about for the moment. 12481da177e4SLinus Torvalds */ 12491da177e4SLinus Torvalds 125042faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 125142faad99SAl Viro sb->s_op->umount_begin(sb); 125242faad99SAl Viro } 12531da177e4SLinus Torvalds 12541da177e4SLinus Torvalds /* 12551da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 12561da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 12571da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 12581da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 12591da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 12601da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 12611da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 12621da177e4SLinus Torvalds */ 12631ab59738SAl Viro if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 12641da177e4SLinus Torvalds /* 12651da177e4SLinus Torvalds * Special case for "unmounting" root ... 12661da177e4SLinus Torvalds * we just try to remount it readonly. 12671da177e4SLinus Torvalds */ 12681da177e4SLinus Torvalds down_write(&sb->s_umount); 12694aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 12701da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 12711da177e4SLinus Torvalds up_write(&sb->s_umount); 12721da177e4SLinus Torvalds return retval; 12731da177e4SLinus Torvalds } 12741da177e4SLinus Torvalds 127597216be0SAl Viro namespace_lock(); 1276962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 12775addc5ddSAl Viro event++; 12781da177e4SLinus Torvalds 1279c35038beSAl Viro if (!(flags & MNT_DETACH)) 1280b54b9be7SAl Viro shrink_submounts(mnt); 1281c35038beSAl Viro 12821da177e4SLinus Torvalds retval = -EBUSY; 1283a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 12841a4eeaf2SAl Viro if (!list_empty(&mnt->mnt_list)) 1285328e6d90SAl Viro umount_tree(mnt, 1); 12861da177e4SLinus Torvalds retval = 0; 12871da177e4SLinus Torvalds } 1288962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 1289e3197d83SAl Viro namespace_unlock(); 12901da177e4SLinus Torvalds return retval; 12911da177e4SLinus Torvalds } 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds /* 12949b40bc90SAl Viro * Is the caller allowed to modify his namespace? 12959b40bc90SAl Viro */ 12969b40bc90SAl Viro static inline bool may_mount(void) 12979b40bc90SAl Viro { 12989b40bc90SAl Viro return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN); 12999b40bc90SAl Viro } 13009b40bc90SAl Viro 13019b40bc90SAl Viro /* 13021da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 13031da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 13041da177e4SLinus Torvalds * 13051da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 13061da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 13071da177e4SLinus Torvalds */ 13081da177e4SLinus Torvalds 1309bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 13101da177e4SLinus Torvalds { 13112d8f3038SAl Viro struct path path; 1312900148dcSAl Viro struct mount *mnt; 13131da177e4SLinus Torvalds int retval; 1314db1f05bbSMiklos Szeredi int lookup_flags = 0; 13151da177e4SLinus Torvalds 1316db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1317db1f05bbSMiklos Szeredi return -EINVAL; 1318db1f05bbSMiklos Szeredi 13199b40bc90SAl Viro if (!may_mount()) 13209b40bc90SAl Viro return -EPERM; 13219b40bc90SAl Viro 1322db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1323db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1324db1f05bbSMiklos Szeredi 1325db1f05bbSMiklos Szeredi retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); 13261da177e4SLinus Torvalds if (retval) 13271da177e4SLinus Torvalds goto out; 1328900148dcSAl Viro mnt = real_mount(path.mnt); 13291da177e4SLinus Torvalds retval = -EINVAL; 13302d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 13311da177e4SLinus Torvalds goto dput_and_out; 1332143c8c91SAl Viro if (!check_mnt(mnt)) 13331da177e4SLinus Torvalds goto dput_and_out; 13345ff9d8a6SEric W. Biederman if (mnt->mnt.mnt_flags & MNT_LOCKED) 13355ff9d8a6SEric W. Biederman goto dput_and_out; 13361da177e4SLinus Torvalds 1337900148dcSAl Viro retval = do_umount(mnt, flags); 13381da177e4SLinus Torvalds dput_and_out: 1339429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 13402d8f3038SAl Viro dput(path.dentry); 1341900148dcSAl Viro mntput_no_expire(mnt); 13421da177e4SLinus Torvalds out: 13431da177e4SLinus Torvalds return retval; 13441da177e4SLinus Torvalds } 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 13471da177e4SLinus Torvalds 13481da177e4SLinus Torvalds /* 13491da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 13501da177e4SLinus Torvalds */ 1351bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 13521da177e4SLinus Torvalds { 13531da177e4SLinus Torvalds return sys_umount(name, 0); 13541da177e4SLinus Torvalds } 13551da177e4SLinus Torvalds 13561da177e4SLinus Torvalds #endif 13571da177e4SLinus Torvalds 13584ce5d2b1SEric W. Biederman static bool is_mnt_ns_file(struct dentry *dentry) 13598823c079SEric W. Biederman { 13604ce5d2b1SEric W. Biederman /* Is this a proxy for a mount namespace? */ 13614ce5d2b1SEric W. Biederman struct inode *inode = dentry->d_inode; 13620bb80f24SDavid Howells struct proc_ns *ei; 13638823c079SEric W. Biederman 13648823c079SEric W. Biederman if (!proc_ns_inode(inode)) 13658823c079SEric W. Biederman return false; 13668823c079SEric W. Biederman 13670bb80f24SDavid Howells ei = get_proc_ns(inode); 13688823c079SEric W. Biederman if (ei->ns_ops != &mntns_operations) 13698823c079SEric W. Biederman return false; 13708823c079SEric W. Biederman 13714ce5d2b1SEric W. Biederman return true; 13724ce5d2b1SEric W. Biederman } 13734ce5d2b1SEric W. Biederman 13744ce5d2b1SEric W. Biederman static bool mnt_ns_loop(struct dentry *dentry) 13754ce5d2b1SEric W. Biederman { 13764ce5d2b1SEric W. Biederman /* Could bind mounting the mount namespace inode cause a 13774ce5d2b1SEric W. Biederman * mount namespace loop? 13784ce5d2b1SEric W. Biederman */ 13794ce5d2b1SEric W. Biederman struct mnt_namespace *mnt_ns; 13804ce5d2b1SEric W. Biederman if (!is_mnt_ns_file(dentry)) 13814ce5d2b1SEric W. Biederman return false; 13824ce5d2b1SEric W. Biederman 13834ce5d2b1SEric W. Biederman mnt_ns = get_proc_ns(dentry->d_inode)->ns; 13848823c079SEric W. Biederman return current->nsproxy->mnt_ns->seq >= mnt_ns->seq; 13858823c079SEric W. Biederman } 13868823c079SEric W. Biederman 138787129cc0SAl Viro struct mount *copy_tree(struct mount *mnt, struct dentry *dentry, 138836341f64SRam Pai int flag) 13891da177e4SLinus Torvalds { 139084d17192SAl Viro struct mount *res, *p, *q, *r, *parent; 13911da177e4SLinus Torvalds 13924ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt)) 13934ce5d2b1SEric W. Biederman return ERR_PTR(-EINVAL); 13944ce5d2b1SEric W. Biederman 13954ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry)) 1396be34d1a3SDavid Howells return ERR_PTR(-EINVAL); 13979676f0c6SRam Pai 139836341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 1399be34d1a3SDavid Howells if (IS_ERR(q)) 1400be34d1a3SDavid Howells return q; 1401be34d1a3SDavid Howells 14025ff9d8a6SEric W. Biederman q->mnt.mnt_flags &= ~MNT_LOCKED; 1403a73324daSAl Viro q->mnt_mountpoint = mnt->mnt_mountpoint; 14041da177e4SLinus Torvalds 14051da177e4SLinus Torvalds p = mnt; 14066b41d536SAl Viro list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 1407315fc83eSAl Viro struct mount *s; 14087ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 14091da177e4SLinus Torvalds continue; 14101da177e4SLinus Torvalds 1411909b0a88SAl Viro for (s = r; s; s = next_mnt(s, r)) { 14124ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_UNBINDABLE) && 14134ce5d2b1SEric W. Biederman IS_MNT_UNBINDABLE(s)) { 14144ce5d2b1SEric W. Biederman s = skip_mnt_tree(s); 14154ce5d2b1SEric W. Biederman continue; 14164ce5d2b1SEric W. Biederman } 14174ce5d2b1SEric W. Biederman if (!(flag & CL_COPY_MNT_NS_FILE) && 14184ce5d2b1SEric W. Biederman is_mnt_ns_file(s->mnt.mnt_root)) { 14199676f0c6SRam Pai s = skip_mnt_tree(s); 14209676f0c6SRam Pai continue; 14219676f0c6SRam Pai } 14220714a533SAl Viro while (p != s->mnt_parent) { 14230714a533SAl Viro p = p->mnt_parent; 14240714a533SAl Viro q = q->mnt_parent; 14251da177e4SLinus Torvalds } 142687129cc0SAl Viro p = s; 142784d17192SAl Viro parent = q; 142887129cc0SAl Viro q = clone_mnt(p, p->mnt.mnt_root, flag); 1429be34d1a3SDavid Howells if (IS_ERR(q)) 1430be34d1a3SDavid Howells goto out; 1431962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 14321a4eeaf2SAl Viro list_add_tail(&q->mnt_list, &res->mnt_list); 143384d17192SAl Viro attach_mnt(q, parent, p->mnt_mp); 1434962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 14351da177e4SLinus Torvalds } 14361da177e4SLinus Torvalds } 14371da177e4SLinus Torvalds return res; 1438be34d1a3SDavid Howells out: 14391da177e4SLinus Torvalds if (res) { 1440962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1441328e6d90SAl Viro umount_tree(res, 0); 1442962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 14431da177e4SLinus Torvalds } 1444be34d1a3SDavid Howells return q; 14451da177e4SLinus Torvalds } 14461da177e4SLinus Torvalds 1447be34d1a3SDavid Howells /* Caller should check returned pointer for errors */ 1448be34d1a3SDavid Howells 1449589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 14508aec0809SAl Viro { 1451cb338d06SAl Viro struct mount *tree; 145297216be0SAl Viro namespace_lock(); 145387129cc0SAl Viro tree = copy_tree(real_mount(path->mnt), path->dentry, 145487129cc0SAl Viro CL_COPY_ALL | CL_PRIVATE); 1455328e6d90SAl Viro namespace_unlock(); 1456be34d1a3SDavid Howells if (IS_ERR(tree)) 1457be34d1a3SDavid Howells return NULL; 1458be34d1a3SDavid Howells return &tree->mnt; 14598aec0809SAl Viro } 14608aec0809SAl Viro 14618aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 14628aec0809SAl Viro { 146397216be0SAl Viro namespace_lock(); 1464962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1465328e6d90SAl Viro umount_tree(real_mount(mnt), 0); 1466962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 14673ab6abeeSAl Viro namespace_unlock(); 14688aec0809SAl Viro } 14698aec0809SAl Viro 14701f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 14711f707137SAl Viro struct vfsmount *root) 14721f707137SAl Viro { 14731a4eeaf2SAl Viro struct mount *mnt; 14741f707137SAl Viro int res = f(root, arg); 14751f707137SAl Viro if (res) 14761f707137SAl Viro return res; 14771a4eeaf2SAl Viro list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) { 14781a4eeaf2SAl Viro res = f(&mnt->mnt, arg); 14791f707137SAl Viro if (res) 14801f707137SAl Viro return res; 14811f707137SAl Viro } 14821f707137SAl Viro return 0; 14831f707137SAl Viro } 14841f707137SAl Viro 14854b8b21f4SAl Viro static void cleanup_group_ids(struct mount *mnt, struct mount *end) 1486719f5d7fSMiklos Szeredi { 1487315fc83eSAl Viro struct mount *p; 1488719f5d7fSMiklos Szeredi 1489909b0a88SAl Viro for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1490fc7be130SAl Viro if (p->mnt_group_id && !IS_MNT_SHARED(p)) 14914b8b21f4SAl Viro mnt_release_group_id(p); 1492719f5d7fSMiklos Szeredi } 1493719f5d7fSMiklos Szeredi } 1494719f5d7fSMiklos Szeredi 14954b8b21f4SAl Viro static int invent_group_ids(struct mount *mnt, bool recurse) 1496719f5d7fSMiklos Szeredi { 1497315fc83eSAl Viro struct mount *p; 1498719f5d7fSMiklos Szeredi 1499909b0a88SAl Viro for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1500fc7be130SAl Viro if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 15014b8b21f4SAl Viro int err = mnt_alloc_group_id(p); 1502719f5d7fSMiklos Szeredi if (err) { 15034b8b21f4SAl Viro cleanup_group_ids(mnt, p); 1504719f5d7fSMiklos Szeredi return err; 1505719f5d7fSMiklos Szeredi } 1506719f5d7fSMiklos Szeredi } 1507719f5d7fSMiklos Szeredi } 1508719f5d7fSMiklos Szeredi 1509719f5d7fSMiklos Szeredi return 0; 1510719f5d7fSMiklos Szeredi } 1511719f5d7fSMiklos Szeredi 1512b90fa9aeSRam Pai /* 1513b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1514b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 151521444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 151621444403SRam Pai * store the parent mount and mountpoint dentry. 151721444403SRam Pai * (done when source_mnt is moved) 1518b90fa9aeSRam Pai * 1519b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1520b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 15219676f0c6SRam Pai * --------------------------------------------------------------------------- 1522b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 15239676f0c6SRam Pai * |************************************************************************** 15249676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15259676f0c6SRam Pai * | dest | | | | | 15269676f0c6SRam Pai * | | | | | | | 15279676f0c6SRam Pai * | v | | | | | 15289676f0c6SRam Pai * |************************************************************************** 15299676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 15305afe0022SRam Pai * | | | | | | 15319676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 15329676f0c6SRam Pai * *************************************************************************** 1533b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1534b90fa9aeSRam Pai * destination mount. 1535b90fa9aeSRam Pai * 1536b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1537b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1538b90fa9aeSRam Pai * the peer group of the source mount. 1539b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1540b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1541b90fa9aeSRam Pai * mount. 15425afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 15435afe0022SRam Pai * of the destination mount and the cloned mount is made slave 15445afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 15455afe0022SRam Pai * is marked as 'shared and slave'. 15465afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 15475afe0022SRam Pai * source mount. 15485afe0022SRam Pai * 15499676f0c6SRam Pai * --------------------------------------------------------------------------- 155021444403SRam Pai * | MOVE 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 (*) | unbindable | 15609676f0c6SRam Pai * *************************************************************************** 15615afe0022SRam Pai * 15625afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 15635afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 156421444403SRam Pai * (+*) the mount is moved to the destination. 15655afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 15665afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 15675afe0022SRam Pai * the mount is marked as 'shared and slave'. 15685afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1569b90fa9aeSRam Pai * 1570b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1571b90fa9aeSRam Pai * applied to each mount in the tree. 1572b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1573b90fa9aeSRam Pai * in allocations. 1574b90fa9aeSRam Pai */ 15750fb54e50SAl Viro static int attach_recursive_mnt(struct mount *source_mnt, 157684d17192SAl Viro struct mount *dest_mnt, 157784d17192SAl Viro struct mountpoint *dest_mp, 157884d17192SAl Viro struct path *parent_path) 1579b90fa9aeSRam Pai { 1580b90fa9aeSRam Pai LIST_HEAD(tree_list); 1581315fc83eSAl Viro struct mount *child, *p; 1582719f5d7fSMiklos Szeredi int err; 1583b90fa9aeSRam Pai 1584fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 15850fb54e50SAl Viro err = invent_group_ids(source_mnt, true); 1586719f5d7fSMiklos Szeredi if (err) 1587719f5d7fSMiklos Szeredi goto out; 1588719f5d7fSMiklos Szeredi } 158984d17192SAl Viro err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list); 1590719f5d7fSMiklos Szeredi if (err) 1591719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1592b90fa9aeSRam Pai 1593962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1594df1a1ad2SAl Viro 1595fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) { 1596909b0a88SAl Viro for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 15970f0afb1dSAl Viro set_mnt_shared(p); 1598b90fa9aeSRam Pai } 15991a390689SAl Viro if (parent_path) { 16000fb54e50SAl Viro detach_mnt(source_mnt, parent_path); 160184d17192SAl Viro attach_mnt(source_mnt, dest_mnt, dest_mp); 1602143c8c91SAl Viro touch_mnt_namespace(source_mnt->mnt_ns); 160321444403SRam Pai } else { 160484d17192SAl Viro mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt); 16050fb54e50SAl Viro commit_tree(source_mnt); 160621444403SRam Pai } 1607b90fa9aeSRam Pai 16081b8e5564SAl Viro list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 16091b8e5564SAl Viro list_del_init(&child->mnt_hash); 16104b2619a5SAl Viro commit_tree(child); 1611b90fa9aeSRam Pai } 1612962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 161399b7db7bSNick Piggin 1614b90fa9aeSRam Pai return 0; 1615719f5d7fSMiklos Szeredi 1616719f5d7fSMiklos Szeredi out_cleanup_ids: 1617fc7be130SAl Viro if (IS_MNT_SHARED(dest_mnt)) 16180fb54e50SAl Viro cleanup_group_ids(source_mnt, NULL); 1619719f5d7fSMiklos Szeredi out: 1620719f5d7fSMiklos Szeredi return err; 1621b90fa9aeSRam Pai } 1622b90fa9aeSRam Pai 162384d17192SAl Viro static struct mountpoint *lock_mount(struct path *path) 1624b12cea91SAl Viro { 1625b12cea91SAl Viro struct vfsmount *mnt; 162684d17192SAl Viro struct dentry *dentry = path->dentry; 1627b12cea91SAl Viro retry: 162884d17192SAl Viro mutex_lock(&dentry->d_inode->i_mutex); 162984d17192SAl Viro if (unlikely(cant_mount(dentry))) { 163084d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 163184d17192SAl Viro return ERR_PTR(-ENOENT); 1632b12cea91SAl Viro } 163397216be0SAl Viro namespace_lock(); 1634b12cea91SAl Viro mnt = lookup_mnt(path); 163584d17192SAl Viro if (likely(!mnt)) { 163684d17192SAl Viro struct mountpoint *mp = new_mountpoint(dentry); 163784d17192SAl Viro if (IS_ERR(mp)) { 163897216be0SAl Viro namespace_unlock(); 163984d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 164084d17192SAl Viro return mp; 164184d17192SAl Viro } 164284d17192SAl Viro return mp; 164384d17192SAl Viro } 164497216be0SAl Viro namespace_unlock(); 1645b12cea91SAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 1646b12cea91SAl Viro path_put(path); 1647b12cea91SAl Viro path->mnt = mnt; 164884d17192SAl Viro dentry = path->dentry = dget(mnt->mnt_root); 1649b12cea91SAl Viro goto retry; 1650b12cea91SAl Viro } 1651b12cea91SAl Viro 165284d17192SAl Viro static void unlock_mount(struct mountpoint *where) 1653b12cea91SAl Viro { 165484d17192SAl Viro struct dentry *dentry = where->m_dentry; 165584d17192SAl Viro put_mountpoint(where); 1656328e6d90SAl Viro namespace_unlock(); 165784d17192SAl Viro mutex_unlock(&dentry->d_inode->i_mutex); 1658b12cea91SAl Viro } 1659b12cea91SAl Viro 166084d17192SAl Viro static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp) 16611da177e4SLinus Torvalds { 166295bc5f25SAl Viro if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER) 16631da177e4SLinus Torvalds return -EINVAL; 16641da177e4SLinus Torvalds 166584d17192SAl Viro if (S_ISDIR(mp->m_dentry->d_inode->i_mode) != 166695bc5f25SAl Viro S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode)) 16671da177e4SLinus Torvalds return -ENOTDIR; 16681da177e4SLinus Torvalds 166984d17192SAl Viro return attach_recursive_mnt(mnt, p, mp, NULL); 16701da177e4SLinus Torvalds } 16711da177e4SLinus Torvalds 16721da177e4SLinus Torvalds /* 16737a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 16747a2e8a8fSValerie Aurora */ 16757a2e8a8fSValerie Aurora 16767a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 16777a2e8a8fSValerie Aurora { 16787c6e984dSRoman Borisov int type = flags & ~(MS_REC | MS_SILENT); 16797a2e8a8fSValerie Aurora 16807a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 16817a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 16827a2e8a8fSValerie Aurora return 0; 16837a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 16847a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 16857a2e8a8fSValerie Aurora return 0; 16867a2e8a8fSValerie Aurora return type; 16877a2e8a8fSValerie Aurora } 16887a2e8a8fSValerie Aurora 16897a2e8a8fSValerie Aurora /* 169007b20889SRam Pai * recursively change the type of the mountpoint. 169107b20889SRam Pai */ 16920a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 169307b20889SRam Pai { 1694315fc83eSAl Viro struct mount *m; 16954b8b21f4SAl Viro struct mount *mnt = real_mount(path->mnt); 169607b20889SRam Pai int recurse = flag & MS_REC; 16977a2e8a8fSValerie Aurora int type; 1698719f5d7fSMiklos Szeredi int err = 0; 169907b20889SRam Pai 17002d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 170107b20889SRam Pai return -EINVAL; 170207b20889SRam Pai 17037a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 17047a2e8a8fSValerie Aurora if (!type) 17057a2e8a8fSValerie Aurora return -EINVAL; 17067a2e8a8fSValerie Aurora 170797216be0SAl Viro namespace_lock(); 1708719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1709719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1710719f5d7fSMiklos Szeredi if (err) 1711719f5d7fSMiklos Szeredi goto out_unlock; 1712719f5d7fSMiklos Szeredi } 1713719f5d7fSMiklos Szeredi 1714962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1715909b0a88SAl Viro for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 17160f0afb1dSAl Viro change_mnt_propagation(m, type); 1717962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 1718719f5d7fSMiklos Szeredi 1719719f5d7fSMiklos Szeredi out_unlock: 172097216be0SAl Viro namespace_unlock(); 1721719f5d7fSMiklos Szeredi return err; 172207b20889SRam Pai } 172307b20889SRam Pai 17245ff9d8a6SEric W. Biederman static bool has_locked_children(struct mount *mnt, struct dentry *dentry) 17255ff9d8a6SEric W. Biederman { 17265ff9d8a6SEric W. Biederman struct mount *child; 17275ff9d8a6SEric W. Biederman list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) { 17285ff9d8a6SEric W. Biederman if (!is_subdir(child->mnt_mountpoint, dentry)) 17295ff9d8a6SEric W. Biederman continue; 17305ff9d8a6SEric W. Biederman 17315ff9d8a6SEric W. Biederman if (child->mnt.mnt_flags & MNT_LOCKED) 17325ff9d8a6SEric W. Biederman return true; 17335ff9d8a6SEric W. Biederman } 17345ff9d8a6SEric W. Biederman return false; 17355ff9d8a6SEric W. Biederman } 17365ff9d8a6SEric W. Biederman 173707b20889SRam Pai /* 17381da177e4SLinus Torvalds * do loopback mount. 17391da177e4SLinus Torvalds */ 1740808d4e3cSAl Viro static int do_loopback(struct path *path, const char *old_name, 17412dafe1c4SEric Sandeen int recurse) 17421da177e4SLinus Torvalds { 17432d92ab3cSAl Viro struct path old_path; 174484d17192SAl Viro struct mount *mnt = NULL, *old, *parent; 174584d17192SAl Viro struct mountpoint *mp; 174657eccb83SAl Viro int err; 17471da177e4SLinus Torvalds if (!old_name || !*old_name) 17481da177e4SLinus Torvalds return -EINVAL; 1749815d405cSTrond Myklebust err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path); 17501da177e4SLinus Torvalds if (err) 17511da177e4SLinus Torvalds return err; 17521da177e4SLinus Torvalds 17538823c079SEric W. Biederman err = -EINVAL; 17544ce5d2b1SEric W. Biederman if (mnt_ns_loop(old_path.dentry)) 17558823c079SEric W. Biederman goto out; 17568823c079SEric W. Biederman 175784d17192SAl Viro mp = lock_mount(path); 175884d17192SAl Viro err = PTR_ERR(mp); 175984d17192SAl Viro if (IS_ERR(mp)) 17609676f0c6SRam Pai goto out; 17619676f0c6SRam Pai 176287129cc0SAl Viro old = real_mount(old_path.mnt); 176384d17192SAl Viro parent = real_mount(path->mnt); 176487129cc0SAl Viro 1765b12cea91SAl Viro err = -EINVAL; 1766fc7be130SAl Viro if (IS_MNT_UNBINDABLE(old)) 1767b12cea91SAl Viro goto out2; 1768b12cea91SAl Viro 176984d17192SAl Viro if (!check_mnt(parent) || !check_mnt(old)) 1770b12cea91SAl Viro goto out2; 1771ccd48bc7SAl Viro 17725ff9d8a6SEric W. Biederman if (!recurse && has_locked_children(old, old_path.dentry)) 17735ff9d8a6SEric W. Biederman goto out2; 17745ff9d8a6SEric W. Biederman 17751da177e4SLinus Torvalds if (recurse) 17764ce5d2b1SEric W. Biederman mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE); 17771da177e4SLinus Torvalds else 177887129cc0SAl Viro mnt = clone_mnt(old, old_path.dentry, 0); 17791da177e4SLinus Torvalds 1780be34d1a3SDavid Howells if (IS_ERR(mnt)) { 1781be34d1a3SDavid Howells err = PTR_ERR(mnt); 1782e9c5d8a5SAndrey Vagin goto out2; 1783be34d1a3SDavid Howells } 1784ccd48bc7SAl Viro 17855ff9d8a6SEric W. Biederman mnt->mnt.mnt_flags &= ~MNT_LOCKED; 17865ff9d8a6SEric W. Biederman 178784d17192SAl Viro err = graft_tree(mnt, parent, mp); 17881da177e4SLinus Torvalds if (err) { 1789962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1790328e6d90SAl Viro umount_tree(mnt, 0); 1791962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 17925b83d2c5SRam Pai } 1793b12cea91SAl Viro out2: 179484d17192SAl Viro unlock_mount(mp); 1795ccd48bc7SAl Viro out: 17962d92ab3cSAl Viro path_put(&old_path); 17971da177e4SLinus Torvalds return err; 17981da177e4SLinus Torvalds } 17991da177e4SLinus Torvalds 18002e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 18012e4b7fcdSDave Hansen { 18022e4b7fcdSDave Hansen int error = 0; 18032e4b7fcdSDave Hansen int readonly_request = 0; 18042e4b7fcdSDave Hansen 18052e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 18062e4b7fcdSDave Hansen readonly_request = 1; 18072e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 18082e4b7fcdSDave Hansen return 0; 18092e4b7fcdSDave Hansen 181090563b19SEric W. Biederman if (mnt->mnt_flags & MNT_LOCK_READONLY) 181190563b19SEric W. Biederman return -EPERM; 181290563b19SEric W. Biederman 18132e4b7fcdSDave Hansen if (readonly_request) 181483adc753SAl Viro error = mnt_make_readonly(real_mount(mnt)); 18152e4b7fcdSDave Hansen else 181683adc753SAl Viro __mnt_unmake_readonly(real_mount(mnt)); 18172e4b7fcdSDave Hansen return error; 18182e4b7fcdSDave Hansen } 18192e4b7fcdSDave Hansen 18201da177e4SLinus Torvalds /* 18211da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 18221da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 18231da177e4SLinus Torvalds * on it - tough luck. 18241da177e4SLinus Torvalds */ 18250a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 18261da177e4SLinus Torvalds void *data) 18271da177e4SLinus Torvalds { 18281da177e4SLinus Torvalds int err; 18292d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 1830143c8c91SAl Viro struct mount *mnt = real_mount(path->mnt); 18311da177e4SLinus Torvalds 1832143c8c91SAl Viro if (!check_mnt(mnt)) 18331da177e4SLinus Torvalds return -EINVAL; 18341da177e4SLinus Torvalds 18352d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 18361da177e4SLinus Torvalds return -EINVAL; 18371da177e4SLinus Torvalds 1838ff36fe2cSEric Paris err = security_sb_remount(sb, data); 1839ff36fe2cSEric Paris if (err) 1840ff36fe2cSEric Paris return err; 1841ff36fe2cSEric Paris 18421da177e4SLinus Torvalds down_write(&sb->s_umount); 18432e4b7fcdSDave Hansen if (flags & MS_BIND) 18442d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 184557eccb83SAl Viro else if (!capable(CAP_SYS_ADMIN)) 184657eccb83SAl Viro err = -EPERM; 18474aa98cf7SAl Viro else 18481da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 18497b43a79fSAl Viro if (!err) { 1850962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1851143c8c91SAl Viro mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK; 1852143c8c91SAl Viro mnt->mnt.mnt_flags = mnt_flags; 1853962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 18547b43a79fSAl Viro } 18551da177e4SLinus Torvalds up_write(&sb->s_umount); 18560e55a7ccSDan Williams if (!err) { 1857962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 1858143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1859962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 18600e55a7ccSDan Williams } 18611da177e4SLinus Torvalds return err; 18621da177e4SLinus Torvalds } 18631da177e4SLinus Torvalds 1864cbbe362cSAl Viro static inline int tree_contains_unbindable(struct mount *mnt) 18659676f0c6SRam Pai { 1866315fc83eSAl Viro struct mount *p; 1867909b0a88SAl Viro for (p = mnt; p; p = next_mnt(p, mnt)) { 1868fc7be130SAl Viro if (IS_MNT_UNBINDABLE(p)) 18699676f0c6SRam Pai return 1; 18709676f0c6SRam Pai } 18719676f0c6SRam Pai return 0; 18729676f0c6SRam Pai } 18739676f0c6SRam Pai 1874808d4e3cSAl Viro static int do_move_mount(struct path *path, const char *old_name) 18751da177e4SLinus Torvalds { 18762d92ab3cSAl Viro struct path old_path, parent_path; 1877676da58dSAl Viro struct mount *p; 18780fb54e50SAl Viro struct mount *old; 187984d17192SAl Viro struct mountpoint *mp; 188057eccb83SAl Viro int err; 18811da177e4SLinus Torvalds if (!old_name || !*old_name) 18821da177e4SLinus Torvalds return -EINVAL; 18832d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 18841da177e4SLinus Torvalds if (err) 18851da177e4SLinus Torvalds return err; 18861da177e4SLinus Torvalds 188784d17192SAl Viro mp = lock_mount(path); 188884d17192SAl Viro err = PTR_ERR(mp); 188984d17192SAl Viro if (IS_ERR(mp)) 1890cc53ce53SDavid Howells goto out; 1891cc53ce53SDavid Howells 1892143c8c91SAl Viro old = real_mount(old_path.mnt); 1893fc7be130SAl Viro p = real_mount(path->mnt); 1894143c8c91SAl Viro 18951da177e4SLinus Torvalds err = -EINVAL; 1896fc7be130SAl Viro if (!check_mnt(p) || !check_mnt(old)) 18971da177e4SLinus Torvalds goto out1; 18981da177e4SLinus Torvalds 18995ff9d8a6SEric W. Biederman if (old->mnt.mnt_flags & MNT_LOCKED) 19005ff9d8a6SEric W. Biederman goto out1; 19015ff9d8a6SEric W. Biederman 19021da177e4SLinus Torvalds err = -EINVAL; 19032d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 190421444403SRam Pai goto out1; 19051da177e4SLinus Torvalds 1906676da58dSAl Viro if (!mnt_has_parent(old)) 190721444403SRam Pai goto out1; 19081da177e4SLinus Torvalds 19092d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 19102d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 191121444403SRam Pai goto out1; 191221444403SRam Pai /* 191321444403SRam Pai * Don't move a mount residing in a shared parent. 191421444403SRam Pai */ 1915fc7be130SAl Viro if (IS_MNT_SHARED(old->mnt_parent)) 191621444403SRam Pai goto out1; 19179676f0c6SRam Pai /* 19189676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 19199676f0c6SRam Pai * mount which is shared. 19209676f0c6SRam Pai */ 1921fc7be130SAl Viro if (IS_MNT_SHARED(p) && tree_contains_unbindable(old)) 19229676f0c6SRam Pai goto out1; 19231da177e4SLinus Torvalds err = -ELOOP; 1924fc7be130SAl Viro for (; mnt_has_parent(p); p = p->mnt_parent) 1925676da58dSAl Viro if (p == old) 192621444403SRam Pai goto out1; 19271da177e4SLinus Torvalds 192884d17192SAl Viro err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path); 19294ac91378SJan Blunck if (err) 193021444403SRam Pai goto out1; 19311da177e4SLinus Torvalds 19321da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 19331da177e4SLinus Torvalds * automatically */ 19346776db3dSAl Viro list_del_init(&old->mnt_expire); 19351da177e4SLinus Torvalds out1: 193684d17192SAl Viro unlock_mount(mp); 19371da177e4SLinus Torvalds out: 19381da177e4SLinus Torvalds if (!err) 19391a390689SAl Viro path_put(&parent_path); 19402d92ab3cSAl Viro path_put(&old_path); 19411da177e4SLinus Torvalds return err; 19421da177e4SLinus Torvalds } 19431da177e4SLinus Torvalds 19449d412a43SAl Viro static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype) 19459d412a43SAl Viro { 19469d412a43SAl Viro int err; 19479d412a43SAl Viro const char *subtype = strchr(fstype, '.'); 19489d412a43SAl Viro if (subtype) { 19499d412a43SAl Viro subtype++; 19509d412a43SAl Viro err = -EINVAL; 19519d412a43SAl Viro if (!subtype[0]) 19529d412a43SAl Viro goto err; 19539d412a43SAl Viro } else 19549d412a43SAl Viro subtype = ""; 19559d412a43SAl Viro 19569d412a43SAl Viro mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL); 19579d412a43SAl Viro err = -ENOMEM; 19589d412a43SAl Viro if (!mnt->mnt_sb->s_subtype) 19599d412a43SAl Viro goto err; 19609d412a43SAl Viro return mnt; 19619d412a43SAl Viro 19629d412a43SAl Viro err: 19639d412a43SAl Viro mntput(mnt); 19649d412a43SAl Viro return ERR_PTR(err); 19659d412a43SAl Viro } 19669d412a43SAl Viro 19679d412a43SAl Viro /* 19689d412a43SAl Viro * add a mount into a namespace's mount tree 19699d412a43SAl Viro */ 197095bc5f25SAl Viro static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags) 19719d412a43SAl Viro { 197284d17192SAl Viro struct mountpoint *mp; 197384d17192SAl Viro struct mount *parent; 19749d412a43SAl Viro int err; 19759d412a43SAl Viro 19769d412a43SAl Viro mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); 19779d412a43SAl Viro 197884d17192SAl Viro mp = lock_mount(path); 197984d17192SAl Viro if (IS_ERR(mp)) 198084d17192SAl Viro return PTR_ERR(mp); 19819d412a43SAl Viro 198284d17192SAl Viro parent = real_mount(path->mnt); 19839d412a43SAl Viro err = -EINVAL; 198484d17192SAl Viro if (unlikely(!check_mnt(parent))) { 1985156cacb1SAl Viro /* that's acceptable only for automounts done in private ns */ 1986156cacb1SAl Viro if (!(mnt_flags & MNT_SHRINKABLE)) 19879d412a43SAl Viro goto unlock; 1988156cacb1SAl Viro /* ... and for those we'd better have mountpoint still alive */ 198984d17192SAl Viro if (!parent->mnt_ns) 1990156cacb1SAl Viro goto unlock; 1991156cacb1SAl Viro } 19929d412a43SAl Viro 19939d412a43SAl Viro /* Refuse the same filesystem on the same mount point */ 19949d412a43SAl Viro err = -EBUSY; 199595bc5f25SAl Viro if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb && 19969d412a43SAl Viro path->mnt->mnt_root == path->dentry) 19979d412a43SAl Viro goto unlock; 19989d412a43SAl Viro 19999d412a43SAl Viro err = -EINVAL; 200095bc5f25SAl Viro if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode)) 20019d412a43SAl Viro goto unlock; 20029d412a43SAl Viro 200395bc5f25SAl Viro newmnt->mnt.mnt_flags = mnt_flags; 200484d17192SAl Viro err = graft_tree(newmnt, parent, mp); 20059d412a43SAl Viro 20069d412a43SAl Viro unlock: 200784d17192SAl Viro unlock_mount(mp); 20089d412a43SAl Viro return err; 20099d412a43SAl Viro } 2010b1e75df4SAl Viro 20111da177e4SLinus Torvalds /* 20121da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 20131da177e4SLinus Torvalds * namespace's tree 20141da177e4SLinus Torvalds */ 20150c55cfc4SEric W. Biederman static int do_new_mount(struct path *path, const char *fstype, int flags, 2016808d4e3cSAl Viro int mnt_flags, const char *name, void *data) 20171da177e4SLinus Torvalds { 20180c55cfc4SEric W. Biederman struct file_system_type *type; 20199b40bc90SAl Viro struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns; 20201da177e4SLinus Torvalds struct vfsmount *mnt; 202115f9a3f3SAl Viro int err; 20221da177e4SLinus Torvalds 20230c55cfc4SEric W. Biederman if (!fstype) 20241da177e4SLinus Torvalds return -EINVAL; 20251da177e4SLinus Torvalds 20260c55cfc4SEric W. Biederman type = get_fs_type(fstype); 20270c55cfc4SEric W. Biederman if (!type) 20280c55cfc4SEric W. Biederman return -ENODEV; 20290c55cfc4SEric W. Biederman 20300c55cfc4SEric W. Biederman if (user_ns != &init_user_ns) { 20310c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_MOUNT)) { 20320c55cfc4SEric W. Biederman put_filesystem(type); 20330c55cfc4SEric W. Biederman return -EPERM; 20340c55cfc4SEric W. Biederman } 20350c55cfc4SEric W. Biederman /* Only in special cases allow devices from mounts 20360c55cfc4SEric W. Biederman * created outside the initial user namespace. 20370c55cfc4SEric W. Biederman */ 20380c55cfc4SEric W. Biederman if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) { 20390c55cfc4SEric W. Biederman flags |= MS_NODEV; 20400c55cfc4SEric W. Biederman mnt_flags |= MNT_NODEV; 20410c55cfc4SEric W. Biederman } 20420c55cfc4SEric W. Biederman } 20430c55cfc4SEric W. Biederman 20440c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, flags, name, data); 20450c55cfc4SEric W. Biederman if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) && 20460c55cfc4SEric W. Biederman !mnt->mnt_sb->s_subtype) 20470c55cfc4SEric W. Biederman mnt = fs_set_subtype(mnt, fstype); 20480c55cfc4SEric W. Biederman 20490c55cfc4SEric W. Biederman put_filesystem(type); 20501da177e4SLinus Torvalds if (IS_ERR(mnt)) 20511da177e4SLinus Torvalds return PTR_ERR(mnt); 20521da177e4SLinus Torvalds 205395bc5f25SAl Viro err = do_add_mount(real_mount(mnt), path, mnt_flags); 205415f9a3f3SAl Viro if (err) 205515f9a3f3SAl Viro mntput(mnt); 205615f9a3f3SAl Viro return err; 20571da177e4SLinus Torvalds } 20581da177e4SLinus Torvalds 205919a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 206019a167afSAl Viro { 20616776db3dSAl Viro struct mount *mnt = real_mount(m); 206219a167afSAl Viro int err; 206319a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 206419a167afSAl Viro * expired before we get a chance to add it 206519a167afSAl Viro */ 20666776db3dSAl Viro BUG_ON(mnt_get_count(mnt) < 2); 206719a167afSAl Viro 206819a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 206919a167afSAl Viro m->mnt_root == path->dentry) { 2070b1e75df4SAl Viro err = -ELOOP; 2071b1e75df4SAl Viro goto fail; 207219a167afSAl Viro } 207319a167afSAl Viro 207495bc5f25SAl Viro err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 2075b1e75df4SAl Viro if (!err) 2076b1e75df4SAl Viro return 0; 2077b1e75df4SAl Viro fail: 2078b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 20796776db3dSAl Viro if (!list_empty(&mnt->mnt_expire)) { 208097216be0SAl Viro namespace_lock(); 2081962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 20826776db3dSAl Viro list_del_init(&mnt->mnt_expire); 2083962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 208497216be0SAl Viro namespace_unlock(); 208519a167afSAl Viro } 2086b1e75df4SAl Viro mntput(m); 2087b1e75df4SAl Viro mntput(m); 208819a167afSAl Viro return err; 208919a167afSAl Viro } 209019a167afSAl Viro 2091ea5b778aSDavid Howells /** 2092ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 2093ea5b778aSDavid Howells * @mnt: The mount to list. 2094ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 2095ea5b778aSDavid Howells */ 2096ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 2097ea5b778aSDavid Howells { 209897216be0SAl Viro namespace_lock(); 2099962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 2100ea5b778aSDavid Howells 21016776db3dSAl Viro list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list); 2102ea5b778aSDavid Howells 2103962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 210497216be0SAl Viro namespace_unlock(); 2105ea5b778aSDavid Howells } 2106ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 2107ea5b778aSDavid Howells 2108ea5b778aSDavid Howells /* 21091da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 21101da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 21111da177e4SLinus Torvalds * here 21121da177e4SLinus Torvalds */ 21131da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 21141da177e4SLinus Torvalds { 2115761d5c38SAl Viro struct mount *mnt, *next; 21161da177e4SLinus Torvalds LIST_HEAD(graveyard); 21171da177e4SLinus Torvalds 21181da177e4SLinus Torvalds if (list_empty(mounts)) 21191da177e4SLinus Torvalds return; 21201da177e4SLinus Torvalds 212197216be0SAl Viro namespace_lock(); 2122962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 21231da177e4SLinus Torvalds 21241da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 21251da177e4SLinus Torvalds * following criteria: 21261da177e4SLinus Torvalds * - only referenced by its parent vfsmount 21271da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 21281da177e4SLinus Torvalds * cleared by mntput()) 21291da177e4SLinus Torvalds */ 21306776db3dSAl Viro list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 2131863d684fSAl Viro if (!xchg(&mnt->mnt_expiry_mark, 1) || 21321ab59738SAl Viro propagate_mount_busy(mnt, 1)) 21331da177e4SLinus Torvalds continue; 21346776db3dSAl Viro list_move(&mnt->mnt_expire, &graveyard); 21351da177e4SLinus Torvalds } 2136bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 21376776db3dSAl Viro mnt = list_first_entry(&graveyard, struct mount, mnt_expire); 2138143c8c91SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2139328e6d90SAl Viro umount_tree(mnt, 1); 2140bcc5c7d2SAl Viro } 2141962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 21423ab6abeeSAl Viro namespace_unlock(); 21431da177e4SLinus Torvalds } 21441da177e4SLinus Torvalds 21451da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 21461da177e4SLinus Torvalds 21471da177e4SLinus Torvalds /* 21485528f911STrond Myklebust * Ripoff of 'select_parent()' 21495528f911STrond Myklebust * 21505528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 21515528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 21525528f911STrond Myklebust */ 2153692afc31SAl Viro static int select_submounts(struct mount *parent, struct list_head *graveyard) 21545528f911STrond Myklebust { 2155692afc31SAl Viro struct mount *this_parent = parent; 21565528f911STrond Myklebust struct list_head *next; 21575528f911STrond Myklebust int found = 0; 21585528f911STrond Myklebust 21595528f911STrond Myklebust repeat: 21606b41d536SAl Viro next = this_parent->mnt_mounts.next; 21615528f911STrond Myklebust resume: 21626b41d536SAl Viro while (next != &this_parent->mnt_mounts) { 21635528f911STrond Myklebust struct list_head *tmp = next; 21646b41d536SAl Viro struct mount *mnt = list_entry(tmp, struct mount, mnt_child); 21655528f911STrond Myklebust 21665528f911STrond Myklebust next = tmp->next; 2167692afc31SAl Viro if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE)) 21685528f911STrond Myklebust continue; 21695528f911STrond Myklebust /* 21705528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 21715528f911STrond Myklebust */ 21726b41d536SAl Viro if (!list_empty(&mnt->mnt_mounts)) { 21735528f911STrond Myklebust this_parent = mnt; 21745528f911STrond Myklebust goto repeat; 21755528f911STrond Myklebust } 21765528f911STrond Myklebust 21771ab59738SAl Viro if (!propagate_mount_busy(mnt, 1)) { 21786776db3dSAl Viro list_move_tail(&mnt->mnt_expire, graveyard); 21795528f911STrond Myklebust found++; 21805528f911STrond Myklebust } 21815528f911STrond Myklebust } 21825528f911STrond Myklebust /* 21835528f911STrond Myklebust * All done at this level ... ascend and resume the search 21845528f911STrond Myklebust */ 21855528f911STrond Myklebust if (this_parent != parent) { 21866b41d536SAl Viro next = this_parent->mnt_child.next; 21870714a533SAl Viro this_parent = this_parent->mnt_parent; 21885528f911STrond Myklebust goto resume; 21895528f911STrond Myklebust } 21905528f911STrond Myklebust return found; 21915528f911STrond Myklebust } 21925528f911STrond Myklebust 21935528f911STrond Myklebust /* 21945528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 21955528f911STrond Myklebust * submounts of a specific parent mountpoint 219699b7db7bSNick Piggin * 219799b7db7bSNick Piggin * vfsmount_lock must be held for write 21985528f911STrond Myklebust */ 2199b54b9be7SAl Viro static void shrink_submounts(struct mount *mnt) 22005528f911STrond Myklebust { 22015528f911STrond Myklebust LIST_HEAD(graveyard); 2202761d5c38SAl Viro struct mount *m; 22035528f911STrond Myklebust 22045528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2205c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2206bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2207761d5c38SAl Viro m = list_first_entry(&graveyard, struct mount, 22086776db3dSAl Viro mnt_expire); 2209143c8c91SAl Viro touch_mnt_namespace(m->mnt_ns); 2210328e6d90SAl Viro umount_tree(m, 1); 2211bcc5c7d2SAl Viro } 2212bcc5c7d2SAl Viro } 22135528f911STrond Myklebust } 22145528f911STrond Myklebust 22155528f911STrond Myklebust /* 22161da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 22171da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 22181da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 22191da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 22201da177e4SLinus Torvalds */ 2221b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2222b58fed8bSRam Pai unsigned long n) 22231da177e4SLinus Torvalds { 22241da177e4SLinus Torvalds char *t = to; 22251da177e4SLinus Torvalds const char __user *f = from; 22261da177e4SLinus Torvalds char c; 22271da177e4SLinus Torvalds 22281da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 22291da177e4SLinus Torvalds return n; 22301da177e4SLinus Torvalds 22311da177e4SLinus Torvalds while (n) { 22321da177e4SLinus Torvalds if (__get_user(c, f)) { 22331da177e4SLinus Torvalds memset(t, 0, n); 22341da177e4SLinus Torvalds break; 22351da177e4SLinus Torvalds } 22361da177e4SLinus Torvalds *t++ = c; 22371da177e4SLinus Torvalds f++; 22381da177e4SLinus Torvalds n--; 22391da177e4SLinus Torvalds } 22401da177e4SLinus Torvalds return n; 22411da177e4SLinus Torvalds } 22421da177e4SLinus Torvalds 22431da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 22441da177e4SLinus Torvalds { 22451da177e4SLinus Torvalds int i; 22461da177e4SLinus Torvalds unsigned long page; 22471da177e4SLinus Torvalds unsigned long size; 22481da177e4SLinus Torvalds 22491da177e4SLinus Torvalds *where = 0; 22501da177e4SLinus Torvalds if (!data) 22511da177e4SLinus Torvalds return 0; 22521da177e4SLinus Torvalds 22531da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 22541da177e4SLinus Torvalds return -ENOMEM; 22551da177e4SLinus Torvalds 22561da177e4SLinus Torvalds /* We only care that *some* data at the address the user 22571da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 22581da177e4SLinus Torvalds * the remainder of the page. 22591da177e4SLinus Torvalds */ 22601da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 22611da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 22621da177e4SLinus Torvalds if (size > PAGE_SIZE) 22631da177e4SLinus Torvalds size = PAGE_SIZE; 22641da177e4SLinus Torvalds 22651da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 22661da177e4SLinus Torvalds if (!i) { 22671da177e4SLinus Torvalds free_page(page); 22681da177e4SLinus Torvalds return -EFAULT; 22691da177e4SLinus Torvalds } 22701da177e4SLinus Torvalds if (i != PAGE_SIZE) 22711da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 22721da177e4SLinus Torvalds *where = page; 22731da177e4SLinus Torvalds return 0; 22741da177e4SLinus Torvalds } 22751da177e4SLinus Torvalds 2276eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 2277eca6f534SVegard Nossum { 2278eca6f534SVegard Nossum char *tmp; 2279eca6f534SVegard Nossum 2280eca6f534SVegard Nossum if (!data) { 2281eca6f534SVegard Nossum *where = NULL; 2282eca6f534SVegard Nossum return 0; 2283eca6f534SVegard Nossum } 2284eca6f534SVegard Nossum 2285eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 2286eca6f534SVegard Nossum if (IS_ERR(tmp)) 2287eca6f534SVegard Nossum return PTR_ERR(tmp); 2288eca6f534SVegard Nossum 2289eca6f534SVegard Nossum *where = tmp; 2290eca6f534SVegard Nossum return 0; 2291eca6f534SVegard Nossum } 2292eca6f534SVegard Nossum 22931da177e4SLinus Torvalds /* 22941da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 22951da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 22961da177e4SLinus Torvalds * 22971da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 22981da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 22991da177e4SLinus Torvalds * information (or be NULL). 23001da177e4SLinus Torvalds * 23011da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 23021da177e4SLinus Torvalds * When the flags word was introduced its top half was required 23031da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 23041da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 23051da177e4SLinus Torvalds * and must be discarded. 23061da177e4SLinus Torvalds */ 2307808d4e3cSAl Viro long do_mount(const char *dev_name, const char *dir_name, 2308808d4e3cSAl Viro const char *type_page, unsigned long flags, void *data_page) 23091da177e4SLinus Torvalds { 23102d92ab3cSAl Viro struct path path; 23111da177e4SLinus Torvalds int retval = 0; 23121da177e4SLinus Torvalds int mnt_flags = 0; 23131da177e4SLinus Torvalds 23141da177e4SLinus Torvalds /* Discard magic */ 23151da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 23161da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 23171da177e4SLinus Torvalds 23181da177e4SLinus Torvalds /* Basic sanity checks */ 23191da177e4SLinus Torvalds 23201da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 23211da177e4SLinus Torvalds return -EINVAL; 23221da177e4SLinus Torvalds 23231da177e4SLinus Torvalds if (data_page) 23241da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 23251da177e4SLinus Torvalds 2326a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 2327a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2328a27ab9f2STetsuo Handa if (retval) 2329a27ab9f2STetsuo Handa return retval; 2330a27ab9f2STetsuo Handa 2331a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2332a27ab9f2STetsuo Handa type_page, flags, data_page); 23330d5cadb8SAl Viro if (!retval && !may_mount()) 23340d5cadb8SAl Viro retval = -EPERM; 2335a27ab9f2STetsuo Handa if (retval) 2336a27ab9f2STetsuo Handa goto dput_out; 2337a27ab9f2STetsuo Handa 2338613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2339613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 23400a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 23410a1c01c9SMatthew Garrett 23421da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 23431da177e4SLinus Torvalds if (flags & MS_NOSUID) 23441da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 23451da177e4SLinus Torvalds if (flags & MS_NODEV) 23461da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 23471da177e4SLinus Torvalds if (flags & MS_NOEXEC) 23481da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2349fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2350fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2351fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2352fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2353d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2354d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 23552e4b7fcdSDave Hansen if (flags & MS_RDONLY) 23562e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2357fc33a7bbSChristoph Hellwig 23587a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2359d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2360d0adde57SMatthew Garrett MS_STRICTATIME); 23611da177e4SLinus Torvalds 23621da177e4SLinus Torvalds if (flags & MS_REMOUNT) 23632d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 23641da177e4SLinus Torvalds data_page); 23651da177e4SLinus Torvalds else if (flags & MS_BIND) 23662d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 23679676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 23682d92ab3cSAl Viro retval = do_change_type(&path, flags); 23691da177e4SLinus Torvalds else if (flags & MS_MOVE) 23702d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 23711da177e4SLinus Torvalds else 23722d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 23731da177e4SLinus Torvalds dev_name, data_page); 23741da177e4SLinus Torvalds dput_out: 23752d92ab3cSAl Viro path_put(&path); 23761da177e4SLinus Torvalds return retval; 23771da177e4SLinus Torvalds } 23781da177e4SLinus Torvalds 2379771b1371SEric W. Biederman static void free_mnt_ns(struct mnt_namespace *ns) 2380771b1371SEric W. Biederman { 238198f842e6SEric W. Biederman proc_free_inum(ns->proc_inum); 2382771b1371SEric W. Biederman put_user_ns(ns->user_ns); 2383771b1371SEric W. Biederman kfree(ns); 2384771b1371SEric W. Biederman } 2385771b1371SEric W. Biederman 23868823c079SEric W. Biederman /* 23878823c079SEric W. Biederman * Assign a sequence number so we can detect when we attempt to bind 23888823c079SEric W. Biederman * mount a reference to an older mount namespace into the current 23898823c079SEric W. Biederman * mount namespace, preventing reference counting loops. A 64bit 23908823c079SEric W. Biederman * number incrementing at 10Ghz will take 12,427 years to wrap which 23918823c079SEric W. Biederman * is effectively never, so we can ignore the possibility. 23928823c079SEric W. Biederman */ 23938823c079SEric W. Biederman static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1); 23948823c079SEric W. Biederman 2395771b1371SEric W. Biederman static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns) 2396cf8d2c11STrond Myklebust { 2397cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 239898f842e6SEric W. Biederman int ret; 2399cf8d2c11STrond Myklebust 2400cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2401cf8d2c11STrond Myklebust if (!new_ns) 2402cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 240398f842e6SEric W. Biederman ret = proc_alloc_inum(&new_ns->proc_inum); 240498f842e6SEric W. Biederman if (ret) { 240598f842e6SEric W. Biederman kfree(new_ns); 240698f842e6SEric W. Biederman return ERR_PTR(ret); 240798f842e6SEric W. Biederman } 24088823c079SEric W. Biederman new_ns->seq = atomic64_add_return(1, &mnt_ns_seq); 2409cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2410cf8d2c11STrond Myklebust new_ns->root = NULL; 2411cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2412cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2413cf8d2c11STrond Myklebust new_ns->event = 0; 2414771b1371SEric W. Biederman new_ns->user_ns = get_user_ns(user_ns); 2415cf8d2c11STrond Myklebust return new_ns; 2416cf8d2c11STrond Myklebust } 2417cf8d2c11STrond Myklebust 2418741a2951SJANAK DESAI /* 2419741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 2420741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 2421741a2951SJANAK DESAI */ 2422e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 2423771b1371SEric W. Biederman struct user_namespace *user_ns, struct fs_struct *fs) 24241da177e4SLinus Torvalds { 24256b3286edSKirill Korotaev struct mnt_namespace *new_ns; 24267f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 2427315fc83eSAl Viro struct mount *p, *q; 2428be08d6d2SAl Viro struct mount *old = mnt_ns->root; 2429cb338d06SAl Viro struct mount *new; 24307a472ef4SEric W. Biederman int copy_flags; 24311da177e4SLinus Torvalds 2432771b1371SEric W. Biederman new_ns = alloc_mnt_ns(user_ns); 2433cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2434cf8d2c11STrond Myklebust return new_ns; 24351da177e4SLinus Torvalds 243697216be0SAl Viro namespace_lock(); 24371da177e4SLinus Torvalds /* First pass: copy the tree topology */ 24384ce5d2b1SEric W. Biederman copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE; 24397a472ef4SEric W. Biederman if (user_ns != mnt_ns->user_ns) 2440132c94e3SEric W. Biederman copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED; 24417a472ef4SEric W. Biederman new = copy_tree(old, old->mnt.mnt_root, copy_flags); 2442be34d1a3SDavid Howells if (IS_ERR(new)) { 2443328e6d90SAl Viro namespace_unlock(); 2444771b1371SEric W. Biederman free_mnt_ns(new_ns); 2445be34d1a3SDavid Howells return ERR_CAST(new); 24461da177e4SLinus Torvalds } 2447be08d6d2SAl Viro new_ns->root = new; 2448962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 24491a4eeaf2SAl Viro list_add_tail(&new_ns->list, &new->mnt_list); 2450962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 24511da177e4SLinus Torvalds 24521da177e4SLinus Torvalds /* 24531da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 24541da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 24551da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 24561da177e4SLinus Torvalds */ 2457909b0a88SAl Viro p = old; 2458cb338d06SAl Viro q = new; 24591da177e4SLinus Torvalds while (p) { 2460143c8c91SAl Viro q->mnt_ns = new_ns; 24611da177e4SLinus Torvalds if (fs) { 2462315fc83eSAl Viro if (&p->mnt == fs->root.mnt) { 2463315fc83eSAl Viro fs->root.mnt = mntget(&q->mnt); 2464315fc83eSAl Viro rootmnt = &p->mnt; 24651da177e4SLinus Torvalds } 2466315fc83eSAl Viro if (&p->mnt == fs->pwd.mnt) { 2467315fc83eSAl Viro fs->pwd.mnt = mntget(&q->mnt); 2468315fc83eSAl Viro pwdmnt = &p->mnt; 24691da177e4SLinus Torvalds } 24701da177e4SLinus Torvalds } 2471909b0a88SAl Viro p = next_mnt(p, old); 2472909b0a88SAl Viro q = next_mnt(q, new); 24734ce5d2b1SEric W. Biederman if (!q) 24744ce5d2b1SEric W. Biederman break; 24754ce5d2b1SEric W. Biederman while (p->mnt.mnt_root != q->mnt.mnt_root) 24764ce5d2b1SEric W. Biederman p = next_mnt(p, old); 24771da177e4SLinus Torvalds } 2478328e6d90SAl Viro namespace_unlock(); 24791da177e4SLinus Torvalds 24801da177e4SLinus Torvalds if (rootmnt) 2481f03c6599SAl Viro mntput(rootmnt); 24821da177e4SLinus Torvalds if (pwdmnt) 2483f03c6599SAl Viro mntput(pwdmnt); 24841da177e4SLinus Torvalds 2485741a2951SJANAK DESAI return new_ns; 2486741a2951SJANAK DESAI } 2487741a2951SJANAK DESAI 2488213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2489771b1371SEric W. Biederman struct user_namespace *user_ns, struct fs_struct *new_fs) 2490741a2951SJANAK DESAI { 24916b3286edSKirill Korotaev struct mnt_namespace *new_ns; 2492741a2951SJANAK DESAI 2493e3222c4eSBadari Pulavarty BUG_ON(!ns); 24946b3286edSKirill Korotaev get_mnt_ns(ns); 2495741a2951SJANAK DESAI 2496741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 2497e3222c4eSBadari Pulavarty return ns; 2498741a2951SJANAK DESAI 2499771b1371SEric W. Biederman new_ns = dup_mnt_ns(ns, user_ns, new_fs); 2500741a2951SJANAK DESAI 25016b3286edSKirill Korotaev put_mnt_ns(ns); 2502e3222c4eSBadari Pulavarty return new_ns; 25031da177e4SLinus Torvalds } 25041da177e4SLinus Torvalds 2505cf8d2c11STrond Myklebust /** 2506cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2507cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2508cf8d2c11STrond Myklebust */ 25091a4eeaf2SAl Viro static struct mnt_namespace *create_mnt_ns(struct vfsmount *m) 2510cf8d2c11STrond Myklebust { 2511771b1371SEric W. Biederman struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns); 2512cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 25131a4eeaf2SAl Viro struct mount *mnt = real_mount(m); 25141a4eeaf2SAl Viro mnt->mnt_ns = new_ns; 2515be08d6d2SAl Viro new_ns->root = mnt; 2516b1983cd8SAl Viro list_add(&mnt->mnt_list, &new_ns->list); 2517c1334495SAl Viro } else { 25181a4eeaf2SAl Viro mntput(m); 2519cf8d2c11STrond Myklebust } 2520cf8d2c11STrond Myklebust return new_ns; 2521cf8d2c11STrond Myklebust } 2522cf8d2c11STrond Myklebust 2523ea441d11SAl Viro struct dentry *mount_subtree(struct vfsmount *mnt, const char *name) 2524ea441d11SAl Viro { 2525ea441d11SAl Viro struct mnt_namespace *ns; 2526d31da0f0SAl Viro struct super_block *s; 2527ea441d11SAl Viro struct path path; 2528ea441d11SAl Viro int err; 2529ea441d11SAl Viro 2530ea441d11SAl Viro ns = create_mnt_ns(mnt); 2531ea441d11SAl Viro if (IS_ERR(ns)) 2532ea441d11SAl Viro return ERR_CAST(ns); 2533ea441d11SAl Viro 2534ea441d11SAl Viro err = vfs_path_lookup(mnt->mnt_root, mnt, 2535ea441d11SAl Viro name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path); 2536ea441d11SAl Viro 2537ea441d11SAl Viro put_mnt_ns(ns); 2538ea441d11SAl Viro 2539ea441d11SAl Viro if (err) 2540ea441d11SAl Viro return ERR_PTR(err); 2541ea441d11SAl Viro 2542ea441d11SAl Viro /* trade a vfsmount reference for active sb one */ 2543d31da0f0SAl Viro s = path.mnt->mnt_sb; 2544d31da0f0SAl Viro atomic_inc(&s->s_active); 2545ea441d11SAl Viro mntput(path.mnt); 2546ea441d11SAl Viro /* lock the sucker */ 2547d31da0f0SAl Viro down_write(&s->s_umount); 2548ea441d11SAl Viro /* ... and return the root of (sub)tree on it */ 2549ea441d11SAl Viro return path.dentry; 2550ea441d11SAl Viro } 2551ea441d11SAl Viro EXPORT_SYMBOL(mount_subtree); 2552ea441d11SAl Viro 2553bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2554bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 25551da177e4SLinus Torvalds { 2556eca6f534SVegard Nossum int ret; 2557eca6f534SVegard Nossum char *kernel_type; 255891a27b2aSJeff Layton struct filename *kernel_dir; 2559eca6f534SVegard Nossum char *kernel_dev; 25601da177e4SLinus Torvalds unsigned long data_page; 25611da177e4SLinus Torvalds 2562eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2563eca6f534SVegard Nossum if (ret < 0) 2564eca6f534SVegard Nossum goto out_type; 25651da177e4SLinus Torvalds 2566eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2567eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2568eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2569eca6f534SVegard Nossum goto out_dir; 2570eca6f534SVegard Nossum } 25711da177e4SLinus Torvalds 2572eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2573eca6f534SVegard Nossum if (ret < 0) 2574eca6f534SVegard Nossum goto out_dev; 25751da177e4SLinus Torvalds 2576eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2577eca6f534SVegard Nossum if (ret < 0) 2578eca6f534SVegard Nossum goto out_data; 25791da177e4SLinus Torvalds 258091a27b2aSJeff Layton ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags, 2581eca6f534SVegard Nossum (void *) data_page); 2582eca6f534SVegard Nossum 25831da177e4SLinus Torvalds free_page(data_page); 2584eca6f534SVegard Nossum out_data: 2585eca6f534SVegard Nossum kfree(kernel_dev); 2586eca6f534SVegard Nossum out_dev: 2587eca6f534SVegard Nossum putname(kernel_dir); 2588eca6f534SVegard Nossum out_dir: 2589eca6f534SVegard Nossum kfree(kernel_type); 2590eca6f534SVegard Nossum out_type: 2591eca6f534SVegard Nossum return ret; 25921da177e4SLinus Torvalds } 25931da177e4SLinus Torvalds 25941da177e4SLinus Torvalds /* 2595afac7cbaSAl Viro * Return true if path is reachable from root 2596afac7cbaSAl Viro * 2597afac7cbaSAl Viro * namespace_sem or vfsmount_lock is held 2598afac7cbaSAl Viro */ 2599643822b4SAl Viro bool is_path_reachable(struct mount *mnt, struct dentry *dentry, 2600afac7cbaSAl Viro const struct path *root) 2601afac7cbaSAl Viro { 2602643822b4SAl Viro while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) { 2603a73324daSAl Viro dentry = mnt->mnt_mountpoint; 26040714a533SAl Viro mnt = mnt->mnt_parent; 2605afac7cbaSAl Viro } 2606643822b4SAl Viro return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry); 2607afac7cbaSAl Viro } 2608afac7cbaSAl Viro 2609afac7cbaSAl Viro int path_is_under(struct path *path1, struct path *path2) 2610afac7cbaSAl Viro { 2611afac7cbaSAl Viro int res; 2612962830dfSAndi Kleen br_read_lock(&vfsmount_lock); 2613643822b4SAl Viro res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2); 2614962830dfSAndi Kleen br_read_unlock(&vfsmount_lock); 2615afac7cbaSAl Viro return res; 2616afac7cbaSAl Viro } 2617afac7cbaSAl Viro EXPORT_SYMBOL(path_is_under); 2618afac7cbaSAl Viro 2619afac7cbaSAl Viro /* 26201da177e4SLinus Torvalds * pivot_root Semantics: 26211da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 26221da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 26231da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 26241da177e4SLinus Torvalds * 26251da177e4SLinus Torvalds * Restrictions: 26261da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 26271da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 26281da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 26291da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 26301da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 26311da177e4SLinus Torvalds * 26324a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 26334a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 26344a0d11faSNeil Brown * in this situation. 26354a0d11faSNeil Brown * 26361da177e4SLinus Torvalds * Notes: 26371da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 26381da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 26391da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 26401da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 26411da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 26421da177e4SLinus Torvalds * first. 26431da177e4SLinus Torvalds */ 26443480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 26453480b257SHeiko Carstens const char __user *, put_old) 26461da177e4SLinus Torvalds { 26472d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 264884d17192SAl Viro struct mount *new_mnt, *root_mnt, *old_mnt; 264984d17192SAl Viro struct mountpoint *old_mp, *root_mp; 26501da177e4SLinus Torvalds int error; 26511da177e4SLinus Torvalds 26529b40bc90SAl Viro if (!may_mount()) 26531da177e4SLinus Torvalds return -EPERM; 26541da177e4SLinus Torvalds 26552d8f3038SAl Viro error = user_path_dir(new_root, &new); 26561da177e4SLinus Torvalds if (error) 26571da177e4SLinus Torvalds goto out0; 26581da177e4SLinus Torvalds 26592d8f3038SAl Viro error = user_path_dir(put_old, &old); 26601da177e4SLinus Torvalds if (error) 26611da177e4SLinus Torvalds goto out1; 26621da177e4SLinus Torvalds 26632d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 2664b12cea91SAl Viro if (error) 2665b12cea91SAl Viro goto out2; 26661da177e4SLinus Torvalds 2667f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 266884d17192SAl Viro old_mp = lock_mount(&old); 266984d17192SAl Viro error = PTR_ERR(old_mp); 267084d17192SAl Viro if (IS_ERR(old_mp)) 2671b12cea91SAl Viro goto out3; 2672b12cea91SAl Viro 26731da177e4SLinus Torvalds error = -EINVAL; 2674419148daSAl Viro new_mnt = real_mount(new.mnt); 2675419148daSAl Viro root_mnt = real_mount(root.mnt); 267684d17192SAl Viro old_mnt = real_mount(old.mnt); 267784d17192SAl Viro if (IS_MNT_SHARED(old_mnt) || 2678fc7be130SAl Viro IS_MNT_SHARED(new_mnt->mnt_parent) || 2679fc7be130SAl Viro IS_MNT_SHARED(root_mnt->mnt_parent)) 2680b12cea91SAl Viro goto out4; 2681143c8c91SAl Viro if (!check_mnt(root_mnt) || !check_mnt(new_mnt)) 2682b12cea91SAl Viro goto out4; 26835ff9d8a6SEric W. Biederman if (new_mnt->mnt.mnt_flags & MNT_LOCKED) 26845ff9d8a6SEric W. Biederman goto out4; 26851da177e4SLinus Torvalds error = -ENOENT; 2686f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 2687b12cea91SAl Viro goto out4; 26881da177e4SLinus Torvalds error = -EBUSY; 268984d17192SAl Viro if (new_mnt == root_mnt || old_mnt == root_mnt) 2690b12cea91SAl Viro goto out4; /* loop, on the same file system */ 26911da177e4SLinus Torvalds error = -EINVAL; 26928c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 2693b12cea91SAl Viro goto out4; /* not a mountpoint */ 2694676da58dSAl Viro if (!mnt_has_parent(root_mnt)) 2695b12cea91SAl Viro goto out4; /* not attached */ 269684d17192SAl Viro root_mp = root_mnt->mnt_mp; 26972d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 2698b12cea91SAl Viro goto out4; /* not a mountpoint */ 2699676da58dSAl Viro if (!mnt_has_parent(new_mnt)) 2700b12cea91SAl Viro goto out4; /* not attached */ 27014ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 270284d17192SAl Viro if (!is_path_reachable(old_mnt, old.dentry, &new)) 2703b12cea91SAl Viro goto out4; 270484d17192SAl Viro root_mp->m_count++; /* pin it so it won't go away */ 2705962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 2706419148daSAl Viro detach_mnt(new_mnt, &parent_path); 2707419148daSAl Viro detach_mnt(root_mnt, &root_parent); 27085ff9d8a6SEric W. Biederman if (root_mnt->mnt.mnt_flags & MNT_LOCKED) { 27095ff9d8a6SEric W. Biederman new_mnt->mnt.mnt_flags |= MNT_LOCKED; 27105ff9d8a6SEric W. Biederman root_mnt->mnt.mnt_flags &= ~MNT_LOCKED; 27115ff9d8a6SEric W. Biederman } 27124ac91378SJan Blunck /* mount old root on put_old */ 271384d17192SAl Viro attach_mnt(root_mnt, old_mnt, old_mp); 27144ac91378SJan Blunck /* mount new_root on / */ 271584d17192SAl Viro attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp); 27166b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 2717962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 27182d8f3038SAl Viro chroot_fs_refs(&root, &new); 271984d17192SAl Viro put_mountpoint(root_mp); 27201da177e4SLinus Torvalds error = 0; 2721b12cea91SAl Viro out4: 272284d17192SAl Viro unlock_mount(old_mp); 2723b12cea91SAl Viro if (!error) { 27241a390689SAl Viro path_put(&root_parent); 27251a390689SAl Viro path_put(&parent_path); 2726b12cea91SAl Viro } 2727b12cea91SAl Viro out3: 27288c3ee42eSAl Viro path_put(&root); 2729b12cea91SAl Viro out2: 27302d8f3038SAl Viro path_put(&old); 27311da177e4SLinus Torvalds out1: 27322d8f3038SAl Viro path_put(&new); 27331da177e4SLinus Torvalds out0: 27341da177e4SLinus Torvalds return error; 27351da177e4SLinus Torvalds } 27361da177e4SLinus Torvalds 27371da177e4SLinus Torvalds static void __init init_mount_tree(void) 27381da177e4SLinus Torvalds { 27391da177e4SLinus Torvalds struct vfsmount *mnt; 27406b3286edSKirill Korotaev struct mnt_namespace *ns; 2741ac748a09SJan Blunck struct path root; 27420c55cfc4SEric W. Biederman struct file_system_type *type; 27431da177e4SLinus Torvalds 27440c55cfc4SEric W. Biederman type = get_fs_type("rootfs"); 27450c55cfc4SEric W. Biederman if (!type) 27460c55cfc4SEric W. Biederman panic("Can't find rootfs type"); 27470c55cfc4SEric W. Biederman mnt = vfs_kern_mount(type, 0, "rootfs", NULL); 27480c55cfc4SEric W. Biederman put_filesystem(type); 27491da177e4SLinus Torvalds if (IS_ERR(mnt)) 27501da177e4SLinus Torvalds panic("Can't create rootfs"); 2751b3e19d92SNick Piggin 27523b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 27533b22edc5STrond Myklebust if (IS_ERR(ns)) 27541da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 27551da177e4SLinus Torvalds 27566b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 27576b3286edSKirill Korotaev get_mnt_ns(ns); 27581da177e4SLinus Torvalds 2759be08d6d2SAl Viro root.mnt = mnt; 2760be08d6d2SAl Viro root.dentry = mnt->mnt_root; 2761ac748a09SJan Blunck 2762ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2763ac748a09SJan Blunck set_fs_root(current->fs, &root); 27641da177e4SLinus Torvalds } 27651da177e4SLinus Torvalds 276674bf17cfSDenis Cheng void __init mnt_init(void) 27671da177e4SLinus Torvalds { 276813f14b4dSEric Dumazet unsigned u; 276915a67dd8SRandy Dunlap int err; 27701da177e4SLinus Torvalds 2771390c6843SRam Pai init_rwsem(&namespace_sem); 2772390c6843SRam Pai 27737d6fec45SAl Viro mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount), 277420c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 27751da177e4SLinus Torvalds 2776b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 277784d17192SAl Viro mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 27781da177e4SLinus Torvalds 277984d17192SAl Viro if (!mount_hashtable || !mountpoint_hashtable) 27801da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 27811da177e4SLinus Torvalds 278280cdc6daSMandeep Singh Baines printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE); 27831da177e4SLinus Torvalds 278413f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 278513f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 278684d17192SAl Viro for (u = 0; u < HASH_SIZE; u++) 278784d17192SAl Viro INIT_LIST_HEAD(&mountpoint_hashtable[u]); 27881da177e4SLinus Torvalds 2789962830dfSAndi Kleen br_lock_init(&vfsmount_lock); 279099b7db7bSNick Piggin 279115a67dd8SRandy Dunlap err = sysfs_init(); 279215a67dd8SRandy Dunlap if (err) 279315a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 27948e24eea7SHarvey Harrison __func__, err); 279500d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 279600d26666SGreg Kroah-Hartman if (!fs_kobj) 27978e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 27981da177e4SLinus Torvalds init_rootfs(); 27991da177e4SLinus Torvalds init_mount_tree(); 28001da177e4SLinus Torvalds } 28011da177e4SLinus Torvalds 2802616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 28031da177e4SLinus Torvalds { 2804d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 2805616511d0STrond Myklebust return; 280697216be0SAl Viro namespace_lock(); 2807962830dfSAndi Kleen br_write_lock(&vfsmount_lock); 2808328e6d90SAl Viro umount_tree(ns->root, 0); 2809962830dfSAndi Kleen br_write_unlock(&vfsmount_lock); 28103ab6abeeSAl Viro namespace_unlock(); 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 br_write_lock(&vfsmount_lock); 2834f7a99c5bSAl Viro real_mount(mnt)->mnt_ns = NULL; 2835f7a99c5bSAl Viro br_write_unlock(&vfsmount_lock); 2836423e0ab0STim Chen mntput(mnt); 2837423e0ab0STim Chen } 2838423e0ab0STim Chen } 2839423e0ab0STim Chen EXPORT_SYMBOL(kern_unmount); 284002125a82SAl Viro 284102125a82SAl Viro bool our_mnt(struct vfsmount *mnt) 284202125a82SAl Viro { 2843143c8c91SAl Viro return check_mnt(real_mount(mnt)); 284402125a82SAl Viro } 28458823c079SEric W. Biederman 28463151527eSEric W. Biederman bool current_chrooted(void) 28473151527eSEric W. Biederman { 28483151527eSEric W. Biederman /* Does the current process have a non-standard root */ 28493151527eSEric W. Biederman struct path ns_root; 28503151527eSEric W. Biederman struct path fs_root; 28513151527eSEric W. Biederman bool chrooted; 28523151527eSEric W. Biederman 28533151527eSEric W. Biederman /* Find the namespace root */ 28543151527eSEric W. Biederman ns_root.mnt = ¤t->nsproxy->mnt_ns->root->mnt; 28553151527eSEric W. Biederman ns_root.dentry = ns_root.mnt->mnt_root; 28563151527eSEric W. Biederman path_get(&ns_root); 28573151527eSEric W. Biederman while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root)) 28583151527eSEric W. Biederman ; 28593151527eSEric W. Biederman 28603151527eSEric W. Biederman get_fs_root(current->fs, &fs_root); 28613151527eSEric W. Biederman 28623151527eSEric W. Biederman chrooted = !path_equal(&fs_root, &ns_root); 28633151527eSEric W. Biederman 28643151527eSEric W. Biederman path_put(&fs_root); 28653151527eSEric W. Biederman path_put(&ns_root); 28663151527eSEric W. Biederman 28673151527eSEric W. Biederman return chrooted; 28683151527eSEric W. Biederman } 28693151527eSEric W. Biederman 287087a8ebd6SEric W. Biederman void update_mnt_policy(struct user_namespace *userns) 287187a8ebd6SEric W. Biederman { 287287a8ebd6SEric W. Biederman struct mnt_namespace *ns = current->nsproxy->mnt_ns; 287387a8ebd6SEric W. Biederman struct mount *mnt; 287487a8ebd6SEric W. Biederman 287587a8ebd6SEric W. Biederman down_read(&namespace_sem); 287687a8ebd6SEric W. Biederman list_for_each_entry(mnt, &ns->list, mnt_list) { 287787a8ebd6SEric W. Biederman switch (mnt->mnt.mnt_sb->s_magic) { 287887a8ebd6SEric W. Biederman case SYSFS_MAGIC: 287987a8ebd6SEric W. Biederman userns->may_mount_sysfs = true; 288087a8ebd6SEric W. Biederman break; 288187a8ebd6SEric W. Biederman case PROC_SUPER_MAGIC: 288287a8ebd6SEric W. Biederman userns->may_mount_proc = true; 288387a8ebd6SEric W. Biederman break; 288487a8ebd6SEric W. Biederman } 288587a8ebd6SEric W. Biederman if (userns->may_mount_sysfs && userns->may_mount_proc) 288687a8ebd6SEric W. Biederman break; 288787a8ebd6SEric W. Biederman } 288887a8ebd6SEric W. Biederman up_read(&namespace_sem); 288987a8ebd6SEric W. Biederman } 289087a8ebd6SEric W. Biederman 28918823c079SEric W. Biederman static void *mntns_get(struct task_struct *task) 28928823c079SEric W. Biederman { 28938823c079SEric W. Biederman struct mnt_namespace *ns = NULL; 28948823c079SEric W. Biederman struct nsproxy *nsproxy; 28958823c079SEric W. Biederman 28968823c079SEric W. Biederman rcu_read_lock(); 28978823c079SEric W. Biederman nsproxy = task_nsproxy(task); 28988823c079SEric W. Biederman if (nsproxy) { 28998823c079SEric W. Biederman ns = nsproxy->mnt_ns; 29008823c079SEric W. Biederman get_mnt_ns(ns); 29018823c079SEric W. Biederman } 29028823c079SEric W. Biederman rcu_read_unlock(); 29038823c079SEric W. Biederman 29048823c079SEric W. Biederman return ns; 29058823c079SEric W. Biederman } 29068823c079SEric W. Biederman 29078823c079SEric W. Biederman static void mntns_put(void *ns) 29088823c079SEric W. Biederman { 29098823c079SEric W. Biederman put_mnt_ns(ns); 29108823c079SEric W. Biederman } 29118823c079SEric W. Biederman 29128823c079SEric W. Biederman static int mntns_install(struct nsproxy *nsproxy, void *ns) 29138823c079SEric W. Biederman { 29148823c079SEric W. Biederman struct fs_struct *fs = current->fs; 29158823c079SEric W. Biederman struct mnt_namespace *mnt_ns = ns; 29168823c079SEric W. Biederman struct path root; 29178823c079SEric W. Biederman 29180c55cfc4SEric W. Biederman if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) || 29195e4a0847SEric W. Biederman !nsown_capable(CAP_SYS_CHROOT) || 29205e4a0847SEric W. Biederman !nsown_capable(CAP_SYS_ADMIN)) 2921ae11e0f1SZhao Hongjiang return -EPERM; 29228823c079SEric W. Biederman 29238823c079SEric W. Biederman if (fs->users != 1) 29248823c079SEric W. Biederman return -EINVAL; 29258823c079SEric W. Biederman 29268823c079SEric W. Biederman get_mnt_ns(mnt_ns); 29278823c079SEric W. Biederman put_mnt_ns(nsproxy->mnt_ns); 29288823c079SEric W. Biederman nsproxy->mnt_ns = mnt_ns; 29298823c079SEric W. Biederman 29308823c079SEric W. Biederman /* Find the root */ 29318823c079SEric W. Biederman root.mnt = &mnt_ns->root->mnt; 29328823c079SEric W. Biederman root.dentry = mnt_ns->root->mnt.mnt_root; 29338823c079SEric W. Biederman path_get(&root); 29348823c079SEric W. Biederman while(d_mountpoint(root.dentry) && follow_down_one(&root)) 29358823c079SEric W. Biederman ; 29368823c079SEric W. Biederman 29378823c079SEric W. Biederman /* Update the pwd and root */ 29388823c079SEric W. Biederman set_fs_pwd(fs, &root); 29398823c079SEric W. Biederman set_fs_root(fs, &root); 29408823c079SEric W. Biederman 29418823c079SEric W. Biederman path_put(&root); 29428823c079SEric W. Biederman return 0; 29438823c079SEric W. Biederman } 29448823c079SEric W. Biederman 294598f842e6SEric W. Biederman static unsigned int mntns_inum(void *ns) 294698f842e6SEric W. Biederman { 294798f842e6SEric W. Biederman struct mnt_namespace *mnt_ns = ns; 294898f842e6SEric W. Biederman return mnt_ns->proc_inum; 294998f842e6SEric W. Biederman } 295098f842e6SEric W. Biederman 29518823c079SEric W. Biederman const struct proc_ns_operations mntns_operations = { 29528823c079SEric W. Biederman .name = "mnt", 29538823c079SEric W. Biederman .type = CLONE_NEWNS, 29548823c079SEric W. Biederman .get = mntns_get, 29558823c079SEric W. Biederman .put = mntns_put, 29568823c079SEric W. Biederman .install = mntns_install, 295798f842e6SEric W. Biederman .inum = mntns_inum, 29588823c079SEric W. Biederman }; 2959