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> 251da177e4SLinus Torvalds #include <linux/security.h> 261da177e4SLinus Torvalds #include <linux/mount.h> 2707f3f05cSDavid Howells #include <linux/ramfs.h> 2813f14b4dSEric Dumazet #include <linux/log2.h> 2973cd49ecSMiklos Szeredi #include <linux/idr.h> 301da177e4SLinus Torvalds #include <asm/uaccess.h> 311da177e4SLinus Torvalds #include <asm/unistd.h> 3207b20889SRam Pai #include "pnode.h" 33948730b0SAdrian Bunk #include "internal.h" 341da177e4SLinus Torvalds 3513f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 3613f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 3713f14b4dSEric Dumazet 381da177e4SLinus Torvalds /* spinlock for vfsmount related operations, inplace of dcache_lock */ 391da177e4SLinus Torvalds __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 401da177e4SLinus Torvalds 415addc5ddSAl Viro static int event; 4273cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 43719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 445addc5ddSAl Viro 45fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 46e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 47390c6843SRam Pai static struct rw_semaphore namespace_sem; 481da177e4SLinus Torvalds 49f87fd4c2SMiklos Szeredi /* /sys/fs */ 5000d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 5100d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 52f87fd4c2SMiklos Szeredi 531da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 541da177e4SLinus Torvalds { 551da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 561da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 5713f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 5813f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 591da177e4SLinus Torvalds } 601da177e4SLinus Torvalds 613d733633SDave Hansen #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 623d733633SDave Hansen 6373cd49ecSMiklos Szeredi /* allocation is serialized by namespace_sem */ 6473cd49ecSMiklos Szeredi static int mnt_alloc_id(struct vfsmount *mnt) 6573cd49ecSMiklos Szeredi { 6673cd49ecSMiklos Szeredi int res; 6773cd49ecSMiklos Szeredi 6873cd49ecSMiklos Szeredi retry: 6973cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 7073cd49ecSMiklos Szeredi spin_lock(&vfsmount_lock); 7173cd49ecSMiklos Szeredi res = ida_get_new(&mnt_id_ida, &mnt->mnt_id); 7273cd49ecSMiklos Szeredi spin_unlock(&vfsmount_lock); 7373cd49ecSMiklos Szeredi if (res == -EAGAIN) 7473cd49ecSMiklos Szeredi goto retry; 7573cd49ecSMiklos Szeredi 7673cd49ecSMiklos Szeredi return res; 7773cd49ecSMiklos Szeredi } 7873cd49ecSMiklos Szeredi 7973cd49ecSMiklos Szeredi static void mnt_free_id(struct vfsmount *mnt) 8073cd49ecSMiklos Szeredi { 8173cd49ecSMiklos Szeredi spin_lock(&vfsmount_lock); 8273cd49ecSMiklos Szeredi ida_remove(&mnt_id_ida, mnt->mnt_id); 8373cd49ecSMiklos Szeredi spin_unlock(&vfsmount_lock); 8473cd49ecSMiklos Szeredi } 8573cd49ecSMiklos Szeredi 86719f5d7fSMiklos Szeredi /* 87719f5d7fSMiklos Szeredi * Allocate a new peer group ID 88719f5d7fSMiklos Szeredi * 89719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 90719f5d7fSMiklos Szeredi */ 91719f5d7fSMiklos Szeredi static int mnt_alloc_group_id(struct vfsmount *mnt) 92719f5d7fSMiklos Szeredi { 93719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 94719f5d7fSMiklos Szeredi return -ENOMEM; 95719f5d7fSMiklos Szeredi 96719f5d7fSMiklos Szeredi return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id); 97719f5d7fSMiklos Szeredi } 98719f5d7fSMiklos Szeredi 99719f5d7fSMiklos Szeredi /* 100719f5d7fSMiklos Szeredi * Release a peer group ID 101719f5d7fSMiklos Szeredi */ 102719f5d7fSMiklos Szeredi void mnt_release_group_id(struct vfsmount *mnt) 103719f5d7fSMiklos Szeredi { 104719f5d7fSMiklos Szeredi ida_remove(&mnt_group_ida, mnt->mnt_group_id); 105719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; 106719f5d7fSMiklos Szeredi } 107719f5d7fSMiklos Szeredi 1081da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 1091da177e4SLinus Torvalds { 110c3762229SRobert P. J. Day struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 1111da177e4SLinus Torvalds if (mnt) { 11273cd49ecSMiklos Szeredi int err; 11373cd49ecSMiklos Szeredi 11473cd49ecSMiklos Szeredi err = mnt_alloc_id(mnt); 11573cd49ecSMiklos Szeredi if (err) { 11673cd49ecSMiklos Szeredi kmem_cache_free(mnt_cache, mnt); 11773cd49ecSMiklos Szeredi return NULL; 11873cd49ecSMiklos Szeredi } 11973cd49ecSMiklos Szeredi 1201da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 1211da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 1221da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 1231da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 1241da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 12555e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 12603e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 127a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 128a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 1293d733633SDave Hansen atomic_set(&mnt->__mnt_writers, 0); 1301da177e4SLinus Torvalds if (name) { 1311da177e4SLinus Torvalds int size = strlen(name) + 1; 1321da177e4SLinus Torvalds char *newname = kmalloc(size, GFP_KERNEL); 1331da177e4SLinus Torvalds if (newname) { 1341da177e4SLinus Torvalds memcpy(newname, name, size); 1351da177e4SLinus Torvalds mnt->mnt_devname = newname; 1361da177e4SLinus Torvalds } 1371da177e4SLinus Torvalds } 1381da177e4SLinus Torvalds } 1391da177e4SLinus Torvalds return mnt; 1401da177e4SLinus Torvalds } 1411da177e4SLinus Torvalds 1428366025eSDave Hansen /* 1438366025eSDave Hansen * Most r/o checks on a fs are for operations that take 1448366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 1458366025eSDave Hansen * We must keep track of when those operations start 1468366025eSDave Hansen * (for permission checks) and when they end, so that 1478366025eSDave Hansen * we can determine when writes are able to occur to 1488366025eSDave Hansen * a filesystem. 1498366025eSDave Hansen */ 1503d733633SDave Hansen /* 1513d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 1523d733633SDave Hansen * @mnt: the mount to check for its write status 1533d733633SDave Hansen * 1543d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 1553d733633SDave Hansen * It does not guarantee that the filesystem will stay 1563d733633SDave Hansen * r/w, just that it is right *now*. This can not and 1573d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 1583d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 1593d733633SDave Hansen * r/w. 1603d733633SDave Hansen */ 1613d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 1623d733633SDave Hansen { 1632e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 1642e4b7fcdSDave Hansen return 1; 1652e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 1662e4b7fcdSDave Hansen return 1; 1672e4b7fcdSDave Hansen return 0; 1683d733633SDave Hansen } 1693d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 1703d733633SDave Hansen 1713d733633SDave Hansen struct mnt_writer { 1723d733633SDave Hansen /* 1733d733633SDave Hansen * If holding multiple instances of this lock, they 1743d733633SDave Hansen * must be ordered by cpu number. 1753d733633SDave Hansen */ 1763d733633SDave Hansen spinlock_t lock; 1773d733633SDave Hansen struct lock_class_key lock_class; /* compiles out with !lockdep */ 1783d733633SDave Hansen unsigned long count; 1793d733633SDave Hansen struct vfsmount *mnt; 1803d733633SDave Hansen } ____cacheline_aligned_in_smp; 1813d733633SDave Hansen static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); 1823d733633SDave Hansen 1833d733633SDave Hansen static int __init init_mnt_writers(void) 1843d733633SDave Hansen { 1853d733633SDave Hansen int cpu; 1863d733633SDave Hansen for_each_possible_cpu(cpu) { 1873d733633SDave Hansen struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); 1883d733633SDave Hansen spin_lock_init(&writer->lock); 1893d733633SDave Hansen lockdep_set_class(&writer->lock, &writer->lock_class); 1903d733633SDave Hansen writer->count = 0; 1913d733633SDave Hansen } 1923d733633SDave Hansen return 0; 1933d733633SDave Hansen } 1943d733633SDave Hansen fs_initcall(init_mnt_writers); 1953d733633SDave Hansen 1963d733633SDave Hansen static void unlock_mnt_writers(void) 1973d733633SDave Hansen { 1983d733633SDave Hansen int cpu; 1993d733633SDave Hansen struct mnt_writer *cpu_writer; 2003d733633SDave Hansen 2013d733633SDave Hansen for_each_possible_cpu(cpu) { 2023d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 2033d733633SDave Hansen spin_unlock(&cpu_writer->lock); 2043d733633SDave Hansen } 2053d733633SDave Hansen } 2063d733633SDave Hansen 2073d733633SDave Hansen static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) 2083d733633SDave Hansen { 2093d733633SDave Hansen if (!cpu_writer->mnt) 2103d733633SDave Hansen return; 2113d733633SDave Hansen /* 2123d733633SDave Hansen * This is in case anyone ever leaves an invalid, 2133d733633SDave Hansen * old ->mnt and a count of 0. 2143d733633SDave Hansen */ 2153d733633SDave Hansen if (!cpu_writer->count) 2163d733633SDave Hansen return; 2173d733633SDave Hansen atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); 2183d733633SDave Hansen cpu_writer->count = 0; 2193d733633SDave Hansen } 2203d733633SDave Hansen /* 2213d733633SDave Hansen * must hold cpu_writer->lock 2223d733633SDave Hansen */ 2233d733633SDave Hansen static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, 2243d733633SDave Hansen struct vfsmount *mnt) 2253d733633SDave Hansen { 2263d733633SDave Hansen if (cpu_writer->mnt == mnt) 2273d733633SDave Hansen return; 2283d733633SDave Hansen __clear_mnt_count(cpu_writer); 2293d733633SDave Hansen cpu_writer->mnt = mnt; 2303d733633SDave Hansen } 2313d733633SDave Hansen 2323d733633SDave Hansen /* 2333d733633SDave Hansen * Most r/o checks on a fs are for operations that take 2343d733633SDave Hansen * discrete amounts of time, like a write() or unlink(). 2353d733633SDave Hansen * We must keep track of when those operations start 2363d733633SDave Hansen * (for permission checks) and when they end, so that 2373d733633SDave Hansen * we can determine when writes are able to occur to 2383d733633SDave Hansen * a filesystem. 2393d733633SDave Hansen */ 2408366025eSDave Hansen /** 2418366025eSDave Hansen * mnt_want_write - get write access to a mount 2428366025eSDave Hansen * @mnt: the mount on which to take a write 2438366025eSDave Hansen * 2448366025eSDave Hansen * This tells the low-level filesystem that a write is 2458366025eSDave Hansen * about to be performed to it, and makes sure that 2468366025eSDave Hansen * writes are allowed before returning success. When 2478366025eSDave Hansen * the write operation is finished, mnt_drop_write() 2488366025eSDave Hansen * must be called. This is effectively a refcount. 2498366025eSDave Hansen */ 2508366025eSDave Hansen int mnt_want_write(struct vfsmount *mnt) 2518366025eSDave Hansen { 2523d733633SDave Hansen int ret = 0; 2533d733633SDave Hansen struct mnt_writer *cpu_writer; 2543d733633SDave Hansen 2553d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 2563d733633SDave Hansen spin_lock(&cpu_writer->lock); 2573d733633SDave Hansen if (__mnt_is_readonly(mnt)) { 2583d733633SDave Hansen ret = -EROFS; 2593d733633SDave Hansen goto out; 2603d733633SDave Hansen } 2613d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 2623d733633SDave Hansen cpu_writer->count++; 2633d733633SDave Hansen out: 2643d733633SDave Hansen spin_unlock(&cpu_writer->lock); 2653d733633SDave Hansen put_cpu_var(mnt_writers); 2663d733633SDave Hansen return ret; 2678366025eSDave Hansen } 2688366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 2698366025eSDave Hansen 2703d733633SDave Hansen static void lock_mnt_writers(void) 2713d733633SDave Hansen { 2723d733633SDave Hansen int cpu; 2733d733633SDave Hansen struct mnt_writer *cpu_writer; 2743d733633SDave Hansen 2753d733633SDave Hansen for_each_possible_cpu(cpu) { 2763d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 2773d733633SDave Hansen spin_lock(&cpu_writer->lock); 2783d733633SDave Hansen __clear_mnt_count(cpu_writer); 2793d733633SDave Hansen cpu_writer->mnt = NULL; 2803d733633SDave Hansen } 2813d733633SDave Hansen } 2823d733633SDave Hansen 2833d733633SDave Hansen /* 2843d733633SDave Hansen * These per-cpu write counts are not guaranteed to have 2853d733633SDave Hansen * matched increments and decrements on any given cpu. 2863d733633SDave Hansen * A file open()ed for write on one cpu and close()d on 2873d733633SDave Hansen * another cpu will imbalance this count. Make sure it 2883d733633SDave Hansen * does not get too far out of whack. 2893d733633SDave Hansen */ 2903d733633SDave Hansen static void handle_write_count_underflow(struct vfsmount *mnt) 2913d733633SDave Hansen { 2923d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) >= 2933d733633SDave Hansen MNT_WRITER_UNDERFLOW_LIMIT) 2943d733633SDave Hansen return; 2953d733633SDave Hansen /* 2963d733633SDave Hansen * It isn't necessary to hold all of the locks 2973d733633SDave Hansen * at the same time, but doing it this way makes 2983d733633SDave Hansen * us share a lot more code. 2993d733633SDave Hansen */ 3003d733633SDave Hansen lock_mnt_writers(); 3013d733633SDave Hansen /* 3023d733633SDave Hansen * vfsmount_lock is for mnt_flags. 3033d733633SDave Hansen */ 3043d733633SDave Hansen spin_lock(&vfsmount_lock); 3053d733633SDave Hansen /* 3063d733633SDave Hansen * If coalescing the per-cpu writer counts did not 3073d733633SDave Hansen * get us back to a positive writer count, we have 3083d733633SDave Hansen * a bug. 3093d733633SDave Hansen */ 3103d733633SDave Hansen if ((atomic_read(&mnt->__mnt_writers) < 0) && 3113d733633SDave Hansen !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { 3123d733633SDave Hansen printk(KERN_DEBUG "leak detected on mount(%p) writers " 3133d733633SDave Hansen "count: %d\n", 3143d733633SDave Hansen mnt, atomic_read(&mnt->__mnt_writers)); 3153d733633SDave Hansen WARN_ON(1); 3163d733633SDave Hansen /* use the flag to keep the dmesg spam down */ 3173d733633SDave Hansen mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT; 3183d733633SDave Hansen } 3193d733633SDave Hansen spin_unlock(&vfsmount_lock); 3203d733633SDave Hansen unlock_mnt_writers(); 3213d733633SDave Hansen } 3223d733633SDave Hansen 3238366025eSDave Hansen /** 3248366025eSDave Hansen * mnt_drop_write - give up write access to a mount 3258366025eSDave Hansen * @mnt: the mount on which to give up write access 3268366025eSDave Hansen * 3278366025eSDave Hansen * Tells the low-level filesystem that we are done 3288366025eSDave Hansen * performing writes to it. Must be matched with 3298366025eSDave Hansen * mnt_want_write() call above. 3308366025eSDave Hansen */ 3318366025eSDave Hansen void mnt_drop_write(struct vfsmount *mnt) 3328366025eSDave Hansen { 3333d733633SDave Hansen int must_check_underflow = 0; 3343d733633SDave Hansen struct mnt_writer *cpu_writer; 3353d733633SDave Hansen 3363d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 3373d733633SDave Hansen spin_lock(&cpu_writer->lock); 3383d733633SDave Hansen 3393d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 3403d733633SDave Hansen if (cpu_writer->count > 0) { 3413d733633SDave Hansen cpu_writer->count--; 3423d733633SDave Hansen } else { 3433d733633SDave Hansen must_check_underflow = 1; 3443d733633SDave Hansen atomic_dec(&mnt->__mnt_writers); 3453d733633SDave Hansen } 3463d733633SDave Hansen 3473d733633SDave Hansen spin_unlock(&cpu_writer->lock); 3483d733633SDave Hansen /* 3493d733633SDave Hansen * Logically, we could call this each time, 3503d733633SDave Hansen * but the __mnt_writers cacheline tends to 3513d733633SDave Hansen * be cold, and makes this expensive. 3523d733633SDave Hansen */ 3533d733633SDave Hansen if (must_check_underflow) 3543d733633SDave Hansen handle_write_count_underflow(mnt); 3553d733633SDave Hansen /* 3563d733633SDave Hansen * This could be done right after the spinlock 3573d733633SDave Hansen * is taken because the spinlock keeps us on 3583d733633SDave Hansen * the cpu, and disables preemption. However, 3593d733633SDave Hansen * putting it here bounds the amount that 3603d733633SDave Hansen * __mnt_writers can underflow. Without it, 3613d733633SDave Hansen * we could theoretically wrap __mnt_writers. 3623d733633SDave Hansen */ 3633d733633SDave Hansen put_cpu_var(mnt_writers); 3648366025eSDave Hansen } 3658366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 3668366025eSDave Hansen 3672e4b7fcdSDave Hansen static int mnt_make_readonly(struct vfsmount *mnt) 3688366025eSDave Hansen { 3693d733633SDave Hansen int ret = 0; 3703d733633SDave Hansen 3713d733633SDave Hansen lock_mnt_writers(); 3723d733633SDave Hansen /* 3733d733633SDave Hansen * With all the locks held, this value is stable 3743d733633SDave Hansen */ 3753d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) > 0) { 3763d733633SDave Hansen ret = -EBUSY; 3773d733633SDave Hansen goto out; 3788366025eSDave Hansen } 3793d733633SDave Hansen /* 3802e4b7fcdSDave Hansen * nobody can do a successful mnt_want_write() with all 3812e4b7fcdSDave Hansen * of the counts in MNT_DENIED_WRITE and the locks held. 3823d733633SDave Hansen */ 3832e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 3842e4b7fcdSDave Hansen if (!ret) 3852e4b7fcdSDave Hansen mnt->mnt_flags |= MNT_READONLY; 3862e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3873d733633SDave Hansen out: 3883d733633SDave Hansen unlock_mnt_writers(); 3893d733633SDave Hansen return ret; 3903d733633SDave Hansen } 3918366025eSDave Hansen 3922e4b7fcdSDave Hansen static void __mnt_unmake_readonly(struct vfsmount *mnt) 3932e4b7fcdSDave Hansen { 3942e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 3952e4b7fcdSDave Hansen mnt->mnt_flags &= ~MNT_READONLY; 3962e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3972e4b7fcdSDave Hansen } 3982e4b7fcdSDave Hansen 399454e2398SDavid Howells int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 400454e2398SDavid Howells { 401454e2398SDavid Howells mnt->mnt_sb = sb; 402454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 403454e2398SDavid Howells return 0; 404454e2398SDavid Howells } 405454e2398SDavid Howells 406454e2398SDavid Howells EXPORT_SYMBOL(simple_set_mnt); 407454e2398SDavid Howells 4081da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 4091da177e4SLinus Torvalds { 4101da177e4SLinus Torvalds kfree(mnt->mnt_devname); 41173cd49ecSMiklos Szeredi mnt_free_id(mnt); 4121da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 4131da177e4SLinus Torvalds } 4141da177e4SLinus Torvalds 4151da177e4SLinus Torvalds /* 416a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 417a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 4181da177e4SLinus Torvalds */ 419a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 420a05964f3SRam Pai int dir) 4211da177e4SLinus Torvalds { 4221da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 4231da177e4SLinus Torvalds struct list_head *tmp = head; 4241da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 4251da177e4SLinus Torvalds 4261da177e4SLinus Torvalds for (;;) { 427a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 4281da177e4SLinus Torvalds p = NULL; 4291da177e4SLinus Torvalds if (tmp == head) 4301da177e4SLinus Torvalds break; 4311da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 4321da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 433a05964f3SRam Pai found = p; 4341da177e4SLinus Torvalds break; 4351da177e4SLinus Torvalds } 4361da177e4SLinus Torvalds } 4371da177e4SLinus Torvalds return found; 4381da177e4SLinus Torvalds } 4391da177e4SLinus Torvalds 440a05964f3SRam Pai /* 441a05964f3SRam Pai * lookup_mnt increments the ref count before returning 442a05964f3SRam Pai * the vfsmount struct. 443a05964f3SRam Pai */ 444a05964f3SRam Pai struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 445a05964f3SRam Pai { 446a05964f3SRam Pai struct vfsmount *child_mnt; 447a05964f3SRam Pai spin_lock(&vfsmount_lock); 448a05964f3SRam Pai if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) 449a05964f3SRam Pai mntget(child_mnt); 450a05964f3SRam Pai spin_unlock(&vfsmount_lock); 451a05964f3SRam Pai return child_mnt; 452a05964f3SRam Pai } 453a05964f3SRam Pai 4541da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 4551da177e4SLinus Torvalds { 4566b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 4571da177e4SLinus Torvalds } 4581da177e4SLinus Torvalds 4596b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 4605addc5ddSAl Viro { 4615addc5ddSAl Viro if (ns) { 4625addc5ddSAl Viro ns->event = ++event; 4635addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4645addc5ddSAl Viro } 4655addc5ddSAl Viro } 4665addc5ddSAl Viro 4676b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 4685addc5ddSAl Viro { 4695addc5ddSAl Viro if (ns && ns->event != event) { 4705addc5ddSAl Viro ns->event = event; 4715addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4725addc5ddSAl Viro } 4735addc5ddSAl Viro } 4745addc5ddSAl Viro 4751a390689SAl Viro static void detach_mnt(struct vfsmount *mnt, struct path *old_path) 4761da177e4SLinus Torvalds { 4771a390689SAl Viro old_path->dentry = mnt->mnt_mountpoint; 4781a390689SAl Viro old_path->mnt = mnt->mnt_parent; 4791da177e4SLinus Torvalds mnt->mnt_parent = mnt; 4801da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 4811da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 4821da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 4831a390689SAl Viro old_path->dentry->d_mounted--; 4841da177e4SLinus Torvalds } 4851da177e4SLinus Torvalds 486b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 487b90fa9aeSRam Pai struct vfsmount *child_mnt) 488b90fa9aeSRam Pai { 489b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 490b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 491b90fa9aeSRam Pai dentry->d_mounted++; 492b90fa9aeSRam Pai } 493b90fa9aeSRam Pai 4941a390689SAl Viro static void attach_mnt(struct vfsmount *mnt, struct path *path) 4951da177e4SLinus Torvalds { 4961a390689SAl Viro mnt_set_mountpoint(path->mnt, path->dentry, mnt); 497b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 4981a390689SAl Viro hash(path->mnt, path->dentry)); 4991a390689SAl Viro list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); 500b90fa9aeSRam Pai } 501b90fa9aeSRam Pai 502b90fa9aeSRam Pai /* 503b90fa9aeSRam Pai * the caller must hold vfsmount_lock 504b90fa9aeSRam Pai */ 505b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 506b90fa9aeSRam Pai { 507b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 508b90fa9aeSRam Pai struct vfsmount *m; 509b90fa9aeSRam Pai LIST_HEAD(head); 5106b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 511b90fa9aeSRam Pai 512b90fa9aeSRam Pai BUG_ON(parent == mnt); 513b90fa9aeSRam Pai 514b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 515b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 5166b3286edSKirill Korotaev m->mnt_ns = n; 517b90fa9aeSRam Pai list_splice(&head, n->list.prev); 518b90fa9aeSRam Pai 519b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 520b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 521b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 5226b3286edSKirill Korotaev touch_mnt_namespace(n); 5231da177e4SLinus Torvalds } 5241da177e4SLinus Torvalds 5251da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 5261da177e4SLinus Torvalds { 5271da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 5281da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 5291da177e4SLinus Torvalds while (1) { 5301da177e4SLinus Torvalds if (p == root) 5311da177e4SLinus Torvalds return NULL; 5321da177e4SLinus Torvalds next = p->mnt_child.next; 5331da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 5341da177e4SLinus Torvalds break; 5351da177e4SLinus Torvalds p = p->mnt_parent; 5361da177e4SLinus Torvalds } 5371da177e4SLinus Torvalds } 5381da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 5391da177e4SLinus Torvalds } 5401da177e4SLinus Torvalds 5419676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 5429676f0c6SRam Pai { 5439676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 5449676f0c6SRam Pai while (prev != &p->mnt_mounts) { 5459676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 5469676f0c6SRam Pai prev = p->mnt_mounts.prev; 5479676f0c6SRam Pai } 5489676f0c6SRam Pai return p; 5499676f0c6SRam Pai } 5509676f0c6SRam Pai 55136341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 55236341f64SRam Pai int flag) 5531da177e4SLinus Torvalds { 5541da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 5551da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 5561da177e4SLinus Torvalds 5571da177e4SLinus Torvalds if (mnt) { 558719f5d7fSMiklos Szeredi if (flag & (CL_SLAVE | CL_PRIVATE)) 559719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; /* not a peer of original */ 560719f5d7fSMiklos Szeredi else 561719f5d7fSMiklos Szeredi mnt->mnt_group_id = old->mnt_group_id; 562719f5d7fSMiklos Szeredi 563719f5d7fSMiklos Szeredi if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 564719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(mnt); 565719f5d7fSMiklos Szeredi if (err) 566719f5d7fSMiklos Szeredi goto out_free; 567719f5d7fSMiklos Szeredi } 568719f5d7fSMiklos Szeredi 5691da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 5701da177e4SLinus Torvalds atomic_inc(&sb->s_active); 5711da177e4SLinus Torvalds mnt->mnt_sb = sb; 5721da177e4SLinus Torvalds mnt->mnt_root = dget(root); 5731da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 5741da177e4SLinus Torvalds mnt->mnt_parent = mnt; 575b90fa9aeSRam Pai 5765afe0022SRam Pai if (flag & CL_SLAVE) { 5775afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 5785afe0022SRam Pai mnt->mnt_master = old; 5795afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 5808aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 581b90fa9aeSRam Pai if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) 582b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 5835afe0022SRam Pai if (IS_MNT_SLAVE(old)) 5845afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 5855afe0022SRam Pai mnt->mnt_master = old->mnt_master; 5865afe0022SRam Pai } 587b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 588b90fa9aeSRam Pai set_mnt_shared(mnt); 5891da177e4SLinus Torvalds 5901da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 5911da177e4SLinus Torvalds * as the original if that was on one */ 59236341f64SRam Pai if (flag & CL_EXPIRE) { 59355e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 59455e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 5951da177e4SLinus Torvalds } 59636341f64SRam Pai } 5971da177e4SLinus Torvalds return mnt; 598719f5d7fSMiklos Szeredi 599719f5d7fSMiklos Szeredi out_free: 600719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 601719f5d7fSMiklos Szeredi return NULL; 6021da177e4SLinus Torvalds } 6031da177e4SLinus Torvalds 6047b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 6051da177e4SLinus Torvalds { 6063d733633SDave Hansen int cpu; 6071da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 6083d733633SDave Hansen /* 6093d733633SDave Hansen * We don't have to hold all of the locks at the 6103d733633SDave Hansen * same time here because we know that we're the 6113d733633SDave Hansen * last reference to mnt and that no new writers 6123d733633SDave Hansen * can come in. 6133d733633SDave Hansen */ 6143d733633SDave Hansen for_each_possible_cpu(cpu) { 6153d733633SDave Hansen struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu); 6163d733633SDave Hansen if (cpu_writer->mnt != mnt) 6173d733633SDave Hansen continue; 6183d733633SDave Hansen spin_lock(&cpu_writer->lock); 6193d733633SDave Hansen atomic_add(cpu_writer->count, &mnt->__mnt_writers); 6203d733633SDave Hansen cpu_writer->count = 0; 6213d733633SDave Hansen /* 6223d733633SDave Hansen * Might as well do this so that no one 6233d733633SDave Hansen * ever sees the pointer and expects 6243d733633SDave Hansen * it to be valid. 6253d733633SDave Hansen */ 6263d733633SDave Hansen cpu_writer->mnt = NULL; 6273d733633SDave Hansen spin_unlock(&cpu_writer->lock); 6283d733633SDave Hansen } 6293d733633SDave Hansen /* 6303d733633SDave Hansen * This probably indicates that somebody messed 6313d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 6323d733633SDave Hansen * happens, the filesystem was probably unable 6333d733633SDave Hansen * to make r/w->r/o transitions. 6343d733633SDave Hansen */ 6353d733633SDave Hansen WARN_ON(atomic_read(&mnt->__mnt_writers)); 6361da177e4SLinus Torvalds dput(mnt->mnt_root); 6371da177e4SLinus Torvalds free_vfsmnt(mnt); 6381da177e4SLinus Torvalds deactivate_super(sb); 6391da177e4SLinus Torvalds } 6401da177e4SLinus Torvalds 6417b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 6427b7b1aceSAl Viro { 6437b7b1aceSAl Viro repeat: 6447b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 6457b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 6467b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6477b7b1aceSAl Viro __mntput(mnt); 6487b7b1aceSAl Viro return; 6497b7b1aceSAl Viro } 6507b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 6517b7b1aceSAl Viro mnt->mnt_pinned = 0; 6527b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6537b7b1aceSAl Viro acct_auto_close_mnt(mnt); 6547b7b1aceSAl Viro security_sb_umount_close(mnt); 6557b7b1aceSAl Viro goto repeat; 6567b7b1aceSAl Viro } 6577b7b1aceSAl Viro } 6587b7b1aceSAl Viro 6597b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 6607b7b1aceSAl Viro 6617b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 6627b7b1aceSAl Viro { 6637b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6647b7b1aceSAl Viro mnt->mnt_pinned++; 6657b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6667b7b1aceSAl Viro } 6677b7b1aceSAl Viro 6687b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 6697b7b1aceSAl Viro 6707b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 6717b7b1aceSAl Viro { 6727b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6737b7b1aceSAl Viro if (mnt->mnt_pinned) { 6747b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 6757b7b1aceSAl Viro mnt->mnt_pinned--; 6767b7b1aceSAl Viro } 6777b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6787b7b1aceSAl Viro } 6797b7b1aceSAl Viro 6807b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 6811da177e4SLinus Torvalds 682b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 683b3b304a2SMiklos Szeredi { 684b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 685b3b304a2SMiklos Szeredi } 686b3b304a2SMiklos Szeredi 687b3b304a2SMiklos Szeredi /* 688b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 689b3b304a2SMiklos Szeredi * implement more complex mount option showing. 690b3b304a2SMiklos Szeredi * 691b3b304a2SMiklos Szeredi * See also save_mount_options(). 692b3b304a2SMiklos Szeredi */ 693b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 694b3b304a2SMiklos Szeredi { 695b3b304a2SMiklos Szeredi const char *options = mnt->mnt_sb->s_options; 696b3b304a2SMiklos Szeredi 697b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 698b3b304a2SMiklos Szeredi seq_putc(m, ','); 699b3b304a2SMiklos Szeredi mangle(m, options); 700b3b304a2SMiklos Szeredi } 701b3b304a2SMiklos Szeredi 702b3b304a2SMiklos Szeredi return 0; 703b3b304a2SMiklos Szeredi } 704b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 705b3b304a2SMiklos Szeredi 706b3b304a2SMiklos Szeredi /* 707b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 708b3b304a2SMiklos Szeredi * called from the fill_super() callback. 709b3b304a2SMiklos Szeredi * 710b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 711b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 712b3b304a2SMiklos Szeredi * remount fails. 713b3b304a2SMiklos Szeredi * 714b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 715b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 716b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 717b3b304a2SMiklos Szeredi * any more. 718b3b304a2SMiklos Szeredi */ 719b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 720b3b304a2SMiklos Szeredi { 721b3b304a2SMiklos Szeredi kfree(sb->s_options); 722b3b304a2SMiklos Szeredi sb->s_options = kstrdup(options, GFP_KERNEL); 723b3b304a2SMiklos Szeredi } 724b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 725b3b304a2SMiklos Szeredi 726a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 7271da177e4SLinus Torvalds /* iterator */ 7281da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 7291da177e4SLinus Torvalds { 730a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 7311da177e4SLinus Torvalds 732390c6843SRam Pai down_read(&namespace_sem); 733a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 7341da177e4SLinus Torvalds } 7351da177e4SLinus Torvalds 7361da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 7371da177e4SLinus Torvalds { 738a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 739b0765fb8SPavel Emelianov 740a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 7411da177e4SLinus Torvalds } 7421da177e4SLinus Torvalds 7431da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 7441da177e4SLinus Torvalds { 745390c6843SRam Pai up_read(&namespace_sem); 7461da177e4SLinus Torvalds } 7471da177e4SLinus Torvalds 7482d4d4864SRam Pai struct proc_fs_info { 7491da177e4SLinus Torvalds int flag; 7502d4d4864SRam Pai const char *str; 7512d4d4864SRam Pai }; 7522d4d4864SRam Pai 7532d4d4864SRam Pai static void show_sb_opts(struct seq_file *m, struct super_block *sb) 7542d4d4864SRam Pai { 7552d4d4864SRam Pai static const struct proc_fs_info fs_info[] = { 7561da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 7571da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 7581da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 7591da177e4SLinus Torvalds { 0, NULL } 7601da177e4SLinus Torvalds }; 7612d4d4864SRam Pai const struct proc_fs_info *fs_infop; 7622d4d4864SRam Pai 7632d4d4864SRam Pai for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 7642d4d4864SRam Pai if (sb->s_flags & fs_infop->flag) 7652d4d4864SRam Pai seq_puts(m, fs_infop->str); 7662d4d4864SRam Pai } 7672d4d4864SRam Pai } 7682d4d4864SRam Pai 7692d4d4864SRam Pai static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) 7702d4d4864SRam Pai { 7712d4d4864SRam Pai static const struct proc_fs_info mnt_info[] = { 7721da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 7731da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 7741da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 775fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 776fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 77747ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 7781da177e4SLinus Torvalds { 0, NULL } 7791da177e4SLinus Torvalds }; 7802d4d4864SRam Pai const struct proc_fs_info *fs_infop; 7812d4d4864SRam Pai 7822d4d4864SRam Pai for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 7832d4d4864SRam Pai if (mnt->mnt_flags & fs_infop->flag) 7842d4d4864SRam Pai seq_puts(m, fs_infop->str); 7852d4d4864SRam Pai } 7862d4d4864SRam Pai } 7872d4d4864SRam Pai 7882d4d4864SRam Pai static void show_type(struct seq_file *m, struct super_block *sb) 7892d4d4864SRam Pai { 7902d4d4864SRam Pai mangle(m, sb->s_type->name); 7912d4d4864SRam Pai if (sb->s_subtype && sb->s_subtype[0]) { 7922d4d4864SRam Pai seq_putc(m, '.'); 7932d4d4864SRam Pai mangle(m, sb->s_subtype); 7942d4d4864SRam Pai } 7952d4d4864SRam Pai } 7962d4d4864SRam Pai 7972d4d4864SRam Pai static int show_vfsmnt(struct seq_file *m, void *v) 7982d4d4864SRam Pai { 7992d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 8002d4d4864SRam Pai int err = 0; 801c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 8021da177e4SLinus Torvalds 8031da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 8041da177e4SLinus Torvalds seq_putc(m, ' '); 805c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 8061da177e4SLinus Torvalds seq_putc(m, ' '); 8072d4d4864SRam Pai show_type(m, mnt->mnt_sb); 8082e4b7fcdSDave Hansen seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 8092d4d4864SRam Pai show_sb_opts(m, mnt->mnt_sb); 8102d4d4864SRam Pai show_mnt_opts(m, mnt); 8111da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 8121da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 8131da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 8141da177e4SLinus Torvalds return err; 8151da177e4SLinus Torvalds } 8161da177e4SLinus Torvalds 817a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 8181da177e4SLinus Torvalds .start = m_start, 8191da177e4SLinus Torvalds .next = m_next, 8201da177e4SLinus Torvalds .stop = m_stop, 8211da177e4SLinus Torvalds .show = show_vfsmnt 8221da177e4SLinus Torvalds }; 8231da177e4SLinus Torvalds 8242d4d4864SRam Pai static int show_mountinfo(struct seq_file *m, void *v) 8252d4d4864SRam Pai { 8262d4d4864SRam Pai struct proc_mounts *p = m->private; 8272d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 8282d4d4864SRam Pai struct super_block *sb = mnt->mnt_sb; 8292d4d4864SRam Pai struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 8302d4d4864SRam Pai struct path root = p->root; 8312d4d4864SRam Pai int err = 0; 8322d4d4864SRam Pai 8332d4d4864SRam Pai seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, 8342d4d4864SRam Pai MAJOR(sb->s_dev), MINOR(sb->s_dev)); 8352d4d4864SRam Pai seq_dentry(m, mnt->mnt_root, " \t\n\\"); 8362d4d4864SRam Pai seq_putc(m, ' '); 8372d4d4864SRam Pai seq_path_root(m, &mnt_path, &root, " \t\n\\"); 8382d4d4864SRam Pai if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { 8392d4d4864SRam Pai /* 8402d4d4864SRam Pai * Mountpoint is outside root, discard that one. Ugly, 8412d4d4864SRam Pai * but less so than trying to do that in iterator in a 8422d4d4864SRam Pai * race-free way (due to renames). 8432d4d4864SRam Pai */ 8442d4d4864SRam Pai return SEQ_SKIP; 8452d4d4864SRam Pai } 8462d4d4864SRam Pai seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); 8472d4d4864SRam Pai show_mnt_opts(m, mnt); 8482d4d4864SRam Pai 8492d4d4864SRam Pai /* Tagged fields ("foo:X" or "bar") */ 8502d4d4864SRam Pai if (IS_MNT_SHARED(mnt)) 8512d4d4864SRam Pai seq_printf(m, " shared:%i", mnt->mnt_group_id); 85297e7e0f7SMiklos Szeredi if (IS_MNT_SLAVE(mnt)) { 85397e7e0f7SMiklos Szeredi int master = mnt->mnt_master->mnt_group_id; 85497e7e0f7SMiklos Szeredi int dom = get_dominating_id(mnt, &p->root); 85597e7e0f7SMiklos Szeredi seq_printf(m, " master:%i", master); 85697e7e0f7SMiklos Szeredi if (dom && dom != master) 85797e7e0f7SMiklos Szeredi seq_printf(m, " propagate_from:%i", dom); 85897e7e0f7SMiklos Szeredi } 8592d4d4864SRam Pai if (IS_MNT_UNBINDABLE(mnt)) 8602d4d4864SRam Pai seq_puts(m, " unbindable"); 8612d4d4864SRam Pai 8622d4d4864SRam Pai /* Filesystem specific data */ 8632d4d4864SRam Pai seq_puts(m, " - "); 8642d4d4864SRam Pai show_type(m, sb); 8652d4d4864SRam Pai seq_putc(m, ' '); 8662d4d4864SRam Pai mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 8672d4d4864SRam Pai seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); 8682d4d4864SRam Pai show_sb_opts(m, sb); 8692d4d4864SRam Pai if (sb->s_op->show_options) 8702d4d4864SRam Pai err = sb->s_op->show_options(m, mnt); 8712d4d4864SRam Pai seq_putc(m, '\n'); 8722d4d4864SRam Pai return err; 8732d4d4864SRam Pai } 8742d4d4864SRam Pai 8752d4d4864SRam Pai const struct seq_operations mountinfo_op = { 8762d4d4864SRam Pai .start = m_start, 8772d4d4864SRam Pai .next = m_next, 8782d4d4864SRam Pai .stop = m_stop, 8792d4d4864SRam Pai .show = show_mountinfo, 8802d4d4864SRam Pai }; 8812d4d4864SRam Pai 882b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 883b4629fe2SChuck Lever { 884b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 885c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 886b4629fe2SChuck Lever int err = 0; 887b4629fe2SChuck Lever 888b4629fe2SChuck Lever /* device */ 889b4629fe2SChuck Lever if (mnt->mnt_devname) { 890b4629fe2SChuck Lever seq_puts(m, "device "); 891b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 892b4629fe2SChuck Lever } else 893b4629fe2SChuck Lever seq_puts(m, "no device"); 894b4629fe2SChuck Lever 895b4629fe2SChuck Lever /* mount point */ 896b4629fe2SChuck Lever seq_puts(m, " mounted on "); 897c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 898b4629fe2SChuck Lever seq_putc(m, ' '); 899b4629fe2SChuck Lever 900b4629fe2SChuck Lever /* file system type */ 901b4629fe2SChuck Lever seq_puts(m, "with fstype "); 9022d4d4864SRam Pai show_type(m, mnt->mnt_sb); 903b4629fe2SChuck Lever 904b4629fe2SChuck Lever /* optional statistics */ 905b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 906b4629fe2SChuck Lever seq_putc(m, ' '); 907b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 908b4629fe2SChuck Lever } 909b4629fe2SChuck Lever 910b4629fe2SChuck Lever seq_putc(m, '\n'); 911b4629fe2SChuck Lever return err; 912b4629fe2SChuck Lever } 913b4629fe2SChuck Lever 914a1a2c409SMiklos Szeredi const struct seq_operations mountstats_op = { 915b4629fe2SChuck Lever .start = m_start, 916b4629fe2SChuck Lever .next = m_next, 917b4629fe2SChuck Lever .stop = m_stop, 918b4629fe2SChuck Lever .show = show_vfsstat, 919b4629fe2SChuck Lever }; 920a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 921b4629fe2SChuck Lever 9221da177e4SLinus Torvalds /** 9231da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 9241da177e4SLinus Torvalds * @mnt: root of mount tree 9251da177e4SLinus Torvalds * 9261da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 9271da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 9281da177e4SLinus Torvalds * busy. 9291da177e4SLinus Torvalds */ 9301da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 9311da177e4SLinus Torvalds { 93236341f64SRam Pai int actual_refs = 0; 93336341f64SRam Pai int minimum_refs = 0; 93436341f64SRam Pai struct vfsmount *p; 9351da177e4SLinus Torvalds 9361da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 93736341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 9381da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 9391da177e4SLinus Torvalds minimum_refs += 2; 9401da177e4SLinus Torvalds } 9411da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 9421da177e4SLinus Torvalds 9431da177e4SLinus Torvalds if (actual_refs > minimum_refs) 9441da177e4SLinus Torvalds return 0; 945e3474a8eSIan Kent 946e3474a8eSIan Kent return 1; 9471da177e4SLinus Torvalds } 9481da177e4SLinus Torvalds 9491da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds /** 9521da177e4SLinus Torvalds * may_umount - check if a mount point is busy 9531da177e4SLinus Torvalds * @mnt: root of mount 9541da177e4SLinus Torvalds * 9551da177e4SLinus Torvalds * This is called to check if a mount point has any 9561da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 9571da177e4SLinus Torvalds * mount has sub mounts this will return busy 9581da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 9591da177e4SLinus Torvalds * 9601da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 9611da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 9621da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 9631da177e4SLinus Torvalds */ 9641da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 9651da177e4SLinus Torvalds { 966e3474a8eSIan Kent int ret = 1; 967a05964f3SRam Pai spin_lock(&vfsmount_lock); 968a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 969e3474a8eSIan Kent ret = 0; 970a05964f3SRam Pai spin_unlock(&vfsmount_lock); 971a05964f3SRam Pai return ret; 9721da177e4SLinus Torvalds } 9731da177e4SLinus Torvalds 9741da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 9751da177e4SLinus Torvalds 976b90fa9aeSRam Pai void release_mounts(struct list_head *head) 9771da177e4SLinus Torvalds { 97870fbcdf4SRam Pai struct vfsmount *mnt; 97970fbcdf4SRam Pai while (!list_empty(head)) { 980b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 98170fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 98270fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 98370fbcdf4SRam Pai struct dentry *dentry; 98470fbcdf4SRam Pai struct vfsmount *m; 98570fbcdf4SRam Pai spin_lock(&vfsmount_lock); 98670fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 98770fbcdf4SRam Pai m = mnt->mnt_parent; 98870fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 98970fbcdf4SRam Pai mnt->mnt_parent = mnt; 9907c4b93d8SAl Viro m->mnt_ghosts--; 9911da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 99270fbcdf4SRam Pai dput(dentry); 99370fbcdf4SRam Pai mntput(m); 9941da177e4SLinus Torvalds } 9951da177e4SLinus Torvalds mntput(mnt); 99670fbcdf4SRam Pai } 99770fbcdf4SRam Pai } 99870fbcdf4SRam Pai 999a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 100070fbcdf4SRam Pai { 100170fbcdf4SRam Pai struct vfsmount *p; 100270fbcdf4SRam Pai 10031bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 10041bfba4e8SAkinobu Mita list_move(&p->mnt_hash, kill); 100570fbcdf4SRam Pai 1006a05964f3SRam Pai if (propagate) 1007a05964f3SRam Pai propagate_umount(kill); 1008a05964f3SRam Pai 100970fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 101070fbcdf4SRam Pai list_del_init(&p->mnt_expire); 101170fbcdf4SRam Pai list_del_init(&p->mnt_list); 10126b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 10136b3286edSKirill Korotaev p->mnt_ns = NULL; 101470fbcdf4SRam Pai list_del_init(&p->mnt_child); 10157c4b93d8SAl Viro if (p->mnt_parent != p) { 10167c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 1017f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 10187c4b93d8SAl Viro } 1019a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 10201da177e4SLinus Torvalds } 10211da177e4SLinus Torvalds } 10221da177e4SLinus Torvalds 1023c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 1024c35038beSAl Viro 10251da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 10261da177e4SLinus Torvalds { 10271da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 10281da177e4SLinus Torvalds int retval; 102970fbcdf4SRam Pai LIST_HEAD(umount_list); 10301da177e4SLinus Torvalds 10311da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 10321da177e4SLinus Torvalds if (retval) 10331da177e4SLinus Torvalds return retval; 10341da177e4SLinus Torvalds 10351da177e4SLinus Torvalds /* 10361da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 10371da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 10381da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 10391da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 10401da177e4SLinus Torvalds */ 10411da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 10426ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 10431da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 10441da177e4SLinus Torvalds return -EINVAL; 10451da177e4SLinus Torvalds 10461da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 10471da177e4SLinus Torvalds return -EBUSY; 10481da177e4SLinus Torvalds 10491da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 10501da177e4SLinus Torvalds return -EAGAIN; 10511da177e4SLinus Torvalds } 10521da177e4SLinus Torvalds 10531da177e4SLinus Torvalds /* 10541da177e4SLinus Torvalds * If we may have to abort operations to get out of this 10551da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 10561da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 10571da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 10581da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 10591da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 10601da177e4SLinus Torvalds * about for the moment. 10611da177e4SLinus Torvalds */ 10621da177e4SLinus Torvalds 106342faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 10641da177e4SLinus Torvalds lock_kernel(); 106542faad99SAl Viro sb->s_op->umount_begin(sb); 10661da177e4SLinus Torvalds unlock_kernel(); 106742faad99SAl Viro } 10681da177e4SLinus Torvalds 10691da177e4SLinus Torvalds /* 10701da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 10711da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 10721da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 10731da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 10741da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 10751da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 10761da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 10771da177e4SLinus Torvalds */ 10786ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 10791da177e4SLinus Torvalds /* 10801da177e4SLinus Torvalds * Special case for "unmounting" root ... 10811da177e4SLinus Torvalds * we just try to remount it readonly. 10821da177e4SLinus Torvalds */ 10831da177e4SLinus Torvalds down_write(&sb->s_umount); 10841da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 10851da177e4SLinus Torvalds lock_kernel(); 10861da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 10871da177e4SLinus Torvalds unlock_kernel(); 10881da177e4SLinus Torvalds } 10891da177e4SLinus Torvalds up_write(&sb->s_umount); 10901da177e4SLinus Torvalds return retval; 10911da177e4SLinus Torvalds } 10921da177e4SLinus Torvalds 1093390c6843SRam Pai down_write(&namespace_sem); 10941da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 10955addc5ddSAl Viro event++; 10961da177e4SLinus Torvalds 1097c35038beSAl Viro if (!(flags & MNT_DETACH)) 1098c35038beSAl Viro shrink_submounts(mnt, &umount_list); 1099c35038beSAl Viro 11001da177e4SLinus Torvalds retval = -EBUSY; 1101a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 11021da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 1103a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 11041da177e4SLinus Torvalds retval = 0; 11051da177e4SLinus Torvalds } 11061da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 11071da177e4SLinus Torvalds if (retval) 11081da177e4SLinus Torvalds security_sb_umount_busy(mnt); 1109390c6843SRam Pai up_write(&namespace_sem); 111070fbcdf4SRam Pai release_mounts(&umount_list); 11111da177e4SLinus Torvalds return retval; 11121da177e4SLinus Torvalds } 11131da177e4SLinus Torvalds 11141da177e4SLinus Torvalds /* 11151da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 11161da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 11171da177e4SLinus Torvalds * 11181da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 11191da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 11201da177e4SLinus Torvalds */ 11211da177e4SLinus Torvalds 11221da177e4SLinus Torvalds asmlinkage long sys_umount(char __user * name, int flags) 11231da177e4SLinus Torvalds { 11241da177e4SLinus Torvalds struct nameidata nd; 11251da177e4SLinus Torvalds int retval; 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 11281da177e4SLinus Torvalds if (retval) 11291da177e4SLinus Torvalds goto out; 11301da177e4SLinus Torvalds retval = -EINVAL; 11314ac91378SJan Blunck if (nd.path.dentry != nd.path.mnt->mnt_root) 11321da177e4SLinus Torvalds goto dput_and_out; 11334ac91378SJan Blunck if (!check_mnt(nd.path.mnt)) 11341da177e4SLinus Torvalds goto dput_and_out; 11351da177e4SLinus Torvalds 11361da177e4SLinus Torvalds retval = -EPERM; 11371da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 11381da177e4SLinus Torvalds goto dput_and_out; 11391da177e4SLinus Torvalds 11404ac91378SJan Blunck retval = do_umount(nd.path.mnt, flags); 11411da177e4SLinus Torvalds dput_and_out: 1142429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 11434ac91378SJan Blunck dput(nd.path.dentry); 11444ac91378SJan Blunck mntput_no_expire(nd.path.mnt); 11451da177e4SLinus Torvalds out: 11461da177e4SLinus Torvalds return retval; 11471da177e4SLinus Torvalds } 11481da177e4SLinus Torvalds 11491da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 11501da177e4SLinus Torvalds 11511da177e4SLinus Torvalds /* 11521da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 11531da177e4SLinus Torvalds */ 11541da177e4SLinus Torvalds asmlinkage long sys_oldumount(char __user * name) 11551da177e4SLinus Torvalds { 11561da177e4SLinus Torvalds return sys_umount(name, 0); 11571da177e4SLinus Torvalds } 11581da177e4SLinus Torvalds 11591da177e4SLinus Torvalds #endif 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds static int mount_is_safe(struct nameidata *nd) 11621da177e4SLinus Torvalds { 11631da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 11641da177e4SLinus Torvalds return 0; 11651da177e4SLinus Torvalds return -EPERM; 11661da177e4SLinus Torvalds #ifdef notyet 11674ac91378SJan Blunck if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) 11681da177e4SLinus Torvalds return -EPERM; 11694ac91378SJan Blunck if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { 11704ac91378SJan Blunck if (current->uid != nd->path.dentry->d_inode->i_uid) 11711da177e4SLinus Torvalds return -EPERM; 11721da177e4SLinus Torvalds } 1173e4543eddSChristoph Hellwig if (vfs_permission(nd, MAY_WRITE)) 11741da177e4SLinus Torvalds return -EPERM; 11751da177e4SLinus Torvalds return 0; 11761da177e4SLinus Torvalds #endif 11771da177e4SLinus Torvalds } 11781da177e4SLinus Torvalds 1179b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 118036341f64SRam Pai int flag) 11811da177e4SLinus Torvalds { 11821da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 11831a390689SAl Viro struct path path; 11841da177e4SLinus Torvalds 11859676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 11869676f0c6SRam Pai return NULL; 11879676f0c6SRam Pai 118836341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 11891da177e4SLinus Torvalds if (!q) 11901da177e4SLinus Torvalds goto Enomem; 11911da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 11921da177e4SLinus Torvalds 11931da177e4SLinus Torvalds p = mnt; 1194fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 11957ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 11961da177e4SLinus Torvalds continue; 11971da177e4SLinus Torvalds 11981da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 11999676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 12009676f0c6SRam Pai s = skip_mnt_tree(s); 12019676f0c6SRam Pai continue; 12029676f0c6SRam Pai } 12031da177e4SLinus Torvalds while (p != s->mnt_parent) { 12041da177e4SLinus Torvalds p = p->mnt_parent; 12051da177e4SLinus Torvalds q = q->mnt_parent; 12061da177e4SLinus Torvalds } 12071da177e4SLinus Torvalds p = s; 12081a390689SAl Viro path.mnt = q; 12091a390689SAl Viro path.dentry = p->mnt_mountpoint; 121036341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 12111da177e4SLinus Torvalds if (!q) 12121da177e4SLinus Torvalds goto Enomem; 12131da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 12141da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 12151a390689SAl Viro attach_mnt(q, &path); 12161da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 12171da177e4SLinus Torvalds } 12181da177e4SLinus Torvalds } 12191da177e4SLinus Torvalds return res; 12201da177e4SLinus Torvalds Enomem: 12211da177e4SLinus Torvalds if (res) { 122270fbcdf4SRam Pai LIST_HEAD(umount_list); 12231da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1224a05964f3SRam Pai umount_tree(res, 0, &umount_list); 12251da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 122670fbcdf4SRam Pai release_mounts(&umount_list); 12271da177e4SLinus Torvalds } 12281da177e4SLinus Torvalds return NULL; 12291da177e4SLinus Torvalds } 12301da177e4SLinus Torvalds 12318aec0809SAl Viro struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) 12328aec0809SAl Viro { 12338aec0809SAl Viro struct vfsmount *tree; 12341a60a280SAl Viro down_write(&namespace_sem); 12358aec0809SAl Viro tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); 12361a60a280SAl Viro up_write(&namespace_sem); 12378aec0809SAl Viro return tree; 12388aec0809SAl Viro } 12398aec0809SAl Viro 12408aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 12418aec0809SAl Viro { 12428aec0809SAl Viro LIST_HEAD(umount_list); 12431a60a280SAl Viro down_write(&namespace_sem); 12448aec0809SAl Viro spin_lock(&vfsmount_lock); 12458aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 12468aec0809SAl Viro spin_unlock(&vfsmount_lock); 12471a60a280SAl Viro up_write(&namespace_sem); 12488aec0809SAl Viro release_mounts(&umount_list); 12498aec0809SAl Viro } 12508aec0809SAl Viro 1251719f5d7fSMiklos Szeredi static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) 1252719f5d7fSMiklos Szeredi { 1253719f5d7fSMiklos Szeredi struct vfsmount *p; 1254719f5d7fSMiklos Szeredi 1255719f5d7fSMiklos Szeredi for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1256719f5d7fSMiklos Szeredi if (p->mnt_group_id && !IS_MNT_SHARED(p)) 1257719f5d7fSMiklos Szeredi mnt_release_group_id(p); 1258719f5d7fSMiklos Szeredi } 1259719f5d7fSMiklos Szeredi } 1260719f5d7fSMiklos Szeredi 1261719f5d7fSMiklos Szeredi static int invent_group_ids(struct vfsmount *mnt, bool recurse) 1262719f5d7fSMiklos Szeredi { 1263719f5d7fSMiklos Szeredi struct vfsmount *p; 1264719f5d7fSMiklos Szeredi 1265719f5d7fSMiklos Szeredi for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1266719f5d7fSMiklos Szeredi if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 1267719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(p); 1268719f5d7fSMiklos Szeredi if (err) { 1269719f5d7fSMiklos Szeredi cleanup_group_ids(mnt, p); 1270719f5d7fSMiklos Szeredi return err; 1271719f5d7fSMiklos Szeredi } 1272719f5d7fSMiklos Szeredi } 1273719f5d7fSMiklos Szeredi } 1274719f5d7fSMiklos Szeredi 1275719f5d7fSMiklos Szeredi return 0; 1276719f5d7fSMiklos Szeredi } 1277719f5d7fSMiklos Szeredi 1278b90fa9aeSRam Pai /* 1279b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1280b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 128121444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 128221444403SRam Pai * store the parent mount and mountpoint dentry. 128321444403SRam Pai * (done when source_mnt is moved) 1284b90fa9aeSRam Pai * 1285b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1286b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 12879676f0c6SRam Pai * --------------------------------------------------------------------------- 1288b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 12899676f0c6SRam Pai * |************************************************************************** 12909676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 12919676f0c6SRam Pai * | dest | | | | | 12929676f0c6SRam Pai * | | | | | | | 12939676f0c6SRam Pai * | v | | | | | 12949676f0c6SRam Pai * |************************************************************************** 12959676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 12965afe0022SRam Pai * | | | | | | 12979676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 12989676f0c6SRam Pai * *************************************************************************** 1299b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1300b90fa9aeSRam Pai * destination mount. 1301b90fa9aeSRam Pai * 1302b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1303b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1304b90fa9aeSRam Pai * the peer group of the source mount. 1305b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1306b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1307b90fa9aeSRam Pai * mount. 13085afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 13095afe0022SRam Pai * of the destination mount and the cloned mount is made slave 13105afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 13115afe0022SRam Pai * is marked as 'shared and slave'. 13125afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 13135afe0022SRam Pai * source mount. 13145afe0022SRam Pai * 13159676f0c6SRam Pai * --------------------------------------------------------------------------- 131621444403SRam Pai * | MOVE 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 (*) | unbindable | 13269676f0c6SRam Pai * *************************************************************************** 13275afe0022SRam Pai * 13285afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 13295afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 133021444403SRam Pai * (+*) the mount is moved to the destination. 13315afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 13325afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 13335afe0022SRam Pai * the mount is marked as 'shared and slave'. 13345afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1335b90fa9aeSRam Pai * 1336b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1337b90fa9aeSRam Pai * applied to each mount in the tree. 1338b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1339b90fa9aeSRam Pai * in allocations. 1340b90fa9aeSRam Pai */ 1341b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 13421a390689SAl Viro struct path *path, struct path *parent_path) 1343b90fa9aeSRam Pai { 1344b90fa9aeSRam Pai LIST_HEAD(tree_list); 13451a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 13461a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1347b90fa9aeSRam Pai struct vfsmount *child, *p; 1348719f5d7fSMiklos Szeredi int err; 1349b90fa9aeSRam Pai 1350719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) { 1351719f5d7fSMiklos Szeredi err = invent_group_ids(source_mnt, true); 1352719f5d7fSMiklos Szeredi if (err) 1353719f5d7fSMiklos Szeredi goto out; 1354719f5d7fSMiklos Szeredi } 1355719f5d7fSMiklos Szeredi err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1356719f5d7fSMiklos Szeredi if (err) 1357719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1358b90fa9aeSRam Pai 1359b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1360b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1361b90fa9aeSRam Pai set_mnt_shared(p); 1362b90fa9aeSRam Pai } 1363b90fa9aeSRam Pai 1364b90fa9aeSRam Pai spin_lock(&vfsmount_lock); 13651a390689SAl Viro if (parent_path) { 13661a390689SAl Viro detach_mnt(source_mnt, parent_path); 13671a390689SAl Viro attach_mnt(source_mnt, path); 13686b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 136921444403SRam Pai } else { 1370b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1371b90fa9aeSRam Pai commit_tree(source_mnt); 137221444403SRam Pai } 1373b90fa9aeSRam Pai 1374b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1375b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1376b90fa9aeSRam Pai commit_tree(child); 1377b90fa9aeSRam Pai } 1378b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 1379b90fa9aeSRam Pai return 0; 1380719f5d7fSMiklos Szeredi 1381719f5d7fSMiklos Szeredi out_cleanup_ids: 1382719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) 1383719f5d7fSMiklos Szeredi cleanup_group_ids(source_mnt, NULL); 1384719f5d7fSMiklos Szeredi out: 1385719f5d7fSMiklos Szeredi return err; 1386b90fa9aeSRam Pai } 1387b90fa9aeSRam Pai 13888c3ee42eSAl Viro static int graft_tree(struct vfsmount *mnt, struct path *path) 13891da177e4SLinus Torvalds { 13901da177e4SLinus Torvalds int err; 13911da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 13921da177e4SLinus Torvalds return -EINVAL; 13931da177e4SLinus Torvalds 13948c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 13951da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 13961da177e4SLinus Torvalds return -ENOTDIR; 13971da177e4SLinus Torvalds 13981da177e4SLinus Torvalds err = -ENOENT; 13998c3ee42eSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 14008c3ee42eSAl Viro if (IS_DEADDIR(path->dentry->d_inode)) 14011da177e4SLinus Torvalds goto out_unlock; 14021da177e4SLinus Torvalds 14038c3ee42eSAl Viro err = security_sb_check_sb(mnt, path); 14041da177e4SLinus Torvalds if (err) 14051da177e4SLinus Torvalds goto out_unlock; 14061da177e4SLinus Torvalds 14071da177e4SLinus Torvalds err = -ENOENT; 14088c3ee42eSAl Viro if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry)) 14098c3ee42eSAl Viro err = attach_recursive_mnt(mnt, path, NULL); 14101da177e4SLinus Torvalds out_unlock: 14118c3ee42eSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 14121da177e4SLinus Torvalds if (!err) 14138c3ee42eSAl Viro security_sb_post_addmount(mnt, path); 14141da177e4SLinus Torvalds return err; 14151da177e4SLinus Torvalds } 14161da177e4SLinus Torvalds 14171da177e4SLinus Torvalds /* 141807b20889SRam Pai * recursively change the type of the mountpoint. 14192dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 142007b20889SRam Pai */ 14212dafe1c4SEric Sandeen static noinline int do_change_type(struct nameidata *nd, int flag) 142207b20889SRam Pai { 14234ac91378SJan Blunck struct vfsmount *m, *mnt = nd->path.mnt; 142407b20889SRam Pai int recurse = flag & MS_REC; 142507b20889SRam Pai int type = flag & ~MS_REC; 1426719f5d7fSMiklos Szeredi int err = 0; 142707b20889SRam Pai 1428ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1429ee6f9582SMiklos Szeredi return -EPERM; 1430ee6f9582SMiklos Szeredi 14314ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 143207b20889SRam Pai return -EINVAL; 143307b20889SRam Pai 143407b20889SRam Pai down_write(&namespace_sem); 1435719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1436719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1437719f5d7fSMiklos Szeredi if (err) 1438719f5d7fSMiklos Szeredi goto out_unlock; 1439719f5d7fSMiklos Szeredi } 1440719f5d7fSMiklos Szeredi 144107b20889SRam Pai spin_lock(&vfsmount_lock); 144207b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 144307b20889SRam Pai change_mnt_propagation(m, type); 144407b20889SRam Pai spin_unlock(&vfsmount_lock); 1445719f5d7fSMiklos Szeredi 1446719f5d7fSMiklos Szeredi out_unlock: 144707b20889SRam Pai up_write(&namespace_sem); 1448719f5d7fSMiklos Szeredi return err; 144907b20889SRam Pai } 145007b20889SRam Pai 145107b20889SRam Pai /* 14521da177e4SLinus Torvalds * do loopback mount. 14532dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 14541da177e4SLinus Torvalds */ 14552dafe1c4SEric Sandeen static noinline int do_loopback(struct nameidata *nd, char *old_name, 14562dafe1c4SEric Sandeen int recurse) 14571da177e4SLinus Torvalds { 14581da177e4SLinus Torvalds struct nameidata old_nd; 14591da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 14601da177e4SLinus Torvalds int err = mount_is_safe(nd); 14611da177e4SLinus Torvalds if (err) 14621da177e4SLinus Torvalds return err; 14631da177e4SLinus Torvalds if (!old_name || !*old_name) 14641da177e4SLinus Torvalds return -EINVAL; 14651da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 14661da177e4SLinus Torvalds if (err) 14671da177e4SLinus Torvalds return err; 14681da177e4SLinus Torvalds 1469390c6843SRam Pai down_write(&namespace_sem); 14701da177e4SLinus Torvalds err = -EINVAL; 14714ac91378SJan Blunck if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) 14729676f0c6SRam Pai goto out; 14739676f0c6SRam Pai 14744ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 1475ccd48bc7SAl Viro goto out; 1476ccd48bc7SAl Viro 14771da177e4SLinus Torvalds err = -ENOMEM; 14781da177e4SLinus Torvalds if (recurse) 14794ac91378SJan Blunck mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); 14801da177e4SLinus Torvalds else 14814ac91378SJan Blunck mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); 14821da177e4SLinus Torvalds 1483ccd48bc7SAl Viro if (!mnt) 1484ccd48bc7SAl Viro goto out; 1485ccd48bc7SAl Viro 14868c3ee42eSAl Viro err = graft_tree(mnt, &nd->path); 14871da177e4SLinus Torvalds if (err) { 148870fbcdf4SRam Pai LIST_HEAD(umount_list); 14891da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1490a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 14911da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 149270fbcdf4SRam Pai release_mounts(&umount_list); 14935b83d2c5SRam Pai } 14941da177e4SLinus Torvalds 1495ccd48bc7SAl Viro out: 1496390c6843SRam Pai up_write(&namespace_sem); 14971d957f9bSJan Blunck path_put(&old_nd.path); 14981da177e4SLinus Torvalds return err; 14991da177e4SLinus Torvalds } 15001da177e4SLinus Torvalds 15012e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 15022e4b7fcdSDave Hansen { 15032e4b7fcdSDave Hansen int error = 0; 15042e4b7fcdSDave Hansen int readonly_request = 0; 15052e4b7fcdSDave Hansen 15062e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 15072e4b7fcdSDave Hansen readonly_request = 1; 15082e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 15092e4b7fcdSDave Hansen return 0; 15102e4b7fcdSDave Hansen 15112e4b7fcdSDave Hansen if (readonly_request) 15122e4b7fcdSDave Hansen error = mnt_make_readonly(mnt); 15132e4b7fcdSDave Hansen else 15142e4b7fcdSDave Hansen __mnt_unmake_readonly(mnt); 15152e4b7fcdSDave Hansen return error; 15162e4b7fcdSDave Hansen } 15172e4b7fcdSDave Hansen 15181da177e4SLinus Torvalds /* 15191da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 15201da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 15211da177e4SLinus Torvalds * on it - tough luck. 15222dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 15231da177e4SLinus Torvalds */ 15242dafe1c4SEric Sandeen static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, 15251da177e4SLinus Torvalds void *data) 15261da177e4SLinus Torvalds { 15271da177e4SLinus Torvalds int err; 15284ac91378SJan Blunck struct super_block *sb = nd->path.mnt->mnt_sb; 15291da177e4SLinus Torvalds 15301da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15311da177e4SLinus Torvalds return -EPERM; 15321da177e4SLinus Torvalds 15334ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 15341da177e4SLinus Torvalds return -EINVAL; 15351da177e4SLinus Torvalds 15364ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 15371da177e4SLinus Torvalds return -EINVAL; 15381da177e4SLinus Torvalds 15391da177e4SLinus Torvalds down_write(&sb->s_umount); 15402e4b7fcdSDave Hansen if (flags & MS_BIND) 15412e4b7fcdSDave Hansen err = change_mount_flags(nd->path.mnt, flags); 15422e4b7fcdSDave Hansen else 15431da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 15441da177e4SLinus Torvalds if (!err) 15454ac91378SJan Blunck nd->path.mnt->mnt_flags = mnt_flags; 15461da177e4SLinus Torvalds up_write(&sb->s_umount); 15471da177e4SLinus Torvalds if (!err) 15484ac91378SJan Blunck security_sb_post_remount(nd->path.mnt, flags, data); 15491da177e4SLinus Torvalds return err; 15501da177e4SLinus Torvalds } 15511da177e4SLinus Torvalds 15529676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 15539676f0c6SRam Pai { 15549676f0c6SRam Pai struct vfsmount *p; 15559676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 15569676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 15579676f0c6SRam Pai return 1; 15589676f0c6SRam Pai } 15599676f0c6SRam Pai return 0; 15609676f0c6SRam Pai } 15619676f0c6SRam Pai 15622dafe1c4SEric Sandeen /* 15632dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 15642dafe1c4SEric Sandeen */ 15652dafe1c4SEric Sandeen static noinline int do_move_mount(struct nameidata *nd, char *old_name) 15661da177e4SLinus Torvalds { 15671a390689SAl Viro struct nameidata old_nd; 15681a390689SAl Viro struct path parent_path; 15691da177e4SLinus Torvalds struct vfsmount *p; 15701da177e4SLinus Torvalds int err = 0; 15711da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15721da177e4SLinus Torvalds return -EPERM; 15731da177e4SLinus Torvalds if (!old_name || !*old_name) 15741da177e4SLinus Torvalds return -EINVAL; 15751da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 15761da177e4SLinus Torvalds if (err) 15771da177e4SLinus Torvalds return err; 15781da177e4SLinus Torvalds 1579390c6843SRam Pai down_write(&namespace_sem); 15804ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 15814ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 15821da177e4SLinus Torvalds ; 15831da177e4SLinus Torvalds err = -EINVAL; 15844ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 15851da177e4SLinus Torvalds goto out; 15861da177e4SLinus Torvalds 15871da177e4SLinus Torvalds err = -ENOENT; 15884ac91378SJan Blunck mutex_lock(&nd->path.dentry->d_inode->i_mutex); 15894ac91378SJan Blunck if (IS_DEADDIR(nd->path.dentry->d_inode)) 15901da177e4SLinus Torvalds goto out1; 15911da177e4SLinus Torvalds 15924ac91378SJan Blunck if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) 159321444403SRam Pai goto out1; 15941da177e4SLinus Torvalds 15951da177e4SLinus Torvalds err = -EINVAL; 15964ac91378SJan Blunck if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) 159721444403SRam Pai goto out1; 15981da177e4SLinus Torvalds 15994ac91378SJan Blunck if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) 160021444403SRam Pai goto out1; 16011da177e4SLinus Torvalds 16024ac91378SJan Blunck if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != 16034ac91378SJan Blunck S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) 160421444403SRam Pai goto out1; 160521444403SRam Pai /* 160621444403SRam Pai * Don't move a mount residing in a shared parent. 160721444403SRam Pai */ 16084ac91378SJan Blunck if (old_nd.path.mnt->mnt_parent && 16094ac91378SJan Blunck IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) 161021444403SRam Pai goto out1; 16119676f0c6SRam Pai /* 16129676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 16139676f0c6SRam Pai * mount which is shared. 16149676f0c6SRam Pai */ 16154ac91378SJan Blunck if (IS_MNT_SHARED(nd->path.mnt) && 16164ac91378SJan Blunck tree_contains_unbindable(old_nd.path.mnt)) 16179676f0c6SRam Pai goto out1; 16181da177e4SLinus Torvalds err = -ELOOP; 16194ac91378SJan Blunck for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) 16204ac91378SJan Blunck if (p == old_nd.path.mnt) 162121444403SRam Pai goto out1; 16221da177e4SLinus Torvalds 16231a390689SAl Viro err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); 16244ac91378SJan Blunck if (err) 162521444403SRam Pai goto out1; 16261da177e4SLinus Torvalds 16271da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 16281da177e4SLinus Torvalds * automatically */ 16294ac91378SJan Blunck list_del_init(&old_nd.path.mnt->mnt_expire); 16301da177e4SLinus Torvalds out1: 16314ac91378SJan Blunck mutex_unlock(&nd->path.dentry->d_inode->i_mutex); 16321da177e4SLinus Torvalds out: 1633390c6843SRam Pai up_write(&namespace_sem); 16341da177e4SLinus Torvalds if (!err) 16351a390689SAl Viro path_put(&parent_path); 16361d957f9bSJan Blunck path_put(&old_nd.path); 16371da177e4SLinus Torvalds return err; 16381da177e4SLinus Torvalds } 16391da177e4SLinus Torvalds 16401da177e4SLinus Torvalds /* 16411da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 16421da177e4SLinus Torvalds * namespace's tree 16432dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 16441da177e4SLinus Torvalds */ 16452dafe1c4SEric Sandeen static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, 16461da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 16471da177e4SLinus Torvalds { 16481da177e4SLinus Torvalds struct vfsmount *mnt; 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 16511da177e4SLinus Torvalds return -EINVAL; 16521da177e4SLinus Torvalds 16531da177e4SLinus Torvalds /* we need capabilities... */ 16541da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 16551da177e4SLinus Torvalds return -EPERM; 16561da177e4SLinus Torvalds 16571da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 16581da177e4SLinus Torvalds if (IS_ERR(mnt)) 16591da177e4SLinus Torvalds return PTR_ERR(mnt); 16601da177e4SLinus Torvalds 16611da177e4SLinus Torvalds return do_add_mount(mnt, nd, mnt_flags, NULL); 16621da177e4SLinus Torvalds } 16631da177e4SLinus Torvalds 16641da177e4SLinus Torvalds /* 16651da177e4SLinus Torvalds * add a mount into a namespace's mount tree 16661da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 16671da177e4SLinus Torvalds */ 16681da177e4SLinus Torvalds int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 16691da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 16701da177e4SLinus Torvalds { 16711da177e4SLinus Torvalds int err; 16721da177e4SLinus Torvalds 1673390c6843SRam Pai down_write(&namespace_sem); 16741da177e4SLinus Torvalds /* Something was mounted here while we slept */ 16754ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 16764ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 16771da177e4SLinus Torvalds ; 16781da177e4SLinus Torvalds err = -EINVAL; 16794ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 16801da177e4SLinus Torvalds goto unlock; 16811da177e4SLinus Torvalds 16821da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 16831da177e4SLinus Torvalds err = -EBUSY; 16844ac91378SJan Blunck if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && 16854ac91378SJan Blunck nd->path.mnt->mnt_root == nd->path.dentry) 16861da177e4SLinus Torvalds goto unlock; 16871da177e4SLinus Torvalds 16881da177e4SLinus Torvalds err = -EINVAL; 16891da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 16901da177e4SLinus Torvalds goto unlock; 16911da177e4SLinus Torvalds 16921da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 16938c3ee42eSAl Viro if ((err = graft_tree(newmnt, &nd->path))) 16945b83d2c5SRam Pai goto unlock; 16951da177e4SLinus Torvalds 16966758f953SAl Viro if (fslist) /* add to the specified expiration list */ 169755e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 16986758f953SAl Viro 1699390c6843SRam Pai up_write(&namespace_sem); 17005b83d2c5SRam Pai return 0; 17011da177e4SLinus Torvalds 17021da177e4SLinus Torvalds unlock: 1703390c6843SRam Pai up_write(&namespace_sem); 17041da177e4SLinus Torvalds mntput(newmnt); 17051da177e4SLinus Torvalds return err; 17061da177e4SLinus Torvalds } 17071da177e4SLinus Torvalds 17081da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 17091da177e4SLinus Torvalds 17105528f911STrond Myklebust /* 17111da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 17121da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 17131da177e4SLinus Torvalds * here 17141da177e4SLinus Torvalds */ 17151da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 17161da177e4SLinus Torvalds { 17171da177e4SLinus Torvalds struct vfsmount *mnt, *next; 17181da177e4SLinus Torvalds LIST_HEAD(graveyard); 1719bcc5c7d2SAl Viro LIST_HEAD(umounts); 17201da177e4SLinus Torvalds 17211da177e4SLinus Torvalds if (list_empty(mounts)) 17221da177e4SLinus Torvalds return; 17231da177e4SLinus Torvalds 1724bcc5c7d2SAl Viro down_write(&namespace_sem); 17251da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 17261da177e4SLinus Torvalds 17271da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 17281da177e4SLinus Torvalds * following criteria: 17291da177e4SLinus Torvalds * - only referenced by its parent vfsmount 17301da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 17311da177e4SLinus Torvalds * cleared by mntput()) 17321da177e4SLinus Torvalds */ 173355e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 17341da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 1735bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 17361da177e4SLinus Torvalds continue; 173755e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 17381da177e4SLinus Torvalds } 1739bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1740bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 1741bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1742bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 1743bcc5c7d2SAl Viro } 17441da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1745bcc5c7d2SAl Viro up_write(&namespace_sem); 1746bcc5c7d2SAl Viro 1747bcc5c7d2SAl Viro release_mounts(&umounts); 17481da177e4SLinus Torvalds } 17491da177e4SLinus Torvalds 17501da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 17511da177e4SLinus Torvalds 17521da177e4SLinus Torvalds /* 17535528f911STrond Myklebust * Ripoff of 'select_parent()' 17545528f911STrond Myklebust * 17555528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 17565528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 17575528f911STrond Myklebust */ 17585528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 17595528f911STrond Myklebust { 17605528f911STrond Myklebust struct vfsmount *this_parent = parent; 17615528f911STrond Myklebust struct list_head *next; 17625528f911STrond Myklebust int found = 0; 17635528f911STrond Myklebust 17645528f911STrond Myklebust repeat: 17655528f911STrond Myklebust next = this_parent->mnt_mounts.next; 17665528f911STrond Myklebust resume: 17675528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 17685528f911STrond Myklebust struct list_head *tmp = next; 17695528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 17705528f911STrond Myklebust 17715528f911STrond Myklebust next = tmp->next; 17725528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 17735528f911STrond Myklebust continue; 17745528f911STrond Myklebust /* 17755528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 17765528f911STrond Myklebust */ 17775528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 17785528f911STrond Myklebust this_parent = mnt; 17795528f911STrond Myklebust goto repeat; 17805528f911STrond Myklebust } 17815528f911STrond Myklebust 17825528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 17835528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 17845528f911STrond Myklebust found++; 17855528f911STrond Myklebust } 17865528f911STrond Myklebust } 17875528f911STrond Myklebust /* 17885528f911STrond Myklebust * All done at this level ... ascend and resume the search 17895528f911STrond Myklebust */ 17905528f911STrond Myklebust if (this_parent != parent) { 17915528f911STrond Myklebust next = this_parent->mnt_child.next; 17925528f911STrond Myklebust this_parent = this_parent->mnt_parent; 17935528f911STrond Myklebust goto resume; 17945528f911STrond Myklebust } 17955528f911STrond Myklebust return found; 17965528f911STrond Myklebust } 17975528f911STrond Myklebust 17985528f911STrond Myklebust /* 17995528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 18005528f911STrond Myklebust * submounts of a specific parent mountpoint 18015528f911STrond Myklebust */ 1802c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 18035528f911STrond Myklebust { 18045528f911STrond Myklebust LIST_HEAD(graveyard); 1805c35038beSAl Viro struct vfsmount *m; 18065528f911STrond Myklebust 18075528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 1808c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 1809bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1810c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 1811bcc5c7d2SAl Viro mnt_expire); 1812bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1813c35038beSAl Viro umount_tree(mnt, 1, umounts); 1814bcc5c7d2SAl Viro } 1815bcc5c7d2SAl Viro } 18165528f911STrond Myklebust } 18175528f911STrond Myklebust 18185528f911STrond Myklebust /* 18191da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 18201da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 18211da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 18221da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 18231da177e4SLinus Torvalds */ 1824b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1825b58fed8bSRam Pai unsigned long n) 18261da177e4SLinus Torvalds { 18271da177e4SLinus Torvalds char *t = to; 18281da177e4SLinus Torvalds const char __user *f = from; 18291da177e4SLinus Torvalds char c; 18301da177e4SLinus Torvalds 18311da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 18321da177e4SLinus Torvalds return n; 18331da177e4SLinus Torvalds 18341da177e4SLinus Torvalds while (n) { 18351da177e4SLinus Torvalds if (__get_user(c, f)) { 18361da177e4SLinus Torvalds memset(t, 0, n); 18371da177e4SLinus Torvalds break; 18381da177e4SLinus Torvalds } 18391da177e4SLinus Torvalds *t++ = c; 18401da177e4SLinus Torvalds f++; 18411da177e4SLinus Torvalds n--; 18421da177e4SLinus Torvalds } 18431da177e4SLinus Torvalds return n; 18441da177e4SLinus Torvalds } 18451da177e4SLinus Torvalds 18461da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 18471da177e4SLinus Torvalds { 18481da177e4SLinus Torvalds int i; 18491da177e4SLinus Torvalds unsigned long page; 18501da177e4SLinus Torvalds unsigned long size; 18511da177e4SLinus Torvalds 18521da177e4SLinus Torvalds *where = 0; 18531da177e4SLinus Torvalds if (!data) 18541da177e4SLinus Torvalds return 0; 18551da177e4SLinus Torvalds 18561da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 18571da177e4SLinus Torvalds return -ENOMEM; 18581da177e4SLinus Torvalds 18591da177e4SLinus Torvalds /* We only care that *some* data at the address the user 18601da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 18611da177e4SLinus Torvalds * the remainder of the page. 18621da177e4SLinus Torvalds */ 18631da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 18641da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 18651da177e4SLinus Torvalds if (size > PAGE_SIZE) 18661da177e4SLinus Torvalds size = PAGE_SIZE; 18671da177e4SLinus Torvalds 18681da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 18691da177e4SLinus Torvalds if (!i) { 18701da177e4SLinus Torvalds free_page(page); 18711da177e4SLinus Torvalds return -EFAULT; 18721da177e4SLinus Torvalds } 18731da177e4SLinus Torvalds if (i != PAGE_SIZE) 18741da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 18751da177e4SLinus Torvalds *where = page; 18761da177e4SLinus Torvalds return 0; 18771da177e4SLinus Torvalds } 18781da177e4SLinus Torvalds 18791da177e4SLinus Torvalds /* 18801da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 18811da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 18821da177e4SLinus Torvalds * 18831da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 18841da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 18851da177e4SLinus Torvalds * information (or be NULL). 18861da177e4SLinus Torvalds * 18871da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 18881da177e4SLinus Torvalds * When the flags word was introduced its top half was required 18891da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 18901da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 18911da177e4SLinus Torvalds * and must be discarded. 18921da177e4SLinus Torvalds */ 18931da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 18941da177e4SLinus Torvalds unsigned long flags, void *data_page) 18951da177e4SLinus Torvalds { 18961da177e4SLinus Torvalds struct nameidata nd; 18971da177e4SLinus Torvalds int retval = 0; 18981da177e4SLinus Torvalds int mnt_flags = 0; 18991da177e4SLinus Torvalds 19001da177e4SLinus Torvalds /* Discard magic */ 19011da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 19021da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds /* Basic sanity checks */ 19051da177e4SLinus Torvalds 19061da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 19071da177e4SLinus Torvalds return -EINVAL; 19081da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 19091da177e4SLinus Torvalds return -EINVAL; 19101da177e4SLinus Torvalds 19111da177e4SLinus Torvalds if (data_page) 19121da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 19131da177e4SLinus Torvalds 19141da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 19151da177e4SLinus Torvalds if (flags & MS_NOSUID) 19161da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 19171da177e4SLinus Torvalds if (flags & MS_NODEV) 19181da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 19191da177e4SLinus Torvalds if (flags & MS_NOEXEC) 19201da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 1921fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 1922fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 1923fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 1924fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 192547ae32d6SValerie Henson if (flags & MS_RELATIME) 192647ae32d6SValerie Henson mnt_flags |= MNT_RELATIME; 19272e4b7fcdSDave Hansen if (flags & MS_RDONLY) 19282e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 1929fc33a7bbSChristoph Hellwig 1930fc33a7bbSChristoph Hellwig flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 19318bf9725cSPavel Emelyanov MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); 19321da177e4SLinus Torvalds 19331da177e4SLinus Torvalds /* ... and get the mountpoint */ 19341da177e4SLinus Torvalds retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 19351da177e4SLinus Torvalds if (retval) 19361da177e4SLinus Torvalds return retval; 19371da177e4SLinus Torvalds 1938b5266eb4SAl Viro retval = security_sb_mount(dev_name, &nd.path, 1939b5266eb4SAl Viro type_page, flags, data_page); 19401da177e4SLinus Torvalds if (retval) 19411da177e4SLinus Torvalds goto dput_out; 19421da177e4SLinus Torvalds 19431da177e4SLinus Torvalds if (flags & MS_REMOUNT) 19441da177e4SLinus Torvalds retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 19451da177e4SLinus Torvalds data_page); 19461da177e4SLinus Torvalds else if (flags & MS_BIND) 1947eee391a6SAndrew Morton retval = do_loopback(&nd, dev_name, flags & MS_REC); 19489676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 194907b20889SRam Pai retval = do_change_type(&nd, flags); 19501da177e4SLinus Torvalds else if (flags & MS_MOVE) 19511da177e4SLinus Torvalds retval = do_move_mount(&nd, dev_name); 19521da177e4SLinus Torvalds else 19531da177e4SLinus Torvalds retval = do_new_mount(&nd, type_page, flags, mnt_flags, 19541da177e4SLinus Torvalds dev_name, data_page); 19551da177e4SLinus Torvalds dput_out: 19561d957f9bSJan Blunck path_put(&nd.path); 19571da177e4SLinus Torvalds return retval; 19581da177e4SLinus Torvalds } 19591da177e4SLinus Torvalds 1960741a2951SJANAK DESAI /* 1961741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 1962741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 1963741a2951SJANAK DESAI */ 1964e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 19656b3286edSKirill Korotaev struct fs_struct *fs) 19661da177e4SLinus Torvalds { 19676b3286edSKirill Korotaev struct mnt_namespace *new_ns; 19681da177e4SLinus Torvalds struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 19691da177e4SLinus Torvalds struct vfsmount *p, *q; 19701da177e4SLinus Torvalds 19716b3286edSKirill Korotaev new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 19721da177e4SLinus Torvalds if (!new_ns) 1973467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM); 19741da177e4SLinus Torvalds 19751da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 19761da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 19775addc5ddSAl Viro init_waitqueue_head(&new_ns->poll); 19785addc5ddSAl Viro new_ns->event = 0; 19791da177e4SLinus Torvalds 1980390c6843SRam Pai down_write(&namespace_sem); 19811da177e4SLinus Torvalds /* First pass: copy the tree topology */ 19826b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 19839676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 19841da177e4SLinus Torvalds if (!new_ns->root) { 1985390c6843SRam Pai up_write(&namespace_sem); 19861da177e4SLinus Torvalds kfree(new_ns); 1987467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM);; 19881da177e4SLinus Torvalds } 19891da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 19901da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 19911da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 19921da177e4SLinus Torvalds 19931da177e4SLinus Torvalds /* 19941da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 19951da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 19961da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 19971da177e4SLinus Torvalds */ 19986b3286edSKirill Korotaev p = mnt_ns->root; 19991da177e4SLinus Torvalds q = new_ns->root; 20001da177e4SLinus Torvalds while (p) { 20016b3286edSKirill Korotaev q->mnt_ns = new_ns; 20021da177e4SLinus Torvalds if (fs) { 20036ac08c39SJan Blunck if (p == fs->root.mnt) { 20041da177e4SLinus Torvalds rootmnt = p; 20056ac08c39SJan Blunck fs->root.mnt = mntget(q); 20061da177e4SLinus Torvalds } 20076ac08c39SJan Blunck if (p == fs->pwd.mnt) { 20081da177e4SLinus Torvalds pwdmnt = p; 20096ac08c39SJan Blunck fs->pwd.mnt = mntget(q); 20101da177e4SLinus Torvalds } 20116ac08c39SJan Blunck if (p == fs->altroot.mnt) { 20121da177e4SLinus Torvalds altrootmnt = p; 20136ac08c39SJan Blunck fs->altroot.mnt = mntget(q); 20141da177e4SLinus Torvalds } 20151da177e4SLinus Torvalds } 20166b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 20171da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 20181da177e4SLinus Torvalds } 2019390c6843SRam Pai up_write(&namespace_sem); 20201da177e4SLinus Torvalds 20211da177e4SLinus Torvalds if (rootmnt) 20221da177e4SLinus Torvalds mntput(rootmnt); 20231da177e4SLinus Torvalds if (pwdmnt) 20241da177e4SLinus Torvalds mntput(pwdmnt); 20251da177e4SLinus Torvalds if (altrootmnt) 20261da177e4SLinus Torvalds mntput(altrootmnt); 20271da177e4SLinus Torvalds 2028741a2951SJANAK DESAI return new_ns; 2029741a2951SJANAK DESAI } 2030741a2951SJANAK DESAI 2031213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2032e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 2033741a2951SJANAK DESAI { 20346b3286edSKirill Korotaev struct mnt_namespace *new_ns; 2035741a2951SJANAK DESAI 2036e3222c4eSBadari Pulavarty BUG_ON(!ns); 20376b3286edSKirill Korotaev get_mnt_ns(ns); 2038741a2951SJANAK DESAI 2039741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 2040e3222c4eSBadari Pulavarty return ns; 2041741a2951SJANAK DESAI 2042e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 2043741a2951SJANAK DESAI 20446b3286edSKirill Korotaev put_mnt_ns(ns); 2045e3222c4eSBadari Pulavarty return new_ns; 20461da177e4SLinus Torvalds } 20471da177e4SLinus Torvalds 20481da177e4SLinus Torvalds asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 20491da177e4SLinus Torvalds char __user * type, unsigned long flags, 20501da177e4SLinus Torvalds void __user * data) 20511da177e4SLinus Torvalds { 20521da177e4SLinus Torvalds int retval; 20531da177e4SLinus Torvalds unsigned long data_page; 20541da177e4SLinus Torvalds unsigned long type_page; 20551da177e4SLinus Torvalds unsigned long dev_page; 20561da177e4SLinus Torvalds char *dir_page; 20571da177e4SLinus Torvalds 20581da177e4SLinus Torvalds retval = copy_mount_options(type, &type_page); 20591da177e4SLinus Torvalds if (retval < 0) 20601da177e4SLinus Torvalds return retval; 20611da177e4SLinus Torvalds 20621da177e4SLinus Torvalds dir_page = getname(dir_name); 20631da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 20641da177e4SLinus Torvalds if (IS_ERR(dir_page)) 20651da177e4SLinus Torvalds goto out1; 20661da177e4SLinus Torvalds 20671da177e4SLinus Torvalds retval = copy_mount_options(dev_name, &dev_page); 20681da177e4SLinus Torvalds if (retval < 0) 20691da177e4SLinus Torvalds goto out2; 20701da177e4SLinus Torvalds 20711da177e4SLinus Torvalds retval = copy_mount_options(data, &data_page); 20721da177e4SLinus Torvalds if (retval < 0) 20731da177e4SLinus Torvalds goto out3; 20741da177e4SLinus Torvalds 20751da177e4SLinus Torvalds lock_kernel(); 20761da177e4SLinus Torvalds retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 20771da177e4SLinus Torvalds flags, (void *)data_page); 20781da177e4SLinus Torvalds unlock_kernel(); 20791da177e4SLinus Torvalds free_page(data_page); 20801da177e4SLinus Torvalds 20811da177e4SLinus Torvalds out3: 20821da177e4SLinus Torvalds free_page(dev_page); 20831da177e4SLinus Torvalds out2: 20841da177e4SLinus Torvalds putname(dir_page); 20851da177e4SLinus Torvalds out1: 20861da177e4SLinus Torvalds free_page(type_page); 20871da177e4SLinus Torvalds return retval; 20881da177e4SLinus Torvalds } 20891da177e4SLinus Torvalds 20901da177e4SLinus Torvalds /* 20911da177e4SLinus Torvalds * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 20921da177e4SLinus Torvalds * It can block. Requires the big lock held. 20931da177e4SLinus Torvalds */ 2094ac748a09SJan Blunck void set_fs_root(struct fs_struct *fs, struct path *path) 20951da177e4SLinus Torvalds { 20966ac08c39SJan Blunck struct path old_root; 20976ac08c39SJan Blunck 20981da177e4SLinus Torvalds write_lock(&fs->lock); 20991da177e4SLinus Torvalds old_root = fs->root; 2100ac748a09SJan Blunck fs->root = *path; 2101ac748a09SJan Blunck path_get(path); 21021da177e4SLinus Torvalds write_unlock(&fs->lock); 21036ac08c39SJan Blunck if (old_root.dentry) 21046ac08c39SJan Blunck path_put(&old_root); 21051da177e4SLinus Torvalds } 21061da177e4SLinus Torvalds 21071da177e4SLinus Torvalds /* 21081da177e4SLinus Torvalds * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 21091da177e4SLinus Torvalds * It can block. Requires the big lock held. 21101da177e4SLinus Torvalds */ 2111ac748a09SJan Blunck void set_fs_pwd(struct fs_struct *fs, struct path *path) 21121da177e4SLinus Torvalds { 21136ac08c39SJan Blunck struct path old_pwd; 21141da177e4SLinus Torvalds 21151da177e4SLinus Torvalds write_lock(&fs->lock); 21161da177e4SLinus Torvalds old_pwd = fs->pwd; 2117ac748a09SJan Blunck fs->pwd = *path; 2118ac748a09SJan Blunck path_get(path); 21191da177e4SLinus Torvalds write_unlock(&fs->lock); 21201da177e4SLinus Torvalds 21216ac08c39SJan Blunck if (old_pwd.dentry) 21226ac08c39SJan Blunck path_put(&old_pwd); 21231da177e4SLinus Torvalds } 21241da177e4SLinus Torvalds 21251a390689SAl Viro static void chroot_fs_refs(struct path *old_root, struct path *new_root) 21261da177e4SLinus Torvalds { 21271da177e4SLinus Torvalds struct task_struct *g, *p; 21281da177e4SLinus Torvalds struct fs_struct *fs; 21291da177e4SLinus Torvalds 21301da177e4SLinus Torvalds read_lock(&tasklist_lock); 21311da177e4SLinus Torvalds do_each_thread(g, p) { 21321da177e4SLinus Torvalds task_lock(p); 21331da177e4SLinus Torvalds fs = p->fs; 21341da177e4SLinus Torvalds if (fs) { 21351da177e4SLinus Torvalds atomic_inc(&fs->count); 21361da177e4SLinus Torvalds task_unlock(p); 21371a390689SAl Viro if (fs->root.dentry == old_root->dentry 21381a390689SAl Viro && fs->root.mnt == old_root->mnt) 21391a390689SAl Viro set_fs_root(fs, new_root); 21401a390689SAl Viro if (fs->pwd.dentry == old_root->dentry 21411a390689SAl Viro && fs->pwd.mnt == old_root->mnt) 21421a390689SAl Viro set_fs_pwd(fs, new_root); 21431da177e4SLinus Torvalds put_fs_struct(fs); 21441da177e4SLinus Torvalds } else 21451da177e4SLinus Torvalds task_unlock(p); 21461da177e4SLinus Torvalds } while_each_thread(g, p); 21471da177e4SLinus Torvalds read_unlock(&tasklist_lock); 21481da177e4SLinus Torvalds } 21491da177e4SLinus Torvalds 21501da177e4SLinus Torvalds /* 21511da177e4SLinus Torvalds * pivot_root Semantics: 21521da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 21531da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 21541da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 21551da177e4SLinus Torvalds * 21561da177e4SLinus Torvalds * Restrictions: 21571da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 21581da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 21591da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 21601da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 21611da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 21621da177e4SLinus Torvalds * 21634a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 21644a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 21654a0d11faSNeil Brown * in this situation. 21664a0d11faSNeil Brown * 21671da177e4SLinus Torvalds * Notes: 21681da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 21691da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 21701da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 21711da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 21721da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 21731da177e4SLinus Torvalds * first. 21741da177e4SLinus Torvalds */ 2175b58fed8bSRam Pai asmlinkage long sys_pivot_root(const char __user * new_root, 2176b58fed8bSRam Pai const char __user * put_old) 21771da177e4SLinus Torvalds { 21781da177e4SLinus Torvalds struct vfsmount *tmp; 21798c3ee42eSAl Viro struct nameidata new_nd, old_nd; 21808c3ee42eSAl Viro struct path parent_path, root_parent, root; 21811da177e4SLinus Torvalds int error; 21821da177e4SLinus Torvalds 21831da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 21841da177e4SLinus Torvalds return -EPERM; 21851da177e4SLinus Torvalds 2186b58fed8bSRam Pai error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, 2187b58fed8bSRam Pai &new_nd); 21881da177e4SLinus Torvalds if (error) 21891da177e4SLinus Torvalds goto out0; 21901da177e4SLinus Torvalds error = -EINVAL; 21914ac91378SJan Blunck if (!check_mnt(new_nd.path.mnt)) 21921da177e4SLinus Torvalds goto out1; 21931da177e4SLinus Torvalds 21941da177e4SLinus Torvalds error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); 21951da177e4SLinus Torvalds if (error) 21961da177e4SLinus Torvalds goto out1; 21971da177e4SLinus Torvalds 2198b5266eb4SAl Viro error = security_sb_pivotroot(&old_nd.path, &new_nd.path); 21991da177e4SLinus Torvalds if (error) { 22001d957f9bSJan Blunck path_put(&old_nd.path); 22011da177e4SLinus Torvalds goto out1; 22021da177e4SLinus Torvalds } 22031da177e4SLinus Torvalds 22041da177e4SLinus Torvalds read_lock(¤t->fs->lock); 22058c3ee42eSAl Viro root = current->fs->root; 22066ac08c39SJan Blunck path_get(¤t->fs->root); 22071da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 2208390c6843SRam Pai down_write(&namespace_sem); 22094ac91378SJan Blunck mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); 22101da177e4SLinus Torvalds error = -EINVAL; 22114ac91378SJan Blunck if (IS_MNT_SHARED(old_nd.path.mnt) || 22124ac91378SJan Blunck IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || 22138c3ee42eSAl Viro IS_MNT_SHARED(root.mnt->mnt_parent)) 221421444403SRam Pai goto out2; 22158c3ee42eSAl Viro if (!check_mnt(root.mnt)) 22161da177e4SLinus Torvalds goto out2; 22171da177e4SLinus Torvalds error = -ENOENT; 22184ac91378SJan Blunck if (IS_DEADDIR(new_nd.path.dentry->d_inode)) 22191da177e4SLinus Torvalds goto out2; 22204ac91378SJan Blunck if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) 22211da177e4SLinus Torvalds goto out2; 22224ac91378SJan Blunck if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) 22231da177e4SLinus Torvalds goto out2; 22241da177e4SLinus Torvalds error = -EBUSY; 22258c3ee42eSAl Viro if (new_nd.path.mnt == root.mnt || 22268c3ee42eSAl Viro old_nd.path.mnt == root.mnt) 22271da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 22281da177e4SLinus Torvalds error = -EINVAL; 22298c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 22301da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 22318c3ee42eSAl Viro if (root.mnt->mnt_parent == root.mnt) 22320bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 22334ac91378SJan Blunck if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) 22341da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 22354ac91378SJan Blunck if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) 22360bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 22374ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 22384ac91378SJan Blunck tmp = old_nd.path.mnt; 22391da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 22404ac91378SJan Blunck if (tmp != new_nd.path.mnt) { 22411da177e4SLinus Torvalds for (;;) { 22421da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 22431da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 22444ac91378SJan Blunck if (tmp->mnt_parent == new_nd.path.mnt) 22451da177e4SLinus Torvalds break; 22461da177e4SLinus Torvalds tmp = tmp->mnt_parent; 22471da177e4SLinus Torvalds } 22484ac91378SJan Blunck if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) 22491da177e4SLinus Torvalds goto out3; 22504ac91378SJan Blunck } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) 22511da177e4SLinus Torvalds goto out3; 22521a390689SAl Viro detach_mnt(new_nd.path.mnt, &parent_path); 22538c3ee42eSAl Viro detach_mnt(root.mnt, &root_parent); 22544ac91378SJan Blunck /* mount old root on put_old */ 22558c3ee42eSAl Viro attach_mnt(root.mnt, &old_nd.path); 22564ac91378SJan Blunck /* mount new_root on / */ 22574ac91378SJan Blunck attach_mnt(new_nd.path.mnt, &root_parent); 22586b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 22591da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 22608c3ee42eSAl Viro chroot_fs_refs(&root, &new_nd.path); 22618c3ee42eSAl Viro security_sb_post_pivotroot(&root, &new_nd.path); 22621da177e4SLinus Torvalds error = 0; 22631a390689SAl Viro path_put(&root_parent); 22641a390689SAl Viro path_put(&parent_path); 22651da177e4SLinus Torvalds out2: 22664ac91378SJan Blunck mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); 2267390c6843SRam Pai up_write(&namespace_sem); 22688c3ee42eSAl Viro path_put(&root); 22691d957f9bSJan Blunck path_put(&old_nd.path); 22701da177e4SLinus Torvalds out1: 22711d957f9bSJan Blunck path_put(&new_nd.path); 22721da177e4SLinus Torvalds out0: 22731da177e4SLinus Torvalds return error; 22741da177e4SLinus Torvalds out3: 22751da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 22761da177e4SLinus Torvalds goto out2; 22771da177e4SLinus Torvalds } 22781da177e4SLinus Torvalds 22791da177e4SLinus Torvalds static void __init init_mount_tree(void) 22801da177e4SLinus Torvalds { 22811da177e4SLinus Torvalds struct vfsmount *mnt; 22826b3286edSKirill Korotaev struct mnt_namespace *ns; 2283ac748a09SJan Blunck struct path root; 22841da177e4SLinus Torvalds 22851da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 22861da177e4SLinus Torvalds if (IS_ERR(mnt)) 22871da177e4SLinus Torvalds panic("Can't create rootfs"); 22886b3286edSKirill Korotaev ns = kmalloc(sizeof(*ns), GFP_KERNEL); 22896b3286edSKirill Korotaev if (!ns) 22901da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 22916b3286edSKirill Korotaev atomic_set(&ns->count, 1); 22926b3286edSKirill Korotaev INIT_LIST_HEAD(&ns->list); 22936b3286edSKirill Korotaev init_waitqueue_head(&ns->poll); 22946b3286edSKirill Korotaev ns->event = 0; 22956b3286edSKirill Korotaev list_add(&mnt->mnt_list, &ns->list); 22966b3286edSKirill Korotaev ns->root = mnt; 22976b3286edSKirill Korotaev mnt->mnt_ns = ns; 22981da177e4SLinus Torvalds 22996b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 23006b3286edSKirill Korotaev get_mnt_ns(ns); 23011da177e4SLinus Torvalds 2302ac748a09SJan Blunck root.mnt = ns->root; 2303ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2304ac748a09SJan Blunck 2305ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2306ac748a09SJan Blunck set_fs_root(current->fs, &root); 23071da177e4SLinus Torvalds } 23081da177e4SLinus Torvalds 230974bf17cfSDenis Cheng void __init mnt_init(void) 23101da177e4SLinus Torvalds { 231113f14b4dSEric Dumazet unsigned u; 231215a67dd8SRandy Dunlap int err; 23131da177e4SLinus Torvalds 2314390c6843SRam Pai init_rwsem(&namespace_sem); 2315390c6843SRam Pai 23161da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 231720c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 23181da177e4SLinus Torvalds 2319b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 23201da177e4SLinus Torvalds 23211da177e4SLinus Torvalds if (!mount_hashtable) 23221da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 23231da177e4SLinus Torvalds 232413f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 23251da177e4SLinus Torvalds 232613f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 232713f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 23281da177e4SLinus Torvalds 232915a67dd8SRandy Dunlap err = sysfs_init(); 233015a67dd8SRandy Dunlap if (err) 233115a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 233215a67dd8SRandy Dunlap __FUNCTION__, err); 233300d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 233400d26666SGreg Kroah-Hartman if (!fs_kobj) 233500d26666SGreg Kroah-Hartman printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); 23361da177e4SLinus Torvalds init_rootfs(); 23371da177e4SLinus Torvalds init_mount_tree(); 23381da177e4SLinus Torvalds } 23391da177e4SLinus Torvalds 23406b3286edSKirill Korotaev void __put_mnt_ns(struct mnt_namespace *ns) 23411da177e4SLinus Torvalds { 23426b3286edSKirill Korotaev struct vfsmount *root = ns->root; 234370fbcdf4SRam Pai LIST_HEAD(umount_list); 23446b3286edSKirill Korotaev ns->root = NULL; 23451ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 2346390c6843SRam Pai down_write(&namespace_sem); 23471da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 2348a05964f3SRam Pai umount_tree(root, 0, &umount_list); 23491da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 2350390c6843SRam Pai up_write(&namespace_sem); 235170fbcdf4SRam Pai release_mounts(&umount_list); 23526b3286edSKirill Korotaev kfree(ns); 23531da177e4SLinus Torvalds } 2354