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> 141da177e4SLinus Torvalds #include <linux/smp_lock.h> 151da177e4SLinus Torvalds #include <linux/init.h> 1615a67dd8SRandy Dunlap #include <linux/kernel.h> 171da177e4SLinus Torvalds #include <linux/acct.h> 1816f7e0feSRandy Dunlap #include <linux/capability.h> 193d733633SDave Hansen #include <linux/cpumask.h> 201da177e4SLinus Torvalds #include <linux/module.h> 21f20a9eadSAndrew Morton #include <linux/sysfs.h> 221da177e4SLinus Torvalds #include <linux/seq_file.h> 236b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 241da177e4SLinus Torvalds #include <linux/namei.h> 25b43f3cbdSAlexey Dobriyan #include <linux/nsproxy.h> 261da177e4SLinus Torvalds #include <linux/security.h> 271da177e4SLinus Torvalds #include <linux/mount.h> 2807f3f05cSDavid Howells #include <linux/ramfs.h> 2913f14b4dSEric Dumazet #include <linux/log2.h> 3073cd49ecSMiklos Szeredi #include <linux/idr.h> 315ad4e53bSAl Viro #include <linux/fs_struct.h> 321da177e4SLinus Torvalds #include <asm/uaccess.h> 331da177e4SLinus Torvalds #include <asm/unistd.h> 3407b20889SRam Pai #include "pnode.h" 35948730b0SAdrian Bunk #include "internal.h" 361da177e4SLinus Torvalds 3713f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 3813f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 3913f14b4dSEric Dumazet 401da177e4SLinus Torvalds /* spinlock for vfsmount related operations, inplace of dcache_lock */ 411da177e4SLinus Torvalds __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 421da177e4SLinus Torvalds 435addc5ddSAl Viro static int event; 4473cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 45719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 46f21f6220SAl Viro static int mnt_id_start = 0; 47f21f6220SAl Viro static int mnt_group_start = 1; 485addc5ddSAl Viro 49fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 50e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 51390c6843SRam Pai static struct rw_semaphore namespace_sem; 521da177e4SLinus Torvalds 53f87fd4c2SMiklos Szeredi /* /sys/fs */ 5400d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 5500d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 56f87fd4c2SMiklos Szeredi 571da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 581da177e4SLinus Torvalds { 591da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 601da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 6113f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 6213f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 631da177e4SLinus Torvalds } 641da177e4SLinus Torvalds 653d733633SDave Hansen #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 663d733633SDave Hansen 6773cd49ecSMiklos Szeredi /* allocation is serialized by namespace_sem */ 6873cd49ecSMiklos Szeredi static int mnt_alloc_id(struct vfsmount *mnt) 6973cd49ecSMiklos Szeredi { 7073cd49ecSMiklos Szeredi int res; 7173cd49ecSMiklos Szeredi 7273cd49ecSMiklos Szeredi retry: 7373cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 7473cd49ecSMiklos Szeredi spin_lock(&vfsmount_lock); 75f21f6220SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 76f21f6220SAl Viro if (!res) 77f21f6220SAl Viro mnt_id_start = mnt->mnt_id + 1; 7873cd49ecSMiklos Szeredi spin_unlock(&vfsmount_lock); 7973cd49ecSMiklos Szeredi if (res == -EAGAIN) 8073cd49ecSMiklos Szeredi goto retry; 8173cd49ecSMiklos Szeredi 8273cd49ecSMiklos Szeredi return res; 8373cd49ecSMiklos Szeredi } 8473cd49ecSMiklos Szeredi 8573cd49ecSMiklos Szeredi static void mnt_free_id(struct vfsmount *mnt) 8673cd49ecSMiklos Szeredi { 87f21f6220SAl Viro int id = mnt->mnt_id; 8873cd49ecSMiklos Szeredi spin_lock(&vfsmount_lock); 89f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 90f21f6220SAl Viro if (mnt_id_start > id) 91f21f6220SAl Viro mnt_id_start = id; 9273cd49ecSMiklos Szeredi spin_unlock(&vfsmount_lock); 9373cd49ecSMiklos Szeredi } 9473cd49ecSMiklos Szeredi 95719f5d7fSMiklos Szeredi /* 96719f5d7fSMiklos Szeredi * Allocate a new peer group ID 97719f5d7fSMiklos Szeredi * 98719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 99719f5d7fSMiklos Szeredi */ 100719f5d7fSMiklos Szeredi static int mnt_alloc_group_id(struct vfsmount *mnt) 101719f5d7fSMiklos Szeredi { 102f21f6220SAl Viro int res; 103f21f6220SAl Viro 104719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 105719f5d7fSMiklos Szeredi return -ENOMEM; 106719f5d7fSMiklos Szeredi 107f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 108f21f6220SAl Viro mnt_group_start, 109f21f6220SAl Viro &mnt->mnt_group_id); 110f21f6220SAl Viro if (!res) 111f21f6220SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 112f21f6220SAl Viro 113f21f6220SAl Viro return res; 114719f5d7fSMiklos Szeredi } 115719f5d7fSMiklos Szeredi 116719f5d7fSMiklos Szeredi /* 117719f5d7fSMiklos Szeredi * Release a peer group ID 118719f5d7fSMiklos Szeredi */ 119719f5d7fSMiklos Szeredi void mnt_release_group_id(struct vfsmount *mnt) 120719f5d7fSMiklos Szeredi { 121f21f6220SAl Viro int id = mnt->mnt_group_id; 122f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 123f21f6220SAl Viro if (mnt_group_start > id) 124f21f6220SAl Viro mnt_group_start = id; 125719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; 126719f5d7fSMiklos Szeredi } 127719f5d7fSMiklos Szeredi 1281da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 1291da177e4SLinus Torvalds { 130c3762229SRobert P. J. Day struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 1311da177e4SLinus Torvalds if (mnt) { 13273cd49ecSMiklos Szeredi int err; 13373cd49ecSMiklos Szeredi 13473cd49ecSMiklos Szeredi err = mnt_alloc_id(mnt); 13588b38782SLi Zefan if (err) 13688b38782SLi Zefan goto out_free_cache; 13788b38782SLi Zefan 13888b38782SLi Zefan if (name) { 13988b38782SLi Zefan mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 14088b38782SLi Zefan if (!mnt->mnt_devname) 14188b38782SLi Zefan goto out_free_id; 14273cd49ecSMiklos Szeredi } 14373cd49ecSMiklos Szeredi 1441da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 1451da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 1461da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 1471da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 1481da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 14955e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 15003e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 151a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 152a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 153d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 154d3ef3d73Snpiggin@suse.de mnt->mnt_writers = alloc_percpu(int); 155d3ef3d73Snpiggin@suse.de if (!mnt->mnt_writers) 156d3ef3d73Snpiggin@suse.de goto out_free_devname; 157d3ef3d73Snpiggin@suse.de #else 158d3ef3d73Snpiggin@suse.de mnt->mnt_writers = 0; 159d3ef3d73Snpiggin@suse.de #endif 1601da177e4SLinus Torvalds } 1611da177e4SLinus Torvalds return mnt; 16288b38782SLi Zefan 163d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 164d3ef3d73Snpiggin@suse.de out_free_devname: 165d3ef3d73Snpiggin@suse.de kfree(mnt->mnt_devname); 166d3ef3d73Snpiggin@suse.de #endif 16788b38782SLi Zefan out_free_id: 16888b38782SLi Zefan mnt_free_id(mnt); 16988b38782SLi Zefan out_free_cache: 17088b38782SLi Zefan kmem_cache_free(mnt_cache, mnt); 17188b38782SLi Zefan return NULL; 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds 1748366025eSDave Hansen /* 1758366025eSDave Hansen * Most r/o checks on a fs are for operations that take 1768366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 1778366025eSDave Hansen * We must keep track of when those operations start 1788366025eSDave Hansen * (for permission checks) and when they end, so that 1798366025eSDave Hansen * we can determine when writes are able to occur to 1808366025eSDave Hansen * a filesystem. 1818366025eSDave Hansen */ 1823d733633SDave Hansen /* 1833d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 1843d733633SDave Hansen * @mnt: the mount to check for its write status 1853d733633SDave Hansen * 1863d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 1873d733633SDave Hansen * It does not guarantee that the filesystem will stay 1883d733633SDave Hansen * r/w, just that it is right *now*. This can not and 1893d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 1903d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 1913d733633SDave Hansen * r/w. 1923d733633SDave Hansen */ 1933d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 1943d733633SDave Hansen { 1952e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 1962e4b7fcdSDave Hansen return 1; 1972e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 1982e4b7fcdSDave Hansen return 1; 1992e4b7fcdSDave Hansen return 0; 2003d733633SDave Hansen } 2013d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2023d733633SDave Hansen 203d3ef3d73Snpiggin@suse.de static inline void inc_mnt_writers(struct vfsmount *mnt) 2043d733633SDave Hansen { 205d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 206d3ef3d73Snpiggin@suse.de (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++; 207d3ef3d73Snpiggin@suse.de #else 208d3ef3d73Snpiggin@suse.de mnt->mnt_writers++; 209d3ef3d73Snpiggin@suse.de #endif 2103d733633SDave Hansen } 2113d733633SDave Hansen 212d3ef3d73Snpiggin@suse.de static inline void dec_mnt_writers(struct vfsmount *mnt) 2133d733633SDave Hansen { 214d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 215d3ef3d73Snpiggin@suse.de (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--; 216d3ef3d73Snpiggin@suse.de #else 217d3ef3d73Snpiggin@suse.de mnt->mnt_writers--; 218d3ef3d73Snpiggin@suse.de #endif 219d3ef3d73Snpiggin@suse.de } 220d3ef3d73Snpiggin@suse.de 221d3ef3d73Snpiggin@suse.de static unsigned int count_mnt_writers(struct vfsmount *mnt) 222d3ef3d73Snpiggin@suse.de { 223d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 224d3ef3d73Snpiggin@suse.de unsigned int count = 0; 2253d733633SDave Hansen int cpu; 2263d733633SDave Hansen 2273d733633SDave Hansen for_each_possible_cpu(cpu) { 228d3ef3d73Snpiggin@suse.de count += *per_cpu_ptr(mnt->mnt_writers, cpu); 2293d733633SDave Hansen } 2303d733633SDave Hansen 231d3ef3d73Snpiggin@suse.de return count; 232d3ef3d73Snpiggin@suse.de #else 233d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 234d3ef3d73Snpiggin@suse.de #endif 2353d733633SDave Hansen } 2363d733633SDave Hansen 2373d733633SDave Hansen /* 2383d733633SDave Hansen * Most r/o checks on a fs are for operations that take 2393d733633SDave Hansen * discrete amounts of time, like a write() or unlink(). 2403d733633SDave Hansen * We must keep track of when those operations start 2413d733633SDave Hansen * (for permission checks) and when they end, so that 2423d733633SDave Hansen * we can determine when writes are able to occur to 2433d733633SDave Hansen * a filesystem. 2443d733633SDave Hansen */ 2458366025eSDave Hansen /** 2468366025eSDave Hansen * mnt_want_write - get write access to a mount 2478366025eSDave Hansen * @mnt: the mount on which to take a write 2488366025eSDave Hansen * 2498366025eSDave Hansen * This tells the low-level filesystem that a write is 2508366025eSDave Hansen * about to be performed to it, and makes sure that 2518366025eSDave Hansen * writes are allowed before returning success. When 2528366025eSDave Hansen * the write operation is finished, mnt_drop_write() 2538366025eSDave Hansen * must be called. This is effectively a refcount. 2548366025eSDave Hansen */ 2558366025eSDave Hansen int mnt_want_write(struct vfsmount *mnt) 2568366025eSDave Hansen { 2573d733633SDave Hansen int ret = 0; 2583d733633SDave Hansen 259d3ef3d73Snpiggin@suse.de preempt_disable(); 260d3ef3d73Snpiggin@suse.de inc_mnt_writers(mnt); 261d3ef3d73Snpiggin@suse.de /* 262d3ef3d73Snpiggin@suse.de * The store to inc_mnt_writers must be visible before we pass 263d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 264d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 265d3ef3d73Snpiggin@suse.de */ 266d3ef3d73Snpiggin@suse.de smp_mb(); 267d3ef3d73Snpiggin@suse.de while (mnt->mnt_flags & MNT_WRITE_HOLD) 268d3ef3d73Snpiggin@suse.de cpu_relax(); 269d3ef3d73Snpiggin@suse.de /* 270d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 271d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 272d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 273d3ef3d73Snpiggin@suse.de */ 274d3ef3d73Snpiggin@suse.de smp_rmb(); 2753d733633SDave Hansen if (__mnt_is_readonly(mnt)) { 276d3ef3d73Snpiggin@suse.de dec_mnt_writers(mnt); 2773d733633SDave Hansen ret = -EROFS; 2783d733633SDave Hansen goto out; 2793d733633SDave Hansen } 2803d733633SDave Hansen out: 281d3ef3d73Snpiggin@suse.de preempt_enable(); 2823d733633SDave Hansen return ret; 2838366025eSDave Hansen } 2848366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 2858366025eSDave Hansen 2868366025eSDave Hansen /** 28796029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 28896029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 28996029c4eSnpiggin@suse.de * 29096029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 29196029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 29296029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 29396029c4eSnpiggin@suse.de * on it. This allows some optimisation. 29496029c4eSnpiggin@suse.de * 29596029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 29696029c4eSnpiggin@suse.de * drop the reference. 29796029c4eSnpiggin@suse.de */ 29896029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 29996029c4eSnpiggin@suse.de { 30096029c4eSnpiggin@suse.de /* superblock may be r/o */ 30196029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 30296029c4eSnpiggin@suse.de return -EROFS; 30396029c4eSnpiggin@suse.de preempt_disable(); 30496029c4eSnpiggin@suse.de inc_mnt_writers(mnt); 30596029c4eSnpiggin@suse.de preempt_enable(); 30696029c4eSnpiggin@suse.de return 0; 30796029c4eSnpiggin@suse.de } 30896029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 30996029c4eSnpiggin@suse.de 31096029c4eSnpiggin@suse.de /** 31196029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 31296029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 31396029c4eSnpiggin@suse.de * 31496029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 31596029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 31696029c4eSnpiggin@suse.de */ 31796029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 31896029c4eSnpiggin@suse.de { 3192d8dd38aSOGAWA Hirofumi struct inode *inode = file->f_dentry->d_inode; 3202d8dd38aSOGAWA Hirofumi if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) 32196029c4eSnpiggin@suse.de return mnt_want_write(file->f_path.mnt); 32296029c4eSnpiggin@suse.de else 32396029c4eSnpiggin@suse.de return mnt_clone_write(file->f_path.mnt); 32496029c4eSnpiggin@suse.de } 32596029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 32696029c4eSnpiggin@suse.de 32796029c4eSnpiggin@suse.de /** 3288366025eSDave Hansen * mnt_drop_write - give up write access to a mount 3298366025eSDave Hansen * @mnt: the mount on which to give up write access 3308366025eSDave Hansen * 3318366025eSDave Hansen * Tells the low-level filesystem that we are done 3328366025eSDave Hansen * performing writes to it. Must be matched with 3338366025eSDave Hansen * mnt_want_write() call above. 3348366025eSDave Hansen */ 3358366025eSDave Hansen void mnt_drop_write(struct vfsmount *mnt) 3368366025eSDave Hansen { 337d3ef3d73Snpiggin@suse.de preempt_disable(); 338d3ef3d73Snpiggin@suse.de dec_mnt_writers(mnt); 339d3ef3d73Snpiggin@suse.de preempt_enable(); 3408366025eSDave Hansen } 3418366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 3428366025eSDave Hansen 3432e4b7fcdSDave Hansen static int mnt_make_readonly(struct vfsmount *mnt) 3448366025eSDave Hansen { 3453d733633SDave Hansen int ret = 0; 3463d733633SDave Hansen 3472e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 348d3ef3d73Snpiggin@suse.de mnt->mnt_flags |= MNT_WRITE_HOLD; 349d3ef3d73Snpiggin@suse.de /* 350d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 351d3ef3d73Snpiggin@suse.de * should be visible before we do. 352d3ef3d73Snpiggin@suse.de */ 353d3ef3d73Snpiggin@suse.de smp_mb(); 354d3ef3d73Snpiggin@suse.de 355d3ef3d73Snpiggin@suse.de /* 356d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 357d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 358d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 359d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 360d3ef3d73Snpiggin@suse.de * 361d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 362d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 363d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 364d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 365d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 366d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 367d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 368d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 369d3ef3d73Snpiggin@suse.de * we're counting up here. 370d3ef3d73Snpiggin@suse.de */ 371d3ef3d73Snpiggin@suse.de if (count_mnt_writers(mnt) > 0) 372d3ef3d73Snpiggin@suse.de ret = -EBUSY; 373d3ef3d73Snpiggin@suse.de else 3742e4b7fcdSDave Hansen mnt->mnt_flags |= MNT_READONLY; 375d3ef3d73Snpiggin@suse.de /* 376d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 377d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 378d3ef3d73Snpiggin@suse.de */ 379d3ef3d73Snpiggin@suse.de smp_wmb(); 380d3ef3d73Snpiggin@suse.de mnt->mnt_flags &= ~MNT_WRITE_HOLD; 3812e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3823d733633SDave Hansen return ret; 3833d733633SDave Hansen } 3848366025eSDave Hansen 3852e4b7fcdSDave Hansen static void __mnt_unmake_readonly(struct vfsmount *mnt) 3862e4b7fcdSDave Hansen { 3872e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 3882e4b7fcdSDave Hansen mnt->mnt_flags &= ~MNT_READONLY; 3892e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3902e4b7fcdSDave Hansen } 3912e4b7fcdSDave Hansen 392a3ec947cSSukadev Bhattiprolu void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 393454e2398SDavid Howells { 394454e2398SDavid Howells mnt->mnt_sb = sb; 395454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 396454e2398SDavid Howells } 397454e2398SDavid Howells 398454e2398SDavid Howells EXPORT_SYMBOL(simple_set_mnt); 399454e2398SDavid Howells 4001da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 4011da177e4SLinus Torvalds { 4021da177e4SLinus Torvalds kfree(mnt->mnt_devname); 40373cd49ecSMiklos Szeredi mnt_free_id(mnt); 404d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 405d3ef3d73Snpiggin@suse.de free_percpu(mnt->mnt_writers); 406d3ef3d73Snpiggin@suse.de #endif 4071da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 4081da177e4SLinus Torvalds } 4091da177e4SLinus Torvalds 4101da177e4SLinus Torvalds /* 411a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 412a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 4131da177e4SLinus Torvalds */ 414a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 415a05964f3SRam Pai int dir) 4161da177e4SLinus Torvalds { 4171da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 4181da177e4SLinus Torvalds struct list_head *tmp = head; 4191da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 4201da177e4SLinus Torvalds 4211da177e4SLinus Torvalds for (;;) { 422a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 4231da177e4SLinus Torvalds p = NULL; 4241da177e4SLinus Torvalds if (tmp == head) 4251da177e4SLinus Torvalds break; 4261da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 4271da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 428a05964f3SRam Pai found = p; 4291da177e4SLinus Torvalds break; 4301da177e4SLinus Torvalds } 4311da177e4SLinus Torvalds } 4321da177e4SLinus Torvalds return found; 4331da177e4SLinus Torvalds } 4341da177e4SLinus Torvalds 435a05964f3SRam Pai /* 436a05964f3SRam Pai * lookup_mnt increments the ref count before returning 437a05964f3SRam Pai * the vfsmount struct. 438a05964f3SRam Pai */ 4391c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 440a05964f3SRam Pai { 441a05964f3SRam Pai struct vfsmount *child_mnt; 442a05964f3SRam Pai spin_lock(&vfsmount_lock); 4431c755af4SAl Viro if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1))) 444a05964f3SRam Pai mntget(child_mnt); 445a05964f3SRam Pai spin_unlock(&vfsmount_lock); 446a05964f3SRam Pai return child_mnt; 447a05964f3SRam Pai } 448a05964f3SRam Pai 4491da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 4501da177e4SLinus Torvalds { 4516b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 4521da177e4SLinus Torvalds } 4531da177e4SLinus Torvalds 4546b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 4555addc5ddSAl Viro { 4565addc5ddSAl Viro if (ns) { 4575addc5ddSAl Viro ns->event = ++event; 4585addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4595addc5ddSAl Viro } 4605addc5ddSAl Viro } 4615addc5ddSAl Viro 4626b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 4635addc5ddSAl Viro { 4645addc5ddSAl Viro if (ns && ns->event != event) { 4655addc5ddSAl Viro ns->event = event; 4665addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4675addc5ddSAl Viro } 4685addc5ddSAl Viro } 4695addc5ddSAl Viro 4701a390689SAl Viro static void detach_mnt(struct vfsmount *mnt, struct path *old_path) 4711da177e4SLinus Torvalds { 4721a390689SAl Viro old_path->dentry = mnt->mnt_mountpoint; 4731a390689SAl Viro old_path->mnt = mnt->mnt_parent; 4741da177e4SLinus Torvalds mnt->mnt_parent = mnt; 4751da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 4761da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 4771da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 4781a390689SAl Viro old_path->dentry->d_mounted--; 4791da177e4SLinus Torvalds } 4801da177e4SLinus Torvalds 481b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 482b90fa9aeSRam Pai struct vfsmount *child_mnt) 483b90fa9aeSRam Pai { 484b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 485b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 486b90fa9aeSRam Pai dentry->d_mounted++; 487b90fa9aeSRam Pai } 488b90fa9aeSRam Pai 4891a390689SAl Viro static void attach_mnt(struct vfsmount *mnt, struct path *path) 4901da177e4SLinus Torvalds { 4911a390689SAl Viro mnt_set_mountpoint(path->mnt, path->dentry, mnt); 492b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 4931a390689SAl Viro hash(path->mnt, path->dentry)); 4941a390689SAl Viro list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); 495b90fa9aeSRam Pai } 496b90fa9aeSRam Pai 497b90fa9aeSRam Pai /* 498b90fa9aeSRam Pai * the caller must hold vfsmount_lock 499b90fa9aeSRam Pai */ 500b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 501b90fa9aeSRam Pai { 502b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 503b90fa9aeSRam Pai struct vfsmount *m; 504b90fa9aeSRam Pai LIST_HEAD(head); 5056b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 506b90fa9aeSRam Pai 507b90fa9aeSRam Pai BUG_ON(parent == mnt); 508b90fa9aeSRam Pai 509b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 510b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 5116b3286edSKirill Korotaev m->mnt_ns = n; 512b90fa9aeSRam Pai list_splice(&head, n->list.prev); 513b90fa9aeSRam Pai 514b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 515b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 516b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 5176b3286edSKirill Korotaev touch_mnt_namespace(n); 5181da177e4SLinus Torvalds } 5191da177e4SLinus Torvalds 5201da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 5211da177e4SLinus Torvalds { 5221da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 5231da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 5241da177e4SLinus Torvalds while (1) { 5251da177e4SLinus Torvalds if (p == root) 5261da177e4SLinus Torvalds return NULL; 5271da177e4SLinus Torvalds next = p->mnt_child.next; 5281da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 5291da177e4SLinus Torvalds break; 5301da177e4SLinus Torvalds p = p->mnt_parent; 5311da177e4SLinus Torvalds } 5321da177e4SLinus Torvalds } 5331da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 5341da177e4SLinus Torvalds } 5351da177e4SLinus Torvalds 5369676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 5379676f0c6SRam Pai { 5389676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 5399676f0c6SRam Pai while (prev != &p->mnt_mounts) { 5409676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 5419676f0c6SRam Pai prev = p->mnt_mounts.prev; 5429676f0c6SRam Pai } 5439676f0c6SRam Pai return p; 5449676f0c6SRam Pai } 5459676f0c6SRam Pai 54636341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 54736341f64SRam Pai int flag) 5481da177e4SLinus Torvalds { 5491da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 5501da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 5511da177e4SLinus Torvalds 5521da177e4SLinus Torvalds if (mnt) { 553719f5d7fSMiklos Szeredi if (flag & (CL_SLAVE | CL_PRIVATE)) 554719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; /* not a peer of original */ 555719f5d7fSMiklos Szeredi else 556719f5d7fSMiklos Szeredi mnt->mnt_group_id = old->mnt_group_id; 557719f5d7fSMiklos Szeredi 558719f5d7fSMiklos Szeredi if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 559719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(mnt); 560719f5d7fSMiklos Szeredi if (err) 561719f5d7fSMiklos Szeredi goto out_free; 562719f5d7fSMiklos Szeredi } 563719f5d7fSMiklos Szeredi 5641da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 5651da177e4SLinus Torvalds atomic_inc(&sb->s_active); 5661da177e4SLinus Torvalds mnt->mnt_sb = sb; 5671da177e4SLinus Torvalds mnt->mnt_root = dget(root); 5681da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 5691da177e4SLinus Torvalds mnt->mnt_parent = mnt; 570b90fa9aeSRam Pai 5715afe0022SRam Pai if (flag & CL_SLAVE) { 5725afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 5735afe0022SRam Pai mnt->mnt_master = old; 5745afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 5758aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 576796a6b52SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 577b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 5785afe0022SRam Pai if (IS_MNT_SLAVE(old)) 5795afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 5805afe0022SRam Pai mnt->mnt_master = old->mnt_master; 5815afe0022SRam Pai } 582b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 583b90fa9aeSRam Pai set_mnt_shared(mnt); 5841da177e4SLinus Torvalds 5851da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 5861da177e4SLinus Torvalds * as the original if that was on one */ 58736341f64SRam Pai if (flag & CL_EXPIRE) { 58855e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 58955e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 5901da177e4SLinus Torvalds } 59136341f64SRam Pai } 5921da177e4SLinus Torvalds return mnt; 593719f5d7fSMiklos Szeredi 594719f5d7fSMiklos Szeredi out_free: 595719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 596719f5d7fSMiklos Szeredi return NULL; 5971da177e4SLinus Torvalds } 5981da177e4SLinus Torvalds 5997b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 6001da177e4SLinus Torvalds { 6011da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 6023d733633SDave Hansen /* 6033d733633SDave Hansen * This probably indicates that somebody messed 6043d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 6053d733633SDave Hansen * happens, the filesystem was probably unable 6063d733633SDave Hansen * to make r/w->r/o transitions. 6073d733633SDave Hansen */ 608d3ef3d73Snpiggin@suse.de /* 609d3ef3d73Snpiggin@suse.de * atomic_dec_and_lock() used to deal with ->mnt_count decrements 610d3ef3d73Snpiggin@suse.de * provides barriers, so count_mnt_writers() below is safe. AV 611d3ef3d73Snpiggin@suse.de */ 612d3ef3d73Snpiggin@suse.de WARN_ON(count_mnt_writers(mnt)); 6131da177e4SLinus Torvalds dput(mnt->mnt_root); 6141da177e4SLinus Torvalds free_vfsmnt(mnt); 6151da177e4SLinus Torvalds deactivate_super(sb); 6161da177e4SLinus Torvalds } 6171da177e4SLinus Torvalds 6187b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 6197b7b1aceSAl Viro { 6207b7b1aceSAl Viro repeat: 6217b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 6227b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 6237b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6247b7b1aceSAl Viro __mntput(mnt); 6257b7b1aceSAl Viro return; 6267b7b1aceSAl Viro } 6277b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 6287b7b1aceSAl Viro mnt->mnt_pinned = 0; 6297b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6307b7b1aceSAl Viro acct_auto_close_mnt(mnt); 6317b7b1aceSAl Viro security_sb_umount_close(mnt); 6327b7b1aceSAl Viro goto repeat; 6337b7b1aceSAl Viro } 6347b7b1aceSAl Viro } 6357b7b1aceSAl Viro 6367b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 6377b7b1aceSAl Viro 6387b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 6397b7b1aceSAl Viro { 6407b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6417b7b1aceSAl Viro mnt->mnt_pinned++; 6427b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6437b7b1aceSAl Viro } 6447b7b1aceSAl Viro 6457b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 6467b7b1aceSAl Viro 6477b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 6487b7b1aceSAl Viro { 6497b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6507b7b1aceSAl Viro if (mnt->mnt_pinned) { 6517b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 6527b7b1aceSAl Viro mnt->mnt_pinned--; 6537b7b1aceSAl Viro } 6547b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6557b7b1aceSAl Viro } 6567b7b1aceSAl Viro 6577b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 6581da177e4SLinus Torvalds 659b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 660b3b304a2SMiklos Szeredi { 661b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 662b3b304a2SMiklos Szeredi } 663b3b304a2SMiklos Szeredi 664b3b304a2SMiklos Szeredi /* 665b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 666b3b304a2SMiklos Szeredi * implement more complex mount option showing. 667b3b304a2SMiklos Szeredi * 668b3b304a2SMiklos Szeredi * See also save_mount_options(). 669b3b304a2SMiklos Szeredi */ 670b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 671b3b304a2SMiklos Szeredi { 6722a32cebdSAl Viro const char *options; 6732a32cebdSAl Viro 6742a32cebdSAl Viro rcu_read_lock(); 6752a32cebdSAl Viro options = rcu_dereference(mnt->mnt_sb->s_options); 676b3b304a2SMiklos Szeredi 677b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 678b3b304a2SMiklos Szeredi seq_putc(m, ','); 679b3b304a2SMiklos Szeredi mangle(m, options); 680b3b304a2SMiklos Szeredi } 6812a32cebdSAl Viro rcu_read_unlock(); 682b3b304a2SMiklos Szeredi 683b3b304a2SMiklos Szeredi return 0; 684b3b304a2SMiklos Szeredi } 685b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 686b3b304a2SMiklos Szeredi 687b3b304a2SMiklos Szeredi /* 688b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 689b3b304a2SMiklos Szeredi * called from the fill_super() callback. 690b3b304a2SMiklos Szeredi * 691b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 692b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 693b3b304a2SMiklos Szeredi * remount fails. 694b3b304a2SMiklos Szeredi * 695b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 696b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 697b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 698b3b304a2SMiklos Szeredi * any more. 699b3b304a2SMiklos Szeredi */ 700b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 701b3b304a2SMiklos Szeredi { 7022a32cebdSAl Viro BUG_ON(sb->s_options); 7032a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 704b3b304a2SMiklos Szeredi } 705b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 706b3b304a2SMiklos Szeredi 7072a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 7082a32cebdSAl Viro { 7092a32cebdSAl Viro char *old = sb->s_options; 7102a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 7112a32cebdSAl Viro if (old) { 7122a32cebdSAl Viro synchronize_rcu(); 7132a32cebdSAl Viro kfree(old); 7142a32cebdSAl Viro } 7152a32cebdSAl Viro } 7162a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 7172a32cebdSAl Viro 718a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 7191da177e4SLinus Torvalds /* iterator */ 7201da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 7211da177e4SLinus Torvalds { 722a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 7231da177e4SLinus Torvalds 724390c6843SRam Pai down_read(&namespace_sem); 725a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 7261da177e4SLinus Torvalds } 7271da177e4SLinus Torvalds 7281da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 7291da177e4SLinus Torvalds { 730a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 731b0765fb8SPavel Emelianov 732a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 7331da177e4SLinus Torvalds } 7341da177e4SLinus Torvalds 7351da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 7361da177e4SLinus Torvalds { 737390c6843SRam Pai up_read(&namespace_sem); 7381da177e4SLinus Torvalds } 7391da177e4SLinus Torvalds 7409f5596afSAl Viro int mnt_had_events(struct proc_mounts *p) 7419f5596afSAl Viro { 7429f5596afSAl Viro struct mnt_namespace *ns = p->ns; 7439f5596afSAl Viro int res = 0; 7449f5596afSAl Viro 7459f5596afSAl Viro spin_lock(&vfsmount_lock); 7469f5596afSAl Viro if (p->event != ns->event) { 7479f5596afSAl Viro p->event = ns->event; 7489f5596afSAl Viro res = 1; 7499f5596afSAl Viro } 7509f5596afSAl Viro spin_unlock(&vfsmount_lock); 7519f5596afSAl Viro 7529f5596afSAl Viro return res; 7539f5596afSAl Viro } 7549f5596afSAl Viro 7552d4d4864SRam Pai struct proc_fs_info { 7561da177e4SLinus Torvalds int flag; 7572d4d4864SRam Pai const char *str; 7582d4d4864SRam Pai }; 7592d4d4864SRam Pai 7602069f457SEric Paris static int show_sb_opts(struct seq_file *m, struct super_block *sb) 7612d4d4864SRam Pai { 7622d4d4864SRam Pai static const struct proc_fs_info fs_info[] = { 7631da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 7641da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 7651da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 7661da177e4SLinus Torvalds { 0, NULL } 7671da177e4SLinus Torvalds }; 7682d4d4864SRam Pai const struct proc_fs_info *fs_infop; 7692d4d4864SRam Pai 7702d4d4864SRam Pai for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 7712d4d4864SRam Pai if (sb->s_flags & fs_infop->flag) 7722d4d4864SRam Pai seq_puts(m, fs_infop->str); 7732d4d4864SRam Pai } 7742069f457SEric Paris 7752069f457SEric Paris return security_sb_show_options(m, sb); 7762d4d4864SRam Pai } 7772d4d4864SRam Pai 7782d4d4864SRam Pai static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) 7792d4d4864SRam Pai { 7802d4d4864SRam Pai static const struct proc_fs_info mnt_info[] = { 7811da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 7821da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 7831da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 784fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 785fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 78647ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 787d0adde57SMatthew Garrett { MNT_STRICTATIME, ",strictatime" }, 7881da177e4SLinus Torvalds { 0, NULL } 7891da177e4SLinus Torvalds }; 7902d4d4864SRam Pai const struct proc_fs_info *fs_infop; 7912d4d4864SRam Pai 7922d4d4864SRam Pai for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 7932d4d4864SRam Pai if (mnt->mnt_flags & fs_infop->flag) 7942d4d4864SRam Pai seq_puts(m, fs_infop->str); 7952d4d4864SRam Pai } 7962d4d4864SRam Pai } 7972d4d4864SRam Pai 7982d4d4864SRam Pai static void show_type(struct seq_file *m, struct super_block *sb) 7992d4d4864SRam Pai { 8002d4d4864SRam Pai mangle(m, sb->s_type->name); 8012d4d4864SRam Pai if (sb->s_subtype && sb->s_subtype[0]) { 8022d4d4864SRam Pai seq_putc(m, '.'); 8032d4d4864SRam Pai mangle(m, sb->s_subtype); 8042d4d4864SRam Pai } 8052d4d4864SRam Pai } 8062d4d4864SRam Pai 8072d4d4864SRam Pai static int show_vfsmnt(struct seq_file *m, void *v) 8082d4d4864SRam Pai { 8092d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 8102d4d4864SRam Pai int err = 0; 811c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 8121da177e4SLinus Torvalds 8131da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 8141da177e4SLinus Torvalds seq_putc(m, ' '); 815c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 8161da177e4SLinus Torvalds seq_putc(m, ' '); 8172d4d4864SRam Pai show_type(m, mnt->mnt_sb); 8182e4b7fcdSDave Hansen seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 8192069f457SEric Paris err = show_sb_opts(m, mnt->mnt_sb); 8202069f457SEric Paris if (err) 8212069f457SEric Paris goto out; 8222d4d4864SRam Pai show_mnt_opts(m, mnt); 8231da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 8241da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 8251da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 8262069f457SEric Paris out: 8271da177e4SLinus Torvalds return err; 8281da177e4SLinus Torvalds } 8291da177e4SLinus Torvalds 830a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 8311da177e4SLinus Torvalds .start = m_start, 8321da177e4SLinus Torvalds .next = m_next, 8331da177e4SLinus Torvalds .stop = m_stop, 8341da177e4SLinus Torvalds .show = show_vfsmnt 8351da177e4SLinus Torvalds }; 8361da177e4SLinus Torvalds 8372d4d4864SRam Pai static int show_mountinfo(struct seq_file *m, void *v) 8382d4d4864SRam Pai { 8392d4d4864SRam Pai struct proc_mounts *p = m->private; 8402d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 8412d4d4864SRam Pai struct super_block *sb = mnt->mnt_sb; 8422d4d4864SRam Pai struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 8432d4d4864SRam Pai struct path root = p->root; 8442d4d4864SRam Pai int err = 0; 8452d4d4864SRam Pai 8462d4d4864SRam Pai seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, 8472d4d4864SRam Pai MAJOR(sb->s_dev), MINOR(sb->s_dev)); 8482d4d4864SRam Pai seq_dentry(m, mnt->mnt_root, " \t\n\\"); 8492d4d4864SRam Pai seq_putc(m, ' '); 8502d4d4864SRam Pai seq_path_root(m, &mnt_path, &root, " \t\n\\"); 8512d4d4864SRam Pai if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { 8522d4d4864SRam Pai /* 8532d4d4864SRam Pai * Mountpoint is outside root, discard that one. Ugly, 8542d4d4864SRam Pai * but less so than trying to do that in iterator in a 8552d4d4864SRam Pai * race-free way (due to renames). 8562d4d4864SRam Pai */ 8572d4d4864SRam Pai return SEQ_SKIP; 8582d4d4864SRam Pai } 8592d4d4864SRam Pai seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); 8602d4d4864SRam Pai show_mnt_opts(m, mnt); 8612d4d4864SRam Pai 8622d4d4864SRam Pai /* Tagged fields ("foo:X" or "bar") */ 8632d4d4864SRam Pai if (IS_MNT_SHARED(mnt)) 8642d4d4864SRam Pai seq_printf(m, " shared:%i", mnt->mnt_group_id); 86597e7e0f7SMiklos Szeredi if (IS_MNT_SLAVE(mnt)) { 86697e7e0f7SMiklos Szeredi int master = mnt->mnt_master->mnt_group_id; 86797e7e0f7SMiklos Szeredi int dom = get_dominating_id(mnt, &p->root); 86897e7e0f7SMiklos Szeredi seq_printf(m, " master:%i", master); 86997e7e0f7SMiklos Szeredi if (dom && dom != master) 87097e7e0f7SMiklos Szeredi seq_printf(m, " propagate_from:%i", dom); 87197e7e0f7SMiklos Szeredi } 8722d4d4864SRam Pai if (IS_MNT_UNBINDABLE(mnt)) 8732d4d4864SRam Pai seq_puts(m, " unbindable"); 8742d4d4864SRam Pai 8752d4d4864SRam Pai /* Filesystem specific data */ 8762d4d4864SRam Pai seq_puts(m, " - "); 8772d4d4864SRam Pai show_type(m, sb); 8782d4d4864SRam Pai seq_putc(m, ' '); 8792d4d4864SRam Pai mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 8802d4d4864SRam Pai seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); 8812069f457SEric Paris err = show_sb_opts(m, sb); 8822069f457SEric Paris if (err) 8832069f457SEric Paris goto out; 8842d4d4864SRam Pai if (sb->s_op->show_options) 8852d4d4864SRam Pai err = sb->s_op->show_options(m, mnt); 8862d4d4864SRam Pai seq_putc(m, '\n'); 8872069f457SEric Paris out: 8882d4d4864SRam Pai return err; 8892d4d4864SRam Pai } 8902d4d4864SRam Pai 8912d4d4864SRam Pai const struct seq_operations mountinfo_op = { 8922d4d4864SRam Pai .start = m_start, 8932d4d4864SRam Pai .next = m_next, 8942d4d4864SRam Pai .stop = m_stop, 8952d4d4864SRam Pai .show = show_mountinfo, 8962d4d4864SRam Pai }; 8972d4d4864SRam Pai 898b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 899b4629fe2SChuck Lever { 900b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 901c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 902b4629fe2SChuck Lever int err = 0; 903b4629fe2SChuck Lever 904b4629fe2SChuck Lever /* device */ 905b4629fe2SChuck Lever if (mnt->mnt_devname) { 906b4629fe2SChuck Lever seq_puts(m, "device "); 907b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 908b4629fe2SChuck Lever } else 909b4629fe2SChuck Lever seq_puts(m, "no device"); 910b4629fe2SChuck Lever 911b4629fe2SChuck Lever /* mount point */ 912b4629fe2SChuck Lever seq_puts(m, " mounted on "); 913c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 914b4629fe2SChuck Lever seq_putc(m, ' '); 915b4629fe2SChuck Lever 916b4629fe2SChuck Lever /* file system type */ 917b4629fe2SChuck Lever seq_puts(m, "with fstype "); 9182d4d4864SRam Pai show_type(m, mnt->mnt_sb); 919b4629fe2SChuck Lever 920b4629fe2SChuck Lever /* optional statistics */ 921b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 922b4629fe2SChuck Lever seq_putc(m, ' '); 923b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 924b4629fe2SChuck Lever } 925b4629fe2SChuck Lever 926b4629fe2SChuck Lever seq_putc(m, '\n'); 927b4629fe2SChuck Lever return err; 928b4629fe2SChuck Lever } 929b4629fe2SChuck Lever 930a1a2c409SMiklos Szeredi const struct seq_operations mountstats_op = { 931b4629fe2SChuck Lever .start = m_start, 932b4629fe2SChuck Lever .next = m_next, 933b4629fe2SChuck Lever .stop = m_stop, 934b4629fe2SChuck Lever .show = show_vfsstat, 935b4629fe2SChuck Lever }; 936a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 937b4629fe2SChuck Lever 9381da177e4SLinus Torvalds /** 9391da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 9401da177e4SLinus Torvalds * @mnt: root of mount tree 9411da177e4SLinus Torvalds * 9421da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 9431da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 9441da177e4SLinus Torvalds * busy. 9451da177e4SLinus Torvalds */ 9461da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 9471da177e4SLinus Torvalds { 94836341f64SRam Pai int actual_refs = 0; 94936341f64SRam Pai int minimum_refs = 0; 95036341f64SRam Pai struct vfsmount *p; 9511da177e4SLinus Torvalds 9521da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 95336341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 9541da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 9551da177e4SLinus Torvalds minimum_refs += 2; 9561da177e4SLinus Torvalds } 9571da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 9581da177e4SLinus Torvalds 9591da177e4SLinus Torvalds if (actual_refs > minimum_refs) 9601da177e4SLinus Torvalds return 0; 961e3474a8eSIan Kent 962e3474a8eSIan Kent return 1; 9631da177e4SLinus Torvalds } 9641da177e4SLinus Torvalds 9651da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 9661da177e4SLinus Torvalds 9671da177e4SLinus Torvalds /** 9681da177e4SLinus Torvalds * may_umount - check if a mount point is busy 9691da177e4SLinus Torvalds * @mnt: root of mount 9701da177e4SLinus Torvalds * 9711da177e4SLinus Torvalds * This is called to check if a mount point has any 9721da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 9731da177e4SLinus Torvalds * mount has sub mounts this will return busy 9741da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 9751da177e4SLinus Torvalds * 9761da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 9771da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 9781da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 9791da177e4SLinus Torvalds */ 9801da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 9811da177e4SLinus Torvalds { 982e3474a8eSIan Kent int ret = 1; 9838ad08d8aSAl Viro down_read(&namespace_sem); 984a05964f3SRam Pai spin_lock(&vfsmount_lock); 985a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 986e3474a8eSIan Kent ret = 0; 987a05964f3SRam Pai spin_unlock(&vfsmount_lock); 9888ad08d8aSAl Viro up_read(&namespace_sem); 989a05964f3SRam Pai return ret; 9901da177e4SLinus Torvalds } 9911da177e4SLinus Torvalds 9921da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 9931da177e4SLinus Torvalds 994b90fa9aeSRam Pai void release_mounts(struct list_head *head) 9951da177e4SLinus Torvalds { 99670fbcdf4SRam Pai struct vfsmount *mnt; 99770fbcdf4SRam Pai while (!list_empty(head)) { 998b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 99970fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 100070fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 100170fbcdf4SRam Pai struct dentry *dentry; 100270fbcdf4SRam Pai struct vfsmount *m; 100370fbcdf4SRam Pai spin_lock(&vfsmount_lock); 100470fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 100570fbcdf4SRam Pai m = mnt->mnt_parent; 100670fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 100770fbcdf4SRam Pai mnt->mnt_parent = mnt; 10087c4b93d8SAl Viro m->mnt_ghosts--; 10091da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 101070fbcdf4SRam Pai dput(dentry); 101170fbcdf4SRam Pai mntput(m); 10121da177e4SLinus Torvalds } 10131da177e4SLinus Torvalds mntput(mnt); 101470fbcdf4SRam Pai } 101570fbcdf4SRam Pai } 101670fbcdf4SRam Pai 1017a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 101870fbcdf4SRam Pai { 101970fbcdf4SRam Pai struct vfsmount *p; 102070fbcdf4SRam Pai 10211bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 10221bfba4e8SAkinobu Mita list_move(&p->mnt_hash, kill); 102370fbcdf4SRam Pai 1024a05964f3SRam Pai if (propagate) 1025a05964f3SRam Pai propagate_umount(kill); 1026a05964f3SRam Pai 102770fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 102870fbcdf4SRam Pai list_del_init(&p->mnt_expire); 102970fbcdf4SRam Pai list_del_init(&p->mnt_list); 10306b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 10316b3286edSKirill Korotaev p->mnt_ns = NULL; 103270fbcdf4SRam Pai list_del_init(&p->mnt_child); 10337c4b93d8SAl Viro if (p->mnt_parent != p) { 10347c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 1035f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 10367c4b93d8SAl Viro } 1037a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 10381da177e4SLinus Torvalds } 10391da177e4SLinus Torvalds } 10401da177e4SLinus Torvalds 1041c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 1042c35038beSAl Viro 10431da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 10441da177e4SLinus Torvalds { 10451da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 10461da177e4SLinus Torvalds int retval; 104770fbcdf4SRam Pai LIST_HEAD(umount_list); 10481da177e4SLinus Torvalds 10491da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 10501da177e4SLinus Torvalds if (retval) 10511da177e4SLinus Torvalds return retval; 10521da177e4SLinus Torvalds 10531da177e4SLinus Torvalds /* 10541da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 10551da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 10561da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 10571da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 10581da177e4SLinus Torvalds */ 10591da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 10606ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 10611da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 10621da177e4SLinus Torvalds return -EINVAL; 10631da177e4SLinus Torvalds 10641da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 10651da177e4SLinus Torvalds return -EBUSY; 10661da177e4SLinus Torvalds 10671da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 10681da177e4SLinus Torvalds return -EAGAIN; 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds 10711da177e4SLinus Torvalds /* 10721da177e4SLinus Torvalds * If we may have to abort operations to get out of this 10731da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 10741da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 10751da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 10761da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 10771da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 10781da177e4SLinus Torvalds * about for the moment. 10791da177e4SLinus Torvalds */ 10801da177e4SLinus Torvalds 108142faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 108242faad99SAl Viro sb->s_op->umount_begin(sb); 108342faad99SAl Viro } 10841da177e4SLinus Torvalds 10851da177e4SLinus Torvalds /* 10861da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 10871da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 10881da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 10891da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 10901da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 10911da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 10921da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 10931da177e4SLinus Torvalds */ 10946ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 10951da177e4SLinus Torvalds /* 10961da177e4SLinus Torvalds * Special case for "unmounting" root ... 10971da177e4SLinus Torvalds * we just try to remount it readonly. 10981da177e4SLinus Torvalds */ 10991da177e4SLinus Torvalds down_write(&sb->s_umount); 11004aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 11011da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 11021da177e4SLinus Torvalds up_write(&sb->s_umount); 11031da177e4SLinus Torvalds return retval; 11041da177e4SLinus Torvalds } 11051da177e4SLinus Torvalds 1106390c6843SRam Pai down_write(&namespace_sem); 11071da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 11085addc5ddSAl Viro event++; 11091da177e4SLinus Torvalds 1110c35038beSAl Viro if (!(flags & MNT_DETACH)) 1111c35038beSAl Viro shrink_submounts(mnt, &umount_list); 1112c35038beSAl Viro 11131da177e4SLinus Torvalds retval = -EBUSY; 1114a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 11151da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 1116a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 11171da177e4SLinus Torvalds retval = 0; 11181da177e4SLinus Torvalds } 11191da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 11201da177e4SLinus Torvalds if (retval) 11211da177e4SLinus Torvalds security_sb_umount_busy(mnt); 1122390c6843SRam Pai up_write(&namespace_sem); 112370fbcdf4SRam Pai release_mounts(&umount_list); 11241da177e4SLinus Torvalds return retval; 11251da177e4SLinus Torvalds } 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds /* 11281da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 11291da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 11301da177e4SLinus Torvalds * 11311da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 11321da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 11331da177e4SLinus Torvalds */ 11341da177e4SLinus Torvalds 1135bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 11361da177e4SLinus Torvalds { 11372d8f3038SAl Viro struct path path; 11381da177e4SLinus Torvalds int retval; 11391da177e4SLinus Torvalds 11402d8f3038SAl Viro retval = user_path(name, &path); 11411da177e4SLinus Torvalds if (retval) 11421da177e4SLinus Torvalds goto out; 11431da177e4SLinus Torvalds retval = -EINVAL; 11442d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 11451da177e4SLinus Torvalds goto dput_and_out; 11462d8f3038SAl Viro if (!check_mnt(path.mnt)) 11471da177e4SLinus Torvalds goto dput_and_out; 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds retval = -EPERM; 11501da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 11511da177e4SLinus Torvalds goto dput_and_out; 11521da177e4SLinus Torvalds 11532d8f3038SAl Viro retval = do_umount(path.mnt, flags); 11541da177e4SLinus Torvalds dput_and_out: 1155429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 11562d8f3038SAl Viro dput(path.dentry); 11572d8f3038SAl Viro mntput_no_expire(path.mnt); 11581da177e4SLinus Torvalds out: 11591da177e4SLinus Torvalds return retval; 11601da177e4SLinus Torvalds } 11611da177e4SLinus Torvalds 11621da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds /* 11651da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 11661da177e4SLinus Torvalds */ 1167bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 11681da177e4SLinus Torvalds { 11691da177e4SLinus Torvalds return sys_umount(name, 0); 11701da177e4SLinus Torvalds } 11711da177e4SLinus Torvalds 11721da177e4SLinus Torvalds #endif 11731da177e4SLinus Torvalds 11742d92ab3cSAl Viro static int mount_is_safe(struct path *path) 11751da177e4SLinus Torvalds { 11761da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 11771da177e4SLinus Torvalds return 0; 11781da177e4SLinus Torvalds return -EPERM; 11791da177e4SLinus Torvalds #ifdef notyet 11802d92ab3cSAl Viro if (S_ISLNK(path->dentry->d_inode->i_mode)) 11811da177e4SLinus Torvalds return -EPERM; 11822d92ab3cSAl Viro if (path->dentry->d_inode->i_mode & S_ISVTX) { 1183da9592edSDavid Howells if (current_uid() != path->dentry->d_inode->i_uid) 11841da177e4SLinus Torvalds return -EPERM; 11851da177e4SLinus Torvalds } 11862d92ab3cSAl Viro if (inode_permission(path->dentry->d_inode, MAY_WRITE)) 11871da177e4SLinus Torvalds return -EPERM; 11881da177e4SLinus Torvalds return 0; 11891da177e4SLinus Torvalds #endif 11901da177e4SLinus Torvalds } 11911da177e4SLinus Torvalds 1192b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 119336341f64SRam Pai int flag) 11941da177e4SLinus Torvalds { 11951da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 11961a390689SAl Viro struct path path; 11971da177e4SLinus Torvalds 11989676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 11999676f0c6SRam Pai return NULL; 12009676f0c6SRam Pai 120136341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 12021da177e4SLinus Torvalds if (!q) 12031da177e4SLinus Torvalds goto Enomem; 12041da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 12051da177e4SLinus Torvalds 12061da177e4SLinus Torvalds p = mnt; 1207fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 12087ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 12091da177e4SLinus Torvalds continue; 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 12129676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 12139676f0c6SRam Pai s = skip_mnt_tree(s); 12149676f0c6SRam Pai continue; 12159676f0c6SRam Pai } 12161da177e4SLinus Torvalds while (p != s->mnt_parent) { 12171da177e4SLinus Torvalds p = p->mnt_parent; 12181da177e4SLinus Torvalds q = q->mnt_parent; 12191da177e4SLinus Torvalds } 12201da177e4SLinus Torvalds p = s; 12211a390689SAl Viro path.mnt = q; 12221a390689SAl Viro path.dentry = p->mnt_mountpoint; 122336341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 12241da177e4SLinus Torvalds if (!q) 12251da177e4SLinus Torvalds goto Enomem; 12261da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 12271da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 12281a390689SAl Viro attach_mnt(q, &path); 12291da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 12301da177e4SLinus Torvalds } 12311da177e4SLinus Torvalds } 12321da177e4SLinus Torvalds return res; 12331da177e4SLinus Torvalds Enomem: 12341da177e4SLinus Torvalds if (res) { 123570fbcdf4SRam Pai LIST_HEAD(umount_list); 12361da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1237a05964f3SRam Pai umount_tree(res, 0, &umount_list); 12381da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 123970fbcdf4SRam Pai release_mounts(&umount_list); 12401da177e4SLinus Torvalds } 12411da177e4SLinus Torvalds return NULL; 12421da177e4SLinus Torvalds } 12431da177e4SLinus Torvalds 1244589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 12458aec0809SAl Viro { 12468aec0809SAl Viro struct vfsmount *tree; 12471a60a280SAl Viro down_write(&namespace_sem); 1248589ff870SAl Viro tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); 12491a60a280SAl Viro up_write(&namespace_sem); 12508aec0809SAl Viro return tree; 12518aec0809SAl Viro } 12528aec0809SAl Viro 12538aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 12548aec0809SAl Viro { 12558aec0809SAl Viro LIST_HEAD(umount_list); 12561a60a280SAl Viro down_write(&namespace_sem); 12578aec0809SAl Viro spin_lock(&vfsmount_lock); 12588aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 12598aec0809SAl Viro spin_unlock(&vfsmount_lock); 12601a60a280SAl Viro up_write(&namespace_sem); 12618aec0809SAl Viro release_mounts(&umount_list); 12628aec0809SAl Viro } 12638aec0809SAl Viro 12641f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 12651f707137SAl Viro struct vfsmount *root) 12661f707137SAl Viro { 12671f707137SAl Viro struct vfsmount *mnt; 12681f707137SAl Viro int res = f(root, arg); 12691f707137SAl Viro if (res) 12701f707137SAl Viro return res; 12711f707137SAl Viro list_for_each_entry(mnt, &root->mnt_list, mnt_list) { 12721f707137SAl Viro res = f(mnt, arg); 12731f707137SAl Viro if (res) 12741f707137SAl Viro return res; 12751f707137SAl Viro } 12761f707137SAl Viro return 0; 12771f707137SAl Viro } 12781f707137SAl Viro 1279719f5d7fSMiklos Szeredi static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) 1280719f5d7fSMiklos Szeredi { 1281719f5d7fSMiklos Szeredi struct vfsmount *p; 1282719f5d7fSMiklos Szeredi 1283719f5d7fSMiklos Szeredi for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1284719f5d7fSMiklos Szeredi if (p->mnt_group_id && !IS_MNT_SHARED(p)) 1285719f5d7fSMiklos Szeredi mnt_release_group_id(p); 1286719f5d7fSMiklos Szeredi } 1287719f5d7fSMiklos Szeredi } 1288719f5d7fSMiklos Szeredi 1289719f5d7fSMiklos Szeredi static int invent_group_ids(struct vfsmount *mnt, bool recurse) 1290719f5d7fSMiklos Szeredi { 1291719f5d7fSMiklos Szeredi struct vfsmount *p; 1292719f5d7fSMiklos Szeredi 1293719f5d7fSMiklos Szeredi for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1294719f5d7fSMiklos Szeredi if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 1295719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(p); 1296719f5d7fSMiklos Szeredi if (err) { 1297719f5d7fSMiklos Szeredi cleanup_group_ids(mnt, p); 1298719f5d7fSMiklos Szeredi return err; 1299719f5d7fSMiklos Szeredi } 1300719f5d7fSMiklos Szeredi } 1301719f5d7fSMiklos Szeredi } 1302719f5d7fSMiklos Szeredi 1303719f5d7fSMiklos Szeredi return 0; 1304719f5d7fSMiklos Szeredi } 1305719f5d7fSMiklos Szeredi 1306b90fa9aeSRam Pai /* 1307b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1308b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 130921444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 131021444403SRam Pai * store the parent mount and mountpoint dentry. 131121444403SRam Pai * (done when source_mnt is moved) 1312b90fa9aeSRam Pai * 1313b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1314b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 13159676f0c6SRam Pai * --------------------------------------------------------------------------- 1316b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 13179676f0c6SRam Pai * |************************************************************************** 13189676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 13199676f0c6SRam Pai * | dest | | | | | 13209676f0c6SRam Pai * | | | | | | | 13219676f0c6SRam Pai * | v | | | | | 13229676f0c6SRam Pai * |************************************************************************** 13239676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 13245afe0022SRam Pai * | | | | | | 13259676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 13269676f0c6SRam Pai * *************************************************************************** 1327b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1328b90fa9aeSRam Pai * destination mount. 1329b90fa9aeSRam Pai * 1330b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1331b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1332b90fa9aeSRam Pai * the peer group of the source mount. 1333b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1334b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1335b90fa9aeSRam Pai * mount. 13365afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 13375afe0022SRam Pai * of the destination mount and the cloned mount is made slave 13385afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 13395afe0022SRam Pai * is marked as 'shared and slave'. 13405afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 13415afe0022SRam Pai * source mount. 13425afe0022SRam Pai * 13439676f0c6SRam Pai * --------------------------------------------------------------------------- 134421444403SRam Pai * | MOVE MOUNT OPERATION | 13459676f0c6SRam Pai * |************************************************************************** 13469676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 13479676f0c6SRam Pai * | dest | | | | | 13489676f0c6SRam Pai * | | | | | | | 13499676f0c6SRam Pai * | v | | | | | 13509676f0c6SRam Pai * |************************************************************************** 13519676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 13525afe0022SRam Pai * | | | | | | 13539676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 13549676f0c6SRam Pai * *************************************************************************** 13555afe0022SRam Pai * 13565afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 13575afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 135821444403SRam Pai * (+*) the mount is moved to the destination. 13595afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 13605afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 13615afe0022SRam Pai * the mount is marked as 'shared and slave'. 13625afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1363b90fa9aeSRam Pai * 1364b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1365b90fa9aeSRam Pai * applied to each mount in the tree. 1366b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1367b90fa9aeSRam Pai * in allocations. 1368b90fa9aeSRam Pai */ 1369b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 13701a390689SAl Viro struct path *path, struct path *parent_path) 1371b90fa9aeSRam Pai { 1372b90fa9aeSRam Pai LIST_HEAD(tree_list); 13731a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 13741a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1375b90fa9aeSRam Pai struct vfsmount *child, *p; 1376719f5d7fSMiklos Szeredi int err; 1377b90fa9aeSRam Pai 1378719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) { 1379719f5d7fSMiklos Szeredi err = invent_group_ids(source_mnt, true); 1380719f5d7fSMiklos Szeredi if (err) 1381719f5d7fSMiklos Szeredi goto out; 1382719f5d7fSMiklos Szeredi } 1383719f5d7fSMiklos Szeredi err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1384719f5d7fSMiklos Szeredi if (err) 1385719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1386b90fa9aeSRam Pai 1387df1a1ad2SAl Viro spin_lock(&vfsmount_lock); 1388df1a1ad2SAl Viro 1389b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1390b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1391b90fa9aeSRam Pai set_mnt_shared(p); 1392b90fa9aeSRam Pai } 13931a390689SAl Viro if (parent_path) { 13941a390689SAl Viro detach_mnt(source_mnt, parent_path); 13951a390689SAl Viro attach_mnt(source_mnt, path); 1396e5d67f07SAl Viro touch_mnt_namespace(parent_path->mnt->mnt_ns); 139721444403SRam Pai } else { 1398b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1399b90fa9aeSRam Pai commit_tree(source_mnt); 140021444403SRam Pai } 1401b90fa9aeSRam Pai 1402b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1403b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1404b90fa9aeSRam Pai commit_tree(child); 1405b90fa9aeSRam Pai } 1406b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 1407b90fa9aeSRam Pai return 0; 1408719f5d7fSMiklos Szeredi 1409719f5d7fSMiklos Szeredi out_cleanup_ids: 1410719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) 1411719f5d7fSMiklos Szeredi cleanup_group_ids(source_mnt, NULL); 1412719f5d7fSMiklos Szeredi out: 1413719f5d7fSMiklos Szeredi return err; 1414b90fa9aeSRam Pai } 1415b90fa9aeSRam Pai 14168c3ee42eSAl Viro static int graft_tree(struct vfsmount *mnt, struct path *path) 14171da177e4SLinus Torvalds { 14181da177e4SLinus Torvalds int err; 14191da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 14201da177e4SLinus Torvalds return -EINVAL; 14211da177e4SLinus Torvalds 14228c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 14231da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 14241da177e4SLinus Torvalds return -ENOTDIR; 14251da177e4SLinus Torvalds 14261da177e4SLinus Torvalds err = -ENOENT; 14278c3ee42eSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 14288c3ee42eSAl Viro if (IS_DEADDIR(path->dentry->d_inode)) 14291da177e4SLinus Torvalds goto out_unlock; 14301da177e4SLinus Torvalds 14318c3ee42eSAl Viro err = security_sb_check_sb(mnt, path); 14321da177e4SLinus Torvalds if (err) 14331da177e4SLinus Torvalds goto out_unlock; 14341da177e4SLinus Torvalds 14351da177e4SLinus Torvalds err = -ENOENT; 1436f3da392eSAlexey Dobriyan if (!d_unlinked(path->dentry)) 14378c3ee42eSAl Viro err = attach_recursive_mnt(mnt, path, NULL); 14381da177e4SLinus Torvalds out_unlock: 14398c3ee42eSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 14401da177e4SLinus Torvalds if (!err) 14418c3ee42eSAl Viro security_sb_post_addmount(mnt, path); 14421da177e4SLinus Torvalds return err; 14431da177e4SLinus Torvalds } 14441da177e4SLinus Torvalds 14451da177e4SLinus Torvalds /* 144607b20889SRam Pai * recursively change the type of the mountpoint. 144707b20889SRam Pai */ 14480a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 144907b20889SRam Pai { 14502d92ab3cSAl Viro struct vfsmount *m, *mnt = path->mnt; 145107b20889SRam Pai int recurse = flag & MS_REC; 145207b20889SRam Pai int type = flag & ~MS_REC; 1453719f5d7fSMiklos Szeredi int err = 0; 145407b20889SRam Pai 1455ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1456ee6f9582SMiklos Szeredi return -EPERM; 1457ee6f9582SMiklos Szeredi 14582d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 145907b20889SRam Pai return -EINVAL; 146007b20889SRam Pai 146107b20889SRam Pai down_write(&namespace_sem); 1462719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1463719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1464719f5d7fSMiklos Szeredi if (err) 1465719f5d7fSMiklos Szeredi goto out_unlock; 1466719f5d7fSMiklos Szeredi } 1467719f5d7fSMiklos Szeredi 146807b20889SRam Pai spin_lock(&vfsmount_lock); 146907b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 147007b20889SRam Pai change_mnt_propagation(m, type); 147107b20889SRam Pai spin_unlock(&vfsmount_lock); 1472719f5d7fSMiklos Szeredi 1473719f5d7fSMiklos Szeredi out_unlock: 147407b20889SRam Pai up_write(&namespace_sem); 1475719f5d7fSMiklos Szeredi return err; 147607b20889SRam Pai } 147707b20889SRam Pai 147807b20889SRam Pai /* 14791da177e4SLinus Torvalds * do loopback mount. 14801da177e4SLinus Torvalds */ 14810a0d8a46SAl Viro static int do_loopback(struct path *path, char *old_name, 14822dafe1c4SEric Sandeen int recurse) 14831da177e4SLinus Torvalds { 14842d92ab3cSAl Viro struct path old_path; 14851da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 14862d92ab3cSAl Viro int err = mount_is_safe(path); 14871da177e4SLinus Torvalds if (err) 14881da177e4SLinus Torvalds return err; 14891da177e4SLinus Torvalds if (!old_name || !*old_name) 14901da177e4SLinus Torvalds return -EINVAL; 14912d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 14921da177e4SLinus Torvalds if (err) 14931da177e4SLinus Torvalds return err; 14941da177e4SLinus Torvalds 1495390c6843SRam Pai down_write(&namespace_sem); 14961da177e4SLinus Torvalds err = -EINVAL; 14972d92ab3cSAl Viro if (IS_MNT_UNBINDABLE(old_path.mnt)) 14989676f0c6SRam Pai goto out; 14999676f0c6SRam Pai 15002d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 1501ccd48bc7SAl Viro goto out; 1502ccd48bc7SAl Viro 15031da177e4SLinus Torvalds err = -ENOMEM; 15041da177e4SLinus Torvalds if (recurse) 15052d92ab3cSAl Viro mnt = copy_tree(old_path.mnt, old_path.dentry, 0); 15061da177e4SLinus Torvalds else 15072d92ab3cSAl Viro mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); 15081da177e4SLinus Torvalds 1509ccd48bc7SAl Viro if (!mnt) 1510ccd48bc7SAl Viro goto out; 1511ccd48bc7SAl Viro 15122d92ab3cSAl Viro err = graft_tree(mnt, path); 15131da177e4SLinus Torvalds if (err) { 151470fbcdf4SRam Pai LIST_HEAD(umount_list); 15151da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1516a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 15171da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 151870fbcdf4SRam Pai release_mounts(&umount_list); 15195b83d2c5SRam Pai } 15201da177e4SLinus Torvalds 1521ccd48bc7SAl Viro out: 1522390c6843SRam Pai up_write(&namespace_sem); 15232d92ab3cSAl Viro path_put(&old_path); 15241da177e4SLinus Torvalds return err; 15251da177e4SLinus Torvalds } 15261da177e4SLinus Torvalds 15272e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 15282e4b7fcdSDave Hansen { 15292e4b7fcdSDave Hansen int error = 0; 15302e4b7fcdSDave Hansen int readonly_request = 0; 15312e4b7fcdSDave Hansen 15322e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 15332e4b7fcdSDave Hansen readonly_request = 1; 15342e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 15352e4b7fcdSDave Hansen return 0; 15362e4b7fcdSDave Hansen 15372e4b7fcdSDave Hansen if (readonly_request) 15382e4b7fcdSDave Hansen error = mnt_make_readonly(mnt); 15392e4b7fcdSDave Hansen else 15402e4b7fcdSDave Hansen __mnt_unmake_readonly(mnt); 15412e4b7fcdSDave Hansen return error; 15422e4b7fcdSDave Hansen } 15432e4b7fcdSDave Hansen 15441da177e4SLinus Torvalds /* 15451da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 15461da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 15471da177e4SLinus Torvalds * on it - tough luck. 15481da177e4SLinus Torvalds */ 15490a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 15501da177e4SLinus Torvalds void *data) 15511da177e4SLinus Torvalds { 15521da177e4SLinus Torvalds int err; 15532d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 15541da177e4SLinus Torvalds 15551da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15561da177e4SLinus Torvalds return -EPERM; 15571da177e4SLinus Torvalds 15582d92ab3cSAl Viro if (!check_mnt(path->mnt)) 15591da177e4SLinus Torvalds return -EINVAL; 15601da177e4SLinus Torvalds 15612d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 15621da177e4SLinus Torvalds return -EINVAL; 15631da177e4SLinus Torvalds 15641da177e4SLinus Torvalds down_write(&sb->s_umount); 15652e4b7fcdSDave Hansen if (flags & MS_BIND) 15662d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 15674aa98cf7SAl Viro else 15681da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 15697b43a79fSAl Viro if (!err) { 15707b43a79fSAl Viro spin_lock(&vfsmount_lock); 1571495d6c9cSValerie Aurora mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK; 15722d92ab3cSAl Viro path->mnt->mnt_flags = mnt_flags; 15737b43a79fSAl Viro spin_unlock(&vfsmount_lock); 15747b43a79fSAl Viro } 15751da177e4SLinus Torvalds up_write(&sb->s_umount); 15760e55a7ccSDan Williams if (!err) { 15772d92ab3cSAl Viro security_sb_post_remount(path->mnt, flags, data); 15780e55a7ccSDan Williams 15790e55a7ccSDan Williams spin_lock(&vfsmount_lock); 15800e55a7ccSDan Williams touch_mnt_namespace(path->mnt->mnt_ns); 15810e55a7ccSDan Williams spin_unlock(&vfsmount_lock); 15820e55a7ccSDan Williams } 15831da177e4SLinus Torvalds return err; 15841da177e4SLinus Torvalds } 15851da177e4SLinus Torvalds 15869676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 15879676f0c6SRam Pai { 15889676f0c6SRam Pai struct vfsmount *p; 15899676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 15909676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 15919676f0c6SRam Pai return 1; 15929676f0c6SRam Pai } 15939676f0c6SRam Pai return 0; 15949676f0c6SRam Pai } 15959676f0c6SRam Pai 15960a0d8a46SAl Viro static int do_move_mount(struct path *path, char *old_name) 15971da177e4SLinus Torvalds { 15982d92ab3cSAl Viro struct path old_path, parent_path; 15991da177e4SLinus Torvalds struct vfsmount *p; 16001da177e4SLinus Torvalds int err = 0; 16011da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 16021da177e4SLinus Torvalds return -EPERM; 16031da177e4SLinus Torvalds if (!old_name || !*old_name) 16041da177e4SLinus Torvalds return -EINVAL; 16052d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 16061da177e4SLinus Torvalds if (err) 16071da177e4SLinus Torvalds return err; 16081da177e4SLinus Torvalds 1609390c6843SRam Pai down_write(&namespace_sem); 16102d92ab3cSAl Viro while (d_mountpoint(path->dentry) && 16119393bd07SAl Viro follow_down(path)) 16121da177e4SLinus Torvalds ; 16131da177e4SLinus Torvalds err = -EINVAL; 16142d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 16151da177e4SLinus Torvalds goto out; 16161da177e4SLinus Torvalds 16171da177e4SLinus Torvalds err = -ENOENT; 16182d92ab3cSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 16192d92ab3cSAl Viro if (IS_DEADDIR(path->dentry->d_inode)) 16201da177e4SLinus Torvalds goto out1; 16211da177e4SLinus Torvalds 1622f3da392eSAlexey Dobriyan if (d_unlinked(path->dentry)) 162321444403SRam Pai goto out1; 16241da177e4SLinus Torvalds 16251da177e4SLinus Torvalds err = -EINVAL; 16262d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 162721444403SRam Pai goto out1; 16281da177e4SLinus Torvalds 16292d92ab3cSAl Viro if (old_path.mnt == old_path.mnt->mnt_parent) 163021444403SRam Pai goto out1; 16311da177e4SLinus Torvalds 16322d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 16332d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 163421444403SRam Pai goto out1; 163521444403SRam Pai /* 163621444403SRam Pai * Don't move a mount residing in a shared parent. 163721444403SRam Pai */ 16382d92ab3cSAl Viro if (old_path.mnt->mnt_parent && 16392d92ab3cSAl Viro IS_MNT_SHARED(old_path.mnt->mnt_parent)) 164021444403SRam Pai goto out1; 16419676f0c6SRam Pai /* 16429676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 16439676f0c6SRam Pai * mount which is shared. 16449676f0c6SRam Pai */ 16452d92ab3cSAl Viro if (IS_MNT_SHARED(path->mnt) && 16462d92ab3cSAl Viro tree_contains_unbindable(old_path.mnt)) 16479676f0c6SRam Pai goto out1; 16481da177e4SLinus Torvalds err = -ELOOP; 16492d92ab3cSAl Viro for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) 16502d92ab3cSAl Viro if (p == old_path.mnt) 165121444403SRam Pai goto out1; 16521da177e4SLinus Torvalds 16532d92ab3cSAl Viro err = attach_recursive_mnt(old_path.mnt, path, &parent_path); 16544ac91378SJan Blunck if (err) 165521444403SRam Pai goto out1; 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 16581da177e4SLinus Torvalds * automatically */ 16592d92ab3cSAl Viro list_del_init(&old_path.mnt->mnt_expire); 16601da177e4SLinus Torvalds out1: 16612d92ab3cSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 16621da177e4SLinus Torvalds out: 1663390c6843SRam Pai up_write(&namespace_sem); 16641da177e4SLinus Torvalds if (!err) 16651a390689SAl Viro path_put(&parent_path); 16662d92ab3cSAl Viro path_put(&old_path); 16671da177e4SLinus Torvalds return err; 16681da177e4SLinus Torvalds } 16691da177e4SLinus Torvalds 16701da177e4SLinus Torvalds /* 16711da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 16721da177e4SLinus Torvalds * namespace's tree 16731da177e4SLinus Torvalds */ 16740a0d8a46SAl Viro static int do_new_mount(struct path *path, char *type, int flags, 16751da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 16761da177e4SLinus Torvalds { 16771da177e4SLinus Torvalds struct vfsmount *mnt; 16781da177e4SLinus Torvalds 1679eca6f534SVegard Nossum if (!type) 16801da177e4SLinus Torvalds return -EINVAL; 16811da177e4SLinus Torvalds 16821da177e4SLinus Torvalds /* we need capabilities... */ 16831da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 16841da177e4SLinus Torvalds return -EPERM; 16851da177e4SLinus Torvalds 16867f78d4cdSAl Viro lock_kernel(); 16871da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 16887f78d4cdSAl Viro unlock_kernel(); 16891da177e4SLinus Torvalds if (IS_ERR(mnt)) 16901da177e4SLinus Torvalds return PTR_ERR(mnt); 16911da177e4SLinus Torvalds 16922d92ab3cSAl Viro return do_add_mount(mnt, path, mnt_flags, NULL); 16931da177e4SLinus Torvalds } 16941da177e4SLinus Torvalds 16951da177e4SLinus Torvalds /* 16961da177e4SLinus Torvalds * add a mount into a namespace's mount tree 16971da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 16981da177e4SLinus Torvalds */ 16998d66bf54SAl Viro int do_add_mount(struct vfsmount *newmnt, struct path *path, 17001da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 17011da177e4SLinus Torvalds { 17021da177e4SLinus Torvalds int err; 17031da177e4SLinus Torvalds 170427d55f1fSAl Viro mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD); 170527d55f1fSAl Viro 1706390c6843SRam Pai down_write(&namespace_sem); 17071da177e4SLinus Torvalds /* Something was mounted here while we slept */ 17088d66bf54SAl Viro while (d_mountpoint(path->dentry) && 17099393bd07SAl Viro follow_down(path)) 17101da177e4SLinus Torvalds ; 17111da177e4SLinus Torvalds err = -EINVAL; 1712dd5cae6eSAl Viro if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) 17131da177e4SLinus Torvalds goto unlock; 17141da177e4SLinus Torvalds 17151da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 17161da177e4SLinus Torvalds err = -EBUSY; 17178d66bf54SAl Viro if (path->mnt->mnt_sb == newmnt->mnt_sb && 17188d66bf54SAl Viro path->mnt->mnt_root == path->dentry) 17191da177e4SLinus Torvalds goto unlock; 17201da177e4SLinus Torvalds 17211da177e4SLinus Torvalds err = -EINVAL; 17221da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 17231da177e4SLinus Torvalds goto unlock; 17241da177e4SLinus Torvalds 17251da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 17268d66bf54SAl Viro if ((err = graft_tree(newmnt, path))) 17275b83d2c5SRam Pai goto unlock; 17281da177e4SLinus Torvalds 17296758f953SAl Viro if (fslist) /* add to the specified expiration list */ 173055e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 17316758f953SAl Viro 1732390c6843SRam Pai up_write(&namespace_sem); 17335b83d2c5SRam Pai return 0; 17341da177e4SLinus Torvalds 17351da177e4SLinus Torvalds unlock: 1736390c6843SRam Pai up_write(&namespace_sem); 17371da177e4SLinus Torvalds mntput(newmnt); 17381da177e4SLinus Torvalds return err; 17391da177e4SLinus Torvalds } 17401da177e4SLinus Torvalds 17411da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 17421da177e4SLinus Torvalds 17435528f911STrond Myklebust /* 17441da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 17451da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 17461da177e4SLinus Torvalds * here 17471da177e4SLinus Torvalds */ 17481da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 17491da177e4SLinus Torvalds { 17501da177e4SLinus Torvalds struct vfsmount *mnt, *next; 17511da177e4SLinus Torvalds LIST_HEAD(graveyard); 1752bcc5c7d2SAl Viro LIST_HEAD(umounts); 17531da177e4SLinus Torvalds 17541da177e4SLinus Torvalds if (list_empty(mounts)) 17551da177e4SLinus Torvalds return; 17561da177e4SLinus Torvalds 1757bcc5c7d2SAl Viro down_write(&namespace_sem); 17581da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 17591da177e4SLinus Torvalds 17601da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 17611da177e4SLinus Torvalds * following criteria: 17621da177e4SLinus Torvalds * - only referenced by its parent vfsmount 17631da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 17641da177e4SLinus Torvalds * cleared by mntput()) 17651da177e4SLinus Torvalds */ 176655e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 17671da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 1768bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 17691da177e4SLinus Torvalds continue; 177055e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 17711da177e4SLinus Torvalds } 1772bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1773bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 1774bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1775bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 1776bcc5c7d2SAl Viro } 17771da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1778bcc5c7d2SAl Viro up_write(&namespace_sem); 1779bcc5c7d2SAl Viro 1780bcc5c7d2SAl Viro release_mounts(&umounts); 17811da177e4SLinus Torvalds } 17821da177e4SLinus Torvalds 17831da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 17841da177e4SLinus Torvalds 17851da177e4SLinus Torvalds /* 17865528f911STrond Myklebust * Ripoff of 'select_parent()' 17875528f911STrond Myklebust * 17885528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 17895528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 17905528f911STrond Myklebust */ 17915528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 17925528f911STrond Myklebust { 17935528f911STrond Myklebust struct vfsmount *this_parent = parent; 17945528f911STrond Myklebust struct list_head *next; 17955528f911STrond Myklebust int found = 0; 17965528f911STrond Myklebust 17975528f911STrond Myklebust repeat: 17985528f911STrond Myklebust next = this_parent->mnt_mounts.next; 17995528f911STrond Myklebust resume: 18005528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 18015528f911STrond Myklebust struct list_head *tmp = next; 18025528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 18035528f911STrond Myklebust 18045528f911STrond Myklebust next = tmp->next; 18055528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 18065528f911STrond Myklebust continue; 18075528f911STrond Myklebust /* 18085528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 18095528f911STrond Myklebust */ 18105528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 18115528f911STrond Myklebust this_parent = mnt; 18125528f911STrond Myklebust goto repeat; 18135528f911STrond Myklebust } 18145528f911STrond Myklebust 18155528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 18165528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 18175528f911STrond Myklebust found++; 18185528f911STrond Myklebust } 18195528f911STrond Myklebust } 18205528f911STrond Myklebust /* 18215528f911STrond Myklebust * All done at this level ... ascend and resume the search 18225528f911STrond Myklebust */ 18235528f911STrond Myklebust if (this_parent != parent) { 18245528f911STrond Myklebust next = this_parent->mnt_child.next; 18255528f911STrond Myklebust this_parent = this_parent->mnt_parent; 18265528f911STrond Myklebust goto resume; 18275528f911STrond Myklebust } 18285528f911STrond Myklebust return found; 18295528f911STrond Myklebust } 18305528f911STrond Myklebust 18315528f911STrond Myklebust /* 18325528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 18335528f911STrond Myklebust * submounts of a specific parent mountpoint 18345528f911STrond Myklebust */ 1835c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 18365528f911STrond Myklebust { 18375528f911STrond Myklebust LIST_HEAD(graveyard); 1838c35038beSAl Viro struct vfsmount *m; 18395528f911STrond Myklebust 18405528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 1841c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 1842bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1843c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 1844bcc5c7d2SAl Viro mnt_expire); 1845afef80b3SEric W. Biederman touch_mnt_namespace(m->mnt_ns); 1846afef80b3SEric W. Biederman umount_tree(m, 1, umounts); 1847bcc5c7d2SAl Viro } 1848bcc5c7d2SAl Viro } 18495528f911STrond Myklebust } 18505528f911STrond Myklebust 18515528f911STrond Myklebust /* 18521da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 18531da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 18541da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 18551da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 18561da177e4SLinus Torvalds */ 1857b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1858b58fed8bSRam Pai unsigned long n) 18591da177e4SLinus Torvalds { 18601da177e4SLinus Torvalds char *t = to; 18611da177e4SLinus Torvalds const char __user *f = from; 18621da177e4SLinus Torvalds char c; 18631da177e4SLinus Torvalds 18641da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 18651da177e4SLinus Torvalds return n; 18661da177e4SLinus Torvalds 18671da177e4SLinus Torvalds while (n) { 18681da177e4SLinus Torvalds if (__get_user(c, f)) { 18691da177e4SLinus Torvalds memset(t, 0, n); 18701da177e4SLinus Torvalds break; 18711da177e4SLinus Torvalds } 18721da177e4SLinus Torvalds *t++ = c; 18731da177e4SLinus Torvalds f++; 18741da177e4SLinus Torvalds n--; 18751da177e4SLinus Torvalds } 18761da177e4SLinus Torvalds return n; 18771da177e4SLinus Torvalds } 18781da177e4SLinus Torvalds 18791da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 18801da177e4SLinus Torvalds { 18811da177e4SLinus Torvalds int i; 18821da177e4SLinus Torvalds unsigned long page; 18831da177e4SLinus Torvalds unsigned long size; 18841da177e4SLinus Torvalds 18851da177e4SLinus Torvalds *where = 0; 18861da177e4SLinus Torvalds if (!data) 18871da177e4SLinus Torvalds return 0; 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 18901da177e4SLinus Torvalds return -ENOMEM; 18911da177e4SLinus Torvalds 18921da177e4SLinus Torvalds /* We only care that *some* data at the address the user 18931da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 18941da177e4SLinus Torvalds * the remainder of the page. 18951da177e4SLinus Torvalds */ 18961da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 18971da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 18981da177e4SLinus Torvalds if (size > PAGE_SIZE) 18991da177e4SLinus Torvalds size = PAGE_SIZE; 19001da177e4SLinus Torvalds 19011da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 19021da177e4SLinus Torvalds if (!i) { 19031da177e4SLinus Torvalds free_page(page); 19041da177e4SLinus Torvalds return -EFAULT; 19051da177e4SLinus Torvalds } 19061da177e4SLinus Torvalds if (i != PAGE_SIZE) 19071da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 19081da177e4SLinus Torvalds *where = page; 19091da177e4SLinus Torvalds return 0; 19101da177e4SLinus Torvalds } 19111da177e4SLinus Torvalds 1912eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 1913eca6f534SVegard Nossum { 1914eca6f534SVegard Nossum char *tmp; 1915eca6f534SVegard Nossum 1916eca6f534SVegard Nossum if (!data) { 1917eca6f534SVegard Nossum *where = NULL; 1918eca6f534SVegard Nossum return 0; 1919eca6f534SVegard Nossum } 1920eca6f534SVegard Nossum 1921eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 1922eca6f534SVegard Nossum if (IS_ERR(tmp)) 1923eca6f534SVegard Nossum return PTR_ERR(tmp); 1924eca6f534SVegard Nossum 1925eca6f534SVegard Nossum *where = tmp; 1926eca6f534SVegard Nossum return 0; 1927eca6f534SVegard Nossum } 1928eca6f534SVegard Nossum 19291da177e4SLinus Torvalds /* 19301da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 19311da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 19321da177e4SLinus Torvalds * 19331da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 19341da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 19351da177e4SLinus Torvalds * information (or be NULL). 19361da177e4SLinus Torvalds * 19371da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 19381da177e4SLinus Torvalds * When the flags word was introduced its top half was required 19391da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 19401da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 19411da177e4SLinus Torvalds * and must be discarded. 19421da177e4SLinus Torvalds */ 19431da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 19441da177e4SLinus Torvalds unsigned long flags, void *data_page) 19451da177e4SLinus Torvalds { 19462d92ab3cSAl Viro struct path path; 19471da177e4SLinus Torvalds int retval = 0; 19481da177e4SLinus Torvalds int mnt_flags = 0; 19491da177e4SLinus Torvalds 19501da177e4SLinus Torvalds /* Discard magic */ 19511da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 19521da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 19531da177e4SLinus Torvalds 19541da177e4SLinus Torvalds /* Basic sanity checks */ 19551da177e4SLinus Torvalds 19561da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 19571da177e4SLinus Torvalds return -EINVAL; 19581da177e4SLinus Torvalds 19591da177e4SLinus Torvalds if (data_page) 19601da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 19611da177e4SLinus Torvalds 1962a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 1963a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 1964a27ab9f2STetsuo Handa if (retval) 1965a27ab9f2STetsuo Handa return retval; 1966a27ab9f2STetsuo Handa 1967a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 1968a27ab9f2STetsuo Handa type_page, flags, data_page); 1969a27ab9f2STetsuo Handa if (retval) 1970a27ab9f2STetsuo Handa goto dput_out; 1971a27ab9f2STetsuo Handa 1972613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 1973613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 19740a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 19750a1c01c9SMatthew Garrett 19761da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 19771da177e4SLinus Torvalds if (flags & MS_NOSUID) 19781da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 19791da177e4SLinus Torvalds if (flags & MS_NODEV) 19801da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 19811da177e4SLinus Torvalds if (flags & MS_NOEXEC) 19821da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 1983fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 1984fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 1985fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 1986fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 1987d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 1988d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 19892e4b7fcdSDave Hansen if (flags & MS_RDONLY) 19902e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 1991fc33a7bbSChristoph Hellwig 1992fc33a7bbSChristoph Hellwig flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 1993d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 1994d0adde57SMatthew Garrett MS_STRICTATIME); 19951da177e4SLinus Torvalds 19961da177e4SLinus Torvalds if (flags & MS_REMOUNT) 19972d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 19981da177e4SLinus Torvalds data_page); 19991da177e4SLinus Torvalds else if (flags & MS_BIND) 20002d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 20019676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 20022d92ab3cSAl Viro retval = do_change_type(&path, flags); 20031da177e4SLinus Torvalds else if (flags & MS_MOVE) 20042d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 20051da177e4SLinus Torvalds else 20062d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 20071da177e4SLinus Torvalds dev_name, data_page); 20081da177e4SLinus Torvalds dput_out: 20092d92ab3cSAl Viro path_put(&path); 20101da177e4SLinus Torvalds return retval; 20111da177e4SLinus Torvalds } 20121da177e4SLinus Torvalds 2013cf8d2c11STrond Myklebust static struct mnt_namespace *alloc_mnt_ns(void) 2014cf8d2c11STrond Myklebust { 2015cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2016cf8d2c11STrond Myklebust 2017cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2018cf8d2c11STrond Myklebust if (!new_ns) 2019cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2020cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2021cf8d2c11STrond Myklebust new_ns->root = NULL; 2022cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2023cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2024cf8d2c11STrond Myklebust new_ns->event = 0; 2025cf8d2c11STrond Myklebust return new_ns; 2026cf8d2c11STrond Myklebust } 2027cf8d2c11STrond Myklebust 2028741a2951SJANAK DESAI /* 2029741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 2030741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 2031741a2951SJANAK DESAI */ 2032e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 20336b3286edSKirill Korotaev struct fs_struct *fs) 20341da177e4SLinus Torvalds { 20356b3286edSKirill Korotaev struct mnt_namespace *new_ns; 20367f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 20371da177e4SLinus Torvalds struct vfsmount *p, *q; 20381da177e4SLinus Torvalds 2039cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2040cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2041cf8d2c11STrond Myklebust return new_ns; 20421da177e4SLinus Torvalds 2043390c6843SRam Pai down_write(&namespace_sem); 20441da177e4SLinus Torvalds /* First pass: copy the tree topology */ 20456b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 20469676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 20471da177e4SLinus Torvalds if (!new_ns->root) { 2048390c6843SRam Pai up_write(&namespace_sem); 20491da177e4SLinus Torvalds kfree(new_ns); 20505cc4a034SJulia Lawall return ERR_PTR(-ENOMEM); 20511da177e4SLinus Torvalds } 20521da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 20531da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 20541da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 20551da177e4SLinus Torvalds 20561da177e4SLinus Torvalds /* 20571da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 20581da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 20591da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 20601da177e4SLinus Torvalds */ 20616b3286edSKirill Korotaev p = mnt_ns->root; 20621da177e4SLinus Torvalds q = new_ns->root; 20631da177e4SLinus Torvalds while (p) { 20646b3286edSKirill Korotaev q->mnt_ns = new_ns; 20651da177e4SLinus Torvalds if (fs) { 20666ac08c39SJan Blunck if (p == fs->root.mnt) { 20671da177e4SLinus Torvalds rootmnt = p; 20686ac08c39SJan Blunck fs->root.mnt = mntget(q); 20691da177e4SLinus Torvalds } 20706ac08c39SJan Blunck if (p == fs->pwd.mnt) { 20711da177e4SLinus Torvalds pwdmnt = p; 20726ac08c39SJan Blunck fs->pwd.mnt = mntget(q); 20731da177e4SLinus Torvalds } 20741da177e4SLinus Torvalds } 20756b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 20761da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 20771da177e4SLinus Torvalds } 2078390c6843SRam Pai up_write(&namespace_sem); 20791da177e4SLinus Torvalds 20801da177e4SLinus Torvalds if (rootmnt) 20811da177e4SLinus Torvalds mntput(rootmnt); 20821da177e4SLinus Torvalds if (pwdmnt) 20831da177e4SLinus Torvalds mntput(pwdmnt); 20841da177e4SLinus Torvalds 2085741a2951SJANAK DESAI return new_ns; 2086741a2951SJANAK DESAI } 2087741a2951SJANAK DESAI 2088213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2089e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 2090741a2951SJANAK DESAI { 20916b3286edSKirill Korotaev struct mnt_namespace *new_ns; 2092741a2951SJANAK DESAI 2093e3222c4eSBadari Pulavarty BUG_ON(!ns); 20946b3286edSKirill Korotaev get_mnt_ns(ns); 2095741a2951SJANAK DESAI 2096741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 2097e3222c4eSBadari Pulavarty return ns; 2098741a2951SJANAK DESAI 2099e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 2100741a2951SJANAK DESAI 21016b3286edSKirill Korotaev put_mnt_ns(ns); 2102e3222c4eSBadari Pulavarty return new_ns; 21031da177e4SLinus Torvalds } 21041da177e4SLinus Torvalds 2105cf8d2c11STrond Myklebust /** 2106cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2107cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2108cf8d2c11STrond Myklebust */ 2109a2770d86SLinus Torvalds struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt) 2110cf8d2c11STrond Myklebust { 2111cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2112cf8d2c11STrond Myklebust 2113cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2114cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 2115cf8d2c11STrond Myklebust mnt->mnt_ns = new_ns; 2116cf8d2c11STrond Myklebust new_ns->root = mnt; 2117cf8d2c11STrond Myklebust list_add(&new_ns->list, &new_ns->root->mnt_list); 2118cf8d2c11STrond Myklebust } 2119cf8d2c11STrond Myklebust return new_ns; 2120cf8d2c11STrond Myklebust } 2121a2770d86SLinus Torvalds EXPORT_SYMBOL(create_mnt_ns); 2122cf8d2c11STrond Myklebust 2123bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2124bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 21251da177e4SLinus Torvalds { 2126eca6f534SVegard Nossum int ret; 2127eca6f534SVegard Nossum char *kernel_type; 2128eca6f534SVegard Nossum char *kernel_dir; 2129eca6f534SVegard Nossum char *kernel_dev; 21301da177e4SLinus Torvalds unsigned long data_page; 21311da177e4SLinus Torvalds 2132eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2133eca6f534SVegard Nossum if (ret < 0) 2134eca6f534SVegard Nossum goto out_type; 21351da177e4SLinus Torvalds 2136eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2137eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2138eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2139eca6f534SVegard Nossum goto out_dir; 2140eca6f534SVegard Nossum } 21411da177e4SLinus Torvalds 2142eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2143eca6f534SVegard Nossum if (ret < 0) 2144eca6f534SVegard Nossum goto out_dev; 21451da177e4SLinus Torvalds 2146eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2147eca6f534SVegard Nossum if (ret < 0) 2148eca6f534SVegard Nossum goto out_data; 21491da177e4SLinus Torvalds 2150eca6f534SVegard Nossum ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags, 2151eca6f534SVegard Nossum (void *) data_page); 2152eca6f534SVegard Nossum 21531da177e4SLinus Torvalds free_page(data_page); 2154eca6f534SVegard Nossum out_data: 2155eca6f534SVegard Nossum kfree(kernel_dev); 2156eca6f534SVegard Nossum out_dev: 2157eca6f534SVegard Nossum putname(kernel_dir); 2158eca6f534SVegard Nossum out_dir: 2159eca6f534SVegard Nossum kfree(kernel_type); 2160eca6f534SVegard Nossum out_type: 2161eca6f534SVegard Nossum return ret; 21621da177e4SLinus Torvalds } 21631da177e4SLinus Torvalds 21641da177e4SLinus Torvalds /* 21651da177e4SLinus Torvalds * pivot_root Semantics: 21661da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 21671da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 21681da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 21691da177e4SLinus Torvalds * 21701da177e4SLinus Torvalds * Restrictions: 21711da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 21721da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 21731da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 21741da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 21751da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 21761da177e4SLinus Torvalds * 21774a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 21784a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 21794a0d11faSNeil Brown * in this situation. 21804a0d11faSNeil Brown * 21811da177e4SLinus Torvalds * Notes: 21821da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 21831da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 21841da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 21851da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 21861da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 21871da177e4SLinus Torvalds * first. 21881da177e4SLinus Torvalds */ 21893480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 21903480b257SHeiko Carstens const char __user *, put_old) 21911da177e4SLinus Torvalds { 21921da177e4SLinus Torvalds struct vfsmount *tmp; 21932d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 21941da177e4SLinus Torvalds int error; 21951da177e4SLinus Torvalds 21961da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 21971da177e4SLinus Torvalds return -EPERM; 21981da177e4SLinus Torvalds 21992d8f3038SAl Viro error = user_path_dir(new_root, &new); 22001da177e4SLinus Torvalds if (error) 22011da177e4SLinus Torvalds goto out0; 22021da177e4SLinus Torvalds error = -EINVAL; 22032d8f3038SAl Viro if (!check_mnt(new.mnt)) 22041da177e4SLinus Torvalds goto out1; 22051da177e4SLinus Torvalds 22062d8f3038SAl Viro error = user_path_dir(put_old, &old); 22071da177e4SLinus Torvalds if (error) 22081da177e4SLinus Torvalds goto out1; 22091da177e4SLinus Torvalds 22102d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 22111da177e4SLinus Torvalds if (error) { 22122d8f3038SAl Viro path_put(&old); 22131da177e4SLinus Torvalds goto out1; 22141da177e4SLinus Torvalds } 22151da177e4SLinus Torvalds 22161da177e4SLinus Torvalds read_lock(¤t->fs->lock); 22178c3ee42eSAl Viro root = current->fs->root; 22186ac08c39SJan Blunck path_get(¤t->fs->root); 22191da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 2220390c6843SRam Pai down_write(&namespace_sem); 22212d8f3038SAl Viro mutex_lock(&old.dentry->d_inode->i_mutex); 22221da177e4SLinus Torvalds error = -EINVAL; 22232d8f3038SAl Viro if (IS_MNT_SHARED(old.mnt) || 22242d8f3038SAl Viro IS_MNT_SHARED(new.mnt->mnt_parent) || 22258c3ee42eSAl Viro IS_MNT_SHARED(root.mnt->mnt_parent)) 222621444403SRam Pai goto out2; 22278c3ee42eSAl Viro if (!check_mnt(root.mnt)) 22281da177e4SLinus Torvalds goto out2; 22291da177e4SLinus Torvalds error = -ENOENT; 22302d8f3038SAl Viro if (IS_DEADDIR(new.dentry->d_inode)) 22311da177e4SLinus Torvalds goto out2; 2232f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 22331da177e4SLinus Torvalds goto out2; 2234f3da392eSAlexey Dobriyan if (d_unlinked(old.dentry)) 22351da177e4SLinus Torvalds goto out2; 22361da177e4SLinus Torvalds error = -EBUSY; 22372d8f3038SAl Viro if (new.mnt == root.mnt || 22382d8f3038SAl Viro old.mnt == root.mnt) 22391da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 22401da177e4SLinus Torvalds error = -EINVAL; 22418c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 22421da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 22438c3ee42eSAl Viro if (root.mnt->mnt_parent == root.mnt) 22440bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 22452d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 22461da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 22472d8f3038SAl Viro if (new.mnt->mnt_parent == new.mnt) 22480bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 22494ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 22502d8f3038SAl Viro tmp = old.mnt; 22511da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 22522d8f3038SAl Viro if (tmp != new.mnt) { 22531da177e4SLinus Torvalds for (;;) { 22541da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 22551da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 22562d8f3038SAl Viro if (tmp->mnt_parent == new.mnt) 22571da177e4SLinus Torvalds break; 22581da177e4SLinus Torvalds tmp = tmp->mnt_parent; 22591da177e4SLinus Torvalds } 22602d8f3038SAl Viro if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) 22611da177e4SLinus Torvalds goto out3; 22622d8f3038SAl Viro } else if (!is_subdir(old.dentry, new.dentry)) 22631da177e4SLinus Torvalds goto out3; 22642d8f3038SAl Viro detach_mnt(new.mnt, &parent_path); 22658c3ee42eSAl Viro detach_mnt(root.mnt, &root_parent); 22664ac91378SJan Blunck /* mount old root on put_old */ 22672d8f3038SAl Viro attach_mnt(root.mnt, &old); 22684ac91378SJan Blunck /* mount new_root on / */ 22692d8f3038SAl Viro attach_mnt(new.mnt, &root_parent); 22706b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 22711da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 22722d8f3038SAl Viro chroot_fs_refs(&root, &new); 22732d8f3038SAl Viro security_sb_post_pivotroot(&root, &new); 22741da177e4SLinus Torvalds error = 0; 22751a390689SAl Viro path_put(&root_parent); 22761a390689SAl Viro path_put(&parent_path); 22771da177e4SLinus Torvalds out2: 22782d8f3038SAl Viro mutex_unlock(&old.dentry->d_inode->i_mutex); 2279390c6843SRam Pai up_write(&namespace_sem); 22808c3ee42eSAl Viro path_put(&root); 22812d8f3038SAl Viro path_put(&old); 22821da177e4SLinus Torvalds out1: 22832d8f3038SAl Viro path_put(&new); 22841da177e4SLinus Torvalds out0: 22851da177e4SLinus Torvalds return error; 22861da177e4SLinus Torvalds out3: 22871da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 22881da177e4SLinus Torvalds goto out2; 22891da177e4SLinus Torvalds } 22901da177e4SLinus Torvalds 22911da177e4SLinus Torvalds static void __init init_mount_tree(void) 22921da177e4SLinus Torvalds { 22931da177e4SLinus Torvalds struct vfsmount *mnt; 22946b3286edSKirill Korotaev struct mnt_namespace *ns; 2295ac748a09SJan Blunck struct path root; 22961da177e4SLinus Torvalds 22971da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 22981da177e4SLinus Torvalds if (IS_ERR(mnt)) 22991da177e4SLinus Torvalds panic("Can't create rootfs"); 23003b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 23013b22edc5STrond Myklebust if (IS_ERR(ns)) 23021da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 23031da177e4SLinus Torvalds 23046b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 23056b3286edSKirill Korotaev get_mnt_ns(ns); 23061da177e4SLinus Torvalds 2307ac748a09SJan Blunck root.mnt = ns->root; 2308ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2309ac748a09SJan Blunck 2310ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2311ac748a09SJan Blunck set_fs_root(current->fs, &root); 23121da177e4SLinus Torvalds } 23131da177e4SLinus Torvalds 231474bf17cfSDenis Cheng void __init mnt_init(void) 23151da177e4SLinus Torvalds { 231613f14b4dSEric Dumazet unsigned u; 231715a67dd8SRandy Dunlap int err; 23181da177e4SLinus Torvalds 2319390c6843SRam Pai init_rwsem(&namespace_sem); 2320390c6843SRam Pai 23211da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 232220c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 23231da177e4SLinus Torvalds 2324b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 23251da177e4SLinus Torvalds 23261da177e4SLinus Torvalds if (!mount_hashtable) 23271da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 23281da177e4SLinus Torvalds 232913f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 23301da177e4SLinus Torvalds 233113f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 233213f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 23331da177e4SLinus Torvalds 233415a67dd8SRandy Dunlap err = sysfs_init(); 233515a67dd8SRandy Dunlap if (err) 233615a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 23378e24eea7SHarvey Harrison __func__, err); 233800d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 233900d26666SGreg Kroah-Hartman if (!fs_kobj) 23408e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 23411da177e4SLinus Torvalds init_rootfs(); 23421da177e4SLinus Torvalds init_mount_tree(); 23431da177e4SLinus Torvalds } 23441da177e4SLinus Torvalds 2345616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 23461da177e4SLinus Torvalds { 2347616511d0STrond Myklebust struct vfsmount *root; 234870fbcdf4SRam Pai LIST_HEAD(umount_list); 2349616511d0STrond Myklebust 2350616511d0STrond Myklebust if (!atomic_dec_and_lock(&ns->count, &vfsmount_lock)) 2351616511d0STrond Myklebust return; 2352616511d0STrond Myklebust root = ns->root; 23536b3286edSKirill Korotaev ns->root = NULL; 23541ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 2355390c6843SRam Pai down_write(&namespace_sem); 23561da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 2357a05964f3SRam Pai umount_tree(root, 0, &umount_list); 23581da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 2359390c6843SRam Pai up_write(&namespace_sem); 236070fbcdf4SRam Pai release_mounts(&umount_list); 23616b3286edSKirill Korotaev kfree(ns); 23621da177e4SLinus Torvalds } 2363cf8d2c11STrond Myklebust EXPORT_SYMBOL(put_mnt_ns); 2364