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> 305ad4e53bSAl Viro #include <linux/fs_struct.h> 311da177e4SLinus Torvalds #include <asm/uaccess.h> 321da177e4SLinus Torvalds #include <asm/unistd.h> 3307b20889SRam Pai #include "pnode.h" 34948730b0SAdrian Bunk #include "internal.h" 351da177e4SLinus Torvalds 3613f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 3713f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 3813f14b4dSEric Dumazet 391da177e4SLinus Torvalds /* spinlock for vfsmount related operations, inplace of dcache_lock */ 401da177e4SLinus Torvalds __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 411da177e4SLinus Torvalds 425addc5ddSAl Viro static int event; 4373cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 44719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 455addc5ddSAl Viro 46fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 47e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 48390c6843SRam Pai static struct rw_semaphore namespace_sem; 491da177e4SLinus Torvalds 50f87fd4c2SMiklos Szeredi /* /sys/fs */ 5100d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 5200d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 53f87fd4c2SMiklos Szeredi 541da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 551da177e4SLinus Torvalds { 561da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 571da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 5813f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 5913f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 601da177e4SLinus Torvalds } 611da177e4SLinus Torvalds 623d733633SDave Hansen #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 633d733633SDave Hansen 6473cd49ecSMiklos Szeredi /* allocation is serialized by namespace_sem */ 6573cd49ecSMiklos Szeredi static int mnt_alloc_id(struct vfsmount *mnt) 6673cd49ecSMiklos Szeredi { 6773cd49ecSMiklos Szeredi int res; 6873cd49ecSMiklos Szeredi 6973cd49ecSMiklos Szeredi retry: 7073cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 7173cd49ecSMiklos Szeredi spin_lock(&vfsmount_lock); 7273cd49ecSMiklos Szeredi res = ida_get_new(&mnt_id_ida, &mnt->mnt_id); 7373cd49ecSMiklos Szeredi spin_unlock(&vfsmount_lock); 7473cd49ecSMiklos Szeredi if (res == -EAGAIN) 7573cd49ecSMiklos Szeredi goto retry; 7673cd49ecSMiklos Szeredi 7773cd49ecSMiklos Szeredi return res; 7873cd49ecSMiklos Szeredi } 7973cd49ecSMiklos Szeredi 8073cd49ecSMiklos Szeredi static void mnt_free_id(struct vfsmount *mnt) 8173cd49ecSMiklos Szeredi { 8273cd49ecSMiklos Szeredi spin_lock(&vfsmount_lock); 8373cd49ecSMiklos Szeredi ida_remove(&mnt_id_ida, mnt->mnt_id); 8473cd49ecSMiklos Szeredi spin_unlock(&vfsmount_lock); 8573cd49ecSMiklos Szeredi } 8673cd49ecSMiklos Szeredi 87719f5d7fSMiklos Szeredi /* 88719f5d7fSMiklos Szeredi * Allocate a new peer group ID 89719f5d7fSMiklos Szeredi * 90719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 91719f5d7fSMiklos Szeredi */ 92719f5d7fSMiklos Szeredi static int mnt_alloc_group_id(struct vfsmount *mnt) 93719f5d7fSMiklos Szeredi { 94719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 95719f5d7fSMiklos Szeredi return -ENOMEM; 96719f5d7fSMiklos Szeredi 97719f5d7fSMiklos Szeredi return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id); 98719f5d7fSMiklos Szeredi } 99719f5d7fSMiklos Szeredi 100719f5d7fSMiklos Szeredi /* 101719f5d7fSMiklos Szeredi * Release a peer group ID 102719f5d7fSMiklos Szeredi */ 103719f5d7fSMiklos Szeredi void mnt_release_group_id(struct vfsmount *mnt) 104719f5d7fSMiklos Szeredi { 105719f5d7fSMiklos Szeredi ida_remove(&mnt_group_ida, mnt->mnt_group_id); 106719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; 107719f5d7fSMiklos Szeredi } 108719f5d7fSMiklos Szeredi 1091da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 1101da177e4SLinus Torvalds { 111c3762229SRobert P. J. Day struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 1121da177e4SLinus Torvalds if (mnt) { 11373cd49ecSMiklos Szeredi int err; 11473cd49ecSMiklos Szeredi 11573cd49ecSMiklos Szeredi err = mnt_alloc_id(mnt); 11688b38782SLi Zefan if (err) 11788b38782SLi Zefan goto out_free_cache; 11888b38782SLi Zefan 11988b38782SLi Zefan if (name) { 12088b38782SLi Zefan mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 12188b38782SLi Zefan if (!mnt->mnt_devname) 12288b38782SLi Zefan goto out_free_id; 12373cd49ecSMiklos Szeredi } 12473cd49ecSMiklos Szeredi 1251da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 1261da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 1271da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 1281da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 1291da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 13055e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 13103e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 132a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 133a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 1343d733633SDave Hansen atomic_set(&mnt->__mnt_writers, 0); 1351da177e4SLinus Torvalds } 1361da177e4SLinus Torvalds return mnt; 13788b38782SLi Zefan 13888b38782SLi Zefan out_free_id: 13988b38782SLi Zefan mnt_free_id(mnt); 14088b38782SLi Zefan out_free_cache: 14188b38782SLi Zefan kmem_cache_free(mnt_cache, mnt); 14288b38782SLi Zefan return NULL; 1431da177e4SLinus Torvalds } 1441da177e4SLinus Torvalds 1458366025eSDave Hansen /* 1468366025eSDave Hansen * Most r/o checks on a fs are for operations that take 1478366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 1488366025eSDave Hansen * We must keep track of when those operations start 1498366025eSDave Hansen * (for permission checks) and when they end, so that 1508366025eSDave Hansen * we can determine when writes are able to occur to 1518366025eSDave Hansen * a filesystem. 1528366025eSDave Hansen */ 1533d733633SDave Hansen /* 1543d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 1553d733633SDave Hansen * @mnt: the mount to check for its write status 1563d733633SDave Hansen * 1573d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 1583d733633SDave Hansen * It does not guarantee that the filesystem will stay 1593d733633SDave Hansen * r/w, just that it is right *now*. This can not and 1603d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 1613d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 1623d733633SDave Hansen * r/w. 1633d733633SDave Hansen */ 1643d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 1653d733633SDave Hansen { 1662e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 1672e4b7fcdSDave Hansen return 1; 1682e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 1692e4b7fcdSDave Hansen return 1; 1702e4b7fcdSDave Hansen return 0; 1713d733633SDave Hansen } 1723d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 1733d733633SDave Hansen 1743d733633SDave Hansen struct mnt_writer { 1753d733633SDave Hansen /* 1763d733633SDave Hansen * If holding multiple instances of this lock, they 1773d733633SDave Hansen * must be ordered by cpu number. 1783d733633SDave Hansen */ 1793d733633SDave Hansen spinlock_t lock; 1803d733633SDave Hansen struct lock_class_key lock_class; /* compiles out with !lockdep */ 1813d733633SDave Hansen unsigned long count; 1823d733633SDave Hansen struct vfsmount *mnt; 1833d733633SDave Hansen } ____cacheline_aligned_in_smp; 1843d733633SDave Hansen static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); 1853d733633SDave Hansen 1863d733633SDave Hansen static int __init init_mnt_writers(void) 1873d733633SDave Hansen { 1883d733633SDave Hansen int cpu; 1893d733633SDave Hansen for_each_possible_cpu(cpu) { 1903d733633SDave Hansen struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); 1913d733633SDave Hansen spin_lock_init(&writer->lock); 1923d733633SDave Hansen lockdep_set_class(&writer->lock, &writer->lock_class); 1933d733633SDave Hansen writer->count = 0; 1943d733633SDave Hansen } 1953d733633SDave Hansen return 0; 1963d733633SDave Hansen } 1973d733633SDave Hansen fs_initcall(init_mnt_writers); 1983d733633SDave Hansen 1993d733633SDave Hansen static void unlock_mnt_writers(void) 2003d733633SDave Hansen { 2013d733633SDave Hansen int cpu; 2023d733633SDave Hansen struct mnt_writer *cpu_writer; 2033d733633SDave Hansen 2043d733633SDave Hansen for_each_possible_cpu(cpu) { 2053d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 2063d733633SDave Hansen spin_unlock(&cpu_writer->lock); 2073d733633SDave Hansen } 2083d733633SDave Hansen } 2093d733633SDave Hansen 2103d733633SDave Hansen static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) 2113d733633SDave Hansen { 2123d733633SDave Hansen if (!cpu_writer->mnt) 2133d733633SDave Hansen return; 2143d733633SDave Hansen /* 2153d733633SDave Hansen * This is in case anyone ever leaves an invalid, 2163d733633SDave Hansen * old ->mnt and a count of 0. 2173d733633SDave Hansen */ 2183d733633SDave Hansen if (!cpu_writer->count) 2193d733633SDave Hansen return; 2203d733633SDave Hansen atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); 2213d733633SDave Hansen cpu_writer->count = 0; 2223d733633SDave Hansen } 2233d733633SDave Hansen /* 2243d733633SDave Hansen * must hold cpu_writer->lock 2253d733633SDave Hansen */ 2263d733633SDave Hansen static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, 2273d733633SDave Hansen struct vfsmount *mnt) 2283d733633SDave Hansen { 2293d733633SDave Hansen if (cpu_writer->mnt == mnt) 2303d733633SDave Hansen return; 2313d733633SDave Hansen __clear_mnt_count(cpu_writer); 2323d733633SDave Hansen cpu_writer->mnt = mnt; 2333d733633SDave Hansen } 2343d733633SDave Hansen 2353d733633SDave Hansen /* 2363d733633SDave Hansen * Most r/o checks on a fs are for operations that take 2373d733633SDave Hansen * discrete amounts of time, like a write() or unlink(). 2383d733633SDave Hansen * We must keep track of when those operations start 2393d733633SDave Hansen * (for permission checks) and when they end, so that 2403d733633SDave Hansen * we can determine when writes are able to occur to 2413d733633SDave Hansen * a filesystem. 2423d733633SDave Hansen */ 2438366025eSDave Hansen /** 2448366025eSDave Hansen * mnt_want_write - get write access to a mount 2458366025eSDave Hansen * @mnt: the mount on which to take a write 2468366025eSDave Hansen * 2478366025eSDave Hansen * This tells the low-level filesystem that a write is 2488366025eSDave Hansen * about to be performed to it, and makes sure that 2498366025eSDave Hansen * writes are allowed before returning success. When 2508366025eSDave Hansen * the write operation is finished, mnt_drop_write() 2518366025eSDave Hansen * must be called. This is effectively a refcount. 2528366025eSDave Hansen */ 2538366025eSDave Hansen int mnt_want_write(struct vfsmount *mnt) 2548366025eSDave Hansen { 2553d733633SDave Hansen int ret = 0; 2563d733633SDave Hansen struct mnt_writer *cpu_writer; 2573d733633SDave Hansen 2583d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 2593d733633SDave Hansen spin_lock(&cpu_writer->lock); 2603d733633SDave Hansen if (__mnt_is_readonly(mnt)) { 2613d733633SDave Hansen ret = -EROFS; 2623d733633SDave Hansen goto out; 2633d733633SDave Hansen } 2643d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 2653d733633SDave Hansen cpu_writer->count++; 2663d733633SDave Hansen out: 2673d733633SDave Hansen spin_unlock(&cpu_writer->lock); 2683d733633SDave Hansen put_cpu_var(mnt_writers); 2693d733633SDave Hansen return ret; 2708366025eSDave Hansen } 2718366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 2728366025eSDave Hansen 2733d733633SDave Hansen static void lock_mnt_writers(void) 2743d733633SDave Hansen { 2753d733633SDave Hansen int cpu; 2763d733633SDave Hansen struct mnt_writer *cpu_writer; 2773d733633SDave Hansen 2783d733633SDave Hansen for_each_possible_cpu(cpu) { 2793d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 2803d733633SDave Hansen spin_lock(&cpu_writer->lock); 2813d733633SDave Hansen __clear_mnt_count(cpu_writer); 2823d733633SDave Hansen cpu_writer->mnt = NULL; 2833d733633SDave Hansen } 2843d733633SDave Hansen } 2853d733633SDave Hansen 2863d733633SDave Hansen /* 2873d733633SDave Hansen * These per-cpu write counts are not guaranteed to have 2883d733633SDave Hansen * matched increments and decrements on any given cpu. 2893d733633SDave Hansen * A file open()ed for write on one cpu and close()d on 2903d733633SDave Hansen * another cpu will imbalance this count. Make sure it 2913d733633SDave Hansen * does not get too far out of whack. 2923d733633SDave Hansen */ 2933d733633SDave Hansen static void handle_write_count_underflow(struct vfsmount *mnt) 2943d733633SDave Hansen { 2953d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) >= 2963d733633SDave Hansen MNT_WRITER_UNDERFLOW_LIMIT) 2973d733633SDave Hansen return; 2983d733633SDave Hansen /* 2993d733633SDave Hansen * It isn't necessary to hold all of the locks 3003d733633SDave Hansen * at the same time, but doing it this way makes 3013d733633SDave Hansen * us share a lot more code. 3023d733633SDave Hansen */ 3033d733633SDave Hansen lock_mnt_writers(); 3043d733633SDave Hansen /* 3053d733633SDave Hansen * vfsmount_lock is for mnt_flags. 3063d733633SDave Hansen */ 3073d733633SDave Hansen spin_lock(&vfsmount_lock); 3083d733633SDave Hansen /* 3093d733633SDave Hansen * If coalescing the per-cpu writer counts did not 3103d733633SDave Hansen * get us back to a positive writer count, we have 3113d733633SDave Hansen * a bug. 3123d733633SDave Hansen */ 3133d733633SDave Hansen if ((atomic_read(&mnt->__mnt_writers) < 0) && 3143d733633SDave Hansen !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { 3155c752ad9SArjan van de Ven WARN(1, KERN_DEBUG "leak detected on mount(%p) writers " 3163d733633SDave Hansen "count: %d\n", 3173d733633SDave Hansen mnt, atomic_read(&mnt->__mnt_writers)); 3183d733633SDave Hansen /* use the flag to keep the dmesg spam down */ 3193d733633SDave Hansen mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT; 3203d733633SDave Hansen } 3213d733633SDave Hansen spin_unlock(&vfsmount_lock); 3223d733633SDave Hansen unlock_mnt_writers(); 3233d733633SDave Hansen } 3243d733633SDave Hansen 3258366025eSDave Hansen /** 3268366025eSDave Hansen * mnt_drop_write - give up write access to a mount 3278366025eSDave Hansen * @mnt: the mount on which to give up write access 3288366025eSDave Hansen * 3298366025eSDave Hansen * Tells the low-level filesystem that we are done 3308366025eSDave Hansen * performing writes to it. Must be matched with 3318366025eSDave Hansen * mnt_want_write() call above. 3328366025eSDave Hansen */ 3338366025eSDave Hansen void mnt_drop_write(struct vfsmount *mnt) 3348366025eSDave Hansen { 3353d733633SDave Hansen int must_check_underflow = 0; 3363d733633SDave Hansen struct mnt_writer *cpu_writer; 3373d733633SDave Hansen 3383d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 3393d733633SDave Hansen spin_lock(&cpu_writer->lock); 3403d733633SDave Hansen 3413d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 3423d733633SDave Hansen if (cpu_writer->count > 0) { 3433d733633SDave Hansen cpu_writer->count--; 3443d733633SDave Hansen } else { 3453d733633SDave Hansen must_check_underflow = 1; 3463d733633SDave Hansen atomic_dec(&mnt->__mnt_writers); 3473d733633SDave Hansen } 3483d733633SDave Hansen 3493d733633SDave Hansen spin_unlock(&cpu_writer->lock); 3503d733633SDave Hansen /* 3513d733633SDave Hansen * Logically, we could call this each time, 3523d733633SDave Hansen * but the __mnt_writers cacheline tends to 3533d733633SDave Hansen * be cold, and makes this expensive. 3543d733633SDave Hansen */ 3553d733633SDave Hansen if (must_check_underflow) 3563d733633SDave Hansen handle_write_count_underflow(mnt); 3573d733633SDave Hansen /* 3583d733633SDave Hansen * This could be done right after the spinlock 3593d733633SDave Hansen * is taken because the spinlock keeps us on 3603d733633SDave Hansen * the cpu, and disables preemption. However, 3613d733633SDave Hansen * putting it here bounds the amount that 3623d733633SDave Hansen * __mnt_writers can underflow. Without it, 3633d733633SDave Hansen * we could theoretically wrap __mnt_writers. 3643d733633SDave Hansen */ 3653d733633SDave Hansen put_cpu_var(mnt_writers); 3668366025eSDave Hansen } 3678366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 3688366025eSDave Hansen 3692e4b7fcdSDave Hansen static int mnt_make_readonly(struct vfsmount *mnt) 3708366025eSDave Hansen { 3713d733633SDave Hansen int ret = 0; 3723d733633SDave Hansen 3733d733633SDave Hansen lock_mnt_writers(); 3743d733633SDave Hansen /* 3753d733633SDave Hansen * With all the locks held, this value is stable 3763d733633SDave Hansen */ 3773d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) > 0) { 3783d733633SDave Hansen ret = -EBUSY; 3793d733633SDave Hansen goto out; 3808366025eSDave Hansen } 3813d733633SDave Hansen /* 3822e4b7fcdSDave Hansen * nobody can do a successful mnt_want_write() with all 3832e4b7fcdSDave Hansen * of the counts in MNT_DENIED_WRITE and the locks held. 3843d733633SDave Hansen */ 3852e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 3862e4b7fcdSDave Hansen if (!ret) 3872e4b7fcdSDave Hansen mnt->mnt_flags |= MNT_READONLY; 3882e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3893d733633SDave Hansen out: 3903d733633SDave Hansen unlock_mnt_writers(); 3913d733633SDave Hansen return ret; 3923d733633SDave Hansen } 3938366025eSDave Hansen 3942e4b7fcdSDave Hansen static void __mnt_unmake_readonly(struct vfsmount *mnt) 3952e4b7fcdSDave Hansen { 3962e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 3972e4b7fcdSDave Hansen mnt->mnt_flags &= ~MNT_READONLY; 3982e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3992e4b7fcdSDave Hansen } 4002e4b7fcdSDave Hansen 401a3ec947cSSukadev Bhattiprolu void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 402454e2398SDavid Howells { 403454e2398SDavid Howells mnt->mnt_sb = sb; 404454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 405454e2398SDavid Howells } 406454e2398SDavid Howells 407454e2398SDavid Howells EXPORT_SYMBOL(simple_set_mnt); 408454e2398SDavid Howells 4091da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 4101da177e4SLinus Torvalds { 4111da177e4SLinus Torvalds kfree(mnt->mnt_devname); 41273cd49ecSMiklos Szeredi mnt_free_id(mnt); 4131da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 4141da177e4SLinus Torvalds } 4151da177e4SLinus Torvalds 4161da177e4SLinus Torvalds /* 417a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 418a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 4191da177e4SLinus Torvalds */ 420a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 421a05964f3SRam Pai int dir) 4221da177e4SLinus Torvalds { 4231da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 4241da177e4SLinus Torvalds struct list_head *tmp = head; 4251da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 4261da177e4SLinus Torvalds 4271da177e4SLinus Torvalds for (;;) { 428a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 4291da177e4SLinus Torvalds p = NULL; 4301da177e4SLinus Torvalds if (tmp == head) 4311da177e4SLinus Torvalds break; 4321da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 4331da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 434a05964f3SRam Pai found = p; 4351da177e4SLinus Torvalds break; 4361da177e4SLinus Torvalds } 4371da177e4SLinus Torvalds } 4381da177e4SLinus Torvalds return found; 4391da177e4SLinus Torvalds } 4401da177e4SLinus Torvalds 441a05964f3SRam Pai /* 442a05964f3SRam Pai * lookup_mnt increments the ref count before returning 443a05964f3SRam Pai * the vfsmount struct. 444a05964f3SRam Pai */ 445a05964f3SRam Pai struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 446a05964f3SRam Pai { 447a05964f3SRam Pai struct vfsmount *child_mnt; 448a05964f3SRam Pai spin_lock(&vfsmount_lock); 449a05964f3SRam Pai if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) 450a05964f3SRam Pai mntget(child_mnt); 451a05964f3SRam Pai spin_unlock(&vfsmount_lock); 452a05964f3SRam Pai return child_mnt; 453a05964f3SRam Pai } 454a05964f3SRam Pai 4551da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 4561da177e4SLinus Torvalds { 4576b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 4581da177e4SLinus Torvalds } 4591da177e4SLinus Torvalds 4606b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 4615addc5ddSAl Viro { 4625addc5ddSAl Viro if (ns) { 4635addc5ddSAl Viro ns->event = ++event; 4645addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4655addc5ddSAl Viro } 4665addc5ddSAl Viro } 4675addc5ddSAl Viro 4686b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 4695addc5ddSAl Viro { 4705addc5ddSAl Viro if (ns && ns->event != event) { 4715addc5ddSAl Viro ns->event = event; 4725addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4735addc5ddSAl Viro } 4745addc5ddSAl Viro } 4755addc5ddSAl Viro 4761a390689SAl Viro static void detach_mnt(struct vfsmount *mnt, struct path *old_path) 4771da177e4SLinus Torvalds { 4781a390689SAl Viro old_path->dentry = mnt->mnt_mountpoint; 4791a390689SAl Viro old_path->mnt = mnt->mnt_parent; 4801da177e4SLinus Torvalds mnt->mnt_parent = mnt; 4811da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 4821da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 4831da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 4841a390689SAl Viro old_path->dentry->d_mounted--; 4851da177e4SLinus Torvalds } 4861da177e4SLinus Torvalds 487b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 488b90fa9aeSRam Pai struct vfsmount *child_mnt) 489b90fa9aeSRam Pai { 490b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 491b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 492b90fa9aeSRam Pai dentry->d_mounted++; 493b90fa9aeSRam Pai } 494b90fa9aeSRam Pai 4951a390689SAl Viro static void attach_mnt(struct vfsmount *mnt, struct path *path) 4961da177e4SLinus Torvalds { 4971a390689SAl Viro mnt_set_mountpoint(path->mnt, path->dentry, mnt); 498b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 4991a390689SAl Viro hash(path->mnt, path->dentry)); 5001a390689SAl Viro list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); 501b90fa9aeSRam Pai } 502b90fa9aeSRam Pai 503b90fa9aeSRam Pai /* 504b90fa9aeSRam Pai * the caller must hold vfsmount_lock 505b90fa9aeSRam Pai */ 506b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 507b90fa9aeSRam Pai { 508b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 509b90fa9aeSRam Pai struct vfsmount *m; 510b90fa9aeSRam Pai LIST_HEAD(head); 5116b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 512b90fa9aeSRam Pai 513b90fa9aeSRam Pai BUG_ON(parent == mnt); 514b90fa9aeSRam Pai 515b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 516b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 5176b3286edSKirill Korotaev m->mnt_ns = n; 518b90fa9aeSRam Pai list_splice(&head, n->list.prev); 519b90fa9aeSRam Pai 520b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 521b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 522b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 5236b3286edSKirill Korotaev touch_mnt_namespace(n); 5241da177e4SLinus Torvalds } 5251da177e4SLinus Torvalds 5261da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 5271da177e4SLinus Torvalds { 5281da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 5291da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 5301da177e4SLinus Torvalds while (1) { 5311da177e4SLinus Torvalds if (p == root) 5321da177e4SLinus Torvalds return NULL; 5331da177e4SLinus Torvalds next = p->mnt_child.next; 5341da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 5351da177e4SLinus Torvalds break; 5361da177e4SLinus Torvalds p = p->mnt_parent; 5371da177e4SLinus Torvalds } 5381da177e4SLinus Torvalds } 5391da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 5401da177e4SLinus Torvalds } 5411da177e4SLinus Torvalds 5429676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 5439676f0c6SRam Pai { 5449676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 5459676f0c6SRam Pai while (prev != &p->mnt_mounts) { 5469676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 5479676f0c6SRam Pai prev = p->mnt_mounts.prev; 5489676f0c6SRam Pai } 5499676f0c6SRam Pai return p; 5509676f0c6SRam Pai } 5519676f0c6SRam Pai 55236341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 55336341f64SRam Pai int flag) 5541da177e4SLinus Torvalds { 5551da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 5561da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 5571da177e4SLinus Torvalds 5581da177e4SLinus Torvalds if (mnt) { 559719f5d7fSMiklos Szeredi if (flag & (CL_SLAVE | CL_PRIVATE)) 560719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; /* not a peer of original */ 561719f5d7fSMiklos Szeredi else 562719f5d7fSMiklos Szeredi mnt->mnt_group_id = old->mnt_group_id; 563719f5d7fSMiklos Szeredi 564719f5d7fSMiklos Szeredi if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 565719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(mnt); 566719f5d7fSMiklos Szeredi if (err) 567719f5d7fSMiklos Szeredi goto out_free; 568719f5d7fSMiklos Szeredi } 569719f5d7fSMiklos Szeredi 5701da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 5711da177e4SLinus Torvalds atomic_inc(&sb->s_active); 5721da177e4SLinus Torvalds mnt->mnt_sb = sb; 5731da177e4SLinus Torvalds mnt->mnt_root = dget(root); 5741da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 5751da177e4SLinus Torvalds mnt->mnt_parent = mnt; 576b90fa9aeSRam Pai 5775afe0022SRam Pai if (flag & CL_SLAVE) { 5785afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 5795afe0022SRam Pai mnt->mnt_master = old; 5805afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 5818aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 582b90fa9aeSRam Pai if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) 583b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 5845afe0022SRam Pai if (IS_MNT_SLAVE(old)) 5855afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 5865afe0022SRam Pai mnt->mnt_master = old->mnt_master; 5875afe0022SRam Pai } 588b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 589b90fa9aeSRam Pai set_mnt_shared(mnt); 5901da177e4SLinus Torvalds 5911da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 5921da177e4SLinus Torvalds * as the original if that was on one */ 59336341f64SRam Pai if (flag & CL_EXPIRE) { 59455e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 59555e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 5961da177e4SLinus Torvalds } 59736341f64SRam Pai } 5981da177e4SLinus Torvalds return mnt; 599719f5d7fSMiklos Szeredi 600719f5d7fSMiklos Szeredi out_free: 601719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 602719f5d7fSMiklos Szeredi return NULL; 6031da177e4SLinus Torvalds } 6041da177e4SLinus Torvalds 6057b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 6061da177e4SLinus Torvalds { 6073d733633SDave Hansen int cpu; 6081da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 6093d733633SDave Hansen /* 6103d733633SDave Hansen * We don't have to hold all of the locks at the 6113d733633SDave Hansen * same time here because we know that we're the 6123d733633SDave Hansen * last reference to mnt and that no new writers 6133d733633SDave Hansen * can come in. 6143d733633SDave Hansen */ 6153d733633SDave Hansen for_each_possible_cpu(cpu) { 6163d733633SDave Hansen struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu); 6173d733633SDave Hansen spin_lock(&cpu_writer->lock); 6181a88b536SAl Viro if (cpu_writer->mnt != mnt) { 6191a88b536SAl Viro spin_unlock(&cpu_writer->lock); 6201a88b536SAl Viro continue; 6211a88b536SAl Viro } 6223d733633SDave Hansen atomic_add(cpu_writer->count, &mnt->__mnt_writers); 6233d733633SDave Hansen cpu_writer->count = 0; 6243d733633SDave Hansen /* 6253d733633SDave Hansen * Might as well do this so that no one 6263d733633SDave Hansen * ever sees the pointer and expects 6273d733633SDave Hansen * it to be valid. 6283d733633SDave Hansen */ 6293d733633SDave Hansen cpu_writer->mnt = NULL; 6303d733633SDave Hansen spin_unlock(&cpu_writer->lock); 6313d733633SDave Hansen } 6323d733633SDave Hansen /* 6333d733633SDave Hansen * This probably indicates that somebody messed 6343d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 6353d733633SDave Hansen * happens, the filesystem was probably unable 6363d733633SDave Hansen * to make r/w->r/o transitions. 6373d733633SDave Hansen */ 6383d733633SDave Hansen WARN_ON(atomic_read(&mnt->__mnt_writers)); 6391da177e4SLinus Torvalds dput(mnt->mnt_root); 6401da177e4SLinus Torvalds free_vfsmnt(mnt); 6411da177e4SLinus Torvalds deactivate_super(sb); 6421da177e4SLinus Torvalds } 6431da177e4SLinus Torvalds 6447b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 6457b7b1aceSAl Viro { 6467b7b1aceSAl Viro repeat: 6477b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 6487b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 6497b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6507b7b1aceSAl Viro __mntput(mnt); 6517b7b1aceSAl Viro return; 6527b7b1aceSAl Viro } 6537b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 6547b7b1aceSAl Viro mnt->mnt_pinned = 0; 6557b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6567b7b1aceSAl Viro acct_auto_close_mnt(mnt); 6577b7b1aceSAl Viro security_sb_umount_close(mnt); 6587b7b1aceSAl Viro goto repeat; 6597b7b1aceSAl Viro } 6607b7b1aceSAl Viro } 6617b7b1aceSAl Viro 6627b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 6637b7b1aceSAl Viro 6647b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 6657b7b1aceSAl Viro { 6667b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6677b7b1aceSAl Viro mnt->mnt_pinned++; 6687b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6697b7b1aceSAl Viro } 6707b7b1aceSAl Viro 6717b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 6727b7b1aceSAl Viro 6737b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 6747b7b1aceSAl Viro { 6757b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6767b7b1aceSAl Viro if (mnt->mnt_pinned) { 6777b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 6787b7b1aceSAl Viro mnt->mnt_pinned--; 6797b7b1aceSAl Viro } 6807b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6817b7b1aceSAl Viro } 6827b7b1aceSAl Viro 6837b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 6841da177e4SLinus Torvalds 685b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 686b3b304a2SMiklos Szeredi { 687b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 688b3b304a2SMiklos Szeredi } 689b3b304a2SMiklos Szeredi 690b3b304a2SMiklos Szeredi /* 691b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 692b3b304a2SMiklos Szeredi * implement more complex mount option showing. 693b3b304a2SMiklos Szeredi * 694b3b304a2SMiklos Szeredi * See also save_mount_options(). 695b3b304a2SMiklos Szeredi */ 696b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 697b3b304a2SMiklos Szeredi { 698*2a32cebdSAl Viro const char *options; 699*2a32cebdSAl Viro 700*2a32cebdSAl Viro rcu_read_lock(); 701*2a32cebdSAl Viro options = rcu_dereference(mnt->mnt_sb->s_options); 702b3b304a2SMiklos Szeredi 703b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 704b3b304a2SMiklos Szeredi seq_putc(m, ','); 705b3b304a2SMiklos Szeredi mangle(m, options); 706b3b304a2SMiklos Szeredi } 707*2a32cebdSAl Viro rcu_read_unlock(); 708b3b304a2SMiklos Szeredi 709b3b304a2SMiklos Szeredi return 0; 710b3b304a2SMiklos Szeredi } 711b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 712b3b304a2SMiklos Szeredi 713b3b304a2SMiklos Szeredi /* 714b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 715b3b304a2SMiklos Szeredi * called from the fill_super() callback. 716b3b304a2SMiklos Szeredi * 717b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 718b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 719b3b304a2SMiklos Szeredi * remount fails. 720b3b304a2SMiklos Szeredi * 721b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 722b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 723b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 724b3b304a2SMiklos Szeredi * any more. 725b3b304a2SMiklos Szeredi */ 726b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 727b3b304a2SMiklos Szeredi { 728*2a32cebdSAl Viro BUG_ON(sb->s_options); 729*2a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 730b3b304a2SMiklos Szeredi } 731b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 732b3b304a2SMiklos Szeredi 733*2a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 734*2a32cebdSAl Viro { 735*2a32cebdSAl Viro char *old = sb->s_options; 736*2a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 737*2a32cebdSAl Viro if (old) { 738*2a32cebdSAl Viro synchronize_rcu(); 739*2a32cebdSAl Viro kfree(old); 740*2a32cebdSAl Viro } 741*2a32cebdSAl Viro } 742*2a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 743*2a32cebdSAl Viro 744a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 7451da177e4SLinus Torvalds /* iterator */ 7461da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 7471da177e4SLinus Torvalds { 748a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 7491da177e4SLinus Torvalds 750390c6843SRam Pai down_read(&namespace_sem); 751a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 7521da177e4SLinus Torvalds } 7531da177e4SLinus Torvalds 7541da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 7551da177e4SLinus Torvalds { 756a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 757b0765fb8SPavel Emelianov 758a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 7591da177e4SLinus Torvalds } 7601da177e4SLinus Torvalds 7611da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 7621da177e4SLinus Torvalds { 763390c6843SRam Pai up_read(&namespace_sem); 7641da177e4SLinus Torvalds } 7651da177e4SLinus Torvalds 7662d4d4864SRam Pai struct proc_fs_info { 7671da177e4SLinus Torvalds int flag; 7682d4d4864SRam Pai const char *str; 7692d4d4864SRam Pai }; 7702d4d4864SRam Pai 7712069f457SEric Paris static int show_sb_opts(struct seq_file *m, struct super_block *sb) 7722d4d4864SRam Pai { 7732d4d4864SRam Pai static const struct proc_fs_info fs_info[] = { 7741da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 7751da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 7761da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 7771da177e4SLinus Torvalds { 0, NULL } 7781da177e4SLinus Torvalds }; 7792d4d4864SRam Pai const struct proc_fs_info *fs_infop; 7802d4d4864SRam Pai 7812d4d4864SRam Pai for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 7822d4d4864SRam Pai if (sb->s_flags & fs_infop->flag) 7832d4d4864SRam Pai seq_puts(m, fs_infop->str); 7842d4d4864SRam Pai } 7852069f457SEric Paris 7862069f457SEric Paris return security_sb_show_options(m, sb); 7872d4d4864SRam Pai } 7882d4d4864SRam Pai 7892d4d4864SRam Pai static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) 7902d4d4864SRam Pai { 7912d4d4864SRam Pai static const struct proc_fs_info mnt_info[] = { 7921da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 7931da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 7941da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 795fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 796fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 79747ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 798d0adde57SMatthew Garrett { MNT_STRICTATIME, ",strictatime" }, 7991da177e4SLinus Torvalds { 0, NULL } 8001da177e4SLinus Torvalds }; 8012d4d4864SRam Pai const struct proc_fs_info *fs_infop; 8022d4d4864SRam Pai 8032d4d4864SRam Pai for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 8042d4d4864SRam Pai if (mnt->mnt_flags & fs_infop->flag) 8052d4d4864SRam Pai seq_puts(m, fs_infop->str); 8062d4d4864SRam Pai } 8072d4d4864SRam Pai } 8082d4d4864SRam Pai 8092d4d4864SRam Pai static void show_type(struct seq_file *m, struct super_block *sb) 8102d4d4864SRam Pai { 8112d4d4864SRam Pai mangle(m, sb->s_type->name); 8122d4d4864SRam Pai if (sb->s_subtype && sb->s_subtype[0]) { 8132d4d4864SRam Pai seq_putc(m, '.'); 8142d4d4864SRam Pai mangle(m, sb->s_subtype); 8152d4d4864SRam Pai } 8162d4d4864SRam Pai } 8172d4d4864SRam Pai 8182d4d4864SRam Pai static int show_vfsmnt(struct seq_file *m, void *v) 8192d4d4864SRam Pai { 8202d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 8212d4d4864SRam Pai int err = 0; 822c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 8231da177e4SLinus Torvalds 8241da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 8251da177e4SLinus Torvalds seq_putc(m, ' '); 826c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 8271da177e4SLinus Torvalds seq_putc(m, ' '); 8282d4d4864SRam Pai show_type(m, mnt->mnt_sb); 8292e4b7fcdSDave Hansen seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 8302069f457SEric Paris err = show_sb_opts(m, mnt->mnt_sb); 8312069f457SEric Paris if (err) 8322069f457SEric Paris goto out; 8332d4d4864SRam Pai show_mnt_opts(m, mnt); 8341da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 8351da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 8361da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 8372069f457SEric Paris out: 8381da177e4SLinus Torvalds return err; 8391da177e4SLinus Torvalds } 8401da177e4SLinus Torvalds 841a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 8421da177e4SLinus Torvalds .start = m_start, 8431da177e4SLinus Torvalds .next = m_next, 8441da177e4SLinus Torvalds .stop = m_stop, 8451da177e4SLinus Torvalds .show = show_vfsmnt 8461da177e4SLinus Torvalds }; 8471da177e4SLinus Torvalds 8482d4d4864SRam Pai static int show_mountinfo(struct seq_file *m, void *v) 8492d4d4864SRam Pai { 8502d4d4864SRam Pai struct proc_mounts *p = m->private; 8512d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 8522d4d4864SRam Pai struct super_block *sb = mnt->mnt_sb; 8532d4d4864SRam Pai struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 8542d4d4864SRam Pai struct path root = p->root; 8552d4d4864SRam Pai int err = 0; 8562d4d4864SRam Pai 8572d4d4864SRam Pai seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, 8582d4d4864SRam Pai MAJOR(sb->s_dev), MINOR(sb->s_dev)); 8592d4d4864SRam Pai seq_dentry(m, mnt->mnt_root, " \t\n\\"); 8602d4d4864SRam Pai seq_putc(m, ' '); 8612d4d4864SRam Pai seq_path_root(m, &mnt_path, &root, " \t\n\\"); 8622d4d4864SRam Pai if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { 8632d4d4864SRam Pai /* 8642d4d4864SRam Pai * Mountpoint is outside root, discard that one. Ugly, 8652d4d4864SRam Pai * but less so than trying to do that in iterator in a 8662d4d4864SRam Pai * race-free way (due to renames). 8672d4d4864SRam Pai */ 8682d4d4864SRam Pai return SEQ_SKIP; 8692d4d4864SRam Pai } 8702d4d4864SRam Pai seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); 8712d4d4864SRam Pai show_mnt_opts(m, mnt); 8722d4d4864SRam Pai 8732d4d4864SRam Pai /* Tagged fields ("foo:X" or "bar") */ 8742d4d4864SRam Pai if (IS_MNT_SHARED(mnt)) 8752d4d4864SRam Pai seq_printf(m, " shared:%i", mnt->mnt_group_id); 87697e7e0f7SMiklos Szeredi if (IS_MNT_SLAVE(mnt)) { 87797e7e0f7SMiklos Szeredi int master = mnt->mnt_master->mnt_group_id; 87897e7e0f7SMiklos Szeredi int dom = get_dominating_id(mnt, &p->root); 87997e7e0f7SMiklos Szeredi seq_printf(m, " master:%i", master); 88097e7e0f7SMiklos Szeredi if (dom && dom != master) 88197e7e0f7SMiklos Szeredi seq_printf(m, " propagate_from:%i", dom); 88297e7e0f7SMiklos Szeredi } 8832d4d4864SRam Pai if (IS_MNT_UNBINDABLE(mnt)) 8842d4d4864SRam Pai seq_puts(m, " unbindable"); 8852d4d4864SRam Pai 8862d4d4864SRam Pai /* Filesystem specific data */ 8872d4d4864SRam Pai seq_puts(m, " - "); 8882d4d4864SRam Pai show_type(m, sb); 8892d4d4864SRam Pai seq_putc(m, ' '); 8902d4d4864SRam Pai mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 8912d4d4864SRam Pai seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); 8922069f457SEric Paris err = show_sb_opts(m, sb); 8932069f457SEric Paris if (err) 8942069f457SEric Paris goto out; 8952d4d4864SRam Pai if (sb->s_op->show_options) 8962d4d4864SRam Pai err = sb->s_op->show_options(m, mnt); 8972d4d4864SRam Pai seq_putc(m, '\n'); 8982069f457SEric Paris out: 8992d4d4864SRam Pai return err; 9002d4d4864SRam Pai } 9012d4d4864SRam Pai 9022d4d4864SRam Pai const struct seq_operations mountinfo_op = { 9032d4d4864SRam Pai .start = m_start, 9042d4d4864SRam Pai .next = m_next, 9052d4d4864SRam Pai .stop = m_stop, 9062d4d4864SRam Pai .show = show_mountinfo, 9072d4d4864SRam Pai }; 9082d4d4864SRam Pai 909b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 910b4629fe2SChuck Lever { 911b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 912c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 913b4629fe2SChuck Lever int err = 0; 914b4629fe2SChuck Lever 915b4629fe2SChuck Lever /* device */ 916b4629fe2SChuck Lever if (mnt->mnt_devname) { 917b4629fe2SChuck Lever seq_puts(m, "device "); 918b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 919b4629fe2SChuck Lever } else 920b4629fe2SChuck Lever seq_puts(m, "no device"); 921b4629fe2SChuck Lever 922b4629fe2SChuck Lever /* mount point */ 923b4629fe2SChuck Lever seq_puts(m, " mounted on "); 924c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 925b4629fe2SChuck Lever seq_putc(m, ' '); 926b4629fe2SChuck Lever 927b4629fe2SChuck Lever /* file system type */ 928b4629fe2SChuck Lever seq_puts(m, "with fstype "); 9292d4d4864SRam Pai show_type(m, mnt->mnt_sb); 930b4629fe2SChuck Lever 931b4629fe2SChuck Lever /* optional statistics */ 932b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 933b4629fe2SChuck Lever seq_putc(m, ' '); 934b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 935b4629fe2SChuck Lever } 936b4629fe2SChuck Lever 937b4629fe2SChuck Lever seq_putc(m, '\n'); 938b4629fe2SChuck Lever return err; 939b4629fe2SChuck Lever } 940b4629fe2SChuck Lever 941a1a2c409SMiklos Szeredi const struct seq_operations mountstats_op = { 942b4629fe2SChuck Lever .start = m_start, 943b4629fe2SChuck Lever .next = m_next, 944b4629fe2SChuck Lever .stop = m_stop, 945b4629fe2SChuck Lever .show = show_vfsstat, 946b4629fe2SChuck Lever }; 947a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 948b4629fe2SChuck Lever 9491da177e4SLinus Torvalds /** 9501da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 9511da177e4SLinus Torvalds * @mnt: root of mount tree 9521da177e4SLinus Torvalds * 9531da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 9541da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 9551da177e4SLinus Torvalds * busy. 9561da177e4SLinus Torvalds */ 9571da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 9581da177e4SLinus Torvalds { 95936341f64SRam Pai int actual_refs = 0; 96036341f64SRam Pai int minimum_refs = 0; 96136341f64SRam Pai struct vfsmount *p; 9621da177e4SLinus Torvalds 9631da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 96436341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 9651da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 9661da177e4SLinus Torvalds minimum_refs += 2; 9671da177e4SLinus Torvalds } 9681da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 9691da177e4SLinus Torvalds 9701da177e4SLinus Torvalds if (actual_refs > minimum_refs) 9711da177e4SLinus Torvalds return 0; 972e3474a8eSIan Kent 973e3474a8eSIan Kent return 1; 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 9761da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 9771da177e4SLinus Torvalds 9781da177e4SLinus Torvalds /** 9791da177e4SLinus Torvalds * may_umount - check if a mount point is busy 9801da177e4SLinus Torvalds * @mnt: root of mount 9811da177e4SLinus Torvalds * 9821da177e4SLinus Torvalds * This is called to check if a mount point has any 9831da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 9841da177e4SLinus Torvalds * mount has sub mounts this will return busy 9851da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 9861da177e4SLinus Torvalds * 9871da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 9881da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 9891da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 9901da177e4SLinus Torvalds */ 9911da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 9921da177e4SLinus Torvalds { 993e3474a8eSIan Kent int ret = 1; 994a05964f3SRam Pai spin_lock(&vfsmount_lock); 995a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 996e3474a8eSIan Kent ret = 0; 997a05964f3SRam Pai spin_unlock(&vfsmount_lock); 998a05964f3SRam Pai return ret; 9991da177e4SLinus Torvalds } 10001da177e4SLinus Torvalds 10011da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 10021da177e4SLinus Torvalds 1003b90fa9aeSRam Pai void release_mounts(struct list_head *head) 10041da177e4SLinus Torvalds { 100570fbcdf4SRam Pai struct vfsmount *mnt; 100670fbcdf4SRam Pai while (!list_empty(head)) { 1007b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 100870fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 100970fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 101070fbcdf4SRam Pai struct dentry *dentry; 101170fbcdf4SRam Pai struct vfsmount *m; 101270fbcdf4SRam Pai spin_lock(&vfsmount_lock); 101370fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 101470fbcdf4SRam Pai m = mnt->mnt_parent; 101570fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 101670fbcdf4SRam Pai mnt->mnt_parent = mnt; 10177c4b93d8SAl Viro m->mnt_ghosts--; 10181da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 101970fbcdf4SRam Pai dput(dentry); 102070fbcdf4SRam Pai mntput(m); 10211da177e4SLinus Torvalds } 10221da177e4SLinus Torvalds mntput(mnt); 102370fbcdf4SRam Pai } 102470fbcdf4SRam Pai } 102570fbcdf4SRam Pai 1026a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 102770fbcdf4SRam Pai { 102870fbcdf4SRam Pai struct vfsmount *p; 102970fbcdf4SRam Pai 10301bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 10311bfba4e8SAkinobu Mita list_move(&p->mnt_hash, kill); 103270fbcdf4SRam Pai 1033a05964f3SRam Pai if (propagate) 1034a05964f3SRam Pai propagate_umount(kill); 1035a05964f3SRam Pai 103670fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 103770fbcdf4SRam Pai list_del_init(&p->mnt_expire); 103870fbcdf4SRam Pai list_del_init(&p->mnt_list); 10396b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 10406b3286edSKirill Korotaev p->mnt_ns = NULL; 104170fbcdf4SRam Pai list_del_init(&p->mnt_child); 10427c4b93d8SAl Viro if (p->mnt_parent != p) { 10437c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 1044f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 10457c4b93d8SAl Viro } 1046a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 10471da177e4SLinus Torvalds } 10481da177e4SLinus Torvalds } 10491da177e4SLinus Torvalds 1050c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 1051c35038beSAl Viro 10521da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 10531da177e4SLinus Torvalds { 10541da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 10551da177e4SLinus Torvalds int retval; 105670fbcdf4SRam Pai LIST_HEAD(umount_list); 10571da177e4SLinus Torvalds 10581da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 10591da177e4SLinus Torvalds if (retval) 10601da177e4SLinus Torvalds return retval; 10611da177e4SLinus Torvalds 10621da177e4SLinus Torvalds /* 10631da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 10641da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 10651da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 10661da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 10671da177e4SLinus Torvalds */ 10681da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 10696ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 10701da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 10711da177e4SLinus Torvalds return -EINVAL; 10721da177e4SLinus Torvalds 10731da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 10741da177e4SLinus Torvalds return -EBUSY; 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 10771da177e4SLinus Torvalds return -EAGAIN; 10781da177e4SLinus Torvalds } 10791da177e4SLinus Torvalds 10801da177e4SLinus Torvalds /* 10811da177e4SLinus Torvalds * If we may have to abort operations to get out of this 10821da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 10831da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 10841da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 10851da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 10861da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 10871da177e4SLinus Torvalds * about for the moment. 10881da177e4SLinus Torvalds */ 10891da177e4SLinus Torvalds 109042faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 109142faad99SAl Viro sb->s_op->umount_begin(sb); 109242faad99SAl Viro } 10931da177e4SLinus Torvalds 10941da177e4SLinus Torvalds /* 10951da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 10961da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 10971da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 10981da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 10991da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 11001da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 11011da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 11021da177e4SLinus Torvalds */ 11036ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 11041da177e4SLinus Torvalds /* 11051da177e4SLinus Torvalds * Special case for "unmounting" root ... 11061da177e4SLinus Torvalds * we just try to remount it readonly. 11071da177e4SLinus Torvalds */ 11081da177e4SLinus Torvalds down_write(&sb->s_umount); 11091da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 11101da177e4SLinus Torvalds lock_kernel(); 11111da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 11121da177e4SLinus Torvalds unlock_kernel(); 11131da177e4SLinus Torvalds } 11141da177e4SLinus Torvalds up_write(&sb->s_umount); 11151da177e4SLinus Torvalds return retval; 11161da177e4SLinus Torvalds } 11171da177e4SLinus Torvalds 1118390c6843SRam Pai down_write(&namespace_sem); 11191da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 11205addc5ddSAl Viro event++; 11211da177e4SLinus Torvalds 1122c35038beSAl Viro if (!(flags & MNT_DETACH)) 1123c35038beSAl Viro shrink_submounts(mnt, &umount_list); 1124c35038beSAl Viro 11251da177e4SLinus Torvalds retval = -EBUSY; 1126a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 11271da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 1128a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 11291da177e4SLinus Torvalds retval = 0; 11301da177e4SLinus Torvalds } 11311da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 11321da177e4SLinus Torvalds if (retval) 11331da177e4SLinus Torvalds security_sb_umount_busy(mnt); 1134390c6843SRam Pai up_write(&namespace_sem); 113570fbcdf4SRam Pai release_mounts(&umount_list); 11361da177e4SLinus Torvalds return retval; 11371da177e4SLinus Torvalds } 11381da177e4SLinus Torvalds 11391da177e4SLinus Torvalds /* 11401da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 11411da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 11421da177e4SLinus Torvalds * 11431da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 11441da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 11451da177e4SLinus Torvalds */ 11461da177e4SLinus Torvalds 1147bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 11481da177e4SLinus Torvalds { 11492d8f3038SAl Viro struct path path; 11501da177e4SLinus Torvalds int retval; 11511da177e4SLinus Torvalds 11522d8f3038SAl Viro retval = user_path(name, &path); 11531da177e4SLinus Torvalds if (retval) 11541da177e4SLinus Torvalds goto out; 11551da177e4SLinus Torvalds retval = -EINVAL; 11562d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 11571da177e4SLinus Torvalds goto dput_and_out; 11582d8f3038SAl Viro if (!check_mnt(path.mnt)) 11591da177e4SLinus Torvalds goto dput_and_out; 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds retval = -EPERM; 11621da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 11631da177e4SLinus Torvalds goto dput_and_out; 11641da177e4SLinus Torvalds 11652d8f3038SAl Viro retval = do_umount(path.mnt, flags); 11661da177e4SLinus Torvalds dput_and_out: 1167429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 11682d8f3038SAl Viro dput(path.dentry); 11692d8f3038SAl Viro mntput_no_expire(path.mnt); 11701da177e4SLinus Torvalds out: 11711da177e4SLinus Torvalds return retval; 11721da177e4SLinus Torvalds } 11731da177e4SLinus Torvalds 11741da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 11751da177e4SLinus Torvalds 11761da177e4SLinus Torvalds /* 11771da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 11781da177e4SLinus Torvalds */ 1179bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 11801da177e4SLinus Torvalds { 11811da177e4SLinus Torvalds return sys_umount(name, 0); 11821da177e4SLinus Torvalds } 11831da177e4SLinus Torvalds 11841da177e4SLinus Torvalds #endif 11851da177e4SLinus Torvalds 11862d92ab3cSAl Viro static int mount_is_safe(struct path *path) 11871da177e4SLinus Torvalds { 11881da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 11891da177e4SLinus Torvalds return 0; 11901da177e4SLinus Torvalds return -EPERM; 11911da177e4SLinus Torvalds #ifdef notyet 11922d92ab3cSAl Viro if (S_ISLNK(path->dentry->d_inode->i_mode)) 11931da177e4SLinus Torvalds return -EPERM; 11942d92ab3cSAl Viro if (path->dentry->d_inode->i_mode & S_ISVTX) { 1195da9592edSDavid Howells if (current_uid() != path->dentry->d_inode->i_uid) 11961da177e4SLinus Torvalds return -EPERM; 11971da177e4SLinus Torvalds } 11982d92ab3cSAl Viro if (inode_permission(path->dentry->d_inode, MAY_WRITE)) 11991da177e4SLinus Torvalds return -EPERM; 12001da177e4SLinus Torvalds return 0; 12011da177e4SLinus Torvalds #endif 12021da177e4SLinus Torvalds } 12031da177e4SLinus Torvalds 1204b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 120536341f64SRam Pai int flag) 12061da177e4SLinus Torvalds { 12071da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 12081a390689SAl Viro struct path path; 12091da177e4SLinus Torvalds 12109676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 12119676f0c6SRam Pai return NULL; 12129676f0c6SRam Pai 121336341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 12141da177e4SLinus Torvalds if (!q) 12151da177e4SLinus Torvalds goto Enomem; 12161da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 12171da177e4SLinus Torvalds 12181da177e4SLinus Torvalds p = mnt; 1219fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 12207ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 12211da177e4SLinus Torvalds continue; 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 12249676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 12259676f0c6SRam Pai s = skip_mnt_tree(s); 12269676f0c6SRam Pai continue; 12279676f0c6SRam Pai } 12281da177e4SLinus Torvalds while (p != s->mnt_parent) { 12291da177e4SLinus Torvalds p = p->mnt_parent; 12301da177e4SLinus Torvalds q = q->mnt_parent; 12311da177e4SLinus Torvalds } 12321da177e4SLinus Torvalds p = s; 12331a390689SAl Viro path.mnt = q; 12341a390689SAl Viro path.dentry = p->mnt_mountpoint; 123536341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 12361da177e4SLinus Torvalds if (!q) 12371da177e4SLinus Torvalds goto Enomem; 12381da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 12391da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 12401a390689SAl Viro attach_mnt(q, &path); 12411da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 12421da177e4SLinus Torvalds } 12431da177e4SLinus Torvalds } 12441da177e4SLinus Torvalds return res; 12451da177e4SLinus Torvalds Enomem: 12461da177e4SLinus Torvalds if (res) { 124770fbcdf4SRam Pai LIST_HEAD(umount_list); 12481da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1249a05964f3SRam Pai umount_tree(res, 0, &umount_list); 12501da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 125170fbcdf4SRam Pai release_mounts(&umount_list); 12521da177e4SLinus Torvalds } 12531da177e4SLinus Torvalds return NULL; 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds 12568aec0809SAl Viro struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) 12578aec0809SAl Viro { 12588aec0809SAl Viro struct vfsmount *tree; 12591a60a280SAl Viro down_write(&namespace_sem); 12608aec0809SAl Viro tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); 12611a60a280SAl Viro up_write(&namespace_sem); 12628aec0809SAl Viro return tree; 12638aec0809SAl Viro } 12648aec0809SAl Viro 12658aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 12668aec0809SAl Viro { 12678aec0809SAl Viro LIST_HEAD(umount_list); 12681a60a280SAl Viro down_write(&namespace_sem); 12698aec0809SAl Viro spin_lock(&vfsmount_lock); 12708aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 12718aec0809SAl Viro spin_unlock(&vfsmount_lock); 12721a60a280SAl Viro up_write(&namespace_sem); 12738aec0809SAl Viro release_mounts(&umount_list); 12748aec0809SAl Viro } 12758aec0809SAl Viro 1276719f5d7fSMiklos Szeredi static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) 1277719f5d7fSMiklos Szeredi { 1278719f5d7fSMiklos Szeredi struct vfsmount *p; 1279719f5d7fSMiklos Szeredi 1280719f5d7fSMiklos Szeredi for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1281719f5d7fSMiklos Szeredi if (p->mnt_group_id && !IS_MNT_SHARED(p)) 1282719f5d7fSMiklos Szeredi mnt_release_group_id(p); 1283719f5d7fSMiklos Szeredi } 1284719f5d7fSMiklos Szeredi } 1285719f5d7fSMiklos Szeredi 1286719f5d7fSMiklos Szeredi static int invent_group_ids(struct vfsmount *mnt, bool recurse) 1287719f5d7fSMiklos Szeredi { 1288719f5d7fSMiklos Szeredi struct vfsmount *p; 1289719f5d7fSMiklos Szeredi 1290719f5d7fSMiklos Szeredi for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1291719f5d7fSMiklos Szeredi if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 1292719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(p); 1293719f5d7fSMiklos Szeredi if (err) { 1294719f5d7fSMiklos Szeredi cleanup_group_ids(mnt, p); 1295719f5d7fSMiklos Szeredi return err; 1296719f5d7fSMiklos Szeredi } 1297719f5d7fSMiklos Szeredi } 1298719f5d7fSMiklos Szeredi } 1299719f5d7fSMiklos Szeredi 1300719f5d7fSMiklos Szeredi return 0; 1301719f5d7fSMiklos Szeredi } 1302719f5d7fSMiklos Szeredi 1303b90fa9aeSRam Pai /* 1304b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1305b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 130621444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 130721444403SRam Pai * store the parent mount and mountpoint dentry. 130821444403SRam Pai * (done when source_mnt is moved) 1309b90fa9aeSRam Pai * 1310b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1311b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 13129676f0c6SRam Pai * --------------------------------------------------------------------------- 1313b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 13149676f0c6SRam Pai * |************************************************************************** 13159676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 13169676f0c6SRam Pai * | dest | | | | | 13179676f0c6SRam Pai * | | | | | | | 13189676f0c6SRam Pai * | v | | | | | 13199676f0c6SRam Pai * |************************************************************************** 13209676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 13215afe0022SRam Pai * | | | | | | 13229676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 13239676f0c6SRam Pai * *************************************************************************** 1324b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1325b90fa9aeSRam Pai * destination mount. 1326b90fa9aeSRam Pai * 1327b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1328b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1329b90fa9aeSRam Pai * the peer group of the source mount. 1330b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1331b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1332b90fa9aeSRam Pai * mount. 13335afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 13345afe0022SRam Pai * of the destination mount and the cloned mount is made slave 13355afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 13365afe0022SRam Pai * is marked as 'shared and slave'. 13375afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 13385afe0022SRam Pai * source mount. 13395afe0022SRam Pai * 13409676f0c6SRam Pai * --------------------------------------------------------------------------- 134121444403SRam Pai * | MOVE MOUNT OPERATION | 13429676f0c6SRam Pai * |************************************************************************** 13439676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 13449676f0c6SRam Pai * | dest | | | | | 13459676f0c6SRam Pai * | | | | | | | 13469676f0c6SRam Pai * | v | | | | | 13479676f0c6SRam Pai * |************************************************************************** 13489676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 13495afe0022SRam Pai * | | | | | | 13509676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 13519676f0c6SRam Pai * *************************************************************************** 13525afe0022SRam Pai * 13535afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 13545afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 135521444403SRam Pai * (+*) the mount is moved to the destination. 13565afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 13575afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 13585afe0022SRam Pai * the mount is marked as 'shared and slave'. 13595afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1360b90fa9aeSRam Pai * 1361b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1362b90fa9aeSRam Pai * applied to each mount in the tree. 1363b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1364b90fa9aeSRam Pai * in allocations. 1365b90fa9aeSRam Pai */ 1366b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 13671a390689SAl Viro struct path *path, struct path *parent_path) 1368b90fa9aeSRam Pai { 1369b90fa9aeSRam Pai LIST_HEAD(tree_list); 13701a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 13711a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1372b90fa9aeSRam Pai struct vfsmount *child, *p; 1373719f5d7fSMiklos Szeredi int err; 1374b90fa9aeSRam Pai 1375719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) { 1376719f5d7fSMiklos Szeredi err = invent_group_ids(source_mnt, true); 1377719f5d7fSMiklos Szeredi if (err) 1378719f5d7fSMiklos Szeredi goto out; 1379719f5d7fSMiklos Szeredi } 1380719f5d7fSMiklos Szeredi err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1381719f5d7fSMiklos Szeredi if (err) 1382719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1383b90fa9aeSRam Pai 1384b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1385b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1386b90fa9aeSRam Pai set_mnt_shared(p); 1387b90fa9aeSRam Pai } 1388b90fa9aeSRam Pai 1389b90fa9aeSRam Pai spin_lock(&vfsmount_lock); 13901a390689SAl Viro if (parent_path) { 13911a390689SAl Viro detach_mnt(source_mnt, parent_path); 13921a390689SAl Viro attach_mnt(source_mnt, path); 1393e5d67f07SAl Viro touch_mnt_namespace(parent_path->mnt->mnt_ns); 139421444403SRam Pai } else { 1395b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1396b90fa9aeSRam Pai commit_tree(source_mnt); 139721444403SRam Pai } 1398b90fa9aeSRam Pai 1399b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1400b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1401b90fa9aeSRam Pai commit_tree(child); 1402b90fa9aeSRam Pai } 1403b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 1404b90fa9aeSRam Pai return 0; 1405719f5d7fSMiklos Szeredi 1406719f5d7fSMiklos Szeredi out_cleanup_ids: 1407719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) 1408719f5d7fSMiklos Szeredi cleanup_group_ids(source_mnt, NULL); 1409719f5d7fSMiklos Szeredi out: 1410719f5d7fSMiklos Szeredi return err; 1411b90fa9aeSRam Pai } 1412b90fa9aeSRam Pai 14138c3ee42eSAl Viro static int graft_tree(struct vfsmount *mnt, struct path *path) 14141da177e4SLinus Torvalds { 14151da177e4SLinus Torvalds int err; 14161da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 14171da177e4SLinus Torvalds return -EINVAL; 14181da177e4SLinus Torvalds 14198c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 14201da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 14211da177e4SLinus Torvalds return -ENOTDIR; 14221da177e4SLinus Torvalds 14231da177e4SLinus Torvalds err = -ENOENT; 14248c3ee42eSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 14258c3ee42eSAl Viro if (IS_DEADDIR(path->dentry->d_inode)) 14261da177e4SLinus Torvalds goto out_unlock; 14271da177e4SLinus Torvalds 14288c3ee42eSAl Viro err = security_sb_check_sb(mnt, path); 14291da177e4SLinus Torvalds if (err) 14301da177e4SLinus Torvalds goto out_unlock; 14311da177e4SLinus Torvalds 14321da177e4SLinus Torvalds err = -ENOENT; 14338c3ee42eSAl Viro if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry)) 14348c3ee42eSAl Viro err = attach_recursive_mnt(mnt, path, NULL); 14351da177e4SLinus Torvalds out_unlock: 14368c3ee42eSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 14371da177e4SLinus Torvalds if (!err) 14388c3ee42eSAl Viro security_sb_post_addmount(mnt, path); 14391da177e4SLinus Torvalds return err; 14401da177e4SLinus Torvalds } 14411da177e4SLinus Torvalds 14421da177e4SLinus Torvalds /* 144307b20889SRam Pai * recursively change the type of the mountpoint. 144407b20889SRam Pai */ 14450a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 144607b20889SRam Pai { 14472d92ab3cSAl Viro struct vfsmount *m, *mnt = path->mnt; 144807b20889SRam Pai int recurse = flag & MS_REC; 144907b20889SRam Pai int type = flag & ~MS_REC; 1450719f5d7fSMiklos Szeredi int err = 0; 145107b20889SRam Pai 1452ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1453ee6f9582SMiklos Szeredi return -EPERM; 1454ee6f9582SMiklos Szeredi 14552d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 145607b20889SRam Pai return -EINVAL; 145707b20889SRam Pai 145807b20889SRam Pai down_write(&namespace_sem); 1459719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1460719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1461719f5d7fSMiklos Szeredi if (err) 1462719f5d7fSMiklos Szeredi goto out_unlock; 1463719f5d7fSMiklos Szeredi } 1464719f5d7fSMiklos Szeredi 146507b20889SRam Pai spin_lock(&vfsmount_lock); 146607b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 146707b20889SRam Pai change_mnt_propagation(m, type); 146807b20889SRam Pai spin_unlock(&vfsmount_lock); 1469719f5d7fSMiklos Szeredi 1470719f5d7fSMiklos Szeredi out_unlock: 147107b20889SRam Pai up_write(&namespace_sem); 1472719f5d7fSMiklos Szeredi return err; 147307b20889SRam Pai } 147407b20889SRam Pai 147507b20889SRam Pai /* 14761da177e4SLinus Torvalds * do loopback mount. 14771da177e4SLinus Torvalds */ 14780a0d8a46SAl Viro static int do_loopback(struct path *path, char *old_name, 14792dafe1c4SEric Sandeen int recurse) 14801da177e4SLinus Torvalds { 14812d92ab3cSAl Viro struct path old_path; 14821da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 14832d92ab3cSAl Viro int err = mount_is_safe(path); 14841da177e4SLinus Torvalds if (err) 14851da177e4SLinus Torvalds return err; 14861da177e4SLinus Torvalds if (!old_name || !*old_name) 14871da177e4SLinus Torvalds return -EINVAL; 14882d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 14891da177e4SLinus Torvalds if (err) 14901da177e4SLinus Torvalds return err; 14911da177e4SLinus Torvalds 1492390c6843SRam Pai down_write(&namespace_sem); 14931da177e4SLinus Torvalds err = -EINVAL; 14942d92ab3cSAl Viro if (IS_MNT_UNBINDABLE(old_path.mnt)) 14959676f0c6SRam Pai goto out; 14969676f0c6SRam Pai 14972d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 1498ccd48bc7SAl Viro goto out; 1499ccd48bc7SAl Viro 15001da177e4SLinus Torvalds err = -ENOMEM; 15011da177e4SLinus Torvalds if (recurse) 15022d92ab3cSAl Viro mnt = copy_tree(old_path.mnt, old_path.dentry, 0); 15031da177e4SLinus Torvalds else 15042d92ab3cSAl Viro mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); 15051da177e4SLinus Torvalds 1506ccd48bc7SAl Viro if (!mnt) 1507ccd48bc7SAl Viro goto out; 1508ccd48bc7SAl Viro 15092d92ab3cSAl Viro err = graft_tree(mnt, path); 15101da177e4SLinus Torvalds if (err) { 151170fbcdf4SRam Pai LIST_HEAD(umount_list); 15121da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1513a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 15141da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 151570fbcdf4SRam Pai release_mounts(&umount_list); 15165b83d2c5SRam Pai } 15171da177e4SLinus Torvalds 1518ccd48bc7SAl Viro out: 1519390c6843SRam Pai up_write(&namespace_sem); 15202d92ab3cSAl Viro path_put(&old_path); 15211da177e4SLinus Torvalds return err; 15221da177e4SLinus Torvalds } 15231da177e4SLinus Torvalds 15242e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 15252e4b7fcdSDave Hansen { 15262e4b7fcdSDave Hansen int error = 0; 15272e4b7fcdSDave Hansen int readonly_request = 0; 15282e4b7fcdSDave Hansen 15292e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 15302e4b7fcdSDave Hansen readonly_request = 1; 15312e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 15322e4b7fcdSDave Hansen return 0; 15332e4b7fcdSDave Hansen 15342e4b7fcdSDave Hansen if (readonly_request) 15352e4b7fcdSDave Hansen error = mnt_make_readonly(mnt); 15362e4b7fcdSDave Hansen else 15372e4b7fcdSDave Hansen __mnt_unmake_readonly(mnt); 15382e4b7fcdSDave Hansen return error; 15392e4b7fcdSDave Hansen } 15402e4b7fcdSDave Hansen 15411da177e4SLinus Torvalds /* 15421da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 15431da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 15441da177e4SLinus Torvalds * on it - tough luck. 15451da177e4SLinus Torvalds */ 15460a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 15471da177e4SLinus Torvalds void *data) 15481da177e4SLinus Torvalds { 15491da177e4SLinus Torvalds int err; 15502d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 15511da177e4SLinus Torvalds 15521da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15531da177e4SLinus Torvalds return -EPERM; 15541da177e4SLinus Torvalds 15552d92ab3cSAl Viro if (!check_mnt(path->mnt)) 15561da177e4SLinus Torvalds return -EINVAL; 15571da177e4SLinus Torvalds 15582d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 15591da177e4SLinus Torvalds return -EINVAL; 15601da177e4SLinus Torvalds 15611da177e4SLinus Torvalds down_write(&sb->s_umount); 15622e4b7fcdSDave Hansen if (flags & MS_BIND) 15632d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 15642e4b7fcdSDave Hansen else 15651da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 15661da177e4SLinus Torvalds if (!err) 15672d92ab3cSAl Viro path->mnt->mnt_flags = mnt_flags; 15681da177e4SLinus Torvalds up_write(&sb->s_umount); 15690e55a7ccSDan Williams if (!err) { 15702d92ab3cSAl Viro security_sb_post_remount(path->mnt, flags, data); 15710e55a7ccSDan Williams 15720e55a7ccSDan Williams spin_lock(&vfsmount_lock); 15730e55a7ccSDan Williams touch_mnt_namespace(path->mnt->mnt_ns); 15740e55a7ccSDan Williams spin_unlock(&vfsmount_lock); 15750e55a7ccSDan Williams } 15761da177e4SLinus Torvalds return err; 15771da177e4SLinus Torvalds } 15781da177e4SLinus Torvalds 15799676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 15809676f0c6SRam Pai { 15819676f0c6SRam Pai struct vfsmount *p; 15829676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 15839676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 15849676f0c6SRam Pai return 1; 15859676f0c6SRam Pai } 15869676f0c6SRam Pai return 0; 15879676f0c6SRam Pai } 15889676f0c6SRam Pai 15890a0d8a46SAl Viro static int do_move_mount(struct path *path, char *old_name) 15901da177e4SLinus Torvalds { 15912d92ab3cSAl Viro struct path old_path, parent_path; 15921da177e4SLinus Torvalds struct vfsmount *p; 15931da177e4SLinus Torvalds int err = 0; 15941da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15951da177e4SLinus Torvalds return -EPERM; 15961da177e4SLinus Torvalds if (!old_name || !*old_name) 15971da177e4SLinus Torvalds return -EINVAL; 15982d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 15991da177e4SLinus Torvalds if (err) 16001da177e4SLinus Torvalds return err; 16011da177e4SLinus Torvalds 1602390c6843SRam Pai down_write(&namespace_sem); 16032d92ab3cSAl Viro while (d_mountpoint(path->dentry) && 16042d92ab3cSAl Viro follow_down(&path->mnt, &path->dentry)) 16051da177e4SLinus Torvalds ; 16061da177e4SLinus Torvalds err = -EINVAL; 16072d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 16081da177e4SLinus Torvalds goto out; 16091da177e4SLinus Torvalds 16101da177e4SLinus Torvalds err = -ENOENT; 16112d92ab3cSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 16122d92ab3cSAl Viro if (IS_DEADDIR(path->dentry->d_inode)) 16131da177e4SLinus Torvalds goto out1; 16141da177e4SLinus Torvalds 16152d92ab3cSAl Viro if (!IS_ROOT(path->dentry) && d_unhashed(path->dentry)) 161621444403SRam Pai goto out1; 16171da177e4SLinus Torvalds 16181da177e4SLinus Torvalds err = -EINVAL; 16192d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 162021444403SRam Pai goto out1; 16211da177e4SLinus Torvalds 16222d92ab3cSAl Viro if (old_path.mnt == old_path.mnt->mnt_parent) 162321444403SRam Pai goto out1; 16241da177e4SLinus Torvalds 16252d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 16262d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 162721444403SRam Pai goto out1; 162821444403SRam Pai /* 162921444403SRam Pai * Don't move a mount residing in a shared parent. 163021444403SRam Pai */ 16312d92ab3cSAl Viro if (old_path.mnt->mnt_parent && 16322d92ab3cSAl Viro IS_MNT_SHARED(old_path.mnt->mnt_parent)) 163321444403SRam Pai goto out1; 16349676f0c6SRam Pai /* 16359676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 16369676f0c6SRam Pai * mount which is shared. 16379676f0c6SRam Pai */ 16382d92ab3cSAl Viro if (IS_MNT_SHARED(path->mnt) && 16392d92ab3cSAl Viro tree_contains_unbindable(old_path.mnt)) 16409676f0c6SRam Pai goto out1; 16411da177e4SLinus Torvalds err = -ELOOP; 16422d92ab3cSAl Viro for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) 16432d92ab3cSAl Viro if (p == old_path.mnt) 164421444403SRam Pai goto out1; 16451da177e4SLinus Torvalds 16462d92ab3cSAl Viro err = attach_recursive_mnt(old_path.mnt, path, &parent_path); 16474ac91378SJan Blunck if (err) 164821444403SRam Pai goto out1; 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 16511da177e4SLinus Torvalds * automatically */ 16522d92ab3cSAl Viro list_del_init(&old_path.mnt->mnt_expire); 16531da177e4SLinus Torvalds out1: 16542d92ab3cSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 16551da177e4SLinus Torvalds out: 1656390c6843SRam Pai up_write(&namespace_sem); 16571da177e4SLinus Torvalds if (!err) 16581a390689SAl Viro path_put(&parent_path); 16592d92ab3cSAl Viro path_put(&old_path); 16601da177e4SLinus Torvalds return err; 16611da177e4SLinus Torvalds } 16621da177e4SLinus Torvalds 16631da177e4SLinus Torvalds /* 16641da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 16651da177e4SLinus Torvalds * namespace's tree 16661da177e4SLinus Torvalds */ 16670a0d8a46SAl Viro static int do_new_mount(struct path *path, char *type, int flags, 16681da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 16691da177e4SLinus Torvalds { 16701da177e4SLinus Torvalds struct vfsmount *mnt; 16711da177e4SLinus Torvalds 16721da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 16731da177e4SLinus Torvalds return -EINVAL; 16741da177e4SLinus Torvalds 16751da177e4SLinus Torvalds /* we need capabilities... */ 16761da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 16771da177e4SLinus Torvalds return -EPERM; 16781da177e4SLinus Torvalds 16791da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 16801da177e4SLinus Torvalds if (IS_ERR(mnt)) 16811da177e4SLinus Torvalds return PTR_ERR(mnt); 16821da177e4SLinus Torvalds 16832d92ab3cSAl Viro return do_add_mount(mnt, path, mnt_flags, NULL); 16841da177e4SLinus Torvalds } 16851da177e4SLinus Torvalds 16861da177e4SLinus Torvalds /* 16871da177e4SLinus Torvalds * add a mount into a namespace's mount tree 16881da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 16891da177e4SLinus Torvalds */ 16908d66bf54SAl Viro int do_add_mount(struct vfsmount *newmnt, struct path *path, 16911da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 16921da177e4SLinus Torvalds { 16931da177e4SLinus Torvalds int err; 16941da177e4SLinus Torvalds 1695390c6843SRam Pai down_write(&namespace_sem); 16961da177e4SLinus Torvalds /* Something was mounted here while we slept */ 16978d66bf54SAl Viro while (d_mountpoint(path->dentry) && 16988d66bf54SAl Viro follow_down(&path->mnt, &path->dentry)) 16991da177e4SLinus Torvalds ; 17001da177e4SLinus Torvalds err = -EINVAL; 17018d66bf54SAl Viro if (!check_mnt(path->mnt)) 17021da177e4SLinus Torvalds goto unlock; 17031da177e4SLinus Torvalds 17041da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 17051da177e4SLinus Torvalds err = -EBUSY; 17068d66bf54SAl Viro if (path->mnt->mnt_sb == newmnt->mnt_sb && 17078d66bf54SAl Viro path->mnt->mnt_root == path->dentry) 17081da177e4SLinus Torvalds goto unlock; 17091da177e4SLinus Torvalds 17101da177e4SLinus Torvalds err = -EINVAL; 17111da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 17121da177e4SLinus Torvalds goto unlock; 17131da177e4SLinus Torvalds 17141da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 17158d66bf54SAl Viro if ((err = graft_tree(newmnt, path))) 17165b83d2c5SRam Pai goto unlock; 17171da177e4SLinus Torvalds 17186758f953SAl Viro if (fslist) /* add to the specified expiration list */ 171955e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 17206758f953SAl Viro 1721390c6843SRam Pai up_write(&namespace_sem); 17225b83d2c5SRam Pai return 0; 17231da177e4SLinus Torvalds 17241da177e4SLinus Torvalds unlock: 1725390c6843SRam Pai up_write(&namespace_sem); 17261da177e4SLinus Torvalds mntput(newmnt); 17271da177e4SLinus Torvalds return err; 17281da177e4SLinus Torvalds } 17291da177e4SLinus Torvalds 17301da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 17311da177e4SLinus Torvalds 17325528f911STrond Myklebust /* 17331da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 17341da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 17351da177e4SLinus Torvalds * here 17361da177e4SLinus Torvalds */ 17371da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 17381da177e4SLinus Torvalds { 17391da177e4SLinus Torvalds struct vfsmount *mnt, *next; 17401da177e4SLinus Torvalds LIST_HEAD(graveyard); 1741bcc5c7d2SAl Viro LIST_HEAD(umounts); 17421da177e4SLinus Torvalds 17431da177e4SLinus Torvalds if (list_empty(mounts)) 17441da177e4SLinus Torvalds return; 17451da177e4SLinus Torvalds 1746bcc5c7d2SAl Viro down_write(&namespace_sem); 17471da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 17481da177e4SLinus Torvalds 17491da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 17501da177e4SLinus Torvalds * following criteria: 17511da177e4SLinus Torvalds * - only referenced by its parent vfsmount 17521da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 17531da177e4SLinus Torvalds * cleared by mntput()) 17541da177e4SLinus Torvalds */ 175555e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 17561da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 1757bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 17581da177e4SLinus Torvalds continue; 175955e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 17601da177e4SLinus Torvalds } 1761bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1762bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 1763bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1764bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 1765bcc5c7d2SAl Viro } 17661da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1767bcc5c7d2SAl Viro up_write(&namespace_sem); 1768bcc5c7d2SAl Viro 1769bcc5c7d2SAl Viro release_mounts(&umounts); 17701da177e4SLinus Torvalds } 17711da177e4SLinus Torvalds 17721da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 17731da177e4SLinus Torvalds 17741da177e4SLinus Torvalds /* 17755528f911STrond Myklebust * Ripoff of 'select_parent()' 17765528f911STrond Myklebust * 17775528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 17785528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 17795528f911STrond Myklebust */ 17805528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 17815528f911STrond Myklebust { 17825528f911STrond Myklebust struct vfsmount *this_parent = parent; 17835528f911STrond Myklebust struct list_head *next; 17845528f911STrond Myklebust int found = 0; 17855528f911STrond Myklebust 17865528f911STrond Myklebust repeat: 17875528f911STrond Myklebust next = this_parent->mnt_mounts.next; 17885528f911STrond Myklebust resume: 17895528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 17905528f911STrond Myklebust struct list_head *tmp = next; 17915528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 17925528f911STrond Myklebust 17935528f911STrond Myklebust next = tmp->next; 17945528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 17955528f911STrond Myklebust continue; 17965528f911STrond Myklebust /* 17975528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 17985528f911STrond Myklebust */ 17995528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 18005528f911STrond Myklebust this_parent = mnt; 18015528f911STrond Myklebust goto repeat; 18025528f911STrond Myklebust } 18035528f911STrond Myklebust 18045528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 18055528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 18065528f911STrond Myklebust found++; 18075528f911STrond Myklebust } 18085528f911STrond Myklebust } 18095528f911STrond Myklebust /* 18105528f911STrond Myklebust * All done at this level ... ascend and resume the search 18115528f911STrond Myklebust */ 18125528f911STrond Myklebust if (this_parent != parent) { 18135528f911STrond Myklebust next = this_parent->mnt_child.next; 18145528f911STrond Myklebust this_parent = this_parent->mnt_parent; 18155528f911STrond Myklebust goto resume; 18165528f911STrond Myklebust } 18175528f911STrond Myklebust return found; 18185528f911STrond Myklebust } 18195528f911STrond Myklebust 18205528f911STrond Myklebust /* 18215528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 18225528f911STrond Myklebust * submounts of a specific parent mountpoint 18235528f911STrond Myklebust */ 1824c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 18255528f911STrond Myklebust { 18265528f911STrond Myklebust LIST_HEAD(graveyard); 1827c35038beSAl Viro struct vfsmount *m; 18285528f911STrond Myklebust 18295528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 1830c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 1831bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1832c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 1833bcc5c7d2SAl Viro mnt_expire); 1834afef80b3SEric W. Biederman touch_mnt_namespace(m->mnt_ns); 1835afef80b3SEric W. Biederman umount_tree(m, 1, umounts); 1836bcc5c7d2SAl Viro } 1837bcc5c7d2SAl Viro } 18385528f911STrond Myklebust } 18395528f911STrond Myklebust 18405528f911STrond Myklebust /* 18411da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 18421da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 18431da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 18441da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 18451da177e4SLinus Torvalds */ 1846b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1847b58fed8bSRam Pai unsigned long n) 18481da177e4SLinus Torvalds { 18491da177e4SLinus Torvalds char *t = to; 18501da177e4SLinus Torvalds const char __user *f = from; 18511da177e4SLinus Torvalds char c; 18521da177e4SLinus Torvalds 18531da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 18541da177e4SLinus Torvalds return n; 18551da177e4SLinus Torvalds 18561da177e4SLinus Torvalds while (n) { 18571da177e4SLinus Torvalds if (__get_user(c, f)) { 18581da177e4SLinus Torvalds memset(t, 0, n); 18591da177e4SLinus Torvalds break; 18601da177e4SLinus Torvalds } 18611da177e4SLinus Torvalds *t++ = c; 18621da177e4SLinus Torvalds f++; 18631da177e4SLinus Torvalds n--; 18641da177e4SLinus Torvalds } 18651da177e4SLinus Torvalds return n; 18661da177e4SLinus Torvalds } 18671da177e4SLinus Torvalds 18681da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 18691da177e4SLinus Torvalds { 18701da177e4SLinus Torvalds int i; 18711da177e4SLinus Torvalds unsigned long page; 18721da177e4SLinus Torvalds unsigned long size; 18731da177e4SLinus Torvalds 18741da177e4SLinus Torvalds *where = 0; 18751da177e4SLinus Torvalds if (!data) 18761da177e4SLinus Torvalds return 0; 18771da177e4SLinus Torvalds 18781da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 18791da177e4SLinus Torvalds return -ENOMEM; 18801da177e4SLinus Torvalds 18811da177e4SLinus Torvalds /* We only care that *some* data at the address the user 18821da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 18831da177e4SLinus Torvalds * the remainder of the page. 18841da177e4SLinus Torvalds */ 18851da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 18861da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 18871da177e4SLinus Torvalds if (size > PAGE_SIZE) 18881da177e4SLinus Torvalds size = PAGE_SIZE; 18891da177e4SLinus Torvalds 18901da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 18911da177e4SLinus Torvalds if (!i) { 18921da177e4SLinus Torvalds free_page(page); 18931da177e4SLinus Torvalds return -EFAULT; 18941da177e4SLinus Torvalds } 18951da177e4SLinus Torvalds if (i != PAGE_SIZE) 18961da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 18971da177e4SLinus Torvalds *where = page; 18981da177e4SLinus Torvalds return 0; 18991da177e4SLinus Torvalds } 19001da177e4SLinus Torvalds 19011da177e4SLinus Torvalds /* 19021da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 19031da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 19041da177e4SLinus Torvalds * 19051da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 19061da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 19071da177e4SLinus Torvalds * information (or be NULL). 19081da177e4SLinus Torvalds * 19091da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 19101da177e4SLinus Torvalds * When the flags word was introduced its top half was required 19111da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 19121da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 19131da177e4SLinus Torvalds * and must be discarded. 19141da177e4SLinus Torvalds */ 19151da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 19161da177e4SLinus Torvalds unsigned long flags, void *data_page) 19171da177e4SLinus Torvalds { 19182d92ab3cSAl Viro struct path path; 19191da177e4SLinus Torvalds int retval = 0; 19201da177e4SLinus Torvalds int mnt_flags = 0; 19211da177e4SLinus Torvalds 19221da177e4SLinus Torvalds /* Discard magic */ 19231da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 19241da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 19251da177e4SLinus Torvalds 19261da177e4SLinus Torvalds /* Basic sanity checks */ 19271da177e4SLinus Torvalds 19281da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 19291da177e4SLinus Torvalds return -EINVAL; 19301da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 19311da177e4SLinus Torvalds return -EINVAL; 19321da177e4SLinus Torvalds 19331da177e4SLinus Torvalds if (data_page) 19341da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 19351da177e4SLinus Torvalds 1936613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 1937613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 19380a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 19390a1c01c9SMatthew Garrett 19401da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 19411da177e4SLinus Torvalds if (flags & MS_NOSUID) 19421da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 19431da177e4SLinus Torvalds if (flags & MS_NODEV) 19441da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 19451da177e4SLinus Torvalds if (flags & MS_NOEXEC) 19461da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 1947fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 1948fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 1949fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 1950fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 1951d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 1952d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 19532e4b7fcdSDave Hansen if (flags & MS_RDONLY) 19542e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 1955fc33a7bbSChristoph Hellwig 1956fc33a7bbSChristoph Hellwig flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 1957d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 1958d0adde57SMatthew Garrett MS_STRICTATIME); 19591da177e4SLinus Torvalds 19601da177e4SLinus Torvalds /* ... and get the mountpoint */ 19612d92ab3cSAl Viro retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 19621da177e4SLinus Torvalds if (retval) 19631da177e4SLinus Torvalds return retval; 19641da177e4SLinus Torvalds 19652d92ab3cSAl Viro retval = security_sb_mount(dev_name, &path, 1966b5266eb4SAl Viro type_page, flags, data_page); 19671da177e4SLinus Torvalds if (retval) 19681da177e4SLinus Torvalds goto dput_out; 19691da177e4SLinus Torvalds 19701da177e4SLinus Torvalds if (flags & MS_REMOUNT) 19712d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 19721da177e4SLinus Torvalds data_page); 19731da177e4SLinus Torvalds else if (flags & MS_BIND) 19742d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 19759676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 19762d92ab3cSAl Viro retval = do_change_type(&path, flags); 19771da177e4SLinus Torvalds else if (flags & MS_MOVE) 19782d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 19791da177e4SLinus Torvalds else 19802d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 19811da177e4SLinus Torvalds dev_name, data_page); 19821da177e4SLinus Torvalds dput_out: 19832d92ab3cSAl Viro path_put(&path); 19841da177e4SLinus Torvalds return retval; 19851da177e4SLinus Torvalds } 19861da177e4SLinus Torvalds 1987741a2951SJANAK DESAI /* 1988741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 1989741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 1990741a2951SJANAK DESAI */ 1991e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 19926b3286edSKirill Korotaev struct fs_struct *fs) 19931da177e4SLinus Torvalds { 19946b3286edSKirill Korotaev struct mnt_namespace *new_ns; 19957f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 19961da177e4SLinus Torvalds struct vfsmount *p, *q; 19971da177e4SLinus Torvalds 19986b3286edSKirill Korotaev new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 19991da177e4SLinus Torvalds if (!new_ns) 2000467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM); 20011da177e4SLinus Torvalds 20021da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 20031da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 20045addc5ddSAl Viro init_waitqueue_head(&new_ns->poll); 20055addc5ddSAl Viro new_ns->event = 0; 20061da177e4SLinus Torvalds 2007390c6843SRam Pai down_write(&namespace_sem); 20081da177e4SLinus Torvalds /* First pass: copy the tree topology */ 20096b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 20109676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 20111da177e4SLinus Torvalds if (!new_ns->root) { 2012390c6843SRam Pai up_write(&namespace_sem); 20131da177e4SLinus Torvalds kfree(new_ns); 20145cc4a034SJulia Lawall return ERR_PTR(-ENOMEM); 20151da177e4SLinus Torvalds } 20161da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 20171da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 20181da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 20191da177e4SLinus Torvalds 20201da177e4SLinus Torvalds /* 20211da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 20221da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 20231da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 20241da177e4SLinus Torvalds */ 20256b3286edSKirill Korotaev p = mnt_ns->root; 20261da177e4SLinus Torvalds q = new_ns->root; 20271da177e4SLinus Torvalds while (p) { 20286b3286edSKirill Korotaev q->mnt_ns = new_ns; 20291da177e4SLinus Torvalds if (fs) { 20306ac08c39SJan Blunck if (p == fs->root.mnt) { 20311da177e4SLinus Torvalds rootmnt = p; 20326ac08c39SJan Blunck fs->root.mnt = mntget(q); 20331da177e4SLinus Torvalds } 20346ac08c39SJan Blunck if (p == fs->pwd.mnt) { 20351da177e4SLinus Torvalds pwdmnt = p; 20366ac08c39SJan Blunck fs->pwd.mnt = mntget(q); 20371da177e4SLinus Torvalds } 20381da177e4SLinus Torvalds } 20396b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 20401da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 20411da177e4SLinus Torvalds } 2042390c6843SRam Pai up_write(&namespace_sem); 20431da177e4SLinus Torvalds 20441da177e4SLinus Torvalds if (rootmnt) 20451da177e4SLinus Torvalds mntput(rootmnt); 20461da177e4SLinus Torvalds if (pwdmnt) 20471da177e4SLinus Torvalds mntput(pwdmnt); 20481da177e4SLinus Torvalds 2049741a2951SJANAK DESAI return new_ns; 2050741a2951SJANAK DESAI } 2051741a2951SJANAK DESAI 2052213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2053e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 2054741a2951SJANAK DESAI { 20556b3286edSKirill Korotaev struct mnt_namespace *new_ns; 2056741a2951SJANAK DESAI 2057e3222c4eSBadari Pulavarty BUG_ON(!ns); 20586b3286edSKirill Korotaev get_mnt_ns(ns); 2059741a2951SJANAK DESAI 2060741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 2061e3222c4eSBadari Pulavarty return ns; 2062741a2951SJANAK DESAI 2063e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 2064741a2951SJANAK DESAI 20656b3286edSKirill Korotaev put_mnt_ns(ns); 2066e3222c4eSBadari Pulavarty return new_ns; 20671da177e4SLinus Torvalds } 20681da177e4SLinus Torvalds 2069bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2070bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 20711da177e4SLinus Torvalds { 20721da177e4SLinus Torvalds int retval; 20731da177e4SLinus Torvalds unsigned long data_page; 20741da177e4SLinus Torvalds unsigned long type_page; 20751da177e4SLinus Torvalds unsigned long dev_page; 20761da177e4SLinus Torvalds char *dir_page; 20771da177e4SLinus Torvalds 20781da177e4SLinus Torvalds retval = copy_mount_options(type, &type_page); 20791da177e4SLinus Torvalds if (retval < 0) 20801da177e4SLinus Torvalds return retval; 20811da177e4SLinus Torvalds 20821da177e4SLinus Torvalds dir_page = getname(dir_name); 20831da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 20841da177e4SLinus Torvalds if (IS_ERR(dir_page)) 20851da177e4SLinus Torvalds goto out1; 20861da177e4SLinus Torvalds 20871da177e4SLinus Torvalds retval = copy_mount_options(dev_name, &dev_page); 20881da177e4SLinus Torvalds if (retval < 0) 20891da177e4SLinus Torvalds goto out2; 20901da177e4SLinus Torvalds 20911da177e4SLinus Torvalds retval = copy_mount_options(data, &data_page); 20921da177e4SLinus Torvalds if (retval < 0) 20931da177e4SLinus Torvalds goto out3; 20941da177e4SLinus Torvalds 20951da177e4SLinus Torvalds lock_kernel(); 20961da177e4SLinus Torvalds retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 20971da177e4SLinus Torvalds flags, (void *)data_page); 20981da177e4SLinus Torvalds unlock_kernel(); 20991da177e4SLinus Torvalds free_page(data_page); 21001da177e4SLinus Torvalds 21011da177e4SLinus Torvalds out3: 21021da177e4SLinus Torvalds free_page(dev_page); 21031da177e4SLinus Torvalds out2: 21041da177e4SLinus Torvalds putname(dir_page); 21051da177e4SLinus Torvalds out1: 21061da177e4SLinus Torvalds free_page(type_page); 21071da177e4SLinus Torvalds return retval; 21081da177e4SLinus Torvalds } 21091da177e4SLinus Torvalds 21101da177e4SLinus Torvalds /* 21111da177e4SLinus Torvalds * pivot_root Semantics: 21121da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 21131da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 21141da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 21151da177e4SLinus Torvalds * 21161da177e4SLinus Torvalds * Restrictions: 21171da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 21181da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 21191da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 21201da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 21211da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 21221da177e4SLinus Torvalds * 21234a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 21244a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 21254a0d11faSNeil Brown * in this situation. 21264a0d11faSNeil Brown * 21271da177e4SLinus Torvalds * Notes: 21281da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 21291da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 21301da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 21311da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 21321da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 21331da177e4SLinus Torvalds * first. 21341da177e4SLinus Torvalds */ 21353480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 21363480b257SHeiko Carstens const char __user *, put_old) 21371da177e4SLinus Torvalds { 21381da177e4SLinus Torvalds struct vfsmount *tmp; 21392d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 21401da177e4SLinus Torvalds int error; 21411da177e4SLinus Torvalds 21421da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 21431da177e4SLinus Torvalds return -EPERM; 21441da177e4SLinus Torvalds 21452d8f3038SAl Viro error = user_path_dir(new_root, &new); 21461da177e4SLinus Torvalds if (error) 21471da177e4SLinus Torvalds goto out0; 21481da177e4SLinus Torvalds error = -EINVAL; 21492d8f3038SAl Viro if (!check_mnt(new.mnt)) 21501da177e4SLinus Torvalds goto out1; 21511da177e4SLinus Torvalds 21522d8f3038SAl Viro error = user_path_dir(put_old, &old); 21531da177e4SLinus Torvalds if (error) 21541da177e4SLinus Torvalds goto out1; 21551da177e4SLinus Torvalds 21562d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 21571da177e4SLinus Torvalds if (error) { 21582d8f3038SAl Viro path_put(&old); 21591da177e4SLinus Torvalds goto out1; 21601da177e4SLinus Torvalds } 21611da177e4SLinus Torvalds 21621da177e4SLinus Torvalds read_lock(¤t->fs->lock); 21638c3ee42eSAl Viro root = current->fs->root; 21646ac08c39SJan Blunck path_get(¤t->fs->root); 21651da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 2166390c6843SRam Pai down_write(&namespace_sem); 21672d8f3038SAl Viro mutex_lock(&old.dentry->d_inode->i_mutex); 21681da177e4SLinus Torvalds error = -EINVAL; 21692d8f3038SAl Viro if (IS_MNT_SHARED(old.mnt) || 21702d8f3038SAl Viro IS_MNT_SHARED(new.mnt->mnt_parent) || 21718c3ee42eSAl Viro IS_MNT_SHARED(root.mnt->mnt_parent)) 217221444403SRam Pai goto out2; 21738c3ee42eSAl Viro if (!check_mnt(root.mnt)) 21741da177e4SLinus Torvalds goto out2; 21751da177e4SLinus Torvalds error = -ENOENT; 21762d8f3038SAl Viro if (IS_DEADDIR(new.dentry->d_inode)) 21771da177e4SLinus Torvalds goto out2; 21782d8f3038SAl Viro if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry)) 21791da177e4SLinus Torvalds goto out2; 21802d8f3038SAl Viro if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry)) 21811da177e4SLinus Torvalds goto out2; 21821da177e4SLinus Torvalds error = -EBUSY; 21832d8f3038SAl Viro if (new.mnt == root.mnt || 21842d8f3038SAl Viro old.mnt == root.mnt) 21851da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 21861da177e4SLinus Torvalds error = -EINVAL; 21878c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 21881da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 21898c3ee42eSAl Viro if (root.mnt->mnt_parent == root.mnt) 21900bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 21912d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 21921da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 21932d8f3038SAl Viro if (new.mnt->mnt_parent == new.mnt) 21940bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 21954ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 21962d8f3038SAl Viro tmp = old.mnt; 21971da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 21982d8f3038SAl Viro if (tmp != new.mnt) { 21991da177e4SLinus Torvalds for (;;) { 22001da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 22011da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 22022d8f3038SAl Viro if (tmp->mnt_parent == new.mnt) 22031da177e4SLinus Torvalds break; 22041da177e4SLinus Torvalds tmp = tmp->mnt_parent; 22051da177e4SLinus Torvalds } 22062d8f3038SAl Viro if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) 22071da177e4SLinus Torvalds goto out3; 22082d8f3038SAl Viro } else if (!is_subdir(old.dentry, new.dentry)) 22091da177e4SLinus Torvalds goto out3; 22102d8f3038SAl Viro detach_mnt(new.mnt, &parent_path); 22118c3ee42eSAl Viro detach_mnt(root.mnt, &root_parent); 22124ac91378SJan Blunck /* mount old root on put_old */ 22132d8f3038SAl Viro attach_mnt(root.mnt, &old); 22144ac91378SJan Blunck /* mount new_root on / */ 22152d8f3038SAl Viro attach_mnt(new.mnt, &root_parent); 22166b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 22171da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 22182d8f3038SAl Viro chroot_fs_refs(&root, &new); 22192d8f3038SAl Viro security_sb_post_pivotroot(&root, &new); 22201da177e4SLinus Torvalds error = 0; 22211a390689SAl Viro path_put(&root_parent); 22221a390689SAl Viro path_put(&parent_path); 22231da177e4SLinus Torvalds out2: 22242d8f3038SAl Viro mutex_unlock(&old.dentry->d_inode->i_mutex); 2225390c6843SRam Pai up_write(&namespace_sem); 22268c3ee42eSAl Viro path_put(&root); 22272d8f3038SAl Viro path_put(&old); 22281da177e4SLinus Torvalds out1: 22292d8f3038SAl Viro path_put(&new); 22301da177e4SLinus Torvalds out0: 22311da177e4SLinus Torvalds return error; 22321da177e4SLinus Torvalds out3: 22331da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 22341da177e4SLinus Torvalds goto out2; 22351da177e4SLinus Torvalds } 22361da177e4SLinus Torvalds 22371da177e4SLinus Torvalds static void __init init_mount_tree(void) 22381da177e4SLinus Torvalds { 22391da177e4SLinus Torvalds struct vfsmount *mnt; 22406b3286edSKirill Korotaev struct mnt_namespace *ns; 2241ac748a09SJan Blunck struct path root; 22421da177e4SLinus Torvalds 22431da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 22441da177e4SLinus Torvalds if (IS_ERR(mnt)) 22451da177e4SLinus Torvalds panic("Can't create rootfs"); 22466b3286edSKirill Korotaev ns = kmalloc(sizeof(*ns), GFP_KERNEL); 22476b3286edSKirill Korotaev if (!ns) 22481da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 22496b3286edSKirill Korotaev atomic_set(&ns->count, 1); 22506b3286edSKirill Korotaev INIT_LIST_HEAD(&ns->list); 22516b3286edSKirill Korotaev init_waitqueue_head(&ns->poll); 22526b3286edSKirill Korotaev ns->event = 0; 22536b3286edSKirill Korotaev list_add(&mnt->mnt_list, &ns->list); 22546b3286edSKirill Korotaev ns->root = mnt; 22556b3286edSKirill Korotaev mnt->mnt_ns = ns; 22561da177e4SLinus Torvalds 22576b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 22586b3286edSKirill Korotaev get_mnt_ns(ns); 22591da177e4SLinus Torvalds 2260ac748a09SJan Blunck root.mnt = ns->root; 2261ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2262ac748a09SJan Blunck 2263ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2264ac748a09SJan Blunck set_fs_root(current->fs, &root); 22651da177e4SLinus Torvalds } 22661da177e4SLinus Torvalds 226774bf17cfSDenis Cheng void __init mnt_init(void) 22681da177e4SLinus Torvalds { 226913f14b4dSEric Dumazet unsigned u; 227015a67dd8SRandy Dunlap int err; 22711da177e4SLinus Torvalds 2272390c6843SRam Pai init_rwsem(&namespace_sem); 2273390c6843SRam Pai 22741da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 227520c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 22761da177e4SLinus Torvalds 2277b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 22781da177e4SLinus Torvalds 22791da177e4SLinus Torvalds if (!mount_hashtable) 22801da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 22811da177e4SLinus Torvalds 228213f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 22831da177e4SLinus Torvalds 228413f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 228513f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 22861da177e4SLinus Torvalds 228715a67dd8SRandy Dunlap err = sysfs_init(); 228815a67dd8SRandy Dunlap if (err) 228915a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 22908e24eea7SHarvey Harrison __func__, err); 229100d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 229200d26666SGreg Kroah-Hartman if (!fs_kobj) 22938e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 22941da177e4SLinus Torvalds init_rootfs(); 22951da177e4SLinus Torvalds init_mount_tree(); 22961da177e4SLinus Torvalds } 22971da177e4SLinus Torvalds 22986b3286edSKirill Korotaev void __put_mnt_ns(struct mnt_namespace *ns) 22991da177e4SLinus Torvalds { 23006b3286edSKirill Korotaev struct vfsmount *root = ns->root; 230170fbcdf4SRam Pai LIST_HEAD(umount_list); 23026b3286edSKirill Korotaev ns->root = NULL; 23031ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 2304390c6843SRam Pai down_write(&namespace_sem); 23051da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 2306a05964f3SRam Pai umount_tree(root, 0, &umount_list); 23071da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 2308390c6843SRam Pai up_write(&namespace_sem); 230970fbcdf4SRam Pai release_mounts(&umount_list); 23106b3286edSKirill Korotaev kfree(ns); 23111da177e4SLinus Torvalds } 2312