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/quotaops.h> 181da177e4SLinus Torvalds #include <linux/acct.h> 1916f7e0feSRandy Dunlap #include <linux/capability.h> 203d733633SDave Hansen #include <linux/cpumask.h> 211da177e4SLinus Torvalds #include <linux/module.h> 22f20a9eadSAndrew Morton #include <linux/sysfs.h> 231da177e4SLinus Torvalds #include <linux/seq_file.h> 246b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 251da177e4SLinus Torvalds #include <linux/namei.h> 261da177e4SLinus Torvalds #include <linux/security.h> 271da177e4SLinus Torvalds #include <linux/mount.h> 2807f3f05cSDavid Howells #include <linux/ramfs.h> 2913f14b4dSEric Dumazet #include <linux/log2.h> 3073cd49ecSMiklos Szeredi #include <linux/idr.h> 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); 11673cd49ecSMiklos Szeredi if (err) { 11773cd49ecSMiklos Szeredi kmem_cache_free(mnt_cache, mnt); 11873cd49ecSMiklos Szeredi return NULL; 11973cd49ecSMiklos Szeredi } 12073cd49ecSMiklos Szeredi 1211da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 1221da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 1231da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 1241da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 1251da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 12655e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 12703e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 128a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 129a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 1303d733633SDave Hansen atomic_set(&mnt->__mnt_writers, 0); 1311da177e4SLinus Torvalds if (name) { 1321da177e4SLinus Torvalds int size = strlen(name) + 1; 1331da177e4SLinus Torvalds char *newname = kmalloc(size, GFP_KERNEL); 1341da177e4SLinus Torvalds if (newname) { 1351da177e4SLinus Torvalds memcpy(newname, name, size); 1361da177e4SLinus Torvalds mnt->mnt_devname = newname; 1371da177e4SLinus Torvalds } 1381da177e4SLinus Torvalds } 1391da177e4SLinus Torvalds } 1401da177e4SLinus Torvalds return mnt; 1411da177e4SLinus Torvalds } 1421da177e4SLinus Torvalds 1438366025eSDave Hansen /* 1448366025eSDave Hansen * Most r/o checks on a fs are for operations that take 1458366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 1468366025eSDave Hansen * We must keep track of when those operations start 1478366025eSDave Hansen * (for permission checks) and when they end, so that 1488366025eSDave Hansen * we can determine when writes are able to occur to 1498366025eSDave Hansen * a filesystem. 1508366025eSDave Hansen */ 1513d733633SDave Hansen /* 1523d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 1533d733633SDave Hansen * @mnt: the mount to check for its write status 1543d733633SDave Hansen * 1553d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 1563d733633SDave Hansen * It does not guarantee that the filesystem will stay 1573d733633SDave Hansen * r/w, just that it is right *now*. This can not and 1583d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 1593d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 1603d733633SDave Hansen * r/w. 1613d733633SDave Hansen */ 1623d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 1633d733633SDave Hansen { 1642e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 1652e4b7fcdSDave Hansen return 1; 1662e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 1672e4b7fcdSDave Hansen return 1; 1682e4b7fcdSDave Hansen return 0; 1693d733633SDave Hansen } 1703d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 1713d733633SDave Hansen 1723d733633SDave Hansen struct mnt_writer { 1733d733633SDave Hansen /* 1743d733633SDave Hansen * If holding multiple instances of this lock, they 1753d733633SDave Hansen * must be ordered by cpu number. 1763d733633SDave Hansen */ 1773d733633SDave Hansen spinlock_t lock; 1783d733633SDave Hansen struct lock_class_key lock_class; /* compiles out with !lockdep */ 1793d733633SDave Hansen unsigned long count; 1803d733633SDave Hansen struct vfsmount *mnt; 1813d733633SDave Hansen } ____cacheline_aligned_in_smp; 1823d733633SDave Hansen static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); 1833d733633SDave Hansen 1843d733633SDave Hansen static int __init init_mnt_writers(void) 1853d733633SDave Hansen { 1863d733633SDave Hansen int cpu; 1873d733633SDave Hansen for_each_possible_cpu(cpu) { 1883d733633SDave Hansen struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); 1893d733633SDave Hansen spin_lock_init(&writer->lock); 1903d733633SDave Hansen lockdep_set_class(&writer->lock, &writer->lock_class); 1913d733633SDave Hansen writer->count = 0; 1923d733633SDave Hansen } 1933d733633SDave Hansen return 0; 1943d733633SDave Hansen } 1953d733633SDave Hansen fs_initcall(init_mnt_writers); 1963d733633SDave Hansen 1973d733633SDave Hansen static void unlock_mnt_writers(void) 1983d733633SDave Hansen { 1993d733633SDave Hansen int cpu; 2003d733633SDave Hansen struct mnt_writer *cpu_writer; 2013d733633SDave Hansen 2023d733633SDave Hansen for_each_possible_cpu(cpu) { 2033d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 2043d733633SDave Hansen spin_unlock(&cpu_writer->lock); 2053d733633SDave Hansen } 2063d733633SDave Hansen } 2073d733633SDave Hansen 2083d733633SDave Hansen static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) 2093d733633SDave Hansen { 2103d733633SDave Hansen if (!cpu_writer->mnt) 2113d733633SDave Hansen return; 2123d733633SDave Hansen /* 2133d733633SDave Hansen * This is in case anyone ever leaves an invalid, 2143d733633SDave Hansen * old ->mnt and a count of 0. 2153d733633SDave Hansen */ 2163d733633SDave Hansen if (!cpu_writer->count) 2173d733633SDave Hansen return; 2183d733633SDave Hansen atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); 2193d733633SDave Hansen cpu_writer->count = 0; 2203d733633SDave Hansen } 2213d733633SDave Hansen /* 2223d733633SDave Hansen * must hold cpu_writer->lock 2233d733633SDave Hansen */ 2243d733633SDave Hansen static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, 2253d733633SDave Hansen struct vfsmount *mnt) 2263d733633SDave Hansen { 2273d733633SDave Hansen if (cpu_writer->mnt == mnt) 2283d733633SDave Hansen return; 2293d733633SDave Hansen __clear_mnt_count(cpu_writer); 2303d733633SDave Hansen cpu_writer->mnt = mnt; 2313d733633SDave Hansen } 2323d733633SDave Hansen 2333d733633SDave Hansen /* 2343d733633SDave Hansen * Most r/o checks on a fs are for operations that take 2353d733633SDave Hansen * discrete amounts of time, like a write() or unlink(). 2363d733633SDave Hansen * We must keep track of when those operations start 2373d733633SDave Hansen * (for permission checks) and when they end, so that 2383d733633SDave Hansen * we can determine when writes are able to occur to 2393d733633SDave Hansen * a filesystem. 2403d733633SDave Hansen */ 2418366025eSDave Hansen /** 2428366025eSDave Hansen * mnt_want_write - get write access to a mount 2438366025eSDave Hansen * @mnt: the mount on which to take a write 2448366025eSDave Hansen * 2458366025eSDave Hansen * This tells the low-level filesystem that a write is 2468366025eSDave Hansen * about to be performed to it, and makes sure that 2478366025eSDave Hansen * writes are allowed before returning success. When 2488366025eSDave Hansen * the write operation is finished, mnt_drop_write() 2498366025eSDave Hansen * must be called. This is effectively a refcount. 2508366025eSDave Hansen */ 2518366025eSDave Hansen int mnt_want_write(struct vfsmount *mnt) 2528366025eSDave Hansen { 2533d733633SDave Hansen int ret = 0; 2543d733633SDave Hansen struct mnt_writer *cpu_writer; 2553d733633SDave Hansen 2563d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 2573d733633SDave Hansen spin_lock(&cpu_writer->lock); 2583d733633SDave Hansen if (__mnt_is_readonly(mnt)) { 2593d733633SDave Hansen ret = -EROFS; 2603d733633SDave Hansen goto out; 2613d733633SDave Hansen } 2623d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 2633d733633SDave Hansen cpu_writer->count++; 2643d733633SDave Hansen out: 2653d733633SDave Hansen spin_unlock(&cpu_writer->lock); 2663d733633SDave Hansen put_cpu_var(mnt_writers); 2673d733633SDave Hansen return ret; 2688366025eSDave Hansen } 2698366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 2708366025eSDave Hansen 2713d733633SDave Hansen static void lock_mnt_writers(void) 2723d733633SDave Hansen { 2733d733633SDave Hansen int cpu; 2743d733633SDave Hansen struct mnt_writer *cpu_writer; 2753d733633SDave Hansen 2763d733633SDave Hansen for_each_possible_cpu(cpu) { 2773d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 2783d733633SDave Hansen spin_lock(&cpu_writer->lock); 2793d733633SDave Hansen __clear_mnt_count(cpu_writer); 2803d733633SDave Hansen cpu_writer->mnt = NULL; 2813d733633SDave Hansen } 2823d733633SDave Hansen } 2833d733633SDave Hansen 2843d733633SDave Hansen /* 2853d733633SDave Hansen * These per-cpu write counts are not guaranteed to have 2863d733633SDave Hansen * matched increments and decrements on any given cpu. 2873d733633SDave Hansen * A file open()ed for write on one cpu and close()d on 2883d733633SDave Hansen * another cpu will imbalance this count. Make sure it 2893d733633SDave Hansen * does not get too far out of whack. 2903d733633SDave Hansen */ 2913d733633SDave Hansen static void handle_write_count_underflow(struct vfsmount *mnt) 2923d733633SDave Hansen { 2933d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) >= 2943d733633SDave Hansen MNT_WRITER_UNDERFLOW_LIMIT) 2953d733633SDave Hansen return; 2963d733633SDave Hansen /* 2973d733633SDave Hansen * It isn't necessary to hold all of the locks 2983d733633SDave Hansen * at the same time, but doing it this way makes 2993d733633SDave Hansen * us share a lot more code. 3003d733633SDave Hansen */ 3013d733633SDave Hansen lock_mnt_writers(); 3023d733633SDave Hansen /* 3033d733633SDave Hansen * vfsmount_lock is for mnt_flags. 3043d733633SDave Hansen */ 3053d733633SDave Hansen spin_lock(&vfsmount_lock); 3063d733633SDave Hansen /* 3073d733633SDave Hansen * If coalescing the per-cpu writer counts did not 3083d733633SDave Hansen * get us back to a positive writer count, we have 3093d733633SDave Hansen * a bug. 3103d733633SDave Hansen */ 3113d733633SDave Hansen if ((atomic_read(&mnt->__mnt_writers) < 0) && 3123d733633SDave Hansen !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { 3133d733633SDave Hansen printk(KERN_DEBUG "leak detected on mount(%p) writers " 3143d733633SDave Hansen "count: %d\n", 3153d733633SDave Hansen mnt, atomic_read(&mnt->__mnt_writers)); 3163d733633SDave Hansen WARN_ON(1); 3173d733633SDave Hansen /* use the flag to keep the dmesg spam down */ 3183d733633SDave Hansen mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT; 3193d733633SDave Hansen } 3203d733633SDave Hansen spin_unlock(&vfsmount_lock); 3213d733633SDave Hansen unlock_mnt_writers(); 3223d733633SDave Hansen } 3233d733633SDave Hansen 3248366025eSDave Hansen /** 3258366025eSDave Hansen * mnt_drop_write - give up write access to a mount 3268366025eSDave Hansen * @mnt: the mount on which to give up write access 3278366025eSDave Hansen * 3288366025eSDave Hansen * Tells the low-level filesystem that we are done 3298366025eSDave Hansen * performing writes to it. Must be matched with 3308366025eSDave Hansen * mnt_want_write() call above. 3318366025eSDave Hansen */ 3328366025eSDave Hansen void mnt_drop_write(struct vfsmount *mnt) 3338366025eSDave Hansen { 3343d733633SDave Hansen int must_check_underflow = 0; 3353d733633SDave Hansen struct mnt_writer *cpu_writer; 3363d733633SDave Hansen 3373d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 3383d733633SDave Hansen spin_lock(&cpu_writer->lock); 3393d733633SDave Hansen 3403d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 3413d733633SDave Hansen if (cpu_writer->count > 0) { 3423d733633SDave Hansen cpu_writer->count--; 3433d733633SDave Hansen } else { 3443d733633SDave Hansen must_check_underflow = 1; 3453d733633SDave Hansen atomic_dec(&mnt->__mnt_writers); 3463d733633SDave Hansen } 3473d733633SDave Hansen 3483d733633SDave Hansen spin_unlock(&cpu_writer->lock); 3493d733633SDave Hansen /* 3503d733633SDave Hansen * Logically, we could call this each time, 3513d733633SDave Hansen * but the __mnt_writers cacheline tends to 3523d733633SDave Hansen * be cold, and makes this expensive. 3533d733633SDave Hansen */ 3543d733633SDave Hansen if (must_check_underflow) 3553d733633SDave Hansen handle_write_count_underflow(mnt); 3563d733633SDave Hansen /* 3573d733633SDave Hansen * This could be done right after the spinlock 3583d733633SDave Hansen * is taken because the spinlock keeps us on 3593d733633SDave Hansen * the cpu, and disables preemption. However, 3603d733633SDave Hansen * putting it here bounds the amount that 3613d733633SDave Hansen * __mnt_writers can underflow. Without it, 3623d733633SDave Hansen * we could theoretically wrap __mnt_writers. 3633d733633SDave Hansen */ 3643d733633SDave Hansen put_cpu_var(mnt_writers); 3658366025eSDave Hansen } 3668366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 3678366025eSDave Hansen 3682e4b7fcdSDave Hansen static int mnt_make_readonly(struct vfsmount *mnt) 3698366025eSDave Hansen { 3703d733633SDave Hansen int ret = 0; 3713d733633SDave Hansen 3723d733633SDave Hansen lock_mnt_writers(); 3733d733633SDave Hansen /* 3743d733633SDave Hansen * With all the locks held, this value is stable 3753d733633SDave Hansen */ 3763d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) > 0) { 3773d733633SDave Hansen ret = -EBUSY; 3783d733633SDave Hansen goto out; 3798366025eSDave Hansen } 3803d733633SDave Hansen /* 3812e4b7fcdSDave Hansen * nobody can do a successful mnt_want_write() with all 3822e4b7fcdSDave Hansen * of the counts in MNT_DENIED_WRITE and the locks held. 3833d733633SDave Hansen */ 3842e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 3852e4b7fcdSDave Hansen if (!ret) 3862e4b7fcdSDave Hansen mnt->mnt_flags |= MNT_READONLY; 3872e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3883d733633SDave Hansen out: 3893d733633SDave Hansen unlock_mnt_writers(); 3903d733633SDave Hansen return ret; 3913d733633SDave Hansen } 3928366025eSDave Hansen 3932e4b7fcdSDave Hansen static void __mnt_unmake_readonly(struct vfsmount *mnt) 3942e4b7fcdSDave Hansen { 3952e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 3962e4b7fcdSDave Hansen mnt->mnt_flags &= ~MNT_READONLY; 3972e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3982e4b7fcdSDave Hansen } 3992e4b7fcdSDave Hansen 400454e2398SDavid Howells int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 401454e2398SDavid Howells { 402454e2398SDavid Howells mnt->mnt_sb = sb; 403454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 404454e2398SDavid Howells return 0; 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 if (cpu_writer->mnt != mnt) 6183d733633SDave Hansen continue; 6193d733633SDave Hansen spin_lock(&cpu_writer->lock); 6203d733633SDave Hansen atomic_add(cpu_writer->count, &mnt->__mnt_writers); 6213d733633SDave Hansen cpu_writer->count = 0; 6223d733633SDave Hansen /* 6233d733633SDave Hansen * Might as well do this so that no one 6243d733633SDave Hansen * ever sees the pointer and expects 6253d733633SDave Hansen * it to be valid. 6263d733633SDave Hansen */ 6273d733633SDave Hansen cpu_writer->mnt = NULL; 6283d733633SDave Hansen spin_unlock(&cpu_writer->lock); 6293d733633SDave Hansen } 6303d733633SDave Hansen /* 6313d733633SDave Hansen * This probably indicates that somebody messed 6323d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 6333d733633SDave Hansen * happens, the filesystem was probably unable 6343d733633SDave Hansen * to make r/w->r/o transitions. 6353d733633SDave Hansen */ 6363d733633SDave Hansen WARN_ON(atomic_read(&mnt->__mnt_writers)); 6371da177e4SLinus Torvalds dput(mnt->mnt_root); 6381da177e4SLinus Torvalds free_vfsmnt(mnt); 6391da177e4SLinus Torvalds deactivate_super(sb); 6401da177e4SLinus Torvalds } 6411da177e4SLinus Torvalds 6427b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 6437b7b1aceSAl Viro { 6447b7b1aceSAl Viro repeat: 6457b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 6467b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 6477b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6487b7b1aceSAl Viro __mntput(mnt); 6497b7b1aceSAl Viro return; 6507b7b1aceSAl Viro } 6517b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 6527b7b1aceSAl Viro mnt->mnt_pinned = 0; 6537b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6547b7b1aceSAl Viro acct_auto_close_mnt(mnt); 6557b7b1aceSAl Viro security_sb_umount_close(mnt); 6567b7b1aceSAl Viro goto repeat; 6577b7b1aceSAl Viro } 6587b7b1aceSAl Viro } 6597b7b1aceSAl Viro 6607b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 6617b7b1aceSAl Viro 6627b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 6637b7b1aceSAl Viro { 6647b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6657b7b1aceSAl Viro mnt->mnt_pinned++; 6667b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6677b7b1aceSAl Viro } 6687b7b1aceSAl Viro 6697b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 6707b7b1aceSAl Viro 6717b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 6727b7b1aceSAl Viro { 6737b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6747b7b1aceSAl Viro if (mnt->mnt_pinned) { 6757b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 6767b7b1aceSAl Viro mnt->mnt_pinned--; 6777b7b1aceSAl Viro } 6787b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6797b7b1aceSAl Viro } 6807b7b1aceSAl Viro 6817b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 6821da177e4SLinus Torvalds 683b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 684b3b304a2SMiklos Szeredi { 685b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 686b3b304a2SMiklos Szeredi } 687b3b304a2SMiklos Szeredi 688b3b304a2SMiklos Szeredi /* 689b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 690b3b304a2SMiklos Szeredi * implement more complex mount option showing. 691b3b304a2SMiklos Szeredi * 692b3b304a2SMiklos Szeredi * See also save_mount_options(). 693b3b304a2SMiklos Szeredi */ 694b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 695b3b304a2SMiklos Szeredi { 696b3b304a2SMiklos Szeredi const char *options = mnt->mnt_sb->s_options; 697b3b304a2SMiklos Szeredi 698b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 699b3b304a2SMiklos Szeredi seq_putc(m, ','); 700b3b304a2SMiklos Szeredi mangle(m, options); 701b3b304a2SMiklos Szeredi } 702b3b304a2SMiklos Szeredi 703b3b304a2SMiklos Szeredi return 0; 704b3b304a2SMiklos Szeredi } 705b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 706b3b304a2SMiklos Szeredi 707b3b304a2SMiklos Szeredi /* 708b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 709b3b304a2SMiklos Szeredi * called from the fill_super() callback. 710b3b304a2SMiklos Szeredi * 711b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 712b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 713b3b304a2SMiklos Szeredi * remount fails. 714b3b304a2SMiklos Szeredi * 715b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 716b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 717b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 718b3b304a2SMiklos Szeredi * any more. 719b3b304a2SMiklos Szeredi */ 720b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 721b3b304a2SMiklos Szeredi { 722b3b304a2SMiklos Szeredi kfree(sb->s_options); 723b3b304a2SMiklos Szeredi sb->s_options = kstrdup(options, GFP_KERNEL); 724b3b304a2SMiklos Szeredi } 725b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 726b3b304a2SMiklos Szeredi 727a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 7281da177e4SLinus Torvalds /* iterator */ 7291da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 7301da177e4SLinus Torvalds { 731a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 7321da177e4SLinus Torvalds 733390c6843SRam Pai down_read(&namespace_sem); 734a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 7351da177e4SLinus Torvalds } 7361da177e4SLinus Torvalds 7371da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 7381da177e4SLinus Torvalds { 739a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 740b0765fb8SPavel Emelianov 741a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 7421da177e4SLinus Torvalds } 7431da177e4SLinus Torvalds 7441da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 7451da177e4SLinus Torvalds { 746390c6843SRam Pai up_read(&namespace_sem); 7471da177e4SLinus Torvalds } 7481da177e4SLinus Torvalds 7491da177e4SLinus Torvalds static int show_vfsmnt(struct seq_file *m, void *v) 7501da177e4SLinus Torvalds { 751b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 7521da177e4SLinus Torvalds int err = 0; 7531da177e4SLinus Torvalds static struct proc_fs_info { 7541da177e4SLinus Torvalds int flag; 7551da177e4SLinus Torvalds char *str; 7561da177e4SLinus Torvalds } fs_info[] = { 7571da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 7581da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 7591da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 7601da177e4SLinus Torvalds { 0, NULL } 7611da177e4SLinus Torvalds }; 7621da177e4SLinus Torvalds static struct proc_fs_info mnt_info[] = { 7631da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 7641da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 7651da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 766fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 767fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 76847ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 7691da177e4SLinus Torvalds { 0, NULL } 7701da177e4SLinus Torvalds }; 7711da177e4SLinus Torvalds struct proc_fs_info *fs_infop; 772c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 7731da177e4SLinus Torvalds 7741da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 7751da177e4SLinus Torvalds seq_putc(m, ' '); 776c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 7771da177e4SLinus Torvalds seq_putc(m, ' '); 7781da177e4SLinus Torvalds mangle(m, mnt->mnt_sb->s_type->name); 77979c0b2dfSMiklos Szeredi if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { 78079c0b2dfSMiklos Szeredi seq_putc(m, '.'); 78179c0b2dfSMiklos Szeredi mangle(m, mnt->mnt_sb->s_subtype); 78279c0b2dfSMiklos Szeredi } 7832e4b7fcdSDave Hansen seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 7841da177e4SLinus Torvalds for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 7851da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & fs_infop->flag) 7861da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 7871da177e4SLinus Torvalds } 7881da177e4SLinus Torvalds for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 7891da177e4SLinus Torvalds if (mnt->mnt_flags & fs_infop->flag) 7901da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 7911da177e4SLinus Torvalds } 7921da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 7931da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 7941da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 7951da177e4SLinus Torvalds return err; 7961da177e4SLinus Torvalds } 7971da177e4SLinus Torvalds 798a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 7991da177e4SLinus Torvalds .start = m_start, 8001da177e4SLinus Torvalds .next = m_next, 8011da177e4SLinus Torvalds .stop = m_stop, 8021da177e4SLinus Torvalds .show = show_vfsmnt 8031da177e4SLinus Torvalds }; 8041da177e4SLinus Torvalds 805b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 806b4629fe2SChuck Lever { 807b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 808c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 809b4629fe2SChuck Lever int err = 0; 810b4629fe2SChuck Lever 811b4629fe2SChuck Lever /* device */ 812b4629fe2SChuck Lever if (mnt->mnt_devname) { 813b4629fe2SChuck Lever seq_puts(m, "device "); 814b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 815b4629fe2SChuck Lever } else 816b4629fe2SChuck Lever seq_puts(m, "no device"); 817b4629fe2SChuck Lever 818b4629fe2SChuck Lever /* mount point */ 819b4629fe2SChuck Lever seq_puts(m, " mounted on "); 820c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 821b4629fe2SChuck Lever seq_putc(m, ' '); 822b4629fe2SChuck Lever 823b4629fe2SChuck Lever /* file system type */ 824b4629fe2SChuck Lever seq_puts(m, "with fstype "); 825b4629fe2SChuck Lever mangle(m, mnt->mnt_sb->s_type->name); 826b4629fe2SChuck Lever 827b4629fe2SChuck Lever /* optional statistics */ 828b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 829b4629fe2SChuck Lever seq_putc(m, ' '); 830b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 831b4629fe2SChuck Lever } 832b4629fe2SChuck Lever 833b4629fe2SChuck Lever seq_putc(m, '\n'); 834b4629fe2SChuck Lever return err; 835b4629fe2SChuck Lever } 836b4629fe2SChuck Lever 837a1a2c409SMiklos Szeredi const struct seq_operations mountstats_op = { 838b4629fe2SChuck Lever .start = m_start, 839b4629fe2SChuck Lever .next = m_next, 840b4629fe2SChuck Lever .stop = m_stop, 841b4629fe2SChuck Lever .show = show_vfsstat, 842b4629fe2SChuck Lever }; 843a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 844b4629fe2SChuck Lever 8451da177e4SLinus Torvalds /** 8461da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 8471da177e4SLinus Torvalds * @mnt: root of mount tree 8481da177e4SLinus Torvalds * 8491da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 8501da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 8511da177e4SLinus Torvalds * busy. 8521da177e4SLinus Torvalds */ 8531da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 8541da177e4SLinus Torvalds { 85536341f64SRam Pai int actual_refs = 0; 85636341f64SRam Pai int minimum_refs = 0; 85736341f64SRam Pai struct vfsmount *p; 8581da177e4SLinus Torvalds 8591da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 86036341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 8611da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 8621da177e4SLinus Torvalds minimum_refs += 2; 8631da177e4SLinus Torvalds } 8641da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 8651da177e4SLinus Torvalds 8661da177e4SLinus Torvalds if (actual_refs > minimum_refs) 8671da177e4SLinus Torvalds return 0; 868e3474a8eSIan Kent 869e3474a8eSIan Kent return 1; 8701da177e4SLinus Torvalds } 8711da177e4SLinus Torvalds 8721da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 8731da177e4SLinus Torvalds 8741da177e4SLinus Torvalds /** 8751da177e4SLinus Torvalds * may_umount - check if a mount point is busy 8761da177e4SLinus Torvalds * @mnt: root of mount 8771da177e4SLinus Torvalds * 8781da177e4SLinus Torvalds * This is called to check if a mount point has any 8791da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 8801da177e4SLinus Torvalds * mount has sub mounts this will return busy 8811da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 8821da177e4SLinus Torvalds * 8831da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 8841da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 8851da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 8861da177e4SLinus Torvalds */ 8871da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 8881da177e4SLinus Torvalds { 889e3474a8eSIan Kent int ret = 1; 890a05964f3SRam Pai spin_lock(&vfsmount_lock); 891a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 892e3474a8eSIan Kent ret = 0; 893a05964f3SRam Pai spin_unlock(&vfsmount_lock); 894a05964f3SRam Pai return ret; 8951da177e4SLinus Torvalds } 8961da177e4SLinus Torvalds 8971da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 8981da177e4SLinus Torvalds 899b90fa9aeSRam Pai void release_mounts(struct list_head *head) 9001da177e4SLinus Torvalds { 90170fbcdf4SRam Pai struct vfsmount *mnt; 90270fbcdf4SRam Pai while (!list_empty(head)) { 903b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 90470fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 90570fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 90670fbcdf4SRam Pai struct dentry *dentry; 90770fbcdf4SRam Pai struct vfsmount *m; 90870fbcdf4SRam Pai spin_lock(&vfsmount_lock); 90970fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 91070fbcdf4SRam Pai m = mnt->mnt_parent; 91170fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 91270fbcdf4SRam Pai mnt->mnt_parent = mnt; 9137c4b93d8SAl Viro m->mnt_ghosts--; 9141da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 91570fbcdf4SRam Pai dput(dentry); 91670fbcdf4SRam Pai mntput(m); 9171da177e4SLinus Torvalds } 9181da177e4SLinus Torvalds mntput(mnt); 91970fbcdf4SRam Pai } 92070fbcdf4SRam Pai } 92170fbcdf4SRam Pai 922a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 92370fbcdf4SRam Pai { 92470fbcdf4SRam Pai struct vfsmount *p; 92570fbcdf4SRam Pai 9261bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 9271bfba4e8SAkinobu Mita list_move(&p->mnt_hash, kill); 92870fbcdf4SRam Pai 929a05964f3SRam Pai if (propagate) 930a05964f3SRam Pai propagate_umount(kill); 931a05964f3SRam Pai 93270fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 93370fbcdf4SRam Pai list_del_init(&p->mnt_expire); 93470fbcdf4SRam Pai list_del_init(&p->mnt_list); 9356b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 9366b3286edSKirill Korotaev p->mnt_ns = NULL; 93770fbcdf4SRam Pai list_del_init(&p->mnt_child); 9387c4b93d8SAl Viro if (p->mnt_parent != p) { 9397c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 940f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 9417c4b93d8SAl Viro } 942a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 9431da177e4SLinus Torvalds } 9441da177e4SLinus Torvalds } 9451da177e4SLinus Torvalds 946c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 947c35038beSAl Viro 9481da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 9491da177e4SLinus Torvalds { 9501da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 9511da177e4SLinus Torvalds int retval; 95270fbcdf4SRam Pai LIST_HEAD(umount_list); 9531da177e4SLinus Torvalds 9541da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 9551da177e4SLinus Torvalds if (retval) 9561da177e4SLinus Torvalds return retval; 9571da177e4SLinus Torvalds 9581da177e4SLinus Torvalds /* 9591da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 9601da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 9611da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 9621da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 9631da177e4SLinus Torvalds */ 9641da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 9656ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 9661da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 9671da177e4SLinus Torvalds return -EINVAL; 9681da177e4SLinus Torvalds 9691da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 9701da177e4SLinus Torvalds return -EBUSY; 9711da177e4SLinus Torvalds 9721da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 9731da177e4SLinus Torvalds return -EAGAIN; 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 9761da177e4SLinus Torvalds /* 9771da177e4SLinus Torvalds * If we may have to abort operations to get out of this 9781da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 9791da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 9801da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 9811da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 9821da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 9831da177e4SLinus Torvalds * about for the moment. 9841da177e4SLinus Torvalds */ 9851da177e4SLinus Torvalds 9861da177e4SLinus Torvalds lock_kernel(); 9878b512d9aSTrond Myklebust if (sb->s_op->umount_begin) 9888b512d9aSTrond Myklebust sb->s_op->umount_begin(mnt, flags); 9891da177e4SLinus Torvalds unlock_kernel(); 9901da177e4SLinus Torvalds 9911da177e4SLinus Torvalds /* 9921da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 9931da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 9941da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 9951da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 9961da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 9971da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 9981da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 9991da177e4SLinus Torvalds */ 10006ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 10011da177e4SLinus Torvalds /* 10021da177e4SLinus Torvalds * Special case for "unmounting" root ... 10031da177e4SLinus Torvalds * we just try to remount it readonly. 10041da177e4SLinus Torvalds */ 10051da177e4SLinus Torvalds down_write(&sb->s_umount); 10061da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 10071da177e4SLinus Torvalds lock_kernel(); 10081da177e4SLinus Torvalds DQUOT_OFF(sb); 10091da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 10101da177e4SLinus Torvalds unlock_kernel(); 10111da177e4SLinus Torvalds } 10121da177e4SLinus Torvalds up_write(&sb->s_umount); 10131da177e4SLinus Torvalds return retval; 10141da177e4SLinus Torvalds } 10151da177e4SLinus Torvalds 1016390c6843SRam Pai down_write(&namespace_sem); 10171da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 10185addc5ddSAl Viro event++; 10191da177e4SLinus Torvalds 1020c35038beSAl Viro if (!(flags & MNT_DETACH)) 1021c35038beSAl Viro shrink_submounts(mnt, &umount_list); 1022c35038beSAl Viro 10231da177e4SLinus Torvalds retval = -EBUSY; 1024a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 10251da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 1026a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 10271da177e4SLinus Torvalds retval = 0; 10281da177e4SLinus Torvalds } 10291da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 10301da177e4SLinus Torvalds if (retval) 10311da177e4SLinus Torvalds security_sb_umount_busy(mnt); 1032390c6843SRam Pai up_write(&namespace_sem); 103370fbcdf4SRam Pai release_mounts(&umount_list); 10341da177e4SLinus Torvalds return retval; 10351da177e4SLinus Torvalds } 10361da177e4SLinus Torvalds 10371da177e4SLinus Torvalds /* 10381da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 10391da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 10401da177e4SLinus Torvalds * 10411da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 10421da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 10431da177e4SLinus Torvalds */ 10441da177e4SLinus Torvalds 10451da177e4SLinus Torvalds asmlinkage long sys_umount(char __user * name, int flags) 10461da177e4SLinus Torvalds { 10471da177e4SLinus Torvalds struct nameidata nd; 10481da177e4SLinus Torvalds int retval; 10491da177e4SLinus Torvalds 10501da177e4SLinus Torvalds retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 10511da177e4SLinus Torvalds if (retval) 10521da177e4SLinus Torvalds goto out; 10531da177e4SLinus Torvalds retval = -EINVAL; 10544ac91378SJan Blunck if (nd.path.dentry != nd.path.mnt->mnt_root) 10551da177e4SLinus Torvalds goto dput_and_out; 10564ac91378SJan Blunck if (!check_mnt(nd.path.mnt)) 10571da177e4SLinus Torvalds goto dput_and_out; 10581da177e4SLinus Torvalds 10591da177e4SLinus Torvalds retval = -EPERM; 10601da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 10611da177e4SLinus Torvalds goto dput_and_out; 10621da177e4SLinus Torvalds 10634ac91378SJan Blunck retval = do_umount(nd.path.mnt, flags); 10641da177e4SLinus Torvalds dput_and_out: 1065429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 10664ac91378SJan Blunck dput(nd.path.dentry); 10674ac91378SJan Blunck mntput_no_expire(nd.path.mnt); 10681da177e4SLinus Torvalds out: 10691da177e4SLinus Torvalds return retval; 10701da177e4SLinus Torvalds } 10711da177e4SLinus Torvalds 10721da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 10731da177e4SLinus Torvalds 10741da177e4SLinus Torvalds /* 10751da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 10761da177e4SLinus Torvalds */ 10771da177e4SLinus Torvalds asmlinkage long sys_oldumount(char __user * name) 10781da177e4SLinus Torvalds { 10791da177e4SLinus Torvalds return sys_umount(name, 0); 10801da177e4SLinus Torvalds } 10811da177e4SLinus Torvalds 10821da177e4SLinus Torvalds #endif 10831da177e4SLinus Torvalds 10841da177e4SLinus Torvalds static int mount_is_safe(struct nameidata *nd) 10851da177e4SLinus Torvalds { 10861da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 10871da177e4SLinus Torvalds return 0; 10881da177e4SLinus Torvalds return -EPERM; 10891da177e4SLinus Torvalds #ifdef notyet 10904ac91378SJan Blunck if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) 10911da177e4SLinus Torvalds return -EPERM; 10924ac91378SJan Blunck if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { 10934ac91378SJan Blunck if (current->uid != nd->path.dentry->d_inode->i_uid) 10941da177e4SLinus Torvalds return -EPERM; 10951da177e4SLinus Torvalds } 1096e4543eddSChristoph Hellwig if (vfs_permission(nd, MAY_WRITE)) 10971da177e4SLinus Torvalds return -EPERM; 10981da177e4SLinus Torvalds return 0; 10991da177e4SLinus Torvalds #endif 11001da177e4SLinus Torvalds } 11011da177e4SLinus Torvalds 1102b58fed8bSRam Pai static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) 11031da177e4SLinus Torvalds { 11041da177e4SLinus Torvalds while (1) { 11051da177e4SLinus Torvalds if (d == dentry) 11061da177e4SLinus Torvalds return 1; 11071da177e4SLinus Torvalds if (d == NULL || d == d->d_parent) 11081da177e4SLinus Torvalds return 0; 11091da177e4SLinus Torvalds d = d->d_parent; 11101da177e4SLinus Torvalds } 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds 1113b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 111436341f64SRam Pai int flag) 11151da177e4SLinus Torvalds { 11161da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 11171a390689SAl Viro struct path path; 11181da177e4SLinus Torvalds 11199676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 11209676f0c6SRam Pai return NULL; 11219676f0c6SRam Pai 112236341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 11231da177e4SLinus Torvalds if (!q) 11241da177e4SLinus Torvalds goto Enomem; 11251da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 11261da177e4SLinus Torvalds 11271da177e4SLinus Torvalds p = mnt; 1128fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 11291da177e4SLinus Torvalds if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) 11301da177e4SLinus Torvalds continue; 11311da177e4SLinus Torvalds 11321da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 11339676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 11349676f0c6SRam Pai s = skip_mnt_tree(s); 11359676f0c6SRam Pai continue; 11369676f0c6SRam Pai } 11371da177e4SLinus Torvalds while (p != s->mnt_parent) { 11381da177e4SLinus Torvalds p = p->mnt_parent; 11391da177e4SLinus Torvalds q = q->mnt_parent; 11401da177e4SLinus Torvalds } 11411da177e4SLinus Torvalds p = s; 11421a390689SAl Viro path.mnt = q; 11431a390689SAl Viro path.dentry = p->mnt_mountpoint; 114436341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 11451da177e4SLinus Torvalds if (!q) 11461da177e4SLinus Torvalds goto Enomem; 11471da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 11481da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 11491a390689SAl Viro attach_mnt(q, &path); 11501da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 11511da177e4SLinus Torvalds } 11521da177e4SLinus Torvalds } 11531da177e4SLinus Torvalds return res; 11541da177e4SLinus Torvalds Enomem: 11551da177e4SLinus Torvalds if (res) { 115670fbcdf4SRam Pai LIST_HEAD(umount_list); 11571da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1158a05964f3SRam Pai umount_tree(res, 0, &umount_list); 11591da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 116070fbcdf4SRam Pai release_mounts(&umount_list); 11611da177e4SLinus Torvalds } 11621da177e4SLinus Torvalds return NULL; 11631da177e4SLinus Torvalds } 11641da177e4SLinus Torvalds 11658aec0809SAl Viro struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) 11668aec0809SAl Viro { 11678aec0809SAl Viro struct vfsmount *tree; 11681a60a280SAl Viro down_write(&namespace_sem); 11698aec0809SAl Viro tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); 11701a60a280SAl Viro up_write(&namespace_sem); 11718aec0809SAl Viro return tree; 11728aec0809SAl Viro } 11738aec0809SAl Viro 11748aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 11758aec0809SAl Viro { 11768aec0809SAl Viro LIST_HEAD(umount_list); 11771a60a280SAl Viro down_write(&namespace_sem); 11788aec0809SAl Viro spin_lock(&vfsmount_lock); 11798aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 11808aec0809SAl Viro spin_unlock(&vfsmount_lock); 11811a60a280SAl Viro up_write(&namespace_sem); 11828aec0809SAl Viro release_mounts(&umount_list); 11838aec0809SAl Viro } 11848aec0809SAl Viro 1185719f5d7fSMiklos Szeredi static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) 1186719f5d7fSMiklos Szeredi { 1187719f5d7fSMiklos Szeredi struct vfsmount *p; 1188719f5d7fSMiklos Szeredi 1189719f5d7fSMiklos Szeredi for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1190719f5d7fSMiklos Szeredi if (p->mnt_group_id && !IS_MNT_SHARED(p)) 1191719f5d7fSMiklos Szeredi mnt_release_group_id(p); 1192719f5d7fSMiklos Szeredi } 1193719f5d7fSMiklos Szeredi } 1194719f5d7fSMiklos Szeredi 1195719f5d7fSMiklos Szeredi static int invent_group_ids(struct vfsmount *mnt, bool recurse) 1196719f5d7fSMiklos Szeredi { 1197719f5d7fSMiklos Szeredi struct vfsmount *p; 1198719f5d7fSMiklos Szeredi 1199719f5d7fSMiklos Szeredi for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1200719f5d7fSMiklos Szeredi if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 1201719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(p); 1202719f5d7fSMiklos Szeredi if (err) { 1203719f5d7fSMiklos Szeredi cleanup_group_ids(mnt, p); 1204719f5d7fSMiklos Szeredi return err; 1205719f5d7fSMiklos Szeredi } 1206719f5d7fSMiklos Szeredi } 1207719f5d7fSMiklos Szeredi } 1208719f5d7fSMiklos Szeredi 1209719f5d7fSMiklos Szeredi return 0; 1210719f5d7fSMiklos Szeredi } 1211719f5d7fSMiklos Szeredi 1212b90fa9aeSRam Pai /* 1213b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1214b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 121521444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 121621444403SRam Pai * store the parent mount and mountpoint dentry. 121721444403SRam Pai * (done when source_mnt is moved) 1218b90fa9aeSRam Pai * 1219b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1220b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 12219676f0c6SRam Pai * --------------------------------------------------------------------------- 1222b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 12239676f0c6SRam Pai * |************************************************************************** 12249676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 12259676f0c6SRam Pai * | dest | | | | | 12269676f0c6SRam Pai * | | | | | | | 12279676f0c6SRam Pai * | v | | | | | 12289676f0c6SRam Pai * |************************************************************************** 12299676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 12305afe0022SRam Pai * | | | | | | 12319676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 12329676f0c6SRam Pai * *************************************************************************** 1233b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1234b90fa9aeSRam Pai * destination mount. 1235b90fa9aeSRam Pai * 1236b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1237b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1238b90fa9aeSRam Pai * the peer group of the source mount. 1239b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1240b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1241b90fa9aeSRam Pai * mount. 12425afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 12435afe0022SRam Pai * of the destination mount and the cloned mount is made slave 12445afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 12455afe0022SRam Pai * is marked as 'shared and slave'. 12465afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 12475afe0022SRam Pai * source mount. 12485afe0022SRam Pai * 12499676f0c6SRam Pai * --------------------------------------------------------------------------- 125021444403SRam Pai * | MOVE MOUNT OPERATION | 12519676f0c6SRam Pai * |************************************************************************** 12529676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 12539676f0c6SRam Pai * | dest | | | | | 12549676f0c6SRam Pai * | | | | | | | 12559676f0c6SRam Pai * | v | | | | | 12569676f0c6SRam Pai * |************************************************************************** 12579676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 12585afe0022SRam Pai * | | | | | | 12599676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 12609676f0c6SRam Pai * *************************************************************************** 12615afe0022SRam Pai * 12625afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 12635afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 126421444403SRam Pai * (+*) the mount is moved to the destination. 12655afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 12665afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 12675afe0022SRam Pai * the mount is marked as 'shared and slave'. 12685afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1269b90fa9aeSRam Pai * 1270b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1271b90fa9aeSRam Pai * applied to each mount in the tree. 1272b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1273b90fa9aeSRam Pai * in allocations. 1274b90fa9aeSRam Pai */ 1275b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 12761a390689SAl Viro struct path *path, struct path *parent_path) 1277b90fa9aeSRam Pai { 1278b90fa9aeSRam Pai LIST_HEAD(tree_list); 12791a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 12801a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1281b90fa9aeSRam Pai struct vfsmount *child, *p; 1282719f5d7fSMiklos Szeredi int err; 1283b90fa9aeSRam Pai 1284719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) { 1285719f5d7fSMiklos Szeredi err = invent_group_ids(source_mnt, true); 1286719f5d7fSMiklos Szeredi if (err) 1287719f5d7fSMiklos Szeredi goto out; 1288719f5d7fSMiklos Szeredi } 1289719f5d7fSMiklos Szeredi err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1290719f5d7fSMiklos Szeredi if (err) 1291719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1292b90fa9aeSRam Pai 1293b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1294b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1295b90fa9aeSRam Pai set_mnt_shared(p); 1296b90fa9aeSRam Pai } 1297b90fa9aeSRam Pai 1298b90fa9aeSRam Pai spin_lock(&vfsmount_lock); 12991a390689SAl Viro if (parent_path) { 13001a390689SAl Viro detach_mnt(source_mnt, parent_path); 13011a390689SAl Viro attach_mnt(source_mnt, path); 13026b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 130321444403SRam Pai } else { 1304b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1305b90fa9aeSRam Pai commit_tree(source_mnt); 130621444403SRam Pai } 1307b90fa9aeSRam Pai 1308b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1309b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1310b90fa9aeSRam Pai commit_tree(child); 1311b90fa9aeSRam Pai } 1312b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 1313b90fa9aeSRam Pai return 0; 1314719f5d7fSMiklos Szeredi 1315719f5d7fSMiklos Szeredi out_cleanup_ids: 1316719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) 1317719f5d7fSMiklos Szeredi cleanup_group_ids(source_mnt, NULL); 1318719f5d7fSMiklos Szeredi out: 1319719f5d7fSMiklos Szeredi return err; 1320b90fa9aeSRam Pai } 1321b90fa9aeSRam Pai 13228c3ee42eSAl Viro static int graft_tree(struct vfsmount *mnt, struct path *path) 13231da177e4SLinus Torvalds { 13241da177e4SLinus Torvalds int err; 13251da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 13261da177e4SLinus Torvalds return -EINVAL; 13271da177e4SLinus Torvalds 13288c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 13291da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 13301da177e4SLinus Torvalds return -ENOTDIR; 13311da177e4SLinus Torvalds 13321da177e4SLinus Torvalds err = -ENOENT; 13338c3ee42eSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 13348c3ee42eSAl Viro if (IS_DEADDIR(path->dentry->d_inode)) 13351da177e4SLinus Torvalds goto out_unlock; 13361da177e4SLinus Torvalds 13378c3ee42eSAl Viro err = security_sb_check_sb(mnt, path); 13381da177e4SLinus Torvalds if (err) 13391da177e4SLinus Torvalds goto out_unlock; 13401da177e4SLinus Torvalds 13411da177e4SLinus Torvalds err = -ENOENT; 13428c3ee42eSAl Viro if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry)) 13438c3ee42eSAl Viro err = attach_recursive_mnt(mnt, path, NULL); 13441da177e4SLinus Torvalds out_unlock: 13458c3ee42eSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 13461da177e4SLinus Torvalds if (!err) 13478c3ee42eSAl Viro security_sb_post_addmount(mnt, path); 13481da177e4SLinus Torvalds return err; 13491da177e4SLinus Torvalds } 13501da177e4SLinus Torvalds 13511da177e4SLinus Torvalds /* 135207b20889SRam Pai * recursively change the type of the mountpoint. 13532dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 135407b20889SRam Pai */ 13552dafe1c4SEric Sandeen static noinline int do_change_type(struct nameidata *nd, int flag) 135607b20889SRam Pai { 13574ac91378SJan Blunck struct vfsmount *m, *mnt = nd->path.mnt; 135807b20889SRam Pai int recurse = flag & MS_REC; 135907b20889SRam Pai int type = flag & ~MS_REC; 1360719f5d7fSMiklos Szeredi int err = 0; 136107b20889SRam Pai 1362ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1363ee6f9582SMiklos Szeredi return -EPERM; 1364ee6f9582SMiklos Szeredi 13654ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 136607b20889SRam Pai return -EINVAL; 136707b20889SRam Pai 136807b20889SRam Pai down_write(&namespace_sem); 1369719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1370719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1371719f5d7fSMiklos Szeredi if (err) 1372719f5d7fSMiklos Szeredi goto out_unlock; 1373719f5d7fSMiklos Szeredi } 1374719f5d7fSMiklos Szeredi 137507b20889SRam Pai spin_lock(&vfsmount_lock); 137607b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 137707b20889SRam Pai change_mnt_propagation(m, type); 137807b20889SRam Pai spin_unlock(&vfsmount_lock); 1379719f5d7fSMiklos Szeredi 1380719f5d7fSMiklos Szeredi out_unlock: 138107b20889SRam Pai up_write(&namespace_sem); 1382719f5d7fSMiklos Szeredi return err; 138307b20889SRam Pai } 138407b20889SRam Pai 138507b20889SRam Pai /* 13861da177e4SLinus Torvalds * do loopback mount. 13872dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 13881da177e4SLinus Torvalds */ 13892dafe1c4SEric Sandeen static noinline int do_loopback(struct nameidata *nd, char *old_name, 13902dafe1c4SEric Sandeen int recurse) 13911da177e4SLinus Torvalds { 13921da177e4SLinus Torvalds struct nameidata old_nd; 13931da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 13941da177e4SLinus Torvalds int err = mount_is_safe(nd); 13951da177e4SLinus Torvalds if (err) 13961da177e4SLinus Torvalds return err; 13971da177e4SLinus Torvalds if (!old_name || !*old_name) 13981da177e4SLinus Torvalds return -EINVAL; 13991da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 14001da177e4SLinus Torvalds if (err) 14011da177e4SLinus Torvalds return err; 14021da177e4SLinus Torvalds 1403390c6843SRam Pai down_write(&namespace_sem); 14041da177e4SLinus Torvalds err = -EINVAL; 14054ac91378SJan Blunck if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) 14069676f0c6SRam Pai goto out; 14079676f0c6SRam Pai 14084ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 1409ccd48bc7SAl Viro goto out; 1410ccd48bc7SAl Viro 14111da177e4SLinus Torvalds err = -ENOMEM; 14121da177e4SLinus Torvalds if (recurse) 14134ac91378SJan Blunck mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); 14141da177e4SLinus Torvalds else 14154ac91378SJan Blunck mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); 14161da177e4SLinus Torvalds 1417ccd48bc7SAl Viro if (!mnt) 1418ccd48bc7SAl Viro goto out; 1419ccd48bc7SAl Viro 14208c3ee42eSAl Viro err = graft_tree(mnt, &nd->path); 14211da177e4SLinus Torvalds if (err) { 142270fbcdf4SRam Pai LIST_HEAD(umount_list); 14231da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1424a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 14251da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 142670fbcdf4SRam Pai release_mounts(&umount_list); 14275b83d2c5SRam Pai } 14281da177e4SLinus Torvalds 1429ccd48bc7SAl Viro out: 1430390c6843SRam Pai up_write(&namespace_sem); 14311d957f9bSJan Blunck path_put(&old_nd.path); 14321da177e4SLinus Torvalds return err; 14331da177e4SLinus Torvalds } 14341da177e4SLinus Torvalds 14352e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 14362e4b7fcdSDave Hansen { 14372e4b7fcdSDave Hansen int error = 0; 14382e4b7fcdSDave Hansen int readonly_request = 0; 14392e4b7fcdSDave Hansen 14402e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 14412e4b7fcdSDave Hansen readonly_request = 1; 14422e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 14432e4b7fcdSDave Hansen return 0; 14442e4b7fcdSDave Hansen 14452e4b7fcdSDave Hansen if (readonly_request) 14462e4b7fcdSDave Hansen error = mnt_make_readonly(mnt); 14472e4b7fcdSDave Hansen else 14482e4b7fcdSDave Hansen __mnt_unmake_readonly(mnt); 14492e4b7fcdSDave Hansen return error; 14502e4b7fcdSDave Hansen } 14512e4b7fcdSDave Hansen 14521da177e4SLinus Torvalds /* 14531da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 14541da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 14551da177e4SLinus Torvalds * on it - tough luck. 14562dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 14571da177e4SLinus Torvalds */ 14582dafe1c4SEric Sandeen static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, 14591da177e4SLinus Torvalds void *data) 14601da177e4SLinus Torvalds { 14611da177e4SLinus Torvalds int err; 14624ac91378SJan Blunck struct super_block *sb = nd->path.mnt->mnt_sb; 14631da177e4SLinus Torvalds 14641da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 14651da177e4SLinus Torvalds return -EPERM; 14661da177e4SLinus Torvalds 14674ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 14681da177e4SLinus Torvalds return -EINVAL; 14691da177e4SLinus Torvalds 14704ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 14711da177e4SLinus Torvalds return -EINVAL; 14721da177e4SLinus Torvalds 14731da177e4SLinus Torvalds down_write(&sb->s_umount); 14742e4b7fcdSDave Hansen if (flags & MS_BIND) 14752e4b7fcdSDave Hansen err = change_mount_flags(nd->path.mnt, flags); 14762e4b7fcdSDave Hansen else 14771da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 14781da177e4SLinus Torvalds if (!err) 14794ac91378SJan Blunck nd->path.mnt->mnt_flags = mnt_flags; 14801da177e4SLinus Torvalds up_write(&sb->s_umount); 14811da177e4SLinus Torvalds if (!err) 14824ac91378SJan Blunck security_sb_post_remount(nd->path.mnt, flags, data); 14831da177e4SLinus Torvalds return err; 14841da177e4SLinus Torvalds } 14851da177e4SLinus Torvalds 14869676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 14879676f0c6SRam Pai { 14889676f0c6SRam Pai struct vfsmount *p; 14899676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 14909676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 14919676f0c6SRam Pai return 1; 14929676f0c6SRam Pai } 14939676f0c6SRam Pai return 0; 14949676f0c6SRam Pai } 14959676f0c6SRam Pai 14962dafe1c4SEric Sandeen /* 14972dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 14982dafe1c4SEric Sandeen */ 14992dafe1c4SEric Sandeen static noinline int do_move_mount(struct nameidata *nd, char *old_name) 15001da177e4SLinus Torvalds { 15011a390689SAl Viro struct nameidata old_nd; 15021a390689SAl Viro struct path parent_path; 15031da177e4SLinus Torvalds struct vfsmount *p; 15041da177e4SLinus Torvalds int err = 0; 15051da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15061da177e4SLinus Torvalds return -EPERM; 15071da177e4SLinus Torvalds if (!old_name || !*old_name) 15081da177e4SLinus Torvalds return -EINVAL; 15091da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 15101da177e4SLinus Torvalds if (err) 15111da177e4SLinus Torvalds return err; 15121da177e4SLinus Torvalds 1513390c6843SRam Pai down_write(&namespace_sem); 15144ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 15154ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 15161da177e4SLinus Torvalds ; 15171da177e4SLinus Torvalds err = -EINVAL; 15184ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 15191da177e4SLinus Torvalds goto out; 15201da177e4SLinus Torvalds 15211da177e4SLinus Torvalds err = -ENOENT; 15224ac91378SJan Blunck mutex_lock(&nd->path.dentry->d_inode->i_mutex); 15234ac91378SJan Blunck if (IS_DEADDIR(nd->path.dentry->d_inode)) 15241da177e4SLinus Torvalds goto out1; 15251da177e4SLinus Torvalds 15264ac91378SJan Blunck if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) 152721444403SRam Pai goto out1; 15281da177e4SLinus Torvalds 15291da177e4SLinus Torvalds err = -EINVAL; 15304ac91378SJan Blunck if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) 153121444403SRam Pai goto out1; 15321da177e4SLinus Torvalds 15334ac91378SJan Blunck if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) 153421444403SRam Pai goto out1; 15351da177e4SLinus Torvalds 15364ac91378SJan Blunck if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != 15374ac91378SJan Blunck S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) 153821444403SRam Pai goto out1; 153921444403SRam Pai /* 154021444403SRam Pai * Don't move a mount residing in a shared parent. 154121444403SRam Pai */ 15424ac91378SJan Blunck if (old_nd.path.mnt->mnt_parent && 15434ac91378SJan Blunck IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) 154421444403SRam Pai goto out1; 15459676f0c6SRam Pai /* 15469676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 15479676f0c6SRam Pai * mount which is shared. 15489676f0c6SRam Pai */ 15494ac91378SJan Blunck if (IS_MNT_SHARED(nd->path.mnt) && 15504ac91378SJan Blunck tree_contains_unbindable(old_nd.path.mnt)) 15519676f0c6SRam Pai goto out1; 15521da177e4SLinus Torvalds err = -ELOOP; 15534ac91378SJan Blunck for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) 15544ac91378SJan Blunck if (p == old_nd.path.mnt) 155521444403SRam Pai goto out1; 15561da177e4SLinus Torvalds 15571a390689SAl Viro err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); 15584ac91378SJan Blunck if (err) 155921444403SRam Pai goto out1; 15601da177e4SLinus Torvalds 15611da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 15621da177e4SLinus Torvalds * automatically */ 15634ac91378SJan Blunck list_del_init(&old_nd.path.mnt->mnt_expire); 15641da177e4SLinus Torvalds out1: 15654ac91378SJan Blunck mutex_unlock(&nd->path.dentry->d_inode->i_mutex); 15661da177e4SLinus Torvalds out: 1567390c6843SRam Pai up_write(&namespace_sem); 15681da177e4SLinus Torvalds if (!err) 15691a390689SAl Viro path_put(&parent_path); 15701d957f9bSJan Blunck path_put(&old_nd.path); 15711da177e4SLinus Torvalds return err; 15721da177e4SLinus Torvalds } 15731da177e4SLinus Torvalds 15741da177e4SLinus Torvalds /* 15751da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 15761da177e4SLinus Torvalds * namespace's tree 15772dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 15781da177e4SLinus Torvalds */ 15792dafe1c4SEric Sandeen static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, 15801da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 15811da177e4SLinus Torvalds { 15821da177e4SLinus Torvalds struct vfsmount *mnt; 15831da177e4SLinus Torvalds 15841da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 15851da177e4SLinus Torvalds return -EINVAL; 15861da177e4SLinus Torvalds 15871da177e4SLinus Torvalds /* we need capabilities... */ 15881da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15891da177e4SLinus Torvalds return -EPERM; 15901da177e4SLinus Torvalds 15911da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 15921da177e4SLinus Torvalds if (IS_ERR(mnt)) 15931da177e4SLinus Torvalds return PTR_ERR(mnt); 15941da177e4SLinus Torvalds 15951da177e4SLinus Torvalds return do_add_mount(mnt, nd, mnt_flags, NULL); 15961da177e4SLinus Torvalds } 15971da177e4SLinus Torvalds 15981da177e4SLinus Torvalds /* 15991da177e4SLinus Torvalds * add a mount into a namespace's mount tree 16001da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 16011da177e4SLinus Torvalds */ 16021da177e4SLinus Torvalds int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 16031da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 16041da177e4SLinus Torvalds { 16051da177e4SLinus Torvalds int err; 16061da177e4SLinus Torvalds 1607390c6843SRam Pai down_write(&namespace_sem); 16081da177e4SLinus Torvalds /* Something was mounted here while we slept */ 16094ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 16104ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 16111da177e4SLinus Torvalds ; 16121da177e4SLinus Torvalds err = -EINVAL; 16134ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 16141da177e4SLinus Torvalds goto unlock; 16151da177e4SLinus Torvalds 16161da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 16171da177e4SLinus Torvalds err = -EBUSY; 16184ac91378SJan Blunck if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && 16194ac91378SJan Blunck nd->path.mnt->mnt_root == nd->path.dentry) 16201da177e4SLinus Torvalds goto unlock; 16211da177e4SLinus Torvalds 16221da177e4SLinus Torvalds err = -EINVAL; 16231da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 16241da177e4SLinus Torvalds goto unlock; 16251da177e4SLinus Torvalds 16261da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 16278c3ee42eSAl Viro if ((err = graft_tree(newmnt, &nd->path))) 16285b83d2c5SRam Pai goto unlock; 16291da177e4SLinus Torvalds 16306758f953SAl Viro if (fslist) /* add to the specified expiration list */ 163155e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 16326758f953SAl Viro 1633390c6843SRam Pai up_write(&namespace_sem); 16345b83d2c5SRam Pai return 0; 16351da177e4SLinus Torvalds 16361da177e4SLinus Torvalds unlock: 1637390c6843SRam Pai up_write(&namespace_sem); 16381da177e4SLinus Torvalds mntput(newmnt); 16391da177e4SLinus Torvalds return err; 16401da177e4SLinus Torvalds } 16411da177e4SLinus Torvalds 16421da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 16431da177e4SLinus Torvalds 16445528f911STrond Myklebust /* 16451da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 16461da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 16471da177e4SLinus Torvalds * here 16481da177e4SLinus Torvalds */ 16491da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 16501da177e4SLinus Torvalds { 16511da177e4SLinus Torvalds struct vfsmount *mnt, *next; 16521da177e4SLinus Torvalds LIST_HEAD(graveyard); 1653bcc5c7d2SAl Viro LIST_HEAD(umounts); 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds if (list_empty(mounts)) 16561da177e4SLinus Torvalds return; 16571da177e4SLinus Torvalds 1658bcc5c7d2SAl Viro down_write(&namespace_sem); 16591da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 16601da177e4SLinus Torvalds 16611da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 16621da177e4SLinus Torvalds * following criteria: 16631da177e4SLinus Torvalds * - only referenced by its parent vfsmount 16641da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 16651da177e4SLinus Torvalds * cleared by mntput()) 16661da177e4SLinus Torvalds */ 166755e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 16681da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 1669bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 16701da177e4SLinus Torvalds continue; 167155e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 16721da177e4SLinus Torvalds } 1673bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1674bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 1675bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1676bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 1677bcc5c7d2SAl Viro } 16781da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1679bcc5c7d2SAl Viro up_write(&namespace_sem); 1680bcc5c7d2SAl Viro 1681bcc5c7d2SAl Viro release_mounts(&umounts); 16821da177e4SLinus Torvalds } 16831da177e4SLinus Torvalds 16841da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 16851da177e4SLinus Torvalds 16861da177e4SLinus Torvalds /* 16875528f911STrond Myklebust * Ripoff of 'select_parent()' 16885528f911STrond Myklebust * 16895528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 16905528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 16915528f911STrond Myklebust */ 16925528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 16935528f911STrond Myklebust { 16945528f911STrond Myklebust struct vfsmount *this_parent = parent; 16955528f911STrond Myklebust struct list_head *next; 16965528f911STrond Myklebust int found = 0; 16975528f911STrond Myklebust 16985528f911STrond Myklebust repeat: 16995528f911STrond Myklebust next = this_parent->mnt_mounts.next; 17005528f911STrond Myklebust resume: 17015528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 17025528f911STrond Myklebust struct list_head *tmp = next; 17035528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 17045528f911STrond Myklebust 17055528f911STrond Myklebust next = tmp->next; 17065528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 17075528f911STrond Myklebust continue; 17085528f911STrond Myklebust /* 17095528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 17105528f911STrond Myklebust */ 17115528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 17125528f911STrond Myklebust this_parent = mnt; 17135528f911STrond Myklebust goto repeat; 17145528f911STrond Myklebust } 17155528f911STrond Myklebust 17165528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 17175528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 17185528f911STrond Myklebust found++; 17195528f911STrond Myklebust } 17205528f911STrond Myklebust } 17215528f911STrond Myklebust /* 17225528f911STrond Myklebust * All done at this level ... ascend and resume the search 17235528f911STrond Myklebust */ 17245528f911STrond Myklebust if (this_parent != parent) { 17255528f911STrond Myklebust next = this_parent->mnt_child.next; 17265528f911STrond Myklebust this_parent = this_parent->mnt_parent; 17275528f911STrond Myklebust goto resume; 17285528f911STrond Myklebust } 17295528f911STrond Myklebust return found; 17305528f911STrond Myklebust } 17315528f911STrond Myklebust 17325528f911STrond Myklebust /* 17335528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 17345528f911STrond Myklebust * submounts of a specific parent mountpoint 17355528f911STrond Myklebust */ 1736c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 17375528f911STrond Myklebust { 17385528f911STrond Myklebust LIST_HEAD(graveyard); 1739c35038beSAl Viro struct vfsmount *m; 17405528f911STrond Myklebust 17415528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 1742c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 1743bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1744c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 1745bcc5c7d2SAl Viro mnt_expire); 1746bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1747c35038beSAl Viro umount_tree(mnt, 1, umounts); 1748bcc5c7d2SAl Viro } 1749bcc5c7d2SAl Viro } 17505528f911STrond Myklebust } 17515528f911STrond Myklebust 17525528f911STrond Myklebust /* 17531da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 17541da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 17551da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 17561da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 17571da177e4SLinus Torvalds */ 1758b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1759b58fed8bSRam Pai unsigned long n) 17601da177e4SLinus Torvalds { 17611da177e4SLinus Torvalds char *t = to; 17621da177e4SLinus Torvalds const char __user *f = from; 17631da177e4SLinus Torvalds char c; 17641da177e4SLinus Torvalds 17651da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 17661da177e4SLinus Torvalds return n; 17671da177e4SLinus Torvalds 17681da177e4SLinus Torvalds while (n) { 17691da177e4SLinus Torvalds if (__get_user(c, f)) { 17701da177e4SLinus Torvalds memset(t, 0, n); 17711da177e4SLinus Torvalds break; 17721da177e4SLinus Torvalds } 17731da177e4SLinus Torvalds *t++ = c; 17741da177e4SLinus Torvalds f++; 17751da177e4SLinus Torvalds n--; 17761da177e4SLinus Torvalds } 17771da177e4SLinus Torvalds return n; 17781da177e4SLinus Torvalds } 17791da177e4SLinus Torvalds 17801da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 17811da177e4SLinus Torvalds { 17821da177e4SLinus Torvalds int i; 17831da177e4SLinus Torvalds unsigned long page; 17841da177e4SLinus Torvalds unsigned long size; 17851da177e4SLinus Torvalds 17861da177e4SLinus Torvalds *where = 0; 17871da177e4SLinus Torvalds if (!data) 17881da177e4SLinus Torvalds return 0; 17891da177e4SLinus Torvalds 17901da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 17911da177e4SLinus Torvalds return -ENOMEM; 17921da177e4SLinus Torvalds 17931da177e4SLinus Torvalds /* We only care that *some* data at the address the user 17941da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 17951da177e4SLinus Torvalds * the remainder of the page. 17961da177e4SLinus Torvalds */ 17971da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 17981da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 17991da177e4SLinus Torvalds if (size > PAGE_SIZE) 18001da177e4SLinus Torvalds size = PAGE_SIZE; 18011da177e4SLinus Torvalds 18021da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 18031da177e4SLinus Torvalds if (!i) { 18041da177e4SLinus Torvalds free_page(page); 18051da177e4SLinus Torvalds return -EFAULT; 18061da177e4SLinus Torvalds } 18071da177e4SLinus Torvalds if (i != PAGE_SIZE) 18081da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 18091da177e4SLinus Torvalds *where = page; 18101da177e4SLinus Torvalds return 0; 18111da177e4SLinus Torvalds } 18121da177e4SLinus Torvalds 18131da177e4SLinus Torvalds /* 18141da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 18151da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 18161da177e4SLinus Torvalds * 18171da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 18181da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 18191da177e4SLinus Torvalds * information (or be NULL). 18201da177e4SLinus Torvalds * 18211da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 18221da177e4SLinus Torvalds * When the flags word was introduced its top half was required 18231da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 18241da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 18251da177e4SLinus Torvalds * and must be discarded. 18261da177e4SLinus Torvalds */ 18271da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 18281da177e4SLinus Torvalds unsigned long flags, void *data_page) 18291da177e4SLinus Torvalds { 18301da177e4SLinus Torvalds struct nameidata nd; 18311da177e4SLinus Torvalds int retval = 0; 18321da177e4SLinus Torvalds int mnt_flags = 0; 18331da177e4SLinus Torvalds 18341da177e4SLinus Torvalds /* Discard magic */ 18351da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 18361da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 18371da177e4SLinus Torvalds 18381da177e4SLinus Torvalds /* Basic sanity checks */ 18391da177e4SLinus Torvalds 18401da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 18411da177e4SLinus Torvalds return -EINVAL; 18421da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 18431da177e4SLinus Torvalds return -EINVAL; 18441da177e4SLinus Torvalds 18451da177e4SLinus Torvalds if (data_page) 18461da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 18471da177e4SLinus Torvalds 18481da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 18491da177e4SLinus Torvalds if (flags & MS_NOSUID) 18501da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 18511da177e4SLinus Torvalds if (flags & MS_NODEV) 18521da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 18531da177e4SLinus Torvalds if (flags & MS_NOEXEC) 18541da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 1855fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 1856fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 1857fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 1858fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 185947ae32d6SValerie Henson if (flags & MS_RELATIME) 186047ae32d6SValerie Henson mnt_flags |= MNT_RELATIME; 18612e4b7fcdSDave Hansen if (flags & MS_RDONLY) 18622e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 1863fc33a7bbSChristoph Hellwig 1864fc33a7bbSChristoph Hellwig flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 18658bf9725cSPavel Emelyanov MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); 18661da177e4SLinus Torvalds 18671da177e4SLinus Torvalds /* ... and get the mountpoint */ 18681da177e4SLinus Torvalds retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 18691da177e4SLinus Torvalds if (retval) 18701da177e4SLinus Torvalds return retval; 18711da177e4SLinus Torvalds 1872b5266eb4SAl Viro retval = security_sb_mount(dev_name, &nd.path, 1873b5266eb4SAl Viro type_page, flags, data_page); 18741da177e4SLinus Torvalds if (retval) 18751da177e4SLinus Torvalds goto dput_out; 18761da177e4SLinus Torvalds 18771da177e4SLinus Torvalds if (flags & MS_REMOUNT) 18781da177e4SLinus Torvalds retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 18791da177e4SLinus Torvalds data_page); 18801da177e4SLinus Torvalds else if (flags & MS_BIND) 1881eee391a6SAndrew Morton retval = do_loopback(&nd, dev_name, flags & MS_REC); 18829676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 188307b20889SRam Pai retval = do_change_type(&nd, flags); 18841da177e4SLinus Torvalds else if (flags & MS_MOVE) 18851da177e4SLinus Torvalds retval = do_move_mount(&nd, dev_name); 18861da177e4SLinus Torvalds else 18871da177e4SLinus Torvalds retval = do_new_mount(&nd, type_page, flags, mnt_flags, 18881da177e4SLinus Torvalds dev_name, data_page); 18891da177e4SLinus Torvalds dput_out: 18901d957f9bSJan Blunck path_put(&nd.path); 18911da177e4SLinus Torvalds return retval; 18921da177e4SLinus Torvalds } 18931da177e4SLinus Torvalds 1894741a2951SJANAK DESAI /* 1895741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 1896741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 1897741a2951SJANAK DESAI */ 1898e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 18996b3286edSKirill Korotaev struct fs_struct *fs) 19001da177e4SLinus Torvalds { 19016b3286edSKirill Korotaev struct mnt_namespace *new_ns; 19021da177e4SLinus Torvalds struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 19031da177e4SLinus Torvalds struct vfsmount *p, *q; 19041da177e4SLinus Torvalds 19056b3286edSKirill Korotaev new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 19061da177e4SLinus Torvalds if (!new_ns) 1907467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM); 19081da177e4SLinus Torvalds 19091da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 19101da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 19115addc5ddSAl Viro init_waitqueue_head(&new_ns->poll); 19125addc5ddSAl Viro new_ns->event = 0; 19131da177e4SLinus Torvalds 1914390c6843SRam Pai down_write(&namespace_sem); 19151da177e4SLinus Torvalds /* First pass: copy the tree topology */ 19166b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 19179676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 19181da177e4SLinus Torvalds if (!new_ns->root) { 1919390c6843SRam Pai up_write(&namespace_sem); 19201da177e4SLinus Torvalds kfree(new_ns); 1921467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM);; 19221da177e4SLinus Torvalds } 19231da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 19241da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 19251da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 19261da177e4SLinus Torvalds 19271da177e4SLinus Torvalds /* 19281da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 19291da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 19301da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 19311da177e4SLinus Torvalds */ 19326b3286edSKirill Korotaev p = mnt_ns->root; 19331da177e4SLinus Torvalds q = new_ns->root; 19341da177e4SLinus Torvalds while (p) { 19356b3286edSKirill Korotaev q->mnt_ns = new_ns; 19361da177e4SLinus Torvalds if (fs) { 19376ac08c39SJan Blunck if (p == fs->root.mnt) { 19381da177e4SLinus Torvalds rootmnt = p; 19396ac08c39SJan Blunck fs->root.mnt = mntget(q); 19401da177e4SLinus Torvalds } 19416ac08c39SJan Blunck if (p == fs->pwd.mnt) { 19421da177e4SLinus Torvalds pwdmnt = p; 19436ac08c39SJan Blunck fs->pwd.mnt = mntget(q); 19441da177e4SLinus Torvalds } 19456ac08c39SJan Blunck if (p == fs->altroot.mnt) { 19461da177e4SLinus Torvalds altrootmnt = p; 19476ac08c39SJan Blunck fs->altroot.mnt = mntget(q); 19481da177e4SLinus Torvalds } 19491da177e4SLinus Torvalds } 19506b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 19511da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 19521da177e4SLinus Torvalds } 1953390c6843SRam Pai up_write(&namespace_sem); 19541da177e4SLinus Torvalds 19551da177e4SLinus Torvalds if (rootmnt) 19561da177e4SLinus Torvalds mntput(rootmnt); 19571da177e4SLinus Torvalds if (pwdmnt) 19581da177e4SLinus Torvalds mntput(pwdmnt); 19591da177e4SLinus Torvalds if (altrootmnt) 19601da177e4SLinus Torvalds mntput(altrootmnt); 19611da177e4SLinus Torvalds 1962741a2951SJANAK DESAI return new_ns; 1963741a2951SJANAK DESAI } 1964741a2951SJANAK DESAI 1965213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 1966e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 1967741a2951SJANAK DESAI { 19686b3286edSKirill Korotaev struct mnt_namespace *new_ns; 1969741a2951SJANAK DESAI 1970e3222c4eSBadari Pulavarty BUG_ON(!ns); 19716b3286edSKirill Korotaev get_mnt_ns(ns); 1972741a2951SJANAK DESAI 1973741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 1974e3222c4eSBadari Pulavarty return ns; 1975741a2951SJANAK DESAI 1976e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 1977741a2951SJANAK DESAI 19786b3286edSKirill Korotaev put_mnt_ns(ns); 1979e3222c4eSBadari Pulavarty return new_ns; 19801da177e4SLinus Torvalds } 19811da177e4SLinus Torvalds 19821da177e4SLinus Torvalds asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 19831da177e4SLinus Torvalds char __user * type, unsigned long flags, 19841da177e4SLinus Torvalds void __user * data) 19851da177e4SLinus Torvalds { 19861da177e4SLinus Torvalds int retval; 19871da177e4SLinus Torvalds unsigned long data_page; 19881da177e4SLinus Torvalds unsigned long type_page; 19891da177e4SLinus Torvalds unsigned long dev_page; 19901da177e4SLinus Torvalds char *dir_page; 19911da177e4SLinus Torvalds 19921da177e4SLinus Torvalds retval = copy_mount_options(type, &type_page); 19931da177e4SLinus Torvalds if (retval < 0) 19941da177e4SLinus Torvalds return retval; 19951da177e4SLinus Torvalds 19961da177e4SLinus Torvalds dir_page = getname(dir_name); 19971da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 19981da177e4SLinus Torvalds if (IS_ERR(dir_page)) 19991da177e4SLinus Torvalds goto out1; 20001da177e4SLinus Torvalds 20011da177e4SLinus Torvalds retval = copy_mount_options(dev_name, &dev_page); 20021da177e4SLinus Torvalds if (retval < 0) 20031da177e4SLinus Torvalds goto out2; 20041da177e4SLinus Torvalds 20051da177e4SLinus Torvalds retval = copy_mount_options(data, &data_page); 20061da177e4SLinus Torvalds if (retval < 0) 20071da177e4SLinus Torvalds goto out3; 20081da177e4SLinus Torvalds 20091da177e4SLinus Torvalds lock_kernel(); 20101da177e4SLinus Torvalds retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 20111da177e4SLinus Torvalds flags, (void *)data_page); 20121da177e4SLinus Torvalds unlock_kernel(); 20131da177e4SLinus Torvalds free_page(data_page); 20141da177e4SLinus Torvalds 20151da177e4SLinus Torvalds out3: 20161da177e4SLinus Torvalds free_page(dev_page); 20171da177e4SLinus Torvalds out2: 20181da177e4SLinus Torvalds putname(dir_page); 20191da177e4SLinus Torvalds out1: 20201da177e4SLinus Torvalds free_page(type_page); 20211da177e4SLinus Torvalds return retval; 20221da177e4SLinus Torvalds } 20231da177e4SLinus Torvalds 20241da177e4SLinus Torvalds /* 20251da177e4SLinus Torvalds * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 20261da177e4SLinus Torvalds * It can block. Requires the big lock held. 20271da177e4SLinus Torvalds */ 2028ac748a09SJan Blunck void set_fs_root(struct fs_struct *fs, struct path *path) 20291da177e4SLinus Torvalds { 20306ac08c39SJan Blunck struct path old_root; 20316ac08c39SJan Blunck 20321da177e4SLinus Torvalds write_lock(&fs->lock); 20331da177e4SLinus Torvalds old_root = fs->root; 2034ac748a09SJan Blunck fs->root = *path; 2035ac748a09SJan Blunck path_get(path); 20361da177e4SLinus Torvalds write_unlock(&fs->lock); 20376ac08c39SJan Blunck if (old_root.dentry) 20386ac08c39SJan Blunck path_put(&old_root); 20391da177e4SLinus Torvalds } 20401da177e4SLinus Torvalds 20411da177e4SLinus Torvalds /* 20421da177e4SLinus Torvalds * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 20431da177e4SLinus Torvalds * It can block. Requires the big lock held. 20441da177e4SLinus Torvalds */ 2045ac748a09SJan Blunck void set_fs_pwd(struct fs_struct *fs, struct path *path) 20461da177e4SLinus Torvalds { 20476ac08c39SJan Blunck struct path old_pwd; 20481da177e4SLinus Torvalds 20491da177e4SLinus Torvalds write_lock(&fs->lock); 20501da177e4SLinus Torvalds old_pwd = fs->pwd; 2051ac748a09SJan Blunck fs->pwd = *path; 2052ac748a09SJan Blunck path_get(path); 20531da177e4SLinus Torvalds write_unlock(&fs->lock); 20541da177e4SLinus Torvalds 20556ac08c39SJan Blunck if (old_pwd.dentry) 20566ac08c39SJan Blunck path_put(&old_pwd); 20571da177e4SLinus Torvalds } 20581da177e4SLinus Torvalds 20591a390689SAl Viro static void chroot_fs_refs(struct path *old_root, struct path *new_root) 20601da177e4SLinus Torvalds { 20611da177e4SLinus Torvalds struct task_struct *g, *p; 20621da177e4SLinus Torvalds struct fs_struct *fs; 20631da177e4SLinus Torvalds 20641da177e4SLinus Torvalds read_lock(&tasklist_lock); 20651da177e4SLinus Torvalds do_each_thread(g, p) { 20661da177e4SLinus Torvalds task_lock(p); 20671da177e4SLinus Torvalds fs = p->fs; 20681da177e4SLinus Torvalds if (fs) { 20691da177e4SLinus Torvalds atomic_inc(&fs->count); 20701da177e4SLinus Torvalds task_unlock(p); 20711a390689SAl Viro if (fs->root.dentry == old_root->dentry 20721a390689SAl Viro && fs->root.mnt == old_root->mnt) 20731a390689SAl Viro set_fs_root(fs, new_root); 20741a390689SAl Viro if (fs->pwd.dentry == old_root->dentry 20751a390689SAl Viro && fs->pwd.mnt == old_root->mnt) 20761a390689SAl Viro set_fs_pwd(fs, new_root); 20771da177e4SLinus Torvalds put_fs_struct(fs); 20781da177e4SLinus Torvalds } else 20791da177e4SLinus Torvalds task_unlock(p); 20801da177e4SLinus Torvalds } while_each_thread(g, p); 20811da177e4SLinus Torvalds read_unlock(&tasklist_lock); 20821da177e4SLinus Torvalds } 20831da177e4SLinus Torvalds 20841da177e4SLinus Torvalds /* 20851da177e4SLinus Torvalds * pivot_root Semantics: 20861da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 20871da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 20881da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 20891da177e4SLinus Torvalds * 20901da177e4SLinus Torvalds * Restrictions: 20911da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 20921da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 20931da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 20941da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 20951da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 20961da177e4SLinus Torvalds * 20974a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 20984a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 20994a0d11faSNeil Brown * in this situation. 21004a0d11faSNeil Brown * 21011da177e4SLinus Torvalds * Notes: 21021da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 21031da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 21041da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 21051da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 21061da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 21071da177e4SLinus Torvalds * first. 21081da177e4SLinus Torvalds */ 2109b58fed8bSRam Pai asmlinkage long sys_pivot_root(const char __user * new_root, 2110b58fed8bSRam Pai const char __user * put_old) 21111da177e4SLinus Torvalds { 21121da177e4SLinus Torvalds struct vfsmount *tmp; 21138c3ee42eSAl Viro struct nameidata new_nd, old_nd; 21148c3ee42eSAl Viro struct path parent_path, root_parent, root; 21151da177e4SLinus Torvalds int error; 21161da177e4SLinus Torvalds 21171da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 21181da177e4SLinus Torvalds return -EPERM; 21191da177e4SLinus Torvalds 2120b58fed8bSRam Pai error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, 2121b58fed8bSRam Pai &new_nd); 21221da177e4SLinus Torvalds if (error) 21231da177e4SLinus Torvalds goto out0; 21241da177e4SLinus Torvalds error = -EINVAL; 21254ac91378SJan Blunck if (!check_mnt(new_nd.path.mnt)) 21261da177e4SLinus Torvalds goto out1; 21271da177e4SLinus Torvalds 21281da177e4SLinus Torvalds error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); 21291da177e4SLinus Torvalds if (error) 21301da177e4SLinus Torvalds goto out1; 21311da177e4SLinus Torvalds 2132b5266eb4SAl Viro error = security_sb_pivotroot(&old_nd.path, &new_nd.path); 21331da177e4SLinus Torvalds if (error) { 21341d957f9bSJan Blunck path_put(&old_nd.path); 21351da177e4SLinus Torvalds goto out1; 21361da177e4SLinus Torvalds } 21371da177e4SLinus Torvalds 21381da177e4SLinus Torvalds read_lock(¤t->fs->lock); 21398c3ee42eSAl Viro root = current->fs->root; 21406ac08c39SJan Blunck path_get(¤t->fs->root); 21411da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 2142390c6843SRam Pai down_write(&namespace_sem); 21434ac91378SJan Blunck mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); 21441da177e4SLinus Torvalds error = -EINVAL; 21454ac91378SJan Blunck if (IS_MNT_SHARED(old_nd.path.mnt) || 21464ac91378SJan Blunck IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || 21478c3ee42eSAl Viro IS_MNT_SHARED(root.mnt->mnt_parent)) 214821444403SRam Pai goto out2; 21498c3ee42eSAl Viro if (!check_mnt(root.mnt)) 21501da177e4SLinus Torvalds goto out2; 21511da177e4SLinus Torvalds error = -ENOENT; 21524ac91378SJan Blunck if (IS_DEADDIR(new_nd.path.dentry->d_inode)) 21531da177e4SLinus Torvalds goto out2; 21544ac91378SJan Blunck if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) 21551da177e4SLinus Torvalds goto out2; 21564ac91378SJan Blunck if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) 21571da177e4SLinus Torvalds goto out2; 21581da177e4SLinus Torvalds error = -EBUSY; 21598c3ee42eSAl Viro if (new_nd.path.mnt == root.mnt || 21608c3ee42eSAl Viro old_nd.path.mnt == root.mnt) 21611da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 21621da177e4SLinus Torvalds error = -EINVAL; 21638c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 21641da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 21658c3ee42eSAl Viro if (root.mnt->mnt_parent == root.mnt) 21660bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 21674ac91378SJan Blunck if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) 21681da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 21694ac91378SJan Blunck if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) 21700bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 21714ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 21724ac91378SJan Blunck tmp = old_nd.path.mnt; 21731da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 21744ac91378SJan Blunck if (tmp != new_nd.path.mnt) { 21751da177e4SLinus Torvalds for (;;) { 21761da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 21771da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 21784ac91378SJan Blunck if (tmp->mnt_parent == new_nd.path.mnt) 21791da177e4SLinus Torvalds break; 21801da177e4SLinus Torvalds tmp = tmp->mnt_parent; 21811da177e4SLinus Torvalds } 21824ac91378SJan Blunck if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) 21831da177e4SLinus Torvalds goto out3; 21844ac91378SJan Blunck } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) 21851da177e4SLinus Torvalds goto out3; 21861a390689SAl Viro detach_mnt(new_nd.path.mnt, &parent_path); 21878c3ee42eSAl Viro detach_mnt(root.mnt, &root_parent); 21884ac91378SJan Blunck /* mount old root on put_old */ 21898c3ee42eSAl Viro attach_mnt(root.mnt, &old_nd.path); 21904ac91378SJan Blunck /* mount new_root on / */ 21914ac91378SJan Blunck attach_mnt(new_nd.path.mnt, &root_parent); 21926b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 21931da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 21948c3ee42eSAl Viro chroot_fs_refs(&root, &new_nd.path); 21958c3ee42eSAl Viro security_sb_post_pivotroot(&root, &new_nd.path); 21961da177e4SLinus Torvalds error = 0; 21971a390689SAl Viro path_put(&root_parent); 21981a390689SAl Viro path_put(&parent_path); 21991da177e4SLinus Torvalds out2: 22004ac91378SJan Blunck mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); 2201390c6843SRam Pai up_write(&namespace_sem); 22028c3ee42eSAl Viro path_put(&root); 22031d957f9bSJan Blunck path_put(&old_nd.path); 22041da177e4SLinus Torvalds out1: 22051d957f9bSJan Blunck path_put(&new_nd.path); 22061da177e4SLinus Torvalds out0: 22071da177e4SLinus Torvalds return error; 22081da177e4SLinus Torvalds out3: 22091da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 22101da177e4SLinus Torvalds goto out2; 22111da177e4SLinus Torvalds } 22121da177e4SLinus Torvalds 22131da177e4SLinus Torvalds static void __init init_mount_tree(void) 22141da177e4SLinus Torvalds { 22151da177e4SLinus Torvalds struct vfsmount *mnt; 22166b3286edSKirill Korotaev struct mnt_namespace *ns; 2217ac748a09SJan Blunck struct path root; 22181da177e4SLinus Torvalds 22191da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 22201da177e4SLinus Torvalds if (IS_ERR(mnt)) 22211da177e4SLinus Torvalds panic("Can't create rootfs"); 22226b3286edSKirill Korotaev ns = kmalloc(sizeof(*ns), GFP_KERNEL); 22236b3286edSKirill Korotaev if (!ns) 22241da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 22256b3286edSKirill Korotaev atomic_set(&ns->count, 1); 22266b3286edSKirill Korotaev INIT_LIST_HEAD(&ns->list); 22276b3286edSKirill Korotaev init_waitqueue_head(&ns->poll); 22286b3286edSKirill Korotaev ns->event = 0; 22296b3286edSKirill Korotaev list_add(&mnt->mnt_list, &ns->list); 22306b3286edSKirill Korotaev ns->root = mnt; 22316b3286edSKirill Korotaev mnt->mnt_ns = ns; 22321da177e4SLinus Torvalds 22336b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 22346b3286edSKirill Korotaev get_mnt_ns(ns); 22351da177e4SLinus Torvalds 2236ac748a09SJan Blunck root.mnt = ns->root; 2237ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2238ac748a09SJan Blunck 2239ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2240ac748a09SJan Blunck set_fs_root(current->fs, &root); 22411da177e4SLinus Torvalds } 22421da177e4SLinus Torvalds 224374bf17cfSDenis Cheng void __init mnt_init(void) 22441da177e4SLinus Torvalds { 224513f14b4dSEric Dumazet unsigned u; 224615a67dd8SRandy Dunlap int err; 22471da177e4SLinus Torvalds 2248390c6843SRam Pai init_rwsem(&namespace_sem); 2249390c6843SRam Pai 22501da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 225120c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 22521da177e4SLinus Torvalds 2253b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 22541da177e4SLinus Torvalds 22551da177e4SLinus Torvalds if (!mount_hashtable) 22561da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 22571da177e4SLinus Torvalds 225813f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 22591da177e4SLinus Torvalds 226013f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 226113f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 22621da177e4SLinus Torvalds 226315a67dd8SRandy Dunlap err = sysfs_init(); 226415a67dd8SRandy Dunlap if (err) 226515a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 226615a67dd8SRandy Dunlap __FUNCTION__, err); 226700d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 226800d26666SGreg Kroah-Hartman if (!fs_kobj) 226900d26666SGreg Kroah-Hartman printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); 22701da177e4SLinus Torvalds init_rootfs(); 22711da177e4SLinus Torvalds init_mount_tree(); 22721da177e4SLinus Torvalds } 22731da177e4SLinus Torvalds 22746b3286edSKirill Korotaev void __put_mnt_ns(struct mnt_namespace *ns) 22751da177e4SLinus Torvalds { 22766b3286edSKirill Korotaev struct vfsmount *root = ns->root; 227770fbcdf4SRam Pai LIST_HEAD(umount_list); 22786b3286edSKirill Korotaev ns->root = NULL; 22791ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 2280390c6843SRam Pai down_write(&namespace_sem); 22811da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 2282a05964f3SRam Pai umount_tree(root, 0, &umount_list); 22831da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 2284390c6843SRam Pai up_write(&namespace_sem); 228570fbcdf4SRam Pai release_mounts(&umount_list); 22866b3286edSKirill Korotaev kfree(ns); 22871da177e4SLinus Torvalds } 2288