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> 121da177e4SLinus Torvalds #include <linux/slab.h> 131da177e4SLinus Torvalds #include <linux/sched.h> 1499b7db7bSNick Piggin #include <linux/spinlock.h> 1599b7db7bSNick Piggin #include <linux/percpu.h> 161da177e4SLinus Torvalds #include <linux/smp_lock.h> 171da177e4SLinus Torvalds #include <linux/init.h> 1815a67dd8SRandy Dunlap #include <linux/kernel.h> 191da177e4SLinus Torvalds #include <linux/acct.h> 2016f7e0feSRandy Dunlap #include <linux/capability.h> 213d733633SDave Hansen #include <linux/cpumask.h> 221da177e4SLinus Torvalds #include <linux/module.h> 23f20a9eadSAndrew Morton #include <linux/sysfs.h> 241da177e4SLinus Torvalds #include <linux/seq_file.h> 256b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 261da177e4SLinus Torvalds #include <linux/namei.h> 27b43f3cbdSAlexey Dobriyan #include <linux/nsproxy.h> 281da177e4SLinus Torvalds #include <linux/security.h> 291da177e4SLinus Torvalds #include <linux/mount.h> 3007f3f05cSDavid Howells #include <linux/ramfs.h> 3113f14b4dSEric Dumazet #include <linux/log2.h> 3273cd49ecSMiklos Szeredi #include <linux/idr.h> 335ad4e53bSAl Viro #include <linux/fs_struct.h> 342504c5d6SAndreas Gruenbacher #include <linux/fsnotify.h> 351da177e4SLinus Torvalds #include <asm/uaccess.h> 361da177e4SLinus Torvalds #include <asm/unistd.h> 3707b20889SRam Pai #include "pnode.h" 38948730b0SAdrian Bunk #include "internal.h" 391da177e4SLinus Torvalds 4013f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 4113f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 4213f14b4dSEric Dumazet 435addc5ddSAl Viro static int event; 4473cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 45719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 4699b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 47f21f6220SAl Viro static int mnt_id_start = 0; 48f21f6220SAl Viro static int mnt_group_start = 1; 495addc5ddSAl Viro 50fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 51e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 52390c6843SRam Pai static struct rw_semaphore namespace_sem; 531da177e4SLinus Torvalds 54f87fd4c2SMiklos Szeredi /* /sys/fs */ 5500d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 5600d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 57f87fd4c2SMiklos Szeredi 5899b7db7bSNick Piggin /* 5999b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 6099b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 6199b7db7bSNick Piggin * up the tree. 6299b7db7bSNick Piggin * 6399b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 6499b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 6599b7db7bSNick Piggin */ 6699b7db7bSNick Piggin DEFINE_BRLOCK(vfsmount_lock); 6799b7db7bSNick Piggin 681da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 691da177e4SLinus Torvalds { 701da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 711da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 7213f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 7313f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 741da177e4SLinus Torvalds } 751da177e4SLinus Torvalds 763d733633SDave Hansen #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 773d733633SDave Hansen 7899b7db7bSNick Piggin /* 7999b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 8099b7db7bSNick Piggin * serialize with freeing. 8199b7db7bSNick Piggin */ 8273cd49ecSMiklos Szeredi static int mnt_alloc_id(struct vfsmount *mnt) 8373cd49ecSMiklos Szeredi { 8473cd49ecSMiklos Szeredi int res; 8573cd49ecSMiklos Szeredi 8673cd49ecSMiklos Szeredi retry: 8773cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 8899b7db7bSNick Piggin spin_lock(&mnt_id_lock); 89f21f6220SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 90f21f6220SAl Viro if (!res) 91f21f6220SAl Viro mnt_id_start = mnt->mnt_id + 1; 9299b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 9373cd49ecSMiklos Szeredi if (res == -EAGAIN) 9473cd49ecSMiklos Szeredi goto retry; 9573cd49ecSMiklos Szeredi 9673cd49ecSMiklos Szeredi return res; 9773cd49ecSMiklos Szeredi } 9873cd49ecSMiklos Szeredi 9973cd49ecSMiklos Szeredi static void mnt_free_id(struct vfsmount *mnt) 10073cd49ecSMiklos Szeredi { 101f21f6220SAl Viro int id = mnt->mnt_id; 10299b7db7bSNick Piggin spin_lock(&mnt_id_lock); 103f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 104f21f6220SAl Viro if (mnt_id_start > id) 105f21f6220SAl Viro mnt_id_start = id; 10699b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 10773cd49ecSMiklos Szeredi } 10873cd49ecSMiklos Szeredi 109719f5d7fSMiklos Szeredi /* 110719f5d7fSMiklos Szeredi * Allocate a new peer group ID 111719f5d7fSMiklos Szeredi * 112719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 113719f5d7fSMiklos Szeredi */ 114719f5d7fSMiklos Szeredi static int mnt_alloc_group_id(struct vfsmount *mnt) 115719f5d7fSMiklos Szeredi { 116f21f6220SAl Viro int res; 117f21f6220SAl Viro 118719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 119719f5d7fSMiklos Szeredi return -ENOMEM; 120719f5d7fSMiklos Szeredi 121f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 122f21f6220SAl Viro mnt_group_start, 123f21f6220SAl Viro &mnt->mnt_group_id); 124f21f6220SAl Viro if (!res) 125f21f6220SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 126f21f6220SAl Viro 127f21f6220SAl Viro return res; 128719f5d7fSMiklos Szeredi } 129719f5d7fSMiklos Szeredi 130719f5d7fSMiklos Szeredi /* 131719f5d7fSMiklos Szeredi * Release a peer group ID 132719f5d7fSMiklos Szeredi */ 133719f5d7fSMiklos Szeredi void mnt_release_group_id(struct vfsmount *mnt) 134719f5d7fSMiklos Szeredi { 135f21f6220SAl Viro int id = mnt->mnt_group_id; 136f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 137f21f6220SAl Viro if (mnt_group_start > id) 138f21f6220SAl Viro mnt_group_start = id; 139719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; 140719f5d7fSMiklos Szeredi } 141719f5d7fSMiklos Szeredi 1421da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 1431da177e4SLinus Torvalds { 144c3762229SRobert P. J. Day struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 1451da177e4SLinus Torvalds if (mnt) { 14673cd49ecSMiklos Szeredi int err; 14773cd49ecSMiklos Szeredi 14873cd49ecSMiklos Szeredi err = mnt_alloc_id(mnt); 14988b38782SLi Zefan if (err) 15088b38782SLi Zefan goto out_free_cache; 15188b38782SLi Zefan 15288b38782SLi Zefan if (name) { 15388b38782SLi Zefan mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 15488b38782SLi Zefan if (!mnt->mnt_devname) 15588b38782SLi Zefan goto out_free_id; 15673cd49ecSMiklos Szeredi } 15773cd49ecSMiklos Szeredi 1581da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 1591da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 1601da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 1611da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 1621da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 16355e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 16403e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 165a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 166a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 1672504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 1682504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 1692504c5d6SAndreas Gruenbacher #endif 170d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 171d3ef3d73Snpiggin@suse.de mnt->mnt_writers = alloc_percpu(int); 172d3ef3d73Snpiggin@suse.de if (!mnt->mnt_writers) 173d3ef3d73Snpiggin@suse.de goto out_free_devname; 174d3ef3d73Snpiggin@suse.de #else 175d3ef3d73Snpiggin@suse.de mnt->mnt_writers = 0; 176d3ef3d73Snpiggin@suse.de #endif 1771da177e4SLinus Torvalds } 1781da177e4SLinus Torvalds return mnt; 17988b38782SLi Zefan 180d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 181d3ef3d73Snpiggin@suse.de out_free_devname: 182d3ef3d73Snpiggin@suse.de kfree(mnt->mnt_devname); 183d3ef3d73Snpiggin@suse.de #endif 18488b38782SLi Zefan out_free_id: 18588b38782SLi Zefan mnt_free_id(mnt); 18688b38782SLi Zefan out_free_cache: 18788b38782SLi Zefan kmem_cache_free(mnt_cache, mnt); 18888b38782SLi Zefan return NULL; 1891da177e4SLinus Torvalds } 1901da177e4SLinus Torvalds 1918366025eSDave Hansen /* 1928366025eSDave Hansen * Most r/o checks on a fs are for operations that take 1938366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 1948366025eSDave Hansen * We must keep track of when those operations start 1958366025eSDave Hansen * (for permission checks) and when they end, so that 1968366025eSDave Hansen * we can determine when writes are able to occur to 1978366025eSDave Hansen * a filesystem. 1988366025eSDave Hansen */ 1993d733633SDave Hansen /* 2003d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2013d733633SDave Hansen * @mnt: the mount to check for its write status 2023d733633SDave Hansen * 2033d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2043d733633SDave Hansen * It does not guarantee that the filesystem will stay 2053d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2063d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2073d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2083d733633SDave Hansen * r/w. 2093d733633SDave Hansen */ 2103d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2113d733633SDave Hansen { 2122e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2132e4b7fcdSDave Hansen return 1; 2142e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2152e4b7fcdSDave Hansen return 1; 2162e4b7fcdSDave Hansen return 0; 2173d733633SDave Hansen } 2183d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2193d733633SDave Hansen 220d3ef3d73Snpiggin@suse.de static inline void inc_mnt_writers(struct vfsmount *mnt) 2213d733633SDave Hansen { 222d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 223d3ef3d73Snpiggin@suse.de (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++; 224d3ef3d73Snpiggin@suse.de #else 225d3ef3d73Snpiggin@suse.de mnt->mnt_writers++; 226d3ef3d73Snpiggin@suse.de #endif 2273d733633SDave Hansen } 2283d733633SDave Hansen 229d3ef3d73Snpiggin@suse.de static inline void dec_mnt_writers(struct vfsmount *mnt) 2303d733633SDave Hansen { 231d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 232d3ef3d73Snpiggin@suse.de (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--; 233d3ef3d73Snpiggin@suse.de #else 234d3ef3d73Snpiggin@suse.de mnt->mnt_writers--; 235d3ef3d73Snpiggin@suse.de #endif 236d3ef3d73Snpiggin@suse.de } 237d3ef3d73Snpiggin@suse.de 238d3ef3d73Snpiggin@suse.de static unsigned int count_mnt_writers(struct vfsmount *mnt) 239d3ef3d73Snpiggin@suse.de { 240d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 241d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2423d733633SDave Hansen int cpu; 2433d733633SDave Hansen 2443d733633SDave Hansen for_each_possible_cpu(cpu) { 245d3ef3d73Snpiggin@suse.de count += *per_cpu_ptr(mnt->mnt_writers, cpu); 2463d733633SDave Hansen } 2473d733633SDave Hansen 248d3ef3d73Snpiggin@suse.de return count; 249d3ef3d73Snpiggin@suse.de #else 250d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 251d3ef3d73Snpiggin@suse.de #endif 2523d733633SDave Hansen } 2533d733633SDave Hansen 2543d733633SDave Hansen /* 2553d733633SDave Hansen * Most r/o checks on a fs are for operations that take 2563d733633SDave Hansen * discrete amounts of time, like a write() or unlink(). 2573d733633SDave Hansen * We must keep track of when those operations start 2583d733633SDave Hansen * (for permission checks) and when they end, so that 2593d733633SDave Hansen * we can determine when writes are able to occur to 2603d733633SDave Hansen * a filesystem. 2613d733633SDave Hansen */ 2628366025eSDave Hansen /** 2638366025eSDave Hansen * mnt_want_write - get write access to a mount 2648366025eSDave Hansen * @mnt: the mount on which to take a write 2658366025eSDave Hansen * 2668366025eSDave Hansen * This tells the low-level filesystem that a write is 2678366025eSDave Hansen * about to be performed to it, and makes sure that 2688366025eSDave Hansen * writes are allowed before returning success. When 2698366025eSDave Hansen * the write operation is finished, mnt_drop_write() 2708366025eSDave Hansen * must be called. This is effectively a refcount. 2718366025eSDave Hansen */ 2728366025eSDave Hansen int mnt_want_write(struct vfsmount *mnt) 2738366025eSDave Hansen { 2743d733633SDave Hansen int ret = 0; 2753d733633SDave Hansen 276d3ef3d73Snpiggin@suse.de preempt_disable(); 277d3ef3d73Snpiggin@suse.de inc_mnt_writers(mnt); 278d3ef3d73Snpiggin@suse.de /* 279d3ef3d73Snpiggin@suse.de * The store to inc_mnt_writers must be visible before we pass 280d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 281d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 282d3ef3d73Snpiggin@suse.de */ 283d3ef3d73Snpiggin@suse.de smp_mb(); 284d3ef3d73Snpiggin@suse.de while (mnt->mnt_flags & MNT_WRITE_HOLD) 285d3ef3d73Snpiggin@suse.de cpu_relax(); 286d3ef3d73Snpiggin@suse.de /* 287d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 288d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 289d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 290d3ef3d73Snpiggin@suse.de */ 291d3ef3d73Snpiggin@suse.de smp_rmb(); 2923d733633SDave Hansen if (__mnt_is_readonly(mnt)) { 293d3ef3d73Snpiggin@suse.de dec_mnt_writers(mnt); 2943d733633SDave Hansen ret = -EROFS; 2953d733633SDave Hansen goto out; 2963d733633SDave Hansen } 2973d733633SDave Hansen out: 298d3ef3d73Snpiggin@suse.de preempt_enable(); 2993d733633SDave Hansen return ret; 3008366025eSDave Hansen } 3018366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3028366025eSDave Hansen 3038366025eSDave Hansen /** 30496029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 30596029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 30696029c4eSnpiggin@suse.de * 30796029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 30896029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 30996029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 31096029c4eSnpiggin@suse.de * on it. This allows some optimisation. 31196029c4eSnpiggin@suse.de * 31296029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 31396029c4eSnpiggin@suse.de * drop the reference. 31496029c4eSnpiggin@suse.de */ 31596029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 31696029c4eSnpiggin@suse.de { 31796029c4eSnpiggin@suse.de /* superblock may be r/o */ 31896029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 31996029c4eSnpiggin@suse.de return -EROFS; 32096029c4eSnpiggin@suse.de preempt_disable(); 32196029c4eSnpiggin@suse.de inc_mnt_writers(mnt); 32296029c4eSnpiggin@suse.de preempt_enable(); 32396029c4eSnpiggin@suse.de return 0; 32496029c4eSnpiggin@suse.de } 32596029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 32696029c4eSnpiggin@suse.de 32796029c4eSnpiggin@suse.de /** 32896029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 32996029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 33096029c4eSnpiggin@suse.de * 33196029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 33296029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 33396029c4eSnpiggin@suse.de */ 33496029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 33596029c4eSnpiggin@suse.de { 3362d8dd38aSOGAWA Hirofumi struct inode *inode = file->f_dentry->d_inode; 3372d8dd38aSOGAWA Hirofumi if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) 33896029c4eSnpiggin@suse.de return mnt_want_write(file->f_path.mnt); 33996029c4eSnpiggin@suse.de else 34096029c4eSnpiggin@suse.de return mnt_clone_write(file->f_path.mnt); 34196029c4eSnpiggin@suse.de } 34296029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 34396029c4eSnpiggin@suse.de 34496029c4eSnpiggin@suse.de /** 3458366025eSDave Hansen * mnt_drop_write - give up write access to a mount 3468366025eSDave Hansen * @mnt: the mount on which to give up write access 3478366025eSDave Hansen * 3488366025eSDave Hansen * Tells the low-level filesystem that we are done 3498366025eSDave Hansen * performing writes to it. Must be matched with 3508366025eSDave Hansen * mnt_want_write() call above. 3518366025eSDave Hansen */ 3528366025eSDave Hansen void mnt_drop_write(struct vfsmount *mnt) 3538366025eSDave Hansen { 354d3ef3d73Snpiggin@suse.de preempt_disable(); 355d3ef3d73Snpiggin@suse.de dec_mnt_writers(mnt); 356d3ef3d73Snpiggin@suse.de preempt_enable(); 3578366025eSDave Hansen } 3588366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 3598366025eSDave Hansen 3602e4b7fcdSDave Hansen static int mnt_make_readonly(struct vfsmount *mnt) 3618366025eSDave Hansen { 3623d733633SDave Hansen int ret = 0; 3633d733633SDave Hansen 36499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 365d3ef3d73Snpiggin@suse.de mnt->mnt_flags |= MNT_WRITE_HOLD; 366d3ef3d73Snpiggin@suse.de /* 367d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 368d3ef3d73Snpiggin@suse.de * should be visible before we do. 369d3ef3d73Snpiggin@suse.de */ 370d3ef3d73Snpiggin@suse.de smp_mb(); 371d3ef3d73Snpiggin@suse.de 372d3ef3d73Snpiggin@suse.de /* 373d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 374d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 375d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 376d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 377d3ef3d73Snpiggin@suse.de * 378d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 379d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 380d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 381d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 382d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 383d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 384d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 385d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 386d3ef3d73Snpiggin@suse.de * we're counting up here. 387d3ef3d73Snpiggin@suse.de */ 388d3ef3d73Snpiggin@suse.de if (count_mnt_writers(mnt) > 0) 389d3ef3d73Snpiggin@suse.de ret = -EBUSY; 390d3ef3d73Snpiggin@suse.de else 3912e4b7fcdSDave Hansen mnt->mnt_flags |= MNT_READONLY; 392d3ef3d73Snpiggin@suse.de /* 393d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 394d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 395d3ef3d73Snpiggin@suse.de */ 396d3ef3d73Snpiggin@suse.de smp_wmb(); 397d3ef3d73Snpiggin@suse.de mnt->mnt_flags &= ~MNT_WRITE_HOLD; 39899b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 3993d733633SDave Hansen return ret; 4003d733633SDave Hansen } 4018366025eSDave Hansen 4022e4b7fcdSDave Hansen static void __mnt_unmake_readonly(struct vfsmount *mnt) 4032e4b7fcdSDave Hansen { 40499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 4052e4b7fcdSDave Hansen mnt->mnt_flags &= ~MNT_READONLY; 40699b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 4072e4b7fcdSDave Hansen } 4082e4b7fcdSDave Hansen 409a3ec947cSSukadev Bhattiprolu void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 410454e2398SDavid Howells { 411454e2398SDavid Howells mnt->mnt_sb = sb; 412454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 413454e2398SDavid Howells } 414454e2398SDavid Howells 415454e2398SDavid Howells EXPORT_SYMBOL(simple_set_mnt); 416454e2398SDavid Howells 4171da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 4181da177e4SLinus Torvalds { 4191da177e4SLinus Torvalds kfree(mnt->mnt_devname); 42073cd49ecSMiklos Szeredi mnt_free_id(mnt); 421d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 422d3ef3d73Snpiggin@suse.de free_percpu(mnt->mnt_writers); 423d3ef3d73Snpiggin@suse.de #endif 4241da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 4251da177e4SLinus Torvalds } 4261da177e4SLinus Torvalds 4271da177e4SLinus Torvalds /* 428a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 429a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 43099b7db7bSNick Piggin * vfsmount_lock must be held for read or write. 4311da177e4SLinus Torvalds */ 432a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 433a05964f3SRam Pai int dir) 4341da177e4SLinus Torvalds { 4351da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 4361da177e4SLinus Torvalds struct list_head *tmp = head; 4371da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 4381da177e4SLinus Torvalds 4391da177e4SLinus Torvalds for (;;) { 440a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 4411da177e4SLinus Torvalds p = NULL; 4421da177e4SLinus Torvalds if (tmp == head) 4431da177e4SLinus Torvalds break; 4441da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 4451da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 446a05964f3SRam Pai found = p; 4471da177e4SLinus Torvalds break; 4481da177e4SLinus Torvalds } 4491da177e4SLinus Torvalds } 4501da177e4SLinus Torvalds return found; 4511da177e4SLinus Torvalds } 4521da177e4SLinus Torvalds 453a05964f3SRam Pai /* 454a05964f3SRam Pai * lookup_mnt increments the ref count before returning 455a05964f3SRam Pai * the vfsmount struct. 456a05964f3SRam Pai */ 4571c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 458a05964f3SRam Pai { 459a05964f3SRam Pai struct vfsmount *child_mnt; 46099b7db7bSNick Piggin 46199b7db7bSNick Piggin br_read_lock(vfsmount_lock); 4621c755af4SAl Viro if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1))) 463a05964f3SRam Pai mntget(child_mnt); 46499b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 465a05964f3SRam Pai return child_mnt; 466a05964f3SRam Pai } 467a05964f3SRam Pai 4681da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 4691da177e4SLinus Torvalds { 4706b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 4711da177e4SLinus Torvalds } 4721da177e4SLinus Torvalds 47399b7db7bSNick Piggin /* 47499b7db7bSNick Piggin * vfsmount lock must be held for write 47599b7db7bSNick Piggin */ 4766b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 4775addc5ddSAl Viro { 4785addc5ddSAl Viro if (ns) { 4795addc5ddSAl Viro ns->event = ++event; 4805addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4815addc5ddSAl Viro } 4825addc5ddSAl Viro } 4835addc5ddSAl Viro 48499b7db7bSNick Piggin /* 48599b7db7bSNick Piggin * vfsmount lock must be held for write 48699b7db7bSNick Piggin */ 4876b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 4885addc5ddSAl Viro { 4895addc5ddSAl Viro if (ns && ns->event != event) { 4905addc5ddSAl Viro ns->event = event; 4915addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4925addc5ddSAl Viro } 4935addc5ddSAl Viro } 4945addc5ddSAl Viro 49599b7db7bSNick Piggin /* 49699b7db7bSNick Piggin * vfsmount lock must be held for write 49799b7db7bSNick Piggin */ 4981a390689SAl Viro static void detach_mnt(struct vfsmount *mnt, struct path *old_path) 4991da177e4SLinus Torvalds { 5001a390689SAl Viro old_path->dentry = mnt->mnt_mountpoint; 5011a390689SAl Viro old_path->mnt = mnt->mnt_parent; 5021da177e4SLinus Torvalds mnt->mnt_parent = mnt; 5031da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 5041da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 5051da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 5061a390689SAl Viro old_path->dentry->d_mounted--; 5071da177e4SLinus Torvalds } 5081da177e4SLinus Torvalds 50999b7db7bSNick Piggin /* 51099b7db7bSNick Piggin * vfsmount lock must be held for write 51199b7db7bSNick Piggin */ 512b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 513b90fa9aeSRam Pai struct vfsmount *child_mnt) 514b90fa9aeSRam Pai { 515b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 516b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 517b90fa9aeSRam Pai dentry->d_mounted++; 518b90fa9aeSRam Pai } 519b90fa9aeSRam Pai 52099b7db7bSNick Piggin /* 52199b7db7bSNick Piggin * vfsmount lock must be held for write 52299b7db7bSNick Piggin */ 5231a390689SAl Viro static void attach_mnt(struct vfsmount *mnt, struct path *path) 5241da177e4SLinus Torvalds { 5251a390689SAl Viro mnt_set_mountpoint(path->mnt, path->dentry, mnt); 526b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 5271a390689SAl Viro hash(path->mnt, path->dentry)); 5281a390689SAl Viro list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); 529b90fa9aeSRam Pai } 530b90fa9aeSRam Pai 531b90fa9aeSRam Pai /* 53299b7db7bSNick Piggin * vfsmount lock must be held for write 533b90fa9aeSRam Pai */ 534b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 535b90fa9aeSRam Pai { 536b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 537b90fa9aeSRam Pai struct vfsmount *m; 538b90fa9aeSRam Pai LIST_HEAD(head); 5396b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 540b90fa9aeSRam Pai 541b90fa9aeSRam Pai BUG_ON(parent == mnt); 542b90fa9aeSRam Pai 543b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 544b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 5456b3286edSKirill Korotaev m->mnt_ns = n; 546b90fa9aeSRam Pai list_splice(&head, n->list.prev); 547b90fa9aeSRam Pai 548b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 549b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 550b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 5516b3286edSKirill Korotaev touch_mnt_namespace(n); 5521da177e4SLinus Torvalds } 5531da177e4SLinus Torvalds 5541da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 5551da177e4SLinus Torvalds { 5561da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 5571da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 5581da177e4SLinus Torvalds while (1) { 5591da177e4SLinus Torvalds if (p == root) 5601da177e4SLinus Torvalds return NULL; 5611da177e4SLinus Torvalds next = p->mnt_child.next; 5621da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 5631da177e4SLinus Torvalds break; 5641da177e4SLinus Torvalds p = p->mnt_parent; 5651da177e4SLinus Torvalds } 5661da177e4SLinus Torvalds } 5671da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 5681da177e4SLinus Torvalds } 5691da177e4SLinus Torvalds 5709676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 5719676f0c6SRam Pai { 5729676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 5739676f0c6SRam Pai while (prev != &p->mnt_mounts) { 5749676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 5759676f0c6SRam Pai prev = p->mnt_mounts.prev; 5769676f0c6SRam Pai } 5779676f0c6SRam Pai return p; 5789676f0c6SRam Pai } 5799676f0c6SRam Pai 58036341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 58136341f64SRam Pai int flag) 5821da177e4SLinus Torvalds { 5831da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 5841da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 5851da177e4SLinus Torvalds 5861da177e4SLinus Torvalds if (mnt) { 587719f5d7fSMiklos Szeredi if (flag & (CL_SLAVE | CL_PRIVATE)) 588719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; /* not a peer of original */ 589719f5d7fSMiklos Szeredi else 590719f5d7fSMiklos Szeredi mnt->mnt_group_id = old->mnt_group_id; 591719f5d7fSMiklos Szeredi 592719f5d7fSMiklos Szeredi if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 593719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(mnt); 594719f5d7fSMiklos Szeredi if (err) 595719f5d7fSMiklos Szeredi goto out_free; 596719f5d7fSMiklos Szeredi } 597719f5d7fSMiklos Szeredi 5981da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 5991da177e4SLinus Torvalds atomic_inc(&sb->s_active); 6001da177e4SLinus Torvalds mnt->mnt_sb = sb; 6011da177e4SLinus Torvalds mnt->mnt_root = dget(root); 6021da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 6031da177e4SLinus Torvalds mnt->mnt_parent = mnt; 604b90fa9aeSRam Pai 6055afe0022SRam Pai if (flag & CL_SLAVE) { 6065afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 6075afe0022SRam Pai mnt->mnt_master = old; 6085afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 6098aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 610796a6b52SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 611b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 6125afe0022SRam Pai if (IS_MNT_SLAVE(old)) 6135afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 6145afe0022SRam Pai mnt->mnt_master = old->mnt_master; 6155afe0022SRam Pai } 616b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 617b90fa9aeSRam Pai set_mnt_shared(mnt); 6181da177e4SLinus Torvalds 6191da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 6201da177e4SLinus Torvalds * as the original if that was on one */ 62136341f64SRam Pai if (flag & CL_EXPIRE) { 62255e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 62355e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 6241da177e4SLinus Torvalds } 62536341f64SRam Pai } 6261da177e4SLinus Torvalds return mnt; 627719f5d7fSMiklos Szeredi 628719f5d7fSMiklos Szeredi out_free: 629719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 630719f5d7fSMiklos Szeredi return NULL; 6311da177e4SLinus Torvalds } 6321da177e4SLinus Torvalds 6337b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 6341da177e4SLinus Torvalds { 6351da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 6363d733633SDave Hansen /* 6373d733633SDave Hansen * This probably indicates that somebody messed 6383d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 6393d733633SDave Hansen * happens, the filesystem was probably unable 6403d733633SDave Hansen * to make r/w->r/o transitions. 6413d733633SDave Hansen */ 642d3ef3d73Snpiggin@suse.de /* 643d3ef3d73Snpiggin@suse.de * atomic_dec_and_lock() used to deal with ->mnt_count decrements 644d3ef3d73Snpiggin@suse.de * provides barriers, so count_mnt_writers() below is safe. AV 645d3ef3d73Snpiggin@suse.de */ 646d3ef3d73Snpiggin@suse.de WARN_ON(count_mnt_writers(mnt)); 647ca9c726eSAndreas Gruenbacher fsnotify_vfsmount_delete(mnt); 6481da177e4SLinus Torvalds dput(mnt->mnt_root); 6491da177e4SLinus Torvalds free_vfsmnt(mnt); 6501da177e4SLinus Torvalds deactivate_super(sb); 6511da177e4SLinus Torvalds } 6521da177e4SLinus Torvalds 6537b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 6547b7b1aceSAl Viro { 6557b7b1aceSAl Viro repeat: 65699b7db7bSNick Piggin if (atomic_add_unless(&mnt->mnt_count, -1, 1)) 65799b7db7bSNick Piggin return; 65899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 65999b7db7bSNick Piggin if (!atomic_dec_and_test(&mnt->mnt_count)) { 66099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 66199b7db7bSNick Piggin return; 66299b7db7bSNick Piggin } 6637b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 66499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 6657b7b1aceSAl Viro __mntput(mnt); 6667b7b1aceSAl Viro return; 6677b7b1aceSAl Viro } 6687b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 6697b7b1aceSAl Viro mnt->mnt_pinned = 0; 67099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 6717b7b1aceSAl Viro acct_auto_close_mnt(mnt); 6727b7b1aceSAl Viro goto repeat; 6737b7b1aceSAl Viro } 6747b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 6757b7b1aceSAl Viro 6767b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 6777b7b1aceSAl Viro { 67899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 6797b7b1aceSAl Viro mnt->mnt_pinned++; 68099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 6817b7b1aceSAl Viro } 6827b7b1aceSAl Viro 6837b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 6847b7b1aceSAl Viro 6857b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 6867b7b1aceSAl Viro { 68799b7db7bSNick Piggin br_write_lock(vfsmount_lock); 6887b7b1aceSAl Viro if (mnt->mnt_pinned) { 6897b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 6907b7b1aceSAl Viro mnt->mnt_pinned--; 6917b7b1aceSAl Viro } 69299b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 6937b7b1aceSAl Viro } 6947b7b1aceSAl Viro 6957b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 6961da177e4SLinus Torvalds 697b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 698b3b304a2SMiklos Szeredi { 699b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 700b3b304a2SMiklos Szeredi } 701b3b304a2SMiklos Szeredi 702b3b304a2SMiklos Szeredi /* 703b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 704b3b304a2SMiklos Szeredi * implement more complex mount option showing. 705b3b304a2SMiklos Szeredi * 706b3b304a2SMiklos Szeredi * See also save_mount_options(). 707b3b304a2SMiklos Szeredi */ 708b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 709b3b304a2SMiklos Szeredi { 7102a32cebdSAl Viro const char *options; 7112a32cebdSAl Viro 7122a32cebdSAl Viro rcu_read_lock(); 7132a32cebdSAl Viro options = rcu_dereference(mnt->mnt_sb->s_options); 714b3b304a2SMiklos Szeredi 715b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 716b3b304a2SMiklos Szeredi seq_putc(m, ','); 717b3b304a2SMiklos Szeredi mangle(m, options); 718b3b304a2SMiklos Szeredi } 7192a32cebdSAl Viro rcu_read_unlock(); 720b3b304a2SMiklos Szeredi 721b3b304a2SMiklos Szeredi return 0; 722b3b304a2SMiklos Szeredi } 723b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 724b3b304a2SMiklos Szeredi 725b3b304a2SMiklos Szeredi /* 726b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 727b3b304a2SMiklos Szeredi * called from the fill_super() callback. 728b3b304a2SMiklos Szeredi * 729b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 730b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 731b3b304a2SMiklos Szeredi * remount fails. 732b3b304a2SMiklos Szeredi * 733b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 734b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 735b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 736b3b304a2SMiklos Szeredi * any more. 737b3b304a2SMiklos Szeredi */ 738b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 739b3b304a2SMiklos Szeredi { 7402a32cebdSAl Viro BUG_ON(sb->s_options); 7412a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 742b3b304a2SMiklos Szeredi } 743b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 744b3b304a2SMiklos Szeredi 7452a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 7462a32cebdSAl Viro { 7472a32cebdSAl Viro char *old = sb->s_options; 7482a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 7492a32cebdSAl Viro if (old) { 7502a32cebdSAl Viro synchronize_rcu(); 7512a32cebdSAl Viro kfree(old); 7522a32cebdSAl Viro } 7532a32cebdSAl Viro } 7542a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 7552a32cebdSAl Viro 756a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 7571da177e4SLinus Torvalds /* iterator */ 7581da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 7591da177e4SLinus Torvalds { 760a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 7611da177e4SLinus Torvalds 762390c6843SRam Pai down_read(&namespace_sem); 763a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 7641da177e4SLinus Torvalds } 7651da177e4SLinus Torvalds 7661da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 7671da177e4SLinus Torvalds { 768a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 769b0765fb8SPavel Emelianov 770a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 7711da177e4SLinus Torvalds } 7721da177e4SLinus Torvalds 7731da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 7741da177e4SLinus Torvalds { 775390c6843SRam Pai up_read(&namespace_sem); 7761da177e4SLinus Torvalds } 7771da177e4SLinus Torvalds 7789f5596afSAl Viro int mnt_had_events(struct proc_mounts *p) 7799f5596afSAl Viro { 7809f5596afSAl Viro struct mnt_namespace *ns = p->ns; 7819f5596afSAl Viro int res = 0; 7829f5596afSAl Viro 78399b7db7bSNick Piggin br_read_lock(vfsmount_lock); 7849f5596afSAl Viro if (p->event != ns->event) { 7859f5596afSAl Viro p->event = ns->event; 7869f5596afSAl Viro res = 1; 7879f5596afSAl Viro } 78899b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 7899f5596afSAl Viro 7909f5596afSAl Viro return res; 7919f5596afSAl Viro } 7929f5596afSAl Viro 7932d4d4864SRam Pai struct proc_fs_info { 7941da177e4SLinus Torvalds int flag; 7952d4d4864SRam Pai const char *str; 7962d4d4864SRam Pai }; 7972d4d4864SRam Pai 7982069f457SEric Paris static int show_sb_opts(struct seq_file *m, struct super_block *sb) 7992d4d4864SRam Pai { 8002d4d4864SRam Pai static const struct proc_fs_info fs_info[] = { 8011da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 8021da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 8031da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 8041da177e4SLinus Torvalds { 0, NULL } 8051da177e4SLinus Torvalds }; 8062d4d4864SRam Pai const struct proc_fs_info *fs_infop; 8072d4d4864SRam Pai 8082d4d4864SRam Pai for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 8092d4d4864SRam Pai if (sb->s_flags & fs_infop->flag) 8102d4d4864SRam Pai seq_puts(m, fs_infop->str); 8112d4d4864SRam Pai } 8122069f457SEric Paris 8132069f457SEric Paris return security_sb_show_options(m, sb); 8142d4d4864SRam Pai } 8152d4d4864SRam Pai 8162d4d4864SRam Pai static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) 8172d4d4864SRam Pai { 8182d4d4864SRam Pai static const struct proc_fs_info mnt_info[] = { 8191da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 8201da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 8211da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 822fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 823fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 82447ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 8251da177e4SLinus Torvalds { 0, NULL } 8261da177e4SLinus Torvalds }; 8272d4d4864SRam Pai const struct proc_fs_info *fs_infop; 8282d4d4864SRam Pai 8292d4d4864SRam Pai for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 8302d4d4864SRam Pai if (mnt->mnt_flags & fs_infop->flag) 8312d4d4864SRam Pai seq_puts(m, fs_infop->str); 8322d4d4864SRam Pai } 8332d4d4864SRam Pai } 8342d4d4864SRam Pai 8352d4d4864SRam Pai static void show_type(struct seq_file *m, struct super_block *sb) 8362d4d4864SRam Pai { 8372d4d4864SRam Pai mangle(m, sb->s_type->name); 8382d4d4864SRam Pai if (sb->s_subtype && sb->s_subtype[0]) { 8392d4d4864SRam Pai seq_putc(m, '.'); 8402d4d4864SRam Pai mangle(m, sb->s_subtype); 8412d4d4864SRam Pai } 8422d4d4864SRam Pai } 8432d4d4864SRam Pai 8442d4d4864SRam Pai static int show_vfsmnt(struct seq_file *m, void *v) 8452d4d4864SRam Pai { 8462d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 8472d4d4864SRam Pai int err = 0; 848c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 8491da177e4SLinus Torvalds 8501da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 8511da177e4SLinus Torvalds seq_putc(m, ' '); 852c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 8531da177e4SLinus Torvalds seq_putc(m, ' '); 8542d4d4864SRam Pai show_type(m, mnt->mnt_sb); 8552e4b7fcdSDave Hansen seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 8562069f457SEric Paris err = show_sb_opts(m, mnt->mnt_sb); 8572069f457SEric Paris if (err) 8582069f457SEric Paris goto out; 8592d4d4864SRam Pai show_mnt_opts(m, mnt); 8601da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 8611da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 8621da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 8632069f457SEric Paris out: 8641da177e4SLinus Torvalds return err; 8651da177e4SLinus Torvalds } 8661da177e4SLinus Torvalds 867a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 8681da177e4SLinus Torvalds .start = m_start, 8691da177e4SLinus Torvalds .next = m_next, 8701da177e4SLinus Torvalds .stop = m_stop, 8711da177e4SLinus Torvalds .show = show_vfsmnt 8721da177e4SLinus Torvalds }; 8731da177e4SLinus Torvalds 8742d4d4864SRam Pai static int show_mountinfo(struct seq_file *m, void *v) 8752d4d4864SRam Pai { 8762d4d4864SRam Pai struct proc_mounts *p = m->private; 8772d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 8782d4d4864SRam Pai struct super_block *sb = mnt->mnt_sb; 8792d4d4864SRam Pai struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 8802d4d4864SRam Pai struct path root = p->root; 8812d4d4864SRam Pai int err = 0; 8822d4d4864SRam Pai 8832d4d4864SRam Pai seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, 8842d4d4864SRam Pai MAJOR(sb->s_dev), MINOR(sb->s_dev)); 8852d4d4864SRam Pai seq_dentry(m, mnt->mnt_root, " \t\n\\"); 8862d4d4864SRam Pai seq_putc(m, ' '); 8872d4d4864SRam Pai seq_path_root(m, &mnt_path, &root, " \t\n\\"); 8882d4d4864SRam Pai if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { 8892d4d4864SRam Pai /* 8902d4d4864SRam Pai * Mountpoint is outside root, discard that one. Ugly, 8912d4d4864SRam Pai * but less so than trying to do that in iterator in a 8922d4d4864SRam Pai * race-free way (due to renames). 8932d4d4864SRam Pai */ 8942d4d4864SRam Pai return SEQ_SKIP; 8952d4d4864SRam Pai } 8962d4d4864SRam Pai seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); 8972d4d4864SRam Pai show_mnt_opts(m, mnt); 8982d4d4864SRam Pai 8992d4d4864SRam Pai /* Tagged fields ("foo:X" or "bar") */ 9002d4d4864SRam Pai if (IS_MNT_SHARED(mnt)) 9012d4d4864SRam Pai seq_printf(m, " shared:%i", mnt->mnt_group_id); 90297e7e0f7SMiklos Szeredi if (IS_MNT_SLAVE(mnt)) { 90397e7e0f7SMiklos Szeredi int master = mnt->mnt_master->mnt_group_id; 90497e7e0f7SMiklos Szeredi int dom = get_dominating_id(mnt, &p->root); 90597e7e0f7SMiklos Szeredi seq_printf(m, " master:%i", master); 90697e7e0f7SMiklos Szeredi if (dom && dom != master) 90797e7e0f7SMiklos Szeredi seq_printf(m, " propagate_from:%i", dom); 90897e7e0f7SMiklos Szeredi } 9092d4d4864SRam Pai if (IS_MNT_UNBINDABLE(mnt)) 9102d4d4864SRam Pai seq_puts(m, " unbindable"); 9112d4d4864SRam Pai 9122d4d4864SRam Pai /* Filesystem specific data */ 9132d4d4864SRam Pai seq_puts(m, " - "); 9142d4d4864SRam Pai show_type(m, sb); 9152d4d4864SRam Pai seq_putc(m, ' '); 9162d4d4864SRam Pai mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 9172d4d4864SRam Pai seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); 9182069f457SEric Paris err = show_sb_opts(m, sb); 9192069f457SEric Paris if (err) 9202069f457SEric Paris goto out; 9212d4d4864SRam Pai if (sb->s_op->show_options) 9222d4d4864SRam Pai err = sb->s_op->show_options(m, mnt); 9232d4d4864SRam Pai seq_putc(m, '\n'); 9242069f457SEric Paris out: 9252d4d4864SRam Pai return err; 9262d4d4864SRam Pai } 9272d4d4864SRam Pai 9282d4d4864SRam Pai const struct seq_operations mountinfo_op = { 9292d4d4864SRam Pai .start = m_start, 9302d4d4864SRam Pai .next = m_next, 9312d4d4864SRam Pai .stop = m_stop, 9322d4d4864SRam Pai .show = show_mountinfo, 9332d4d4864SRam Pai }; 9342d4d4864SRam Pai 935b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 936b4629fe2SChuck Lever { 937b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 938c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 939b4629fe2SChuck Lever int err = 0; 940b4629fe2SChuck Lever 941b4629fe2SChuck Lever /* device */ 942b4629fe2SChuck Lever if (mnt->mnt_devname) { 943b4629fe2SChuck Lever seq_puts(m, "device "); 944b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 945b4629fe2SChuck Lever } else 946b4629fe2SChuck Lever seq_puts(m, "no device"); 947b4629fe2SChuck Lever 948b4629fe2SChuck Lever /* mount point */ 949b4629fe2SChuck Lever seq_puts(m, " mounted on "); 950c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 951b4629fe2SChuck Lever seq_putc(m, ' '); 952b4629fe2SChuck Lever 953b4629fe2SChuck Lever /* file system type */ 954b4629fe2SChuck Lever seq_puts(m, "with fstype "); 9552d4d4864SRam Pai show_type(m, mnt->mnt_sb); 956b4629fe2SChuck Lever 957b4629fe2SChuck Lever /* optional statistics */ 958b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 959b4629fe2SChuck Lever seq_putc(m, ' '); 960b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 961b4629fe2SChuck Lever } 962b4629fe2SChuck Lever 963b4629fe2SChuck Lever seq_putc(m, '\n'); 964b4629fe2SChuck Lever return err; 965b4629fe2SChuck Lever } 966b4629fe2SChuck Lever 967a1a2c409SMiklos Szeredi const struct seq_operations mountstats_op = { 968b4629fe2SChuck Lever .start = m_start, 969b4629fe2SChuck Lever .next = m_next, 970b4629fe2SChuck Lever .stop = m_stop, 971b4629fe2SChuck Lever .show = show_vfsstat, 972b4629fe2SChuck Lever }; 973a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 974b4629fe2SChuck Lever 9751da177e4SLinus Torvalds /** 9761da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 9771da177e4SLinus Torvalds * @mnt: root of mount tree 9781da177e4SLinus Torvalds * 9791da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 9801da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 9811da177e4SLinus Torvalds * busy. 9821da177e4SLinus Torvalds */ 9831da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 9841da177e4SLinus Torvalds { 98536341f64SRam Pai int actual_refs = 0; 98636341f64SRam Pai int minimum_refs = 0; 98736341f64SRam Pai struct vfsmount *p; 9881da177e4SLinus Torvalds 98999b7db7bSNick Piggin br_read_lock(vfsmount_lock); 99036341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 9911da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 9921da177e4SLinus Torvalds minimum_refs += 2; 9931da177e4SLinus Torvalds } 99499b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 9951da177e4SLinus Torvalds 9961da177e4SLinus Torvalds if (actual_refs > minimum_refs) 9971da177e4SLinus Torvalds return 0; 998e3474a8eSIan Kent 999e3474a8eSIan Kent return 1; 10001da177e4SLinus Torvalds } 10011da177e4SLinus Torvalds 10021da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 10031da177e4SLinus Torvalds 10041da177e4SLinus Torvalds /** 10051da177e4SLinus Torvalds * may_umount - check if a mount point is busy 10061da177e4SLinus Torvalds * @mnt: root of mount 10071da177e4SLinus Torvalds * 10081da177e4SLinus Torvalds * This is called to check if a mount point has any 10091da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 10101da177e4SLinus Torvalds * mount has sub mounts this will return busy 10111da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 10121da177e4SLinus Torvalds * 10131da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 10141da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 10151da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 10161da177e4SLinus Torvalds */ 10171da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 10181da177e4SLinus Torvalds { 1019e3474a8eSIan Kent int ret = 1; 10208ad08d8aSAl Viro down_read(&namespace_sem); 102199b7db7bSNick Piggin br_read_lock(vfsmount_lock); 1022a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 1023e3474a8eSIan Kent ret = 0; 102499b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 10258ad08d8aSAl Viro up_read(&namespace_sem); 1026a05964f3SRam Pai return ret; 10271da177e4SLinus Torvalds } 10281da177e4SLinus Torvalds 10291da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 10301da177e4SLinus Torvalds 1031b90fa9aeSRam Pai void release_mounts(struct list_head *head) 10321da177e4SLinus Torvalds { 103370fbcdf4SRam Pai struct vfsmount *mnt; 103470fbcdf4SRam Pai while (!list_empty(head)) { 1035b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 103670fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 103770fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 103870fbcdf4SRam Pai struct dentry *dentry; 103970fbcdf4SRam Pai struct vfsmount *m; 104099b7db7bSNick Piggin 104199b7db7bSNick Piggin br_write_lock(vfsmount_lock); 104270fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 104370fbcdf4SRam Pai m = mnt->mnt_parent; 104470fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 104570fbcdf4SRam Pai mnt->mnt_parent = mnt; 10467c4b93d8SAl Viro m->mnt_ghosts--; 104799b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 104870fbcdf4SRam Pai dput(dentry); 104970fbcdf4SRam Pai mntput(m); 10501da177e4SLinus Torvalds } 10511da177e4SLinus Torvalds mntput(mnt); 105270fbcdf4SRam Pai } 105370fbcdf4SRam Pai } 105470fbcdf4SRam Pai 105599b7db7bSNick Piggin /* 105699b7db7bSNick Piggin * vfsmount lock must be held for write 105799b7db7bSNick Piggin * namespace_sem must be held for write 105899b7db7bSNick Piggin */ 1059a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 106070fbcdf4SRam Pai { 106170fbcdf4SRam Pai struct vfsmount *p; 106270fbcdf4SRam Pai 10631bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 10641bfba4e8SAkinobu Mita list_move(&p->mnt_hash, kill); 106570fbcdf4SRam Pai 1066a05964f3SRam Pai if (propagate) 1067a05964f3SRam Pai propagate_umount(kill); 1068a05964f3SRam Pai 106970fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 107070fbcdf4SRam Pai list_del_init(&p->mnt_expire); 107170fbcdf4SRam Pai list_del_init(&p->mnt_list); 10726b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 10736b3286edSKirill Korotaev p->mnt_ns = NULL; 107470fbcdf4SRam Pai list_del_init(&p->mnt_child); 10757c4b93d8SAl Viro if (p->mnt_parent != p) { 10767c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 1077f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 10787c4b93d8SAl Viro } 1079a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds } 10821da177e4SLinus Torvalds 1083c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 1084c35038beSAl Viro 10851da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 10861da177e4SLinus Torvalds { 10871da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 10881da177e4SLinus Torvalds int retval; 108970fbcdf4SRam Pai LIST_HEAD(umount_list); 10901da177e4SLinus Torvalds 10911da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 10921da177e4SLinus Torvalds if (retval) 10931da177e4SLinus Torvalds return retval; 10941da177e4SLinus Torvalds 10951da177e4SLinus Torvalds /* 10961da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 10971da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 10981da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 10991da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 11001da177e4SLinus Torvalds */ 11011da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 11026ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 11031da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 11041da177e4SLinus Torvalds return -EINVAL; 11051da177e4SLinus Torvalds 11061da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 11071da177e4SLinus Torvalds return -EBUSY; 11081da177e4SLinus Torvalds 11091da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 11101da177e4SLinus Torvalds return -EAGAIN; 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds 11131da177e4SLinus Torvalds /* 11141da177e4SLinus Torvalds * If we may have to abort operations to get out of this 11151da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 11161da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 11171da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 11181da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 11191da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 11201da177e4SLinus Torvalds * about for the moment. 11211da177e4SLinus Torvalds */ 11221da177e4SLinus Torvalds 112342faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 112442faad99SAl Viro sb->s_op->umount_begin(sb); 112542faad99SAl Viro } 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds /* 11281da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 11291da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 11301da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 11311da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 11321da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 11331da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 11341da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 11351da177e4SLinus Torvalds */ 11366ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 11371da177e4SLinus Torvalds /* 11381da177e4SLinus Torvalds * Special case for "unmounting" root ... 11391da177e4SLinus Torvalds * we just try to remount it readonly. 11401da177e4SLinus Torvalds */ 11411da177e4SLinus Torvalds down_write(&sb->s_umount); 11424aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 11431da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 11441da177e4SLinus Torvalds up_write(&sb->s_umount); 11451da177e4SLinus Torvalds return retval; 11461da177e4SLinus Torvalds } 11471da177e4SLinus Torvalds 1148390c6843SRam Pai down_write(&namespace_sem); 114999b7db7bSNick Piggin br_write_lock(vfsmount_lock); 11505addc5ddSAl Viro event++; 11511da177e4SLinus Torvalds 1152c35038beSAl Viro if (!(flags & MNT_DETACH)) 1153c35038beSAl Viro shrink_submounts(mnt, &umount_list); 1154c35038beSAl Viro 11551da177e4SLinus Torvalds retval = -EBUSY; 1156a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 11571da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 1158a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 11591da177e4SLinus Torvalds retval = 0; 11601da177e4SLinus Torvalds } 116199b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1162390c6843SRam Pai up_write(&namespace_sem); 116370fbcdf4SRam Pai release_mounts(&umount_list); 11641da177e4SLinus Torvalds return retval; 11651da177e4SLinus Torvalds } 11661da177e4SLinus Torvalds 11671da177e4SLinus Torvalds /* 11681da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 11691da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 11701da177e4SLinus Torvalds * 11711da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 11721da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 11731da177e4SLinus Torvalds */ 11741da177e4SLinus Torvalds 1175bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 11761da177e4SLinus Torvalds { 11772d8f3038SAl Viro struct path path; 11781da177e4SLinus Torvalds int retval; 1179db1f05bbSMiklos Szeredi int lookup_flags = 0; 11801da177e4SLinus Torvalds 1181db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1182db1f05bbSMiklos Szeredi return -EINVAL; 1183db1f05bbSMiklos Szeredi 1184db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1185db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1186db1f05bbSMiklos Szeredi 1187db1f05bbSMiklos Szeredi retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); 11881da177e4SLinus Torvalds if (retval) 11891da177e4SLinus Torvalds goto out; 11901da177e4SLinus Torvalds retval = -EINVAL; 11912d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 11921da177e4SLinus Torvalds goto dput_and_out; 11932d8f3038SAl Viro if (!check_mnt(path.mnt)) 11941da177e4SLinus Torvalds goto dput_and_out; 11951da177e4SLinus Torvalds 11961da177e4SLinus Torvalds retval = -EPERM; 11971da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 11981da177e4SLinus Torvalds goto dput_and_out; 11991da177e4SLinus Torvalds 12002d8f3038SAl Viro retval = do_umount(path.mnt, flags); 12011da177e4SLinus Torvalds dput_and_out: 1202429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 12032d8f3038SAl Viro dput(path.dentry); 12042d8f3038SAl Viro mntput_no_expire(path.mnt); 12051da177e4SLinus Torvalds out: 12061da177e4SLinus Torvalds return retval; 12071da177e4SLinus Torvalds } 12081da177e4SLinus Torvalds 12091da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds /* 12121da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 12131da177e4SLinus Torvalds */ 1214bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 12151da177e4SLinus Torvalds { 12161da177e4SLinus Torvalds return sys_umount(name, 0); 12171da177e4SLinus Torvalds } 12181da177e4SLinus Torvalds 12191da177e4SLinus Torvalds #endif 12201da177e4SLinus Torvalds 12212d92ab3cSAl Viro static int mount_is_safe(struct path *path) 12221da177e4SLinus Torvalds { 12231da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 12241da177e4SLinus Torvalds return 0; 12251da177e4SLinus Torvalds return -EPERM; 12261da177e4SLinus Torvalds #ifdef notyet 12272d92ab3cSAl Viro if (S_ISLNK(path->dentry->d_inode->i_mode)) 12281da177e4SLinus Torvalds return -EPERM; 12292d92ab3cSAl Viro if (path->dentry->d_inode->i_mode & S_ISVTX) { 1230da9592edSDavid Howells if (current_uid() != path->dentry->d_inode->i_uid) 12311da177e4SLinus Torvalds return -EPERM; 12321da177e4SLinus Torvalds } 12332d92ab3cSAl Viro if (inode_permission(path->dentry->d_inode, MAY_WRITE)) 12341da177e4SLinus Torvalds return -EPERM; 12351da177e4SLinus Torvalds return 0; 12361da177e4SLinus Torvalds #endif 12371da177e4SLinus Torvalds } 12381da177e4SLinus Torvalds 1239b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 124036341f64SRam Pai int flag) 12411da177e4SLinus Torvalds { 12421da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 12431a390689SAl Viro struct path path; 12441da177e4SLinus Torvalds 12459676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 12469676f0c6SRam Pai return NULL; 12479676f0c6SRam Pai 124836341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 12491da177e4SLinus Torvalds if (!q) 12501da177e4SLinus Torvalds goto Enomem; 12511da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 12521da177e4SLinus Torvalds 12531da177e4SLinus Torvalds p = mnt; 1254fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 12557ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 12561da177e4SLinus Torvalds continue; 12571da177e4SLinus Torvalds 12581da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 12599676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 12609676f0c6SRam Pai s = skip_mnt_tree(s); 12619676f0c6SRam Pai continue; 12629676f0c6SRam Pai } 12631da177e4SLinus Torvalds while (p != s->mnt_parent) { 12641da177e4SLinus Torvalds p = p->mnt_parent; 12651da177e4SLinus Torvalds q = q->mnt_parent; 12661da177e4SLinus Torvalds } 12671da177e4SLinus Torvalds p = s; 12681a390689SAl Viro path.mnt = q; 12691a390689SAl Viro path.dentry = p->mnt_mountpoint; 127036341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 12711da177e4SLinus Torvalds if (!q) 12721da177e4SLinus Torvalds goto Enomem; 127399b7db7bSNick Piggin br_write_lock(vfsmount_lock); 12741da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 12751a390689SAl Viro attach_mnt(q, &path); 127699b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 12771da177e4SLinus Torvalds } 12781da177e4SLinus Torvalds } 12791da177e4SLinus Torvalds return res; 12801da177e4SLinus Torvalds Enomem: 12811da177e4SLinus Torvalds if (res) { 128270fbcdf4SRam Pai LIST_HEAD(umount_list); 128399b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1284a05964f3SRam Pai umount_tree(res, 0, &umount_list); 128599b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 128670fbcdf4SRam Pai release_mounts(&umount_list); 12871da177e4SLinus Torvalds } 12881da177e4SLinus Torvalds return NULL; 12891da177e4SLinus Torvalds } 12901da177e4SLinus Torvalds 1291589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 12928aec0809SAl Viro { 12938aec0809SAl Viro struct vfsmount *tree; 12941a60a280SAl Viro down_write(&namespace_sem); 1295589ff870SAl Viro tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); 12961a60a280SAl Viro up_write(&namespace_sem); 12978aec0809SAl Viro return tree; 12988aec0809SAl Viro } 12998aec0809SAl Viro 13008aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 13018aec0809SAl Viro { 13028aec0809SAl Viro LIST_HEAD(umount_list); 13031a60a280SAl Viro down_write(&namespace_sem); 130499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 13058aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 130699b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 13071a60a280SAl Viro up_write(&namespace_sem); 13088aec0809SAl Viro release_mounts(&umount_list); 13098aec0809SAl Viro } 13108aec0809SAl Viro 13111f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 13121f707137SAl Viro struct vfsmount *root) 13131f707137SAl Viro { 13141f707137SAl Viro struct vfsmount *mnt; 13151f707137SAl Viro int res = f(root, arg); 13161f707137SAl Viro if (res) 13171f707137SAl Viro return res; 13181f707137SAl Viro list_for_each_entry(mnt, &root->mnt_list, mnt_list) { 13191f707137SAl Viro res = f(mnt, arg); 13201f707137SAl Viro if (res) 13211f707137SAl Viro return res; 13221f707137SAl Viro } 13231f707137SAl Viro return 0; 13241f707137SAl Viro } 13251f707137SAl Viro 1326719f5d7fSMiklos Szeredi static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) 1327719f5d7fSMiklos Szeredi { 1328719f5d7fSMiklos Szeredi struct vfsmount *p; 1329719f5d7fSMiklos Szeredi 1330719f5d7fSMiklos Szeredi for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1331719f5d7fSMiklos Szeredi if (p->mnt_group_id && !IS_MNT_SHARED(p)) 1332719f5d7fSMiklos Szeredi mnt_release_group_id(p); 1333719f5d7fSMiklos Szeredi } 1334719f5d7fSMiklos Szeredi } 1335719f5d7fSMiklos Szeredi 1336719f5d7fSMiklos Szeredi static int invent_group_ids(struct vfsmount *mnt, bool recurse) 1337719f5d7fSMiklos Szeredi { 1338719f5d7fSMiklos Szeredi struct vfsmount *p; 1339719f5d7fSMiklos Szeredi 1340719f5d7fSMiklos Szeredi for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1341719f5d7fSMiklos Szeredi if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 1342719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(p); 1343719f5d7fSMiklos Szeredi if (err) { 1344719f5d7fSMiklos Szeredi cleanup_group_ids(mnt, p); 1345719f5d7fSMiklos Szeredi return err; 1346719f5d7fSMiklos Szeredi } 1347719f5d7fSMiklos Szeredi } 1348719f5d7fSMiklos Szeredi } 1349719f5d7fSMiklos Szeredi 1350719f5d7fSMiklos Szeredi return 0; 1351719f5d7fSMiklos Szeredi } 1352719f5d7fSMiklos Szeredi 1353b90fa9aeSRam Pai /* 1354b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1355b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 135621444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 135721444403SRam Pai * store the parent mount and mountpoint dentry. 135821444403SRam Pai * (done when source_mnt is moved) 1359b90fa9aeSRam Pai * 1360b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1361b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 13629676f0c6SRam Pai * --------------------------------------------------------------------------- 1363b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 13649676f0c6SRam Pai * |************************************************************************** 13659676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 13669676f0c6SRam Pai * | dest | | | | | 13679676f0c6SRam Pai * | | | | | | | 13689676f0c6SRam Pai * | v | | | | | 13699676f0c6SRam Pai * |************************************************************************** 13709676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 13715afe0022SRam Pai * | | | | | | 13729676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 13739676f0c6SRam Pai * *************************************************************************** 1374b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1375b90fa9aeSRam Pai * destination mount. 1376b90fa9aeSRam Pai * 1377b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1378b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1379b90fa9aeSRam Pai * the peer group of the source mount. 1380b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1381b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1382b90fa9aeSRam Pai * mount. 13835afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 13845afe0022SRam Pai * of the destination mount and the cloned mount is made slave 13855afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 13865afe0022SRam Pai * is marked as 'shared and slave'. 13875afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 13885afe0022SRam Pai * source mount. 13895afe0022SRam Pai * 13909676f0c6SRam Pai * --------------------------------------------------------------------------- 139121444403SRam Pai * | MOVE MOUNT OPERATION | 13929676f0c6SRam Pai * |************************************************************************** 13939676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 13949676f0c6SRam Pai * | dest | | | | | 13959676f0c6SRam Pai * | | | | | | | 13969676f0c6SRam Pai * | v | | | | | 13979676f0c6SRam Pai * |************************************************************************** 13989676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 13995afe0022SRam Pai * | | | | | | 14009676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 14019676f0c6SRam Pai * *************************************************************************** 14025afe0022SRam Pai * 14035afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 14045afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 140521444403SRam Pai * (+*) the mount is moved to the destination. 14065afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 14075afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 14085afe0022SRam Pai * the mount is marked as 'shared and slave'. 14095afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1410b90fa9aeSRam Pai * 1411b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1412b90fa9aeSRam Pai * applied to each mount in the tree. 1413b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1414b90fa9aeSRam Pai * in allocations. 1415b90fa9aeSRam Pai */ 1416b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 14171a390689SAl Viro struct path *path, struct path *parent_path) 1418b90fa9aeSRam Pai { 1419b90fa9aeSRam Pai LIST_HEAD(tree_list); 14201a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 14211a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1422b90fa9aeSRam Pai struct vfsmount *child, *p; 1423719f5d7fSMiklos Szeredi int err; 1424b90fa9aeSRam Pai 1425719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) { 1426719f5d7fSMiklos Szeredi err = invent_group_ids(source_mnt, true); 1427719f5d7fSMiklos Szeredi if (err) 1428719f5d7fSMiklos Szeredi goto out; 1429719f5d7fSMiklos Szeredi } 1430719f5d7fSMiklos Szeredi err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1431719f5d7fSMiklos Szeredi if (err) 1432719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1433b90fa9aeSRam Pai 143499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1435df1a1ad2SAl Viro 1436b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1437b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1438b90fa9aeSRam Pai set_mnt_shared(p); 1439b90fa9aeSRam Pai } 14401a390689SAl Viro if (parent_path) { 14411a390689SAl Viro detach_mnt(source_mnt, parent_path); 14421a390689SAl Viro attach_mnt(source_mnt, path); 1443e5d67f07SAl Viro touch_mnt_namespace(parent_path->mnt->mnt_ns); 144421444403SRam Pai } else { 1445b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1446b90fa9aeSRam Pai commit_tree(source_mnt); 144721444403SRam Pai } 1448b90fa9aeSRam Pai 1449b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1450b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1451b90fa9aeSRam Pai commit_tree(child); 1452b90fa9aeSRam Pai } 145399b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 145499b7db7bSNick Piggin 1455b90fa9aeSRam Pai return 0; 1456719f5d7fSMiklos Szeredi 1457719f5d7fSMiklos Szeredi out_cleanup_ids: 1458719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) 1459719f5d7fSMiklos Szeredi cleanup_group_ids(source_mnt, NULL); 1460719f5d7fSMiklos Szeredi out: 1461719f5d7fSMiklos Szeredi return err; 1462b90fa9aeSRam Pai } 1463b90fa9aeSRam Pai 14648c3ee42eSAl Viro static int graft_tree(struct vfsmount *mnt, struct path *path) 14651da177e4SLinus Torvalds { 14661da177e4SLinus Torvalds int err; 14671da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 14681da177e4SLinus Torvalds return -EINVAL; 14691da177e4SLinus Torvalds 14708c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 14711da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 14721da177e4SLinus Torvalds return -ENOTDIR; 14731da177e4SLinus Torvalds 14741da177e4SLinus Torvalds err = -ENOENT; 14758c3ee42eSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 1476d83c49f3SAl Viro if (cant_mount(path->dentry)) 14771da177e4SLinus Torvalds goto out_unlock; 14781da177e4SLinus Torvalds 1479f3da392eSAlexey Dobriyan if (!d_unlinked(path->dentry)) 14808c3ee42eSAl Viro err = attach_recursive_mnt(mnt, path, NULL); 14811da177e4SLinus Torvalds out_unlock: 14828c3ee42eSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 14831da177e4SLinus Torvalds return err; 14841da177e4SLinus Torvalds } 14851da177e4SLinus Torvalds 14861da177e4SLinus Torvalds /* 148707b20889SRam Pai * recursively change the type of the mountpoint. 148807b20889SRam Pai */ 14890a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 149007b20889SRam Pai { 14912d92ab3cSAl Viro struct vfsmount *m, *mnt = path->mnt; 149207b20889SRam Pai int recurse = flag & MS_REC; 149307b20889SRam Pai int type = flag & ~MS_REC; 1494719f5d7fSMiklos Szeredi int err = 0; 149507b20889SRam Pai 1496ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1497ee6f9582SMiklos Szeredi return -EPERM; 1498ee6f9582SMiklos Szeredi 14992d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 150007b20889SRam Pai return -EINVAL; 150107b20889SRam Pai 150207b20889SRam Pai down_write(&namespace_sem); 1503719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1504719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1505719f5d7fSMiklos Szeredi if (err) 1506719f5d7fSMiklos Szeredi goto out_unlock; 1507719f5d7fSMiklos Szeredi } 1508719f5d7fSMiklos Szeredi 150999b7db7bSNick Piggin br_write_lock(vfsmount_lock); 151007b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 151107b20889SRam Pai change_mnt_propagation(m, type); 151299b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1513719f5d7fSMiklos Szeredi 1514719f5d7fSMiklos Szeredi out_unlock: 151507b20889SRam Pai up_write(&namespace_sem); 1516719f5d7fSMiklos Szeredi return err; 151707b20889SRam Pai } 151807b20889SRam Pai 151907b20889SRam Pai /* 15201da177e4SLinus Torvalds * do loopback mount. 15211da177e4SLinus Torvalds */ 15220a0d8a46SAl Viro static int do_loopback(struct path *path, char *old_name, 15232dafe1c4SEric Sandeen int recurse) 15241da177e4SLinus Torvalds { 15252d92ab3cSAl Viro struct path old_path; 15261da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 15272d92ab3cSAl Viro int err = mount_is_safe(path); 15281da177e4SLinus Torvalds if (err) 15291da177e4SLinus Torvalds return err; 15301da177e4SLinus Torvalds if (!old_name || !*old_name) 15311da177e4SLinus Torvalds return -EINVAL; 15322d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 15331da177e4SLinus Torvalds if (err) 15341da177e4SLinus Torvalds return err; 15351da177e4SLinus Torvalds 1536390c6843SRam Pai down_write(&namespace_sem); 15371da177e4SLinus Torvalds err = -EINVAL; 15382d92ab3cSAl Viro if (IS_MNT_UNBINDABLE(old_path.mnt)) 15399676f0c6SRam Pai goto out; 15409676f0c6SRam Pai 15412d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 1542ccd48bc7SAl Viro goto out; 1543ccd48bc7SAl Viro 15441da177e4SLinus Torvalds err = -ENOMEM; 15451da177e4SLinus Torvalds if (recurse) 15462d92ab3cSAl Viro mnt = copy_tree(old_path.mnt, old_path.dentry, 0); 15471da177e4SLinus Torvalds else 15482d92ab3cSAl Viro mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); 15491da177e4SLinus Torvalds 1550ccd48bc7SAl Viro if (!mnt) 1551ccd48bc7SAl Viro goto out; 1552ccd48bc7SAl Viro 15532d92ab3cSAl Viro err = graft_tree(mnt, path); 15541da177e4SLinus Torvalds if (err) { 155570fbcdf4SRam Pai LIST_HEAD(umount_list); 155699b7db7bSNick Piggin 155799b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1558a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 155999b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 156070fbcdf4SRam Pai release_mounts(&umount_list); 15615b83d2c5SRam Pai } 15621da177e4SLinus Torvalds 1563ccd48bc7SAl Viro out: 1564390c6843SRam Pai up_write(&namespace_sem); 15652d92ab3cSAl Viro path_put(&old_path); 15661da177e4SLinus Torvalds return err; 15671da177e4SLinus Torvalds } 15681da177e4SLinus Torvalds 15692e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 15702e4b7fcdSDave Hansen { 15712e4b7fcdSDave Hansen int error = 0; 15722e4b7fcdSDave Hansen int readonly_request = 0; 15732e4b7fcdSDave Hansen 15742e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 15752e4b7fcdSDave Hansen readonly_request = 1; 15762e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 15772e4b7fcdSDave Hansen return 0; 15782e4b7fcdSDave Hansen 15792e4b7fcdSDave Hansen if (readonly_request) 15802e4b7fcdSDave Hansen error = mnt_make_readonly(mnt); 15812e4b7fcdSDave Hansen else 15822e4b7fcdSDave Hansen __mnt_unmake_readonly(mnt); 15832e4b7fcdSDave Hansen return error; 15842e4b7fcdSDave Hansen } 15852e4b7fcdSDave Hansen 15861da177e4SLinus Torvalds /* 15871da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 15881da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 15891da177e4SLinus Torvalds * on it - tough luck. 15901da177e4SLinus Torvalds */ 15910a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 15921da177e4SLinus Torvalds void *data) 15931da177e4SLinus Torvalds { 15941da177e4SLinus Torvalds int err; 15952d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 15961da177e4SLinus Torvalds 15971da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15981da177e4SLinus Torvalds return -EPERM; 15991da177e4SLinus Torvalds 16002d92ab3cSAl Viro if (!check_mnt(path->mnt)) 16011da177e4SLinus Torvalds return -EINVAL; 16021da177e4SLinus Torvalds 16032d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 16041da177e4SLinus Torvalds return -EINVAL; 16051da177e4SLinus Torvalds 16061da177e4SLinus Torvalds down_write(&sb->s_umount); 16072e4b7fcdSDave Hansen if (flags & MS_BIND) 16082d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 16094aa98cf7SAl Viro else 16101da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 16117b43a79fSAl Viro if (!err) { 161299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1613495d6c9cSValerie Aurora mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK; 16142d92ab3cSAl Viro path->mnt->mnt_flags = mnt_flags; 161599b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 16167b43a79fSAl Viro } 16171da177e4SLinus Torvalds up_write(&sb->s_umount); 16180e55a7ccSDan Williams if (!err) { 161999b7db7bSNick Piggin br_write_lock(vfsmount_lock); 16200e55a7ccSDan Williams touch_mnt_namespace(path->mnt->mnt_ns); 162199b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 16220e55a7ccSDan Williams } 16231da177e4SLinus Torvalds return err; 16241da177e4SLinus Torvalds } 16251da177e4SLinus Torvalds 16269676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 16279676f0c6SRam Pai { 16289676f0c6SRam Pai struct vfsmount *p; 16299676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 16309676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 16319676f0c6SRam Pai return 1; 16329676f0c6SRam Pai } 16339676f0c6SRam Pai return 0; 16349676f0c6SRam Pai } 16359676f0c6SRam Pai 16360a0d8a46SAl Viro static int do_move_mount(struct path *path, char *old_name) 16371da177e4SLinus Torvalds { 16382d92ab3cSAl Viro struct path old_path, parent_path; 16391da177e4SLinus Torvalds struct vfsmount *p; 16401da177e4SLinus Torvalds int err = 0; 16411da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 16421da177e4SLinus Torvalds return -EPERM; 16431da177e4SLinus Torvalds if (!old_name || !*old_name) 16441da177e4SLinus Torvalds return -EINVAL; 16452d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 16461da177e4SLinus Torvalds if (err) 16471da177e4SLinus Torvalds return err; 16481da177e4SLinus Torvalds 1649390c6843SRam Pai down_write(&namespace_sem); 16502d92ab3cSAl Viro while (d_mountpoint(path->dentry) && 16519393bd07SAl Viro follow_down(path)) 16521da177e4SLinus Torvalds ; 16531da177e4SLinus Torvalds err = -EINVAL; 16542d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 16551da177e4SLinus Torvalds goto out; 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds err = -ENOENT; 16582d92ab3cSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 1659d83c49f3SAl Viro if (cant_mount(path->dentry)) 16601da177e4SLinus Torvalds goto out1; 16611da177e4SLinus Torvalds 1662f3da392eSAlexey Dobriyan if (d_unlinked(path->dentry)) 166321444403SRam Pai goto out1; 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds err = -EINVAL; 16662d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 166721444403SRam Pai goto out1; 16681da177e4SLinus Torvalds 16692d92ab3cSAl Viro if (old_path.mnt == old_path.mnt->mnt_parent) 167021444403SRam Pai goto out1; 16711da177e4SLinus Torvalds 16722d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 16732d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 167421444403SRam Pai goto out1; 167521444403SRam Pai /* 167621444403SRam Pai * Don't move a mount residing in a shared parent. 167721444403SRam Pai */ 16782d92ab3cSAl Viro if (old_path.mnt->mnt_parent && 16792d92ab3cSAl Viro IS_MNT_SHARED(old_path.mnt->mnt_parent)) 168021444403SRam Pai goto out1; 16819676f0c6SRam Pai /* 16829676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 16839676f0c6SRam Pai * mount which is shared. 16849676f0c6SRam Pai */ 16852d92ab3cSAl Viro if (IS_MNT_SHARED(path->mnt) && 16862d92ab3cSAl Viro tree_contains_unbindable(old_path.mnt)) 16879676f0c6SRam Pai goto out1; 16881da177e4SLinus Torvalds err = -ELOOP; 16892d92ab3cSAl Viro for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) 16902d92ab3cSAl Viro if (p == old_path.mnt) 169121444403SRam Pai goto out1; 16921da177e4SLinus Torvalds 16932d92ab3cSAl Viro err = attach_recursive_mnt(old_path.mnt, path, &parent_path); 16944ac91378SJan Blunck if (err) 169521444403SRam Pai goto out1; 16961da177e4SLinus Torvalds 16971da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 16981da177e4SLinus Torvalds * automatically */ 16992d92ab3cSAl Viro list_del_init(&old_path.mnt->mnt_expire); 17001da177e4SLinus Torvalds out1: 17012d92ab3cSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 17021da177e4SLinus Torvalds out: 1703390c6843SRam Pai up_write(&namespace_sem); 17041da177e4SLinus Torvalds if (!err) 17051a390689SAl Viro path_put(&parent_path); 17062d92ab3cSAl Viro path_put(&old_path); 17071da177e4SLinus Torvalds return err; 17081da177e4SLinus Torvalds } 17091da177e4SLinus Torvalds 17101da177e4SLinus Torvalds /* 17111da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 17121da177e4SLinus Torvalds * namespace's tree 17131da177e4SLinus Torvalds */ 17140a0d8a46SAl Viro static int do_new_mount(struct path *path, char *type, int flags, 17151da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 17161da177e4SLinus Torvalds { 17171da177e4SLinus Torvalds struct vfsmount *mnt; 17181da177e4SLinus Torvalds 1719eca6f534SVegard Nossum if (!type) 17201da177e4SLinus Torvalds return -EINVAL; 17211da177e4SLinus Torvalds 17221da177e4SLinus Torvalds /* we need capabilities... */ 17231da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 17241da177e4SLinus Torvalds return -EPERM; 17251da177e4SLinus Torvalds 17267f78d4cdSAl Viro lock_kernel(); 17271da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 17287f78d4cdSAl Viro unlock_kernel(); 17291da177e4SLinus Torvalds if (IS_ERR(mnt)) 17301da177e4SLinus Torvalds return PTR_ERR(mnt); 17311da177e4SLinus Torvalds 17322d92ab3cSAl Viro return do_add_mount(mnt, path, mnt_flags, NULL); 17331da177e4SLinus Torvalds } 17341da177e4SLinus Torvalds 17351da177e4SLinus Torvalds /* 17361da177e4SLinus Torvalds * add a mount into a namespace's mount tree 17371da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 17381da177e4SLinus Torvalds */ 17398d66bf54SAl Viro int do_add_mount(struct vfsmount *newmnt, struct path *path, 17401da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 17411da177e4SLinus Torvalds { 17421da177e4SLinus Torvalds int err; 17431da177e4SLinus Torvalds 17448089352aSAl Viro mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); 174527d55f1fSAl Viro 1746390c6843SRam Pai down_write(&namespace_sem); 17471da177e4SLinus Torvalds /* Something was mounted here while we slept */ 17488d66bf54SAl Viro while (d_mountpoint(path->dentry) && 17499393bd07SAl Viro follow_down(path)) 17501da177e4SLinus Torvalds ; 17511da177e4SLinus Torvalds err = -EINVAL; 1752dd5cae6eSAl Viro if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) 17531da177e4SLinus Torvalds goto unlock; 17541da177e4SLinus Torvalds 17551da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 17561da177e4SLinus Torvalds err = -EBUSY; 17578d66bf54SAl Viro if (path->mnt->mnt_sb == newmnt->mnt_sb && 17588d66bf54SAl Viro path->mnt->mnt_root == path->dentry) 17591da177e4SLinus Torvalds goto unlock; 17601da177e4SLinus Torvalds 17611da177e4SLinus Torvalds err = -EINVAL; 17621da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 17631da177e4SLinus Torvalds goto unlock; 17641da177e4SLinus Torvalds 17651da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 17668d66bf54SAl Viro if ((err = graft_tree(newmnt, path))) 17675b83d2c5SRam Pai goto unlock; 17681da177e4SLinus Torvalds 17696758f953SAl Viro if (fslist) /* add to the specified expiration list */ 177055e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 17716758f953SAl Viro 1772390c6843SRam Pai up_write(&namespace_sem); 17735b83d2c5SRam Pai return 0; 17741da177e4SLinus Torvalds 17751da177e4SLinus Torvalds unlock: 1776390c6843SRam Pai up_write(&namespace_sem); 17771da177e4SLinus Torvalds mntput(newmnt); 17781da177e4SLinus Torvalds return err; 17791da177e4SLinus Torvalds } 17801da177e4SLinus Torvalds 17811da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 17821da177e4SLinus Torvalds 17835528f911STrond Myklebust /* 17841da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 17851da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 17861da177e4SLinus Torvalds * here 17871da177e4SLinus Torvalds */ 17881da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 17891da177e4SLinus Torvalds { 17901da177e4SLinus Torvalds struct vfsmount *mnt, *next; 17911da177e4SLinus Torvalds LIST_HEAD(graveyard); 1792bcc5c7d2SAl Viro LIST_HEAD(umounts); 17931da177e4SLinus Torvalds 17941da177e4SLinus Torvalds if (list_empty(mounts)) 17951da177e4SLinus Torvalds return; 17961da177e4SLinus Torvalds 1797bcc5c7d2SAl Viro down_write(&namespace_sem); 179899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 17991da177e4SLinus Torvalds 18001da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 18011da177e4SLinus Torvalds * following criteria: 18021da177e4SLinus Torvalds * - only referenced by its parent vfsmount 18031da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 18041da177e4SLinus Torvalds * cleared by mntput()) 18051da177e4SLinus Torvalds */ 180655e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 18071da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 1808bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 18091da177e4SLinus Torvalds continue; 181055e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 18111da177e4SLinus Torvalds } 1812bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1813bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 1814bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1815bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 1816bcc5c7d2SAl Viro } 181799b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1818bcc5c7d2SAl Viro up_write(&namespace_sem); 1819bcc5c7d2SAl Viro 1820bcc5c7d2SAl Viro release_mounts(&umounts); 18211da177e4SLinus Torvalds } 18221da177e4SLinus Torvalds 18231da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 18241da177e4SLinus Torvalds 18251da177e4SLinus Torvalds /* 18265528f911STrond Myklebust * Ripoff of 'select_parent()' 18275528f911STrond Myklebust * 18285528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 18295528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 18305528f911STrond Myklebust */ 18315528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 18325528f911STrond Myklebust { 18335528f911STrond Myklebust struct vfsmount *this_parent = parent; 18345528f911STrond Myklebust struct list_head *next; 18355528f911STrond Myklebust int found = 0; 18365528f911STrond Myklebust 18375528f911STrond Myklebust repeat: 18385528f911STrond Myklebust next = this_parent->mnt_mounts.next; 18395528f911STrond Myklebust resume: 18405528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 18415528f911STrond Myklebust struct list_head *tmp = next; 18425528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 18435528f911STrond Myklebust 18445528f911STrond Myklebust next = tmp->next; 18455528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 18465528f911STrond Myklebust continue; 18475528f911STrond Myklebust /* 18485528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 18495528f911STrond Myklebust */ 18505528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 18515528f911STrond Myklebust this_parent = mnt; 18525528f911STrond Myklebust goto repeat; 18535528f911STrond Myklebust } 18545528f911STrond Myklebust 18555528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 18565528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 18575528f911STrond Myklebust found++; 18585528f911STrond Myklebust } 18595528f911STrond Myklebust } 18605528f911STrond Myklebust /* 18615528f911STrond Myklebust * All done at this level ... ascend and resume the search 18625528f911STrond Myklebust */ 18635528f911STrond Myklebust if (this_parent != parent) { 18645528f911STrond Myklebust next = this_parent->mnt_child.next; 18655528f911STrond Myklebust this_parent = this_parent->mnt_parent; 18665528f911STrond Myklebust goto resume; 18675528f911STrond Myklebust } 18685528f911STrond Myklebust return found; 18695528f911STrond Myklebust } 18705528f911STrond Myklebust 18715528f911STrond Myklebust /* 18725528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 18735528f911STrond Myklebust * submounts of a specific parent mountpoint 187499b7db7bSNick Piggin * 187599b7db7bSNick Piggin * vfsmount_lock must be held for write 18765528f911STrond Myklebust */ 1877c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 18785528f911STrond Myklebust { 18795528f911STrond Myklebust LIST_HEAD(graveyard); 1880c35038beSAl Viro struct vfsmount *m; 18815528f911STrond Myklebust 18825528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 1883c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 1884bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1885c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 1886bcc5c7d2SAl Viro mnt_expire); 1887afef80b3SEric W. Biederman touch_mnt_namespace(m->mnt_ns); 1888afef80b3SEric W. Biederman umount_tree(m, 1, umounts); 1889bcc5c7d2SAl Viro } 1890bcc5c7d2SAl Viro } 18915528f911STrond Myklebust } 18925528f911STrond Myklebust 18935528f911STrond Myklebust /* 18941da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 18951da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 18961da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 18971da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 18981da177e4SLinus Torvalds */ 1899b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1900b58fed8bSRam Pai unsigned long n) 19011da177e4SLinus Torvalds { 19021da177e4SLinus Torvalds char *t = to; 19031da177e4SLinus Torvalds const char __user *f = from; 19041da177e4SLinus Torvalds char c; 19051da177e4SLinus Torvalds 19061da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 19071da177e4SLinus Torvalds return n; 19081da177e4SLinus Torvalds 19091da177e4SLinus Torvalds while (n) { 19101da177e4SLinus Torvalds if (__get_user(c, f)) { 19111da177e4SLinus Torvalds memset(t, 0, n); 19121da177e4SLinus Torvalds break; 19131da177e4SLinus Torvalds } 19141da177e4SLinus Torvalds *t++ = c; 19151da177e4SLinus Torvalds f++; 19161da177e4SLinus Torvalds n--; 19171da177e4SLinus Torvalds } 19181da177e4SLinus Torvalds return n; 19191da177e4SLinus Torvalds } 19201da177e4SLinus Torvalds 19211da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 19221da177e4SLinus Torvalds { 19231da177e4SLinus Torvalds int i; 19241da177e4SLinus Torvalds unsigned long page; 19251da177e4SLinus Torvalds unsigned long size; 19261da177e4SLinus Torvalds 19271da177e4SLinus Torvalds *where = 0; 19281da177e4SLinus Torvalds if (!data) 19291da177e4SLinus Torvalds return 0; 19301da177e4SLinus Torvalds 19311da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 19321da177e4SLinus Torvalds return -ENOMEM; 19331da177e4SLinus Torvalds 19341da177e4SLinus Torvalds /* We only care that *some* data at the address the user 19351da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 19361da177e4SLinus Torvalds * the remainder of the page. 19371da177e4SLinus Torvalds */ 19381da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 19391da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 19401da177e4SLinus Torvalds if (size > PAGE_SIZE) 19411da177e4SLinus Torvalds size = PAGE_SIZE; 19421da177e4SLinus Torvalds 19431da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 19441da177e4SLinus Torvalds if (!i) { 19451da177e4SLinus Torvalds free_page(page); 19461da177e4SLinus Torvalds return -EFAULT; 19471da177e4SLinus Torvalds } 19481da177e4SLinus Torvalds if (i != PAGE_SIZE) 19491da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 19501da177e4SLinus Torvalds *where = page; 19511da177e4SLinus Torvalds return 0; 19521da177e4SLinus Torvalds } 19531da177e4SLinus Torvalds 1954eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 1955eca6f534SVegard Nossum { 1956eca6f534SVegard Nossum char *tmp; 1957eca6f534SVegard Nossum 1958eca6f534SVegard Nossum if (!data) { 1959eca6f534SVegard Nossum *where = NULL; 1960eca6f534SVegard Nossum return 0; 1961eca6f534SVegard Nossum } 1962eca6f534SVegard Nossum 1963eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 1964eca6f534SVegard Nossum if (IS_ERR(tmp)) 1965eca6f534SVegard Nossum return PTR_ERR(tmp); 1966eca6f534SVegard Nossum 1967eca6f534SVegard Nossum *where = tmp; 1968eca6f534SVegard Nossum return 0; 1969eca6f534SVegard Nossum } 1970eca6f534SVegard Nossum 19711da177e4SLinus Torvalds /* 19721da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 19731da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 19741da177e4SLinus Torvalds * 19751da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 19761da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 19771da177e4SLinus Torvalds * information (or be NULL). 19781da177e4SLinus Torvalds * 19791da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 19801da177e4SLinus Torvalds * When the flags word was introduced its top half was required 19811da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 19821da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 19831da177e4SLinus Torvalds * and must be discarded. 19841da177e4SLinus Torvalds */ 19851da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 19861da177e4SLinus Torvalds unsigned long flags, void *data_page) 19871da177e4SLinus Torvalds { 19882d92ab3cSAl Viro struct path path; 19891da177e4SLinus Torvalds int retval = 0; 19901da177e4SLinus Torvalds int mnt_flags = 0; 19911da177e4SLinus Torvalds 19921da177e4SLinus Torvalds /* Discard magic */ 19931da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 19941da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 19951da177e4SLinus Torvalds 19961da177e4SLinus Torvalds /* Basic sanity checks */ 19971da177e4SLinus Torvalds 19981da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 19991da177e4SLinus Torvalds return -EINVAL; 20001da177e4SLinus Torvalds 20011da177e4SLinus Torvalds if (data_page) 20021da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 20031da177e4SLinus Torvalds 2004a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 2005a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2006a27ab9f2STetsuo Handa if (retval) 2007a27ab9f2STetsuo Handa return retval; 2008a27ab9f2STetsuo Handa 2009a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2010a27ab9f2STetsuo Handa type_page, flags, data_page); 2011a27ab9f2STetsuo Handa if (retval) 2012a27ab9f2STetsuo Handa goto dput_out; 2013a27ab9f2STetsuo Handa 2014613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2015613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 20160a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 20170a1c01c9SMatthew Garrett 20181da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 20191da177e4SLinus Torvalds if (flags & MS_NOSUID) 20201da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 20211da177e4SLinus Torvalds if (flags & MS_NODEV) 20221da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 20231da177e4SLinus Torvalds if (flags & MS_NOEXEC) 20241da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2025fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2026fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2027fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2028fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2029d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2030d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 20312e4b7fcdSDave Hansen if (flags & MS_RDONLY) 20322e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2033fc33a7bbSChristoph Hellwig 20347a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2035d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2036d0adde57SMatthew Garrett MS_STRICTATIME); 20371da177e4SLinus Torvalds 20381da177e4SLinus Torvalds if (flags & MS_REMOUNT) 20392d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 20401da177e4SLinus Torvalds data_page); 20411da177e4SLinus Torvalds else if (flags & MS_BIND) 20422d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 20439676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 20442d92ab3cSAl Viro retval = do_change_type(&path, flags); 20451da177e4SLinus Torvalds else if (flags & MS_MOVE) 20462d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 20471da177e4SLinus Torvalds else 20482d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 20491da177e4SLinus Torvalds dev_name, data_page); 20501da177e4SLinus Torvalds dput_out: 20512d92ab3cSAl Viro path_put(&path); 20521da177e4SLinus Torvalds return retval; 20531da177e4SLinus Torvalds } 20541da177e4SLinus Torvalds 2055cf8d2c11STrond Myklebust static struct mnt_namespace *alloc_mnt_ns(void) 2056cf8d2c11STrond Myklebust { 2057cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2058cf8d2c11STrond Myklebust 2059cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2060cf8d2c11STrond Myklebust if (!new_ns) 2061cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2062cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2063cf8d2c11STrond Myklebust new_ns->root = NULL; 2064cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2065cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2066cf8d2c11STrond Myklebust new_ns->event = 0; 2067cf8d2c11STrond Myklebust return new_ns; 2068cf8d2c11STrond Myklebust } 2069cf8d2c11STrond Myklebust 2070741a2951SJANAK DESAI /* 2071741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 2072741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 2073741a2951SJANAK DESAI */ 2074e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 20756b3286edSKirill Korotaev struct fs_struct *fs) 20761da177e4SLinus Torvalds { 20776b3286edSKirill Korotaev struct mnt_namespace *new_ns; 20787f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 20791da177e4SLinus Torvalds struct vfsmount *p, *q; 20801da177e4SLinus Torvalds 2081cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2082cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2083cf8d2c11STrond Myklebust return new_ns; 20841da177e4SLinus Torvalds 2085390c6843SRam Pai down_write(&namespace_sem); 20861da177e4SLinus Torvalds /* First pass: copy the tree topology */ 20876b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 20889676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 20891da177e4SLinus Torvalds if (!new_ns->root) { 2090390c6843SRam Pai up_write(&namespace_sem); 20911da177e4SLinus Torvalds kfree(new_ns); 20925cc4a034SJulia Lawall return ERR_PTR(-ENOMEM); 20931da177e4SLinus Torvalds } 209499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 20951da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 209699b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 20971da177e4SLinus Torvalds 20981da177e4SLinus Torvalds /* 20991da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 21001da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 21011da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 21021da177e4SLinus Torvalds */ 21036b3286edSKirill Korotaev p = mnt_ns->root; 21041da177e4SLinus Torvalds q = new_ns->root; 21051da177e4SLinus Torvalds while (p) { 21066b3286edSKirill Korotaev q->mnt_ns = new_ns; 21071da177e4SLinus Torvalds if (fs) { 21086ac08c39SJan Blunck if (p == fs->root.mnt) { 21091da177e4SLinus Torvalds rootmnt = p; 21106ac08c39SJan Blunck fs->root.mnt = mntget(q); 21111da177e4SLinus Torvalds } 21126ac08c39SJan Blunck if (p == fs->pwd.mnt) { 21131da177e4SLinus Torvalds pwdmnt = p; 21146ac08c39SJan Blunck fs->pwd.mnt = mntget(q); 21151da177e4SLinus Torvalds } 21161da177e4SLinus Torvalds } 21176b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 21181da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 21191da177e4SLinus Torvalds } 2120390c6843SRam Pai up_write(&namespace_sem); 21211da177e4SLinus Torvalds 21221da177e4SLinus Torvalds if (rootmnt) 21231da177e4SLinus Torvalds mntput(rootmnt); 21241da177e4SLinus Torvalds if (pwdmnt) 21251da177e4SLinus Torvalds mntput(pwdmnt); 21261da177e4SLinus Torvalds 2127741a2951SJANAK DESAI return new_ns; 2128741a2951SJANAK DESAI } 2129741a2951SJANAK DESAI 2130213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2131e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 2132741a2951SJANAK DESAI { 21336b3286edSKirill Korotaev struct mnt_namespace *new_ns; 2134741a2951SJANAK DESAI 2135e3222c4eSBadari Pulavarty BUG_ON(!ns); 21366b3286edSKirill Korotaev get_mnt_ns(ns); 2137741a2951SJANAK DESAI 2138741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 2139e3222c4eSBadari Pulavarty return ns; 2140741a2951SJANAK DESAI 2141e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 2142741a2951SJANAK DESAI 21436b3286edSKirill Korotaev put_mnt_ns(ns); 2144e3222c4eSBadari Pulavarty return new_ns; 21451da177e4SLinus Torvalds } 21461da177e4SLinus Torvalds 2147cf8d2c11STrond Myklebust /** 2148cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2149cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2150cf8d2c11STrond Myklebust */ 2151a2770d86SLinus Torvalds struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt) 2152cf8d2c11STrond Myklebust { 2153cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2154cf8d2c11STrond Myklebust 2155cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2156cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 2157cf8d2c11STrond Myklebust mnt->mnt_ns = new_ns; 2158cf8d2c11STrond Myklebust new_ns->root = mnt; 2159cf8d2c11STrond Myklebust list_add(&new_ns->list, &new_ns->root->mnt_list); 2160cf8d2c11STrond Myklebust } 2161cf8d2c11STrond Myklebust return new_ns; 2162cf8d2c11STrond Myklebust } 2163a2770d86SLinus Torvalds EXPORT_SYMBOL(create_mnt_ns); 2164cf8d2c11STrond Myklebust 2165bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2166bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 21671da177e4SLinus Torvalds { 2168eca6f534SVegard Nossum int ret; 2169eca6f534SVegard Nossum char *kernel_type; 2170eca6f534SVegard Nossum char *kernel_dir; 2171eca6f534SVegard Nossum char *kernel_dev; 21721da177e4SLinus Torvalds unsigned long data_page; 21731da177e4SLinus Torvalds 2174eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2175eca6f534SVegard Nossum if (ret < 0) 2176eca6f534SVegard Nossum goto out_type; 21771da177e4SLinus Torvalds 2178eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2179eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2180eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2181eca6f534SVegard Nossum goto out_dir; 2182eca6f534SVegard Nossum } 21831da177e4SLinus Torvalds 2184eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2185eca6f534SVegard Nossum if (ret < 0) 2186eca6f534SVegard Nossum goto out_dev; 21871da177e4SLinus Torvalds 2188eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2189eca6f534SVegard Nossum if (ret < 0) 2190eca6f534SVegard Nossum goto out_data; 21911da177e4SLinus Torvalds 2192eca6f534SVegard Nossum ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags, 2193eca6f534SVegard Nossum (void *) data_page); 2194eca6f534SVegard Nossum 21951da177e4SLinus Torvalds free_page(data_page); 2196eca6f534SVegard Nossum out_data: 2197eca6f534SVegard Nossum kfree(kernel_dev); 2198eca6f534SVegard Nossum out_dev: 2199eca6f534SVegard Nossum putname(kernel_dir); 2200eca6f534SVegard Nossum out_dir: 2201eca6f534SVegard Nossum kfree(kernel_type); 2202eca6f534SVegard Nossum out_type: 2203eca6f534SVegard Nossum return ret; 22041da177e4SLinus Torvalds } 22051da177e4SLinus Torvalds 22061da177e4SLinus Torvalds /* 22071da177e4SLinus Torvalds * pivot_root Semantics: 22081da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 22091da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 22101da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 22111da177e4SLinus Torvalds * 22121da177e4SLinus Torvalds * Restrictions: 22131da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 22141da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 22151da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 22161da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 22171da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 22181da177e4SLinus Torvalds * 22194a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 22204a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 22214a0d11faSNeil Brown * in this situation. 22224a0d11faSNeil Brown * 22231da177e4SLinus Torvalds * Notes: 22241da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 22251da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 22261da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 22271da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 22281da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 22291da177e4SLinus Torvalds * first. 22301da177e4SLinus Torvalds */ 22313480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 22323480b257SHeiko Carstens const char __user *, put_old) 22331da177e4SLinus Torvalds { 22341da177e4SLinus Torvalds struct vfsmount *tmp; 22352d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 22361da177e4SLinus Torvalds int error; 22371da177e4SLinus Torvalds 22381da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 22391da177e4SLinus Torvalds return -EPERM; 22401da177e4SLinus Torvalds 22412d8f3038SAl Viro error = user_path_dir(new_root, &new); 22421da177e4SLinus Torvalds if (error) 22431da177e4SLinus Torvalds goto out0; 22441da177e4SLinus Torvalds error = -EINVAL; 22452d8f3038SAl Viro if (!check_mnt(new.mnt)) 22461da177e4SLinus Torvalds goto out1; 22471da177e4SLinus Torvalds 22482d8f3038SAl Viro error = user_path_dir(put_old, &old); 22491da177e4SLinus Torvalds if (error) 22501da177e4SLinus Torvalds goto out1; 22511da177e4SLinus Torvalds 22522d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 22531da177e4SLinus Torvalds if (error) { 22542d8f3038SAl Viro path_put(&old); 22551da177e4SLinus Torvalds goto out1; 22561da177e4SLinus Torvalds } 22571da177e4SLinus Torvalds 2258f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 2259390c6843SRam Pai down_write(&namespace_sem); 22602d8f3038SAl Viro mutex_lock(&old.dentry->d_inode->i_mutex); 22611da177e4SLinus Torvalds error = -EINVAL; 22622d8f3038SAl Viro if (IS_MNT_SHARED(old.mnt) || 22632d8f3038SAl Viro IS_MNT_SHARED(new.mnt->mnt_parent) || 22648c3ee42eSAl Viro IS_MNT_SHARED(root.mnt->mnt_parent)) 226521444403SRam Pai goto out2; 22668c3ee42eSAl Viro if (!check_mnt(root.mnt)) 22671da177e4SLinus Torvalds goto out2; 22681da177e4SLinus Torvalds error = -ENOENT; 2269d83c49f3SAl Viro if (cant_mount(old.dentry)) 22701da177e4SLinus Torvalds goto out2; 2271f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 22721da177e4SLinus Torvalds goto out2; 2273f3da392eSAlexey Dobriyan if (d_unlinked(old.dentry)) 22741da177e4SLinus Torvalds goto out2; 22751da177e4SLinus Torvalds error = -EBUSY; 22762d8f3038SAl Viro if (new.mnt == root.mnt || 22772d8f3038SAl Viro old.mnt == root.mnt) 22781da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 22791da177e4SLinus Torvalds error = -EINVAL; 22808c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 22811da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 22828c3ee42eSAl Viro if (root.mnt->mnt_parent == root.mnt) 22830bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 22842d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 22851da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 22862d8f3038SAl Viro if (new.mnt->mnt_parent == new.mnt) 22870bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 22884ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 22892d8f3038SAl Viro tmp = old.mnt; 229099b7db7bSNick Piggin br_write_lock(vfsmount_lock); 22912d8f3038SAl Viro if (tmp != new.mnt) { 22921da177e4SLinus Torvalds for (;;) { 22931da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 22941da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 22952d8f3038SAl Viro if (tmp->mnt_parent == new.mnt) 22961da177e4SLinus Torvalds break; 22971da177e4SLinus Torvalds tmp = tmp->mnt_parent; 22981da177e4SLinus Torvalds } 22992d8f3038SAl Viro if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) 23001da177e4SLinus Torvalds goto out3; 23012d8f3038SAl Viro } else if (!is_subdir(old.dentry, new.dentry)) 23021da177e4SLinus Torvalds goto out3; 23032d8f3038SAl Viro detach_mnt(new.mnt, &parent_path); 23048c3ee42eSAl Viro detach_mnt(root.mnt, &root_parent); 23054ac91378SJan Blunck /* mount old root on put_old */ 23062d8f3038SAl Viro attach_mnt(root.mnt, &old); 23074ac91378SJan Blunck /* mount new_root on / */ 23082d8f3038SAl Viro attach_mnt(new.mnt, &root_parent); 23096b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 231099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 23112d8f3038SAl Viro chroot_fs_refs(&root, &new); 23121da177e4SLinus Torvalds error = 0; 23131a390689SAl Viro path_put(&root_parent); 23141a390689SAl Viro path_put(&parent_path); 23151da177e4SLinus Torvalds out2: 23162d8f3038SAl Viro mutex_unlock(&old.dentry->d_inode->i_mutex); 2317390c6843SRam Pai up_write(&namespace_sem); 23188c3ee42eSAl Viro path_put(&root); 23192d8f3038SAl Viro path_put(&old); 23201da177e4SLinus Torvalds out1: 23212d8f3038SAl Viro path_put(&new); 23221da177e4SLinus Torvalds out0: 23231da177e4SLinus Torvalds return error; 23241da177e4SLinus Torvalds out3: 232599b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 23261da177e4SLinus Torvalds goto out2; 23271da177e4SLinus Torvalds } 23281da177e4SLinus Torvalds 23291da177e4SLinus Torvalds static void __init init_mount_tree(void) 23301da177e4SLinus Torvalds { 23311da177e4SLinus Torvalds struct vfsmount *mnt; 23326b3286edSKirill Korotaev struct mnt_namespace *ns; 2333ac748a09SJan Blunck struct path root; 23341da177e4SLinus Torvalds 23351da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 23361da177e4SLinus Torvalds if (IS_ERR(mnt)) 23371da177e4SLinus Torvalds panic("Can't create rootfs"); 23383b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 23393b22edc5STrond Myklebust if (IS_ERR(ns)) 23401da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 23411da177e4SLinus Torvalds 23426b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 23436b3286edSKirill Korotaev get_mnt_ns(ns); 23441da177e4SLinus Torvalds 2345ac748a09SJan Blunck root.mnt = ns->root; 2346ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2347ac748a09SJan Blunck 2348ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2349ac748a09SJan Blunck set_fs_root(current->fs, &root); 23501da177e4SLinus Torvalds } 23511da177e4SLinus Torvalds 235274bf17cfSDenis Cheng void __init mnt_init(void) 23531da177e4SLinus Torvalds { 235413f14b4dSEric Dumazet unsigned u; 235515a67dd8SRandy Dunlap int err; 23561da177e4SLinus Torvalds 2357390c6843SRam Pai init_rwsem(&namespace_sem); 2358390c6843SRam Pai 23591da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 236020c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 23611da177e4SLinus Torvalds 2362b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 23631da177e4SLinus Torvalds 23641da177e4SLinus Torvalds if (!mount_hashtable) 23651da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 23661da177e4SLinus Torvalds 236713f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 23681da177e4SLinus Torvalds 236913f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 237013f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 23711da177e4SLinus Torvalds 237299b7db7bSNick Piggin br_lock_init(vfsmount_lock); 237399b7db7bSNick Piggin 237415a67dd8SRandy Dunlap err = sysfs_init(); 237515a67dd8SRandy Dunlap if (err) 237615a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 23778e24eea7SHarvey Harrison __func__, err); 237800d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 237900d26666SGreg Kroah-Hartman if (!fs_kobj) 23808e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 23811da177e4SLinus Torvalds init_rootfs(); 23821da177e4SLinus Torvalds init_mount_tree(); 23831da177e4SLinus Torvalds } 23841da177e4SLinus Torvalds 2385616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 23861da177e4SLinus Torvalds { 238770fbcdf4SRam Pai LIST_HEAD(umount_list); 2388616511d0STrond Myklebust 2389d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 2390616511d0STrond Myklebust return; 2391390c6843SRam Pai down_write(&namespace_sem); 239299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 2393d498b25aSAl Viro umount_tree(ns->root, 0, &umount_list); 239499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 2395390c6843SRam Pai up_write(&namespace_sem); 239670fbcdf4SRam Pai release_mounts(&umount_list); 23976b3286edSKirill Korotaev kfree(ns); 23981da177e4SLinus Torvalds } 2399cf8d2c11STrond Myklebust EXPORT_SYMBOL(put_mnt_ns); 2400