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); 445addc5ddSAl Viro 45fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 46e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 47390c6843SRam Pai static struct rw_semaphore namespace_sem; 481da177e4SLinus Torvalds 49f87fd4c2SMiklos Szeredi /* /sys/fs */ 5000d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 5100d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 52f87fd4c2SMiklos Szeredi 531da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 541da177e4SLinus Torvalds { 551da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 561da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 5713f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 5813f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 591da177e4SLinus Torvalds } 601da177e4SLinus Torvalds 613d733633SDave Hansen #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 623d733633SDave Hansen 6373cd49ecSMiklos Szeredi /* allocation is serialized by namespace_sem */ 6473cd49ecSMiklos Szeredi static int mnt_alloc_id(struct vfsmount *mnt) 6573cd49ecSMiklos Szeredi { 6673cd49ecSMiklos Szeredi int res; 6773cd49ecSMiklos Szeredi 6873cd49ecSMiklos Szeredi retry: 6973cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 7073cd49ecSMiklos Szeredi spin_lock(&vfsmount_lock); 7173cd49ecSMiklos Szeredi res = ida_get_new(&mnt_id_ida, &mnt->mnt_id); 7273cd49ecSMiklos Szeredi spin_unlock(&vfsmount_lock); 7373cd49ecSMiklos Szeredi if (res == -EAGAIN) 7473cd49ecSMiklos Szeredi goto retry; 7573cd49ecSMiklos Szeredi 7673cd49ecSMiklos Szeredi return res; 7773cd49ecSMiklos Szeredi } 7873cd49ecSMiklos Szeredi 7973cd49ecSMiklos Szeredi static void mnt_free_id(struct vfsmount *mnt) 8073cd49ecSMiklos Szeredi { 8173cd49ecSMiklos Szeredi spin_lock(&vfsmount_lock); 8273cd49ecSMiklos Szeredi ida_remove(&mnt_id_ida, mnt->mnt_id); 8373cd49ecSMiklos Szeredi spin_unlock(&vfsmount_lock); 8473cd49ecSMiklos Szeredi } 8573cd49ecSMiklos Szeredi 861da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 871da177e4SLinus Torvalds { 88c3762229SRobert P. J. Day struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 891da177e4SLinus Torvalds if (mnt) { 9073cd49ecSMiklos Szeredi int err; 9173cd49ecSMiklos Szeredi 9273cd49ecSMiklos Szeredi err = mnt_alloc_id(mnt); 9373cd49ecSMiklos Szeredi if (err) { 9473cd49ecSMiklos Szeredi kmem_cache_free(mnt_cache, mnt); 9573cd49ecSMiklos Szeredi return NULL; 9673cd49ecSMiklos Szeredi } 9773cd49ecSMiklos Szeredi 981da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 991da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 1001da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 1011da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 1021da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 10355e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 10403e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 105a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 106a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 1073d733633SDave Hansen atomic_set(&mnt->__mnt_writers, 0); 1081da177e4SLinus Torvalds if (name) { 1091da177e4SLinus Torvalds int size = strlen(name) + 1; 1101da177e4SLinus Torvalds char *newname = kmalloc(size, GFP_KERNEL); 1111da177e4SLinus Torvalds if (newname) { 1121da177e4SLinus Torvalds memcpy(newname, name, size); 1131da177e4SLinus Torvalds mnt->mnt_devname = newname; 1141da177e4SLinus Torvalds } 1151da177e4SLinus Torvalds } 1161da177e4SLinus Torvalds } 1171da177e4SLinus Torvalds return mnt; 1181da177e4SLinus Torvalds } 1191da177e4SLinus Torvalds 1208366025eSDave Hansen /* 1218366025eSDave Hansen * Most r/o checks on a fs are for operations that take 1228366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 1238366025eSDave Hansen * We must keep track of when those operations start 1248366025eSDave Hansen * (for permission checks) and when they end, so that 1258366025eSDave Hansen * we can determine when writes are able to occur to 1268366025eSDave Hansen * a filesystem. 1278366025eSDave Hansen */ 1283d733633SDave Hansen /* 1293d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 1303d733633SDave Hansen * @mnt: the mount to check for its write status 1313d733633SDave Hansen * 1323d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 1333d733633SDave Hansen * It does not guarantee that the filesystem will stay 1343d733633SDave Hansen * r/w, just that it is right *now*. This can not and 1353d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 1363d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 1373d733633SDave Hansen * r/w. 1383d733633SDave Hansen */ 1393d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 1403d733633SDave Hansen { 1412e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 1422e4b7fcdSDave Hansen return 1; 1432e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 1442e4b7fcdSDave Hansen return 1; 1452e4b7fcdSDave Hansen return 0; 1463d733633SDave Hansen } 1473d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 1483d733633SDave Hansen 1493d733633SDave Hansen struct mnt_writer { 1503d733633SDave Hansen /* 1513d733633SDave Hansen * If holding multiple instances of this lock, they 1523d733633SDave Hansen * must be ordered by cpu number. 1533d733633SDave Hansen */ 1543d733633SDave Hansen spinlock_t lock; 1553d733633SDave Hansen struct lock_class_key lock_class; /* compiles out with !lockdep */ 1563d733633SDave Hansen unsigned long count; 1573d733633SDave Hansen struct vfsmount *mnt; 1583d733633SDave Hansen } ____cacheline_aligned_in_smp; 1593d733633SDave Hansen static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); 1603d733633SDave Hansen 1613d733633SDave Hansen static int __init init_mnt_writers(void) 1623d733633SDave Hansen { 1633d733633SDave Hansen int cpu; 1643d733633SDave Hansen for_each_possible_cpu(cpu) { 1653d733633SDave Hansen struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); 1663d733633SDave Hansen spin_lock_init(&writer->lock); 1673d733633SDave Hansen lockdep_set_class(&writer->lock, &writer->lock_class); 1683d733633SDave Hansen writer->count = 0; 1693d733633SDave Hansen } 1703d733633SDave Hansen return 0; 1713d733633SDave Hansen } 1723d733633SDave Hansen fs_initcall(init_mnt_writers); 1733d733633SDave Hansen 1743d733633SDave Hansen static void unlock_mnt_writers(void) 1753d733633SDave Hansen { 1763d733633SDave Hansen int cpu; 1773d733633SDave Hansen struct mnt_writer *cpu_writer; 1783d733633SDave Hansen 1793d733633SDave Hansen for_each_possible_cpu(cpu) { 1803d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 1813d733633SDave Hansen spin_unlock(&cpu_writer->lock); 1823d733633SDave Hansen } 1833d733633SDave Hansen } 1843d733633SDave Hansen 1853d733633SDave Hansen static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) 1863d733633SDave Hansen { 1873d733633SDave Hansen if (!cpu_writer->mnt) 1883d733633SDave Hansen return; 1893d733633SDave Hansen /* 1903d733633SDave Hansen * This is in case anyone ever leaves an invalid, 1913d733633SDave Hansen * old ->mnt and a count of 0. 1923d733633SDave Hansen */ 1933d733633SDave Hansen if (!cpu_writer->count) 1943d733633SDave Hansen return; 1953d733633SDave Hansen atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); 1963d733633SDave Hansen cpu_writer->count = 0; 1973d733633SDave Hansen } 1983d733633SDave Hansen /* 1993d733633SDave Hansen * must hold cpu_writer->lock 2003d733633SDave Hansen */ 2013d733633SDave Hansen static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, 2023d733633SDave Hansen struct vfsmount *mnt) 2033d733633SDave Hansen { 2043d733633SDave Hansen if (cpu_writer->mnt == mnt) 2053d733633SDave Hansen return; 2063d733633SDave Hansen __clear_mnt_count(cpu_writer); 2073d733633SDave Hansen cpu_writer->mnt = mnt; 2083d733633SDave Hansen } 2093d733633SDave Hansen 2103d733633SDave Hansen /* 2113d733633SDave Hansen * Most r/o checks on a fs are for operations that take 2123d733633SDave Hansen * discrete amounts of time, like a write() or unlink(). 2133d733633SDave Hansen * We must keep track of when those operations start 2143d733633SDave Hansen * (for permission checks) and when they end, so that 2153d733633SDave Hansen * we can determine when writes are able to occur to 2163d733633SDave Hansen * a filesystem. 2173d733633SDave Hansen */ 2188366025eSDave Hansen /** 2198366025eSDave Hansen * mnt_want_write - get write access to a mount 2208366025eSDave Hansen * @mnt: the mount on which to take a write 2218366025eSDave Hansen * 2228366025eSDave Hansen * This tells the low-level filesystem that a write is 2238366025eSDave Hansen * about to be performed to it, and makes sure that 2248366025eSDave Hansen * writes are allowed before returning success. When 2258366025eSDave Hansen * the write operation is finished, mnt_drop_write() 2268366025eSDave Hansen * must be called. This is effectively a refcount. 2278366025eSDave Hansen */ 2288366025eSDave Hansen int mnt_want_write(struct vfsmount *mnt) 2298366025eSDave Hansen { 2303d733633SDave Hansen int ret = 0; 2313d733633SDave Hansen struct mnt_writer *cpu_writer; 2323d733633SDave Hansen 2333d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 2343d733633SDave Hansen spin_lock(&cpu_writer->lock); 2353d733633SDave Hansen if (__mnt_is_readonly(mnt)) { 2363d733633SDave Hansen ret = -EROFS; 2373d733633SDave Hansen goto out; 2383d733633SDave Hansen } 2393d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 2403d733633SDave Hansen cpu_writer->count++; 2413d733633SDave Hansen out: 2423d733633SDave Hansen spin_unlock(&cpu_writer->lock); 2433d733633SDave Hansen put_cpu_var(mnt_writers); 2443d733633SDave Hansen return ret; 2458366025eSDave Hansen } 2468366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 2478366025eSDave Hansen 2483d733633SDave Hansen static void lock_mnt_writers(void) 2493d733633SDave Hansen { 2503d733633SDave Hansen int cpu; 2513d733633SDave Hansen struct mnt_writer *cpu_writer; 2523d733633SDave Hansen 2533d733633SDave Hansen for_each_possible_cpu(cpu) { 2543d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 2553d733633SDave Hansen spin_lock(&cpu_writer->lock); 2563d733633SDave Hansen __clear_mnt_count(cpu_writer); 2573d733633SDave Hansen cpu_writer->mnt = NULL; 2583d733633SDave Hansen } 2593d733633SDave Hansen } 2603d733633SDave Hansen 2613d733633SDave Hansen /* 2623d733633SDave Hansen * These per-cpu write counts are not guaranteed to have 2633d733633SDave Hansen * matched increments and decrements on any given cpu. 2643d733633SDave Hansen * A file open()ed for write on one cpu and close()d on 2653d733633SDave Hansen * another cpu will imbalance this count. Make sure it 2663d733633SDave Hansen * does not get too far out of whack. 2673d733633SDave Hansen */ 2683d733633SDave Hansen static void handle_write_count_underflow(struct vfsmount *mnt) 2693d733633SDave Hansen { 2703d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) >= 2713d733633SDave Hansen MNT_WRITER_UNDERFLOW_LIMIT) 2723d733633SDave Hansen return; 2733d733633SDave Hansen /* 2743d733633SDave Hansen * It isn't necessary to hold all of the locks 2753d733633SDave Hansen * at the same time, but doing it this way makes 2763d733633SDave Hansen * us share a lot more code. 2773d733633SDave Hansen */ 2783d733633SDave Hansen lock_mnt_writers(); 2793d733633SDave Hansen /* 2803d733633SDave Hansen * vfsmount_lock is for mnt_flags. 2813d733633SDave Hansen */ 2823d733633SDave Hansen spin_lock(&vfsmount_lock); 2833d733633SDave Hansen /* 2843d733633SDave Hansen * If coalescing the per-cpu writer counts did not 2853d733633SDave Hansen * get us back to a positive writer count, we have 2863d733633SDave Hansen * a bug. 2873d733633SDave Hansen */ 2883d733633SDave Hansen if ((atomic_read(&mnt->__mnt_writers) < 0) && 2893d733633SDave Hansen !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { 2903d733633SDave Hansen printk(KERN_DEBUG "leak detected on mount(%p) writers " 2913d733633SDave Hansen "count: %d\n", 2923d733633SDave Hansen mnt, atomic_read(&mnt->__mnt_writers)); 2933d733633SDave Hansen WARN_ON(1); 2943d733633SDave Hansen /* use the flag to keep the dmesg spam down */ 2953d733633SDave Hansen mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT; 2963d733633SDave Hansen } 2973d733633SDave Hansen spin_unlock(&vfsmount_lock); 2983d733633SDave Hansen unlock_mnt_writers(); 2993d733633SDave Hansen } 3003d733633SDave Hansen 3018366025eSDave Hansen /** 3028366025eSDave Hansen * mnt_drop_write - give up write access to a mount 3038366025eSDave Hansen * @mnt: the mount on which to give up write access 3048366025eSDave Hansen * 3058366025eSDave Hansen * Tells the low-level filesystem that we are done 3068366025eSDave Hansen * performing writes to it. Must be matched with 3078366025eSDave Hansen * mnt_want_write() call above. 3088366025eSDave Hansen */ 3098366025eSDave Hansen void mnt_drop_write(struct vfsmount *mnt) 3108366025eSDave Hansen { 3113d733633SDave Hansen int must_check_underflow = 0; 3123d733633SDave Hansen struct mnt_writer *cpu_writer; 3133d733633SDave Hansen 3143d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 3153d733633SDave Hansen spin_lock(&cpu_writer->lock); 3163d733633SDave Hansen 3173d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 3183d733633SDave Hansen if (cpu_writer->count > 0) { 3193d733633SDave Hansen cpu_writer->count--; 3203d733633SDave Hansen } else { 3213d733633SDave Hansen must_check_underflow = 1; 3223d733633SDave Hansen atomic_dec(&mnt->__mnt_writers); 3233d733633SDave Hansen } 3243d733633SDave Hansen 3253d733633SDave Hansen spin_unlock(&cpu_writer->lock); 3263d733633SDave Hansen /* 3273d733633SDave Hansen * Logically, we could call this each time, 3283d733633SDave Hansen * but the __mnt_writers cacheline tends to 3293d733633SDave Hansen * be cold, and makes this expensive. 3303d733633SDave Hansen */ 3313d733633SDave Hansen if (must_check_underflow) 3323d733633SDave Hansen handle_write_count_underflow(mnt); 3333d733633SDave Hansen /* 3343d733633SDave Hansen * This could be done right after the spinlock 3353d733633SDave Hansen * is taken because the spinlock keeps us on 3363d733633SDave Hansen * the cpu, and disables preemption. However, 3373d733633SDave Hansen * putting it here bounds the amount that 3383d733633SDave Hansen * __mnt_writers can underflow. Without it, 3393d733633SDave Hansen * we could theoretically wrap __mnt_writers. 3403d733633SDave Hansen */ 3413d733633SDave Hansen put_cpu_var(mnt_writers); 3428366025eSDave Hansen } 3438366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 3448366025eSDave Hansen 3452e4b7fcdSDave Hansen static int mnt_make_readonly(struct vfsmount *mnt) 3468366025eSDave Hansen { 3473d733633SDave Hansen int ret = 0; 3483d733633SDave Hansen 3493d733633SDave Hansen lock_mnt_writers(); 3503d733633SDave Hansen /* 3513d733633SDave Hansen * With all the locks held, this value is stable 3523d733633SDave Hansen */ 3533d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) > 0) { 3543d733633SDave Hansen ret = -EBUSY; 3553d733633SDave Hansen goto out; 3568366025eSDave Hansen } 3573d733633SDave Hansen /* 3582e4b7fcdSDave Hansen * nobody can do a successful mnt_want_write() with all 3592e4b7fcdSDave Hansen * of the counts in MNT_DENIED_WRITE and the locks held. 3603d733633SDave Hansen */ 3612e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 3622e4b7fcdSDave Hansen if (!ret) 3632e4b7fcdSDave Hansen mnt->mnt_flags |= MNT_READONLY; 3642e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3653d733633SDave Hansen out: 3663d733633SDave Hansen unlock_mnt_writers(); 3673d733633SDave Hansen return ret; 3683d733633SDave Hansen } 3698366025eSDave Hansen 3702e4b7fcdSDave Hansen static void __mnt_unmake_readonly(struct vfsmount *mnt) 3712e4b7fcdSDave Hansen { 3722e4b7fcdSDave Hansen spin_lock(&vfsmount_lock); 3732e4b7fcdSDave Hansen mnt->mnt_flags &= ~MNT_READONLY; 3742e4b7fcdSDave Hansen spin_unlock(&vfsmount_lock); 3752e4b7fcdSDave Hansen } 3762e4b7fcdSDave Hansen 377454e2398SDavid Howells int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 378454e2398SDavid Howells { 379454e2398SDavid Howells mnt->mnt_sb = sb; 380454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 381454e2398SDavid Howells return 0; 382454e2398SDavid Howells } 383454e2398SDavid Howells 384454e2398SDavid Howells EXPORT_SYMBOL(simple_set_mnt); 385454e2398SDavid Howells 3861da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 3871da177e4SLinus Torvalds { 3881da177e4SLinus Torvalds kfree(mnt->mnt_devname); 38973cd49ecSMiklos Szeredi mnt_free_id(mnt); 3901da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 3911da177e4SLinus Torvalds } 3921da177e4SLinus Torvalds 3931da177e4SLinus Torvalds /* 394a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 395a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 3961da177e4SLinus Torvalds */ 397a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 398a05964f3SRam Pai int dir) 3991da177e4SLinus Torvalds { 4001da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 4011da177e4SLinus Torvalds struct list_head *tmp = head; 4021da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 4031da177e4SLinus Torvalds 4041da177e4SLinus Torvalds for (;;) { 405a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 4061da177e4SLinus Torvalds p = NULL; 4071da177e4SLinus Torvalds if (tmp == head) 4081da177e4SLinus Torvalds break; 4091da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 4101da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 411a05964f3SRam Pai found = p; 4121da177e4SLinus Torvalds break; 4131da177e4SLinus Torvalds } 4141da177e4SLinus Torvalds } 4151da177e4SLinus Torvalds return found; 4161da177e4SLinus Torvalds } 4171da177e4SLinus Torvalds 418a05964f3SRam Pai /* 419a05964f3SRam Pai * lookup_mnt increments the ref count before returning 420a05964f3SRam Pai * the vfsmount struct. 421a05964f3SRam Pai */ 422a05964f3SRam Pai struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 423a05964f3SRam Pai { 424a05964f3SRam Pai struct vfsmount *child_mnt; 425a05964f3SRam Pai spin_lock(&vfsmount_lock); 426a05964f3SRam Pai if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) 427a05964f3SRam Pai mntget(child_mnt); 428a05964f3SRam Pai spin_unlock(&vfsmount_lock); 429a05964f3SRam Pai return child_mnt; 430a05964f3SRam Pai } 431a05964f3SRam Pai 4321da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 4331da177e4SLinus Torvalds { 4346b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 4351da177e4SLinus Torvalds } 4361da177e4SLinus Torvalds 4376b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 4385addc5ddSAl Viro { 4395addc5ddSAl Viro if (ns) { 4405addc5ddSAl Viro ns->event = ++event; 4415addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4425addc5ddSAl Viro } 4435addc5ddSAl Viro } 4445addc5ddSAl Viro 4456b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 4465addc5ddSAl Viro { 4475addc5ddSAl Viro if (ns && ns->event != event) { 4485addc5ddSAl Viro ns->event = event; 4495addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4505addc5ddSAl Viro } 4515addc5ddSAl Viro } 4525addc5ddSAl Viro 4531a390689SAl Viro static void detach_mnt(struct vfsmount *mnt, struct path *old_path) 4541da177e4SLinus Torvalds { 4551a390689SAl Viro old_path->dentry = mnt->mnt_mountpoint; 4561a390689SAl Viro old_path->mnt = mnt->mnt_parent; 4571da177e4SLinus Torvalds mnt->mnt_parent = mnt; 4581da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 4591da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 4601da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 4611a390689SAl Viro old_path->dentry->d_mounted--; 4621da177e4SLinus Torvalds } 4631da177e4SLinus Torvalds 464b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 465b90fa9aeSRam Pai struct vfsmount *child_mnt) 466b90fa9aeSRam Pai { 467b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 468b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 469b90fa9aeSRam Pai dentry->d_mounted++; 470b90fa9aeSRam Pai } 471b90fa9aeSRam Pai 4721a390689SAl Viro static void attach_mnt(struct vfsmount *mnt, struct path *path) 4731da177e4SLinus Torvalds { 4741a390689SAl Viro mnt_set_mountpoint(path->mnt, path->dentry, mnt); 475b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 4761a390689SAl Viro hash(path->mnt, path->dentry)); 4771a390689SAl Viro list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); 478b90fa9aeSRam Pai } 479b90fa9aeSRam Pai 480b90fa9aeSRam Pai /* 481b90fa9aeSRam Pai * the caller must hold vfsmount_lock 482b90fa9aeSRam Pai */ 483b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 484b90fa9aeSRam Pai { 485b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 486b90fa9aeSRam Pai struct vfsmount *m; 487b90fa9aeSRam Pai LIST_HEAD(head); 4886b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 489b90fa9aeSRam Pai 490b90fa9aeSRam Pai BUG_ON(parent == mnt); 491b90fa9aeSRam Pai 492b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 493b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 4946b3286edSKirill Korotaev m->mnt_ns = n; 495b90fa9aeSRam Pai list_splice(&head, n->list.prev); 496b90fa9aeSRam Pai 497b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 498b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 499b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 5006b3286edSKirill Korotaev touch_mnt_namespace(n); 5011da177e4SLinus Torvalds } 5021da177e4SLinus Torvalds 5031da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 5041da177e4SLinus Torvalds { 5051da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 5061da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 5071da177e4SLinus Torvalds while (1) { 5081da177e4SLinus Torvalds if (p == root) 5091da177e4SLinus Torvalds return NULL; 5101da177e4SLinus Torvalds next = p->mnt_child.next; 5111da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 5121da177e4SLinus Torvalds break; 5131da177e4SLinus Torvalds p = p->mnt_parent; 5141da177e4SLinus Torvalds } 5151da177e4SLinus Torvalds } 5161da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 5171da177e4SLinus Torvalds } 5181da177e4SLinus Torvalds 5199676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 5209676f0c6SRam Pai { 5219676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 5229676f0c6SRam Pai while (prev != &p->mnt_mounts) { 5239676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 5249676f0c6SRam Pai prev = p->mnt_mounts.prev; 5259676f0c6SRam Pai } 5269676f0c6SRam Pai return p; 5279676f0c6SRam Pai } 5289676f0c6SRam Pai 52936341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 53036341f64SRam Pai int flag) 5311da177e4SLinus Torvalds { 5321da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 5331da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 5341da177e4SLinus Torvalds 5351da177e4SLinus Torvalds if (mnt) { 5361da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 5371da177e4SLinus Torvalds atomic_inc(&sb->s_active); 5381da177e4SLinus Torvalds mnt->mnt_sb = sb; 5391da177e4SLinus Torvalds mnt->mnt_root = dget(root); 5401da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 5411da177e4SLinus Torvalds mnt->mnt_parent = mnt; 542b90fa9aeSRam Pai 5435afe0022SRam Pai if (flag & CL_SLAVE) { 5445afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 5455afe0022SRam Pai mnt->mnt_master = old; 5465afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 5478aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 548b90fa9aeSRam Pai if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) 549b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 5505afe0022SRam Pai if (IS_MNT_SLAVE(old)) 5515afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 5525afe0022SRam Pai mnt->mnt_master = old->mnt_master; 5535afe0022SRam Pai } 554b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 555b90fa9aeSRam Pai set_mnt_shared(mnt); 5561da177e4SLinus Torvalds 5571da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 5581da177e4SLinus Torvalds * as the original if that was on one */ 55936341f64SRam Pai if (flag & CL_EXPIRE) { 56055e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 56155e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 5621da177e4SLinus Torvalds } 56336341f64SRam Pai } 5641da177e4SLinus Torvalds return mnt; 5651da177e4SLinus Torvalds } 5661da177e4SLinus Torvalds 5677b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 5681da177e4SLinus Torvalds { 5693d733633SDave Hansen int cpu; 5701da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 5713d733633SDave Hansen /* 5723d733633SDave Hansen * We don't have to hold all of the locks at the 5733d733633SDave Hansen * same time here because we know that we're the 5743d733633SDave Hansen * last reference to mnt and that no new writers 5753d733633SDave Hansen * can come in. 5763d733633SDave Hansen */ 5773d733633SDave Hansen for_each_possible_cpu(cpu) { 5783d733633SDave Hansen struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu); 5793d733633SDave Hansen if (cpu_writer->mnt != mnt) 5803d733633SDave Hansen continue; 5813d733633SDave Hansen spin_lock(&cpu_writer->lock); 5823d733633SDave Hansen atomic_add(cpu_writer->count, &mnt->__mnt_writers); 5833d733633SDave Hansen cpu_writer->count = 0; 5843d733633SDave Hansen /* 5853d733633SDave Hansen * Might as well do this so that no one 5863d733633SDave Hansen * ever sees the pointer and expects 5873d733633SDave Hansen * it to be valid. 5883d733633SDave Hansen */ 5893d733633SDave Hansen cpu_writer->mnt = NULL; 5903d733633SDave Hansen spin_unlock(&cpu_writer->lock); 5913d733633SDave Hansen } 5923d733633SDave Hansen /* 5933d733633SDave Hansen * This probably indicates that somebody messed 5943d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 5953d733633SDave Hansen * happens, the filesystem was probably unable 5963d733633SDave Hansen * to make r/w->r/o transitions. 5973d733633SDave Hansen */ 5983d733633SDave Hansen WARN_ON(atomic_read(&mnt->__mnt_writers)); 5991da177e4SLinus Torvalds dput(mnt->mnt_root); 6001da177e4SLinus Torvalds free_vfsmnt(mnt); 6011da177e4SLinus Torvalds deactivate_super(sb); 6021da177e4SLinus Torvalds } 6031da177e4SLinus Torvalds 6047b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 6057b7b1aceSAl Viro { 6067b7b1aceSAl Viro repeat: 6077b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 6087b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 6097b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6107b7b1aceSAl Viro __mntput(mnt); 6117b7b1aceSAl Viro return; 6127b7b1aceSAl Viro } 6137b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 6147b7b1aceSAl Viro mnt->mnt_pinned = 0; 6157b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6167b7b1aceSAl Viro acct_auto_close_mnt(mnt); 6177b7b1aceSAl Viro security_sb_umount_close(mnt); 6187b7b1aceSAl Viro goto repeat; 6197b7b1aceSAl Viro } 6207b7b1aceSAl Viro } 6217b7b1aceSAl Viro 6227b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 6237b7b1aceSAl Viro 6247b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 6257b7b1aceSAl Viro { 6267b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6277b7b1aceSAl Viro mnt->mnt_pinned++; 6287b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6297b7b1aceSAl Viro } 6307b7b1aceSAl Viro 6317b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 6327b7b1aceSAl Viro 6337b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 6347b7b1aceSAl Viro { 6357b7b1aceSAl Viro spin_lock(&vfsmount_lock); 6367b7b1aceSAl Viro if (mnt->mnt_pinned) { 6377b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 6387b7b1aceSAl Viro mnt->mnt_pinned--; 6397b7b1aceSAl Viro } 6407b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 6417b7b1aceSAl Viro } 6427b7b1aceSAl Viro 6437b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 6441da177e4SLinus Torvalds 645b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 646b3b304a2SMiklos Szeredi { 647b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 648b3b304a2SMiklos Szeredi } 649b3b304a2SMiklos Szeredi 650b3b304a2SMiklos Szeredi /* 651b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 652b3b304a2SMiklos Szeredi * implement more complex mount option showing. 653b3b304a2SMiklos Szeredi * 654b3b304a2SMiklos Szeredi * See also save_mount_options(). 655b3b304a2SMiklos Szeredi */ 656b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 657b3b304a2SMiklos Szeredi { 658b3b304a2SMiklos Szeredi const char *options = mnt->mnt_sb->s_options; 659b3b304a2SMiklos Szeredi 660b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 661b3b304a2SMiklos Szeredi seq_putc(m, ','); 662b3b304a2SMiklos Szeredi mangle(m, options); 663b3b304a2SMiklos Szeredi } 664b3b304a2SMiklos Szeredi 665b3b304a2SMiklos Szeredi return 0; 666b3b304a2SMiklos Szeredi } 667b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 668b3b304a2SMiklos Szeredi 669b3b304a2SMiklos Szeredi /* 670b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 671b3b304a2SMiklos Szeredi * called from the fill_super() callback. 672b3b304a2SMiklos Szeredi * 673b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 674b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 675b3b304a2SMiklos Szeredi * remount fails. 676b3b304a2SMiklos Szeredi * 677b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 678b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 679b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 680b3b304a2SMiklos Szeredi * any more. 681b3b304a2SMiklos Szeredi */ 682b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 683b3b304a2SMiklos Szeredi { 684b3b304a2SMiklos Szeredi kfree(sb->s_options); 685b3b304a2SMiklos Szeredi sb->s_options = kstrdup(options, GFP_KERNEL); 686b3b304a2SMiklos Szeredi } 687b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 688b3b304a2SMiklos Szeredi 6891da177e4SLinus Torvalds /* iterator */ 6901da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 6911da177e4SLinus Torvalds { 6926b3286edSKirill Korotaev struct mnt_namespace *n = m->private; 6931da177e4SLinus Torvalds 694390c6843SRam Pai down_read(&namespace_sem); 695b0765fb8SPavel Emelianov return seq_list_start(&n->list, *pos); 6961da177e4SLinus Torvalds } 6971da177e4SLinus Torvalds 6981da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 6991da177e4SLinus Torvalds { 7006b3286edSKirill Korotaev struct mnt_namespace *n = m->private; 701b0765fb8SPavel Emelianov 702b0765fb8SPavel Emelianov return seq_list_next(v, &n->list, pos); 7031da177e4SLinus Torvalds } 7041da177e4SLinus Torvalds 7051da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 7061da177e4SLinus Torvalds { 707390c6843SRam Pai up_read(&namespace_sem); 7081da177e4SLinus Torvalds } 7091da177e4SLinus Torvalds 7101da177e4SLinus Torvalds static int show_vfsmnt(struct seq_file *m, void *v) 7111da177e4SLinus Torvalds { 712b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 7131da177e4SLinus Torvalds int err = 0; 7141da177e4SLinus Torvalds static struct proc_fs_info { 7151da177e4SLinus Torvalds int flag; 7161da177e4SLinus Torvalds char *str; 7171da177e4SLinus Torvalds } fs_info[] = { 7181da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 7191da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 7201da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 7211da177e4SLinus Torvalds { 0, NULL } 7221da177e4SLinus Torvalds }; 7231da177e4SLinus Torvalds static struct proc_fs_info mnt_info[] = { 7241da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 7251da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 7261da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 727fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 728fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 72947ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 7301da177e4SLinus Torvalds { 0, NULL } 7311da177e4SLinus Torvalds }; 7321da177e4SLinus Torvalds struct proc_fs_info *fs_infop; 733c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 7341da177e4SLinus Torvalds 7351da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 7361da177e4SLinus Torvalds seq_putc(m, ' '); 737c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 7381da177e4SLinus Torvalds seq_putc(m, ' '); 7391da177e4SLinus Torvalds mangle(m, mnt->mnt_sb->s_type->name); 74079c0b2dfSMiklos Szeredi if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { 74179c0b2dfSMiklos Szeredi seq_putc(m, '.'); 74279c0b2dfSMiklos Szeredi mangle(m, mnt->mnt_sb->s_subtype); 74379c0b2dfSMiklos Szeredi } 7442e4b7fcdSDave Hansen seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 7451da177e4SLinus Torvalds for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 7461da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & fs_infop->flag) 7471da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 7481da177e4SLinus Torvalds } 7491da177e4SLinus Torvalds for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 7501da177e4SLinus Torvalds if (mnt->mnt_flags & fs_infop->flag) 7511da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 7521da177e4SLinus Torvalds } 7531da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 7541da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 7551da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 7561da177e4SLinus Torvalds return err; 7571da177e4SLinus Torvalds } 7581da177e4SLinus Torvalds 7591da177e4SLinus Torvalds struct seq_operations mounts_op = { 7601da177e4SLinus Torvalds .start = m_start, 7611da177e4SLinus Torvalds .next = m_next, 7621da177e4SLinus Torvalds .stop = m_stop, 7631da177e4SLinus Torvalds .show = show_vfsmnt 7641da177e4SLinus Torvalds }; 7651da177e4SLinus Torvalds 766b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 767b4629fe2SChuck Lever { 768b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 769c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 770b4629fe2SChuck Lever int err = 0; 771b4629fe2SChuck Lever 772b4629fe2SChuck Lever /* device */ 773b4629fe2SChuck Lever if (mnt->mnt_devname) { 774b4629fe2SChuck Lever seq_puts(m, "device "); 775b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 776b4629fe2SChuck Lever } else 777b4629fe2SChuck Lever seq_puts(m, "no device"); 778b4629fe2SChuck Lever 779b4629fe2SChuck Lever /* mount point */ 780b4629fe2SChuck Lever seq_puts(m, " mounted on "); 781c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 782b4629fe2SChuck Lever seq_putc(m, ' '); 783b4629fe2SChuck Lever 784b4629fe2SChuck Lever /* file system type */ 785b4629fe2SChuck Lever seq_puts(m, "with fstype "); 786b4629fe2SChuck Lever mangle(m, mnt->mnt_sb->s_type->name); 787b4629fe2SChuck Lever 788b4629fe2SChuck Lever /* optional statistics */ 789b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 790b4629fe2SChuck Lever seq_putc(m, ' '); 791b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 792b4629fe2SChuck Lever } 793b4629fe2SChuck Lever 794b4629fe2SChuck Lever seq_putc(m, '\n'); 795b4629fe2SChuck Lever return err; 796b4629fe2SChuck Lever } 797b4629fe2SChuck Lever 798b4629fe2SChuck Lever struct seq_operations mountstats_op = { 799b4629fe2SChuck Lever .start = m_start, 800b4629fe2SChuck Lever .next = m_next, 801b4629fe2SChuck Lever .stop = m_stop, 802b4629fe2SChuck Lever .show = show_vfsstat, 803b4629fe2SChuck Lever }; 804b4629fe2SChuck Lever 8051da177e4SLinus Torvalds /** 8061da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 8071da177e4SLinus Torvalds * @mnt: root of mount tree 8081da177e4SLinus Torvalds * 8091da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 8101da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 8111da177e4SLinus Torvalds * busy. 8121da177e4SLinus Torvalds */ 8131da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 8141da177e4SLinus Torvalds { 81536341f64SRam Pai int actual_refs = 0; 81636341f64SRam Pai int minimum_refs = 0; 81736341f64SRam Pai struct vfsmount *p; 8181da177e4SLinus Torvalds 8191da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 82036341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 8211da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 8221da177e4SLinus Torvalds minimum_refs += 2; 8231da177e4SLinus Torvalds } 8241da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 8251da177e4SLinus Torvalds 8261da177e4SLinus Torvalds if (actual_refs > minimum_refs) 8271da177e4SLinus Torvalds return 0; 828e3474a8eSIan Kent 829e3474a8eSIan Kent return 1; 8301da177e4SLinus Torvalds } 8311da177e4SLinus Torvalds 8321da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 8331da177e4SLinus Torvalds 8341da177e4SLinus Torvalds /** 8351da177e4SLinus Torvalds * may_umount - check if a mount point is busy 8361da177e4SLinus Torvalds * @mnt: root of mount 8371da177e4SLinus Torvalds * 8381da177e4SLinus Torvalds * This is called to check if a mount point has any 8391da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 8401da177e4SLinus Torvalds * mount has sub mounts this will return busy 8411da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 8421da177e4SLinus Torvalds * 8431da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 8441da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 8451da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 8461da177e4SLinus Torvalds */ 8471da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 8481da177e4SLinus Torvalds { 849e3474a8eSIan Kent int ret = 1; 850a05964f3SRam Pai spin_lock(&vfsmount_lock); 851a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 852e3474a8eSIan Kent ret = 0; 853a05964f3SRam Pai spin_unlock(&vfsmount_lock); 854a05964f3SRam Pai return ret; 8551da177e4SLinus Torvalds } 8561da177e4SLinus Torvalds 8571da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 8581da177e4SLinus Torvalds 859b90fa9aeSRam Pai void release_mounts(struct list_head *head) 8601da177e4SLinus Torvalds { 86170fbcdf4SRam Pai struct vfsmount *mnt; 86270fbcdf4SRam Pai while (!list_empty(head)) { 863b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 86470fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 86570fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 86670fbcdf4SRam Pai struct dentry *dentry; 86770fbcdf4SRam Pai struct vfsmount *m; 86870fbcdf4SRam Pai spin_lock(&vfsmount_lock); 86970fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 87070fbcdf4SRam Pai m = mnt->mnt_parent; 87170fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 87270fbcdf4SRam Pai mnt->mnt_parent = mnt; 8737c4b93d8SAl Viro m->mnt_ghosts--; 8741da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 87570fbcdf4SRam Pai dput(dentry); 87670fbcdf4SRam Pai mntput(m); 8771da177e4SLinus Torvalds } 8781da177e4SLinus Torvalds mntput(mnt); 87970fbcdf4SRam Pai } 88070fbcdf4SRam Pai } 88170fbcdf4SRam Pai 882a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 88370fbcdf4SRam Pai { 88470fbcdf4SRam Pai struct vfsmount *p; 88570fbcdf4SRam Pai 8861bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 8871bfba4e8SAkinobu Mita list_move(&p->mnt_hash, kill); 88870fbcdf4SRam Pai 889a05964f3SRam Pai if (propagate) 890a05964f3SRam Pai propagate_umount(kill); 891a05964f3SRam Pai 89270fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 89370fbcdf4SRam Pai list_del_init(&p->mnt_expire); 89470fbcdf4SRam Pai list_del_init(&p->mnt_list); 8956b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 8966b3286edSKirill Korotaev p->mnt_ns = NULL; 89770fbcdf4SRam Pai list_del_init(&p->mnt_child); 8987c4b93d8SAl Viro if (p->mnt_parent != p) { 8997c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 900f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 9017c4b93d8SAl Viro } 902a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 9031da177e4SLinus Torvalds } 9041da177e4SLinus Torvalds } 9051da177e4SLinus Torvalds 906c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 907c35038beSAl Viro 9081da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 9091da177e4SLinus Torvalds { 9101da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 9111da177e4SLinus Torvalds int retval; 91270fbcdf4SRam Pai LIST_HEAD(umount_list); 9131da177e4SLinus Torvalds 9141da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 9151da177e4SLinus Torvalds if (retval) 9161da177e4SLinus Torvalds return retval; 9171da177e4SLinus Torvalds 9181da177e4SLinus Torvalds /* 9191da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 9201da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 9211da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 9221da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 9231da177e4SLinus Torvalds */ 9241da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 9256ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 9261da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 9271da177e4SLinus Torvalds return -EINVAL; 9281da177e4SLinus Torvalds 9291da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 9301da177e4SLinus Torvalds return -EBUSY; 9311da177e4SLinus Torvalds 9321da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 9331da177e4SLinus Torvalds return -EAGAIN; 9341da177e4SLinus Torvalds } 9351da177e4SLinus Torvalds 9361da177e4SLinus Torvalds /* 9371da177e4SLinus Torvalds * If we may have to abort operations to get out of this 9381da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 9391da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 9401da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 9411da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 9421da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 9431da177e4SLinus Torvalds * about for the moment. 9441da177e4SLinus Torvalds */ 9451da177e4SLinus Torvalds 9461da177e4SLinus Torvalds lock_kernel(); 9478b512d9aSTrond Myklebust if (sb->s_op->umount_begin) 9488b512d9aSTrond Myklebust sb->s_op->umount_begin(mnt, flags); 9491da177e4SLinus Torvalds unlock_kernel(); 9501da177e4SLinus Torvalds 9511da177e4SLinus Torvalds /* 9521da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 9531da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 9541da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 9551da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 9561da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 9571da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 9581da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 9591da177e4SLinus Torvalds */ 9606ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 9611da177e4SLinus Torvalds /* 9621da177e4SLinus Torvalds * Special case for "unmounting" root ... 9631da177e4SLinus Torvalds * we just try to remount it readonly. 9641da177e4SLinus Torvalds */ 9651da177e4SLinus Torvalds down_write(&sb->s_umount); 9661da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 9671da177e4SLinus Torvalds lock_kernel(); 9681da177e4SLinus Torvalds DQUOT_OFF(sb); 9691da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 9701da177e4SLinus Torvalds unlock_kernel(); 9711da177e4SLinus Torvalds } 9721da177e4SLinus Torvalds up_write(&sb->s_umount); 9731da177e4SLinus Torvalds return retval; 9741da177e4SLinus Torvalds } 9751da177e4SLinus Torvalds 976390c6843SRam Pai down_write(&namespace_sem); 9771da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 9785addc5ddSAl Viro event++; 9791da177e4SLinus Torvalds 980c35038beSAl Viro if (!(flags & MNT_DETACH)) 981c35038beSAl Viro shrink_submounts(mnt, &umount_list); 982c35038beSAl Viro 9831da177e4SLinus Torvalds retval = -EBUSY; 984a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 9851da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 986a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 9871da177e4SLinus Torvalds retval = 0; 9881da177e4SLinus Torvalds } 9891da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 9901da177e4SLinus Torvalds if (retval) 9911da177e4SLinus Torvalds security_sb_umount_busy(mnt); 992390c6843SRam Pai up_write(&namespace_sem); 99370fbcdf4SRam Pai release_mounts(&umount_list); 9941da177e4SLinus Torvalds return retval; 9951da177e4SLinus Torvalds } 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds /* 9981da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 9991da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 10001da177e4SLinus Torvalds * 10011da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 10021da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 10031da177e4SLinus Torvalds */ 10041da177e4SLinus Torvalds 10051da177e4SLinus Torvalds asmlinkage long sys_umount(char __user * name, int flags) 10061da177e4SLinus Torvalds { 10071da177e4SLinus Torvalds struct nameidata nd; 10081da177e4SLinus Torvalds int retval; 10091da177e4SLinus Torvalds 10101da177e4SLinus Torvalds retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 10111da177e4SLinus Torvalds if (retval) 10121da177e4SLinus Torvalds goto out; 10131da177e4SLinus Torvalds retval = -EINVAL; 10144ac91378SJan Blunck if (nd.path.dentry != nd.path.mnt->mnt_root) 10151da177e4SLinus Torvalds goto dput_and_out; 10164ac91378SJan Blunck if (!check_mnt(nd.path.mnt)) 10171da177e4SLinus Torvalds goto dput_and_out; 10181da177e4SLinus Torvalds 10191da177e4SLinus Torvalds retval = -EPERM; 10201da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 10211da177e4SLinus Torvalds goto dput_and_out; 10221da177e4SLinus Torvalds 10234ac91378SJan Blunck retval = do_umount(nd.path.mnt, flags); 10241da177e4SLinus Torvalds dput_and_out: 1025429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 10264ac91378SJan Blunck dput(nd.path.dentry); 10274ac91378SJan Blunck mntput_no_expire(nd.path.mnt); 10281da177e4SLinus Torvalds out: 10291da177e4SLinus Torvalds return retval; 10301da177e4SLinus Torvalds } 10311da177e4SLinus Torvalds 10321da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 10331da177e4SLinus Torvalds 10341da177e4SLinus Torvalds /* 10351da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 10361da177e4SLinus Torvalds */ 10371da177e4SLinus Torvalds asmlinkage long sys_oldumount(char __user * name) 10381da177e4SLinus Torvalds { 10391da177e4SLinus Torvalds return sys_umount(name, 0); 10401da177e4SLinus Torvalds } 10411da177e4SLinus Torvalds 10421da177e4SLinus Torvalds #endif 10431da177e4SLinus Torvalds 10441da177e4SLinus Torvalds static int mount_is_safe(struct nameidata *nd) 10451da177e4SLinus Torvalds { 10461da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 10471da177e4SLinus Torvalds return 0; 10481da177e4SLinus Torvalds return -EPERM; 10491da177e4SLinus Torvalds #ifdef notyet 10504ac91378SJan Blunck if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) 10511da177e4SLinus Torvalds return -EPERM; 10524ac91378SJan Blunck if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { 10534ac91378SJan Blunck if (current->uid != nd->path.dentry->d_inode->i_uid) 10541da177e4SLinus Torvalds return -EPERM; 10551da177e4SLinus Torvalds } 1056e4543eddSChristoph Hellwig if (vfs_permission(nd, MAY_WRITE)) 10571da177e4SLinus Torvalds return -EPERM; 10581da177e4SLinus Torvalds return 0; 10591da177e4SLinus Torvalds #endif 10601da177e4SLinus Torvalds } 10611da177e4SLinus Torvalds 1062b58fed8bSRam Pai static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) 10631da177e4SLinus Torvalds { 10641da177e4SLinus Torvalds while (1) { 10651da177e4SLinus Torvalds if (d == dentry) 10661da177e4SLinus Torvalds return 1; 10671da177e4SLinus Torvalds if (d == NULL || d == d->d_parent) 10681da177e4SLinus Torvalds return 0; 10691da177e4SLinus Torvalds d = d->d_parent; 10701da177e4SLinus Torvalds } 10711da177e4SLinus Torvalds } 10721da177e4SLinus Torvalds 1073b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 107436341f64SRam Pai int flag) 10751da177e4SLinus Torvalds { 10761da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 10771a390689SAl Viro struct path path; 10781da177e4SLinus Torvalds 10799676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 10809676f0c6SRam Pai return NULL; 10819676f0c6SRam Pai 108236341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 10831da177e4SLinus Torvalds if (!q) 10841da177e4SLinus Torvalds goto Enomem; 10851da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 10861da177e4SLinus Torvalds 10871da177e4SLinus Torvalds p = mnt; 1088fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 10891da177e4SLinus Torvalds if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) 10901da177e4SLinus Torvalds continue; 10911da177e4SLinus Torvalds 10921da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 10939676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 10949676f0c6SRam Pai s = skip_mnt_tree(s); 10959676f0c6SRam Pai continue; 10969676f0c6SRam Pai } 10971da177e4SLinus Torvalds while (p != s->mnt_parent) { 10981da177e4SLinus Torvalds p = p->mnt_parent; 10991da177e4SLinus Torvalds q = q->mnt_parent; 11001da177e4SLinus Torvalds } 11011da177e4SLinus Torvalds p = s; 11021a390689SAl Viro path.mnt = q; 11031a390689SAl Viro path.dentry = p->mnt_mountpoint; 110436341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 11051da177e4SLinus Torvalds if (!q) 11061da177e4SLinus Torvalds goto Enomem; 11071da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 11081da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 11091a390689SAl Viro attach_mnt(q, &path); 11101da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 11111da177e4SLinus Torvalds } 11121da177e4SLinus Torvalds } 11131da177e4SLinus Torvalds return res; 11141da177e4SLinus Torvalds Enomem: 11151da177e4SLinus Torvalds if (res) { 111670fbcdf4SRam Pai LIST_HEAD(umount_list); 11171da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1118a05964f3SRam Pai umount_tree(res, 0, &umount_list); 11191da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 112070fbcdf4SRam Pai release_mounts(&umount_list); 11211da177e4SLinus Torvalds } 11221da177e4SLinus Torvalds return NULL; 11231da177e4SLinus Torvalds } 11241da177e4SLinus Torvalds 11258aec0809SAl Viro struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) 11268aec0809SAl Viro { 11278aec0809SAl Viro struct vfsmount *tree; 11281a60a280SAl Viro down_write(&namespace_sem); 11298aec0809SAl Viro tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); 11301a60a280SAl Viro up_write(&namespace_sem); 11318aec0809SAl Viro return tree; 11328aec0809SAl Viro } 11338aec0809SAl Viro 11348aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 11358aec0809SAl Viro { 11368aec0809SAl Viro LIST_HEAD(umount_list); 11371a60a280SAl Viro down_write(&namespace_sem); 11388aec0809SAl Viro spin_lock(&vfsmount_lock); 11398aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 11408aec0809SAl Viro spin_unlock(&vfsmount_lock); 11411a60a280SAl Viro up_write(&namespace_sem); 11428aec0809SAl Viro release_mounts(&umount_list); 11438aec0809SAl Viro } 11448aec0809SAl Viro 1145b90fa9aeSRam Pai /* 1146b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1147b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 114821444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 114921444403SRam Pai * store the parent mount and mountpoint dentry. 115021444403SRam Pai * (done when source_mnt is moved) 1151b90fa9aeSRam Pai * 1152b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1153b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 11549676f0c6SRam Pai * --------------------------------------------------------------------------- 1155b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 11569676f0c6SRam Pai * |************************************************************************** 11579676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 11589676f0c6SRam Pai * | dest | | | | | 11599676f0c6SRam Pai * | | | | | | | 11609676f0c6SRam Pai * | v | | | | | 11619676f0c6SRam Pai * |************************************************************************** 11629676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 11635afe0022SRam Pai * | | | | | | 11649676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 11659676f0c6SRam Pai * *************************************************************************** 1166b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1167b90fa9aeSRam Pai * destination mount. 1168b90fa9aeSRam Pai * 1169b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1170b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1171b90fa9aeSRam Pai * the peer group of the source mount. 1172b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1173b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1174b90fa9aeSRam Pai * mount. 11755afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 11765afe0022SRam Pai * of the destination mount and the cloned mount is made slave 11775afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 11785afe0022SRam Pai * is marked as 'shared and slave'. 11795afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 11805afe0022SRam Pai * source mount. 11815afe0022SRam Pai * 11829676f0c6SRam Pai * --------------------------------------------------------------------------- 118321444403SRam Pai * | MOVE MOUNT OPERATION | 11849676f0c6SRam Pai * |************************************************************************** 11859676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 11869676f0c6SRam Pai * | dest | | | | | 11879676f0c6SRam Pai * | | | | | | | 11889676f0c6SRam Pai * | v | | | | | 11899676f0c6SRam Pai * |************************************************************************** 11909676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 11915afe0022SRam Pai * | | | | | | 11929676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 11939676f0c6SRam Pai * *************************************************************************** 11945afe0022SRam Pai * 11955afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 11965afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 119721444403SRam Pai * (+*) the mount is moved to the destination. 11985afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 11995afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 12005afe0022SRam Pai * the mount is marked as 'shared and slave'. 12015afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1202b90fa9aeSRam Pai * 1203b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1204b90fa9aeSRam Pai * applied to each mount in the tree. 1205b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1206b90fa9aeSRam Pai * in allocations. 1207b90fa9aeSRam Pai */ 1208b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 12091a390689SAl Viro struct path *path, struct path *parent_path) 1210b90fa9aeSRam Pai { 1211b90fa9aeSRam Pai LIST_HEAD(tree_list); 12121a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 12131a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1214b90fa9aeSRam Pai struct vfsmount *child, *p; 1215b90fa9aeSRam Pai 1216b90fa9aeSRam Pai if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) 1217b90fa9aeSRam Pai return -EINVAL; 1218b90fa9aeSRam Pai 1219b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1220b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1221b90fa9aeSRam Pai set_mnt_shared(p); 1222b90fa9aeSRam Pai } 1223b90fa9aeSRam Pai 1224b90fa9aeSRam Pai spin_lock(&vfsmount_lock); 12251a390689SAl Viro if (parent_path) { 12261a390689SAl Viro detach_mnt(source_mnt, parent_path); 12271a390689SAl Viro attach_mnt(source_mnt, path); 12286b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 122921444403SRam Pai } else { 1230b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1231b90fa9aeSRam Pai commit_tree(source_mnt); 123221444403SRam Pai } 1233b90fa9aeSRam Pai 1234b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1235b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1236b90fa9aeSRam Pai commit_tree(child); 1237b90fa9aeSRam Pai } 1238b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 1239b90fa9aeSRam Pai return 0; 1240b90fa9aeSRam Pai } 1241b90fa9aeSRam Pai 12428c3ee42eSAl Viro static int graft_tree(struct vfsmount *mnt, struct path *path) 12431da177e4SLinus Torvalds { 12441da177e4SLinus Torvalds int err; 12451da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 12461da177e4SLinus Torvalds return -EINVAL; 12471da177e4SLinus Torvalds 12488c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 12491da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 12501da177e4SLinus Torvalds return -ENOTDIR; 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds err = -ENOENT; 12538c3ee42eSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 12548c3ee42eSAl Viro if (IS_DEADDIR(path->dentry->d_inode)) 12551da177e4SLinus Torvalds goto out_unlock; 12561da177e4SLinus Torvalds 12578c3ee42eSAl Viro err = security_sb_check_sb(mnt, path); 12581da177e4SLinus Torvalds if (err) 12591da177e4SLinus Torvalds goto out_unlock; 12601da177e4SLinus Torvalds 12611da177e4SLinus Torvalds err = -ENOENT; 12628c3ee42eSAl Viro if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry)) 12638c3ee42eSAl Viro err = attach_recursive_mnt(mnt, path, NULL); 12641da177e4SLinus Torvalds out_unlock: 12658c3ee42eSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 12661da177e4SLinus Torvalds if (!err) 12678c3ee42eSAl Viro security_sb_post_addmount(mnt, path); 12681da177e4SLinus Torvalds return err; 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 12711da177e4SLinus Torvalds /* 127207b20889SRam Pai * recursively change the type of the mountpoint. 12732dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 127407b20889SRam Pai */ 12752dafe1c4SEric Sandeen static noinline int do_change_type(struct nameidata *nd, int flag) 127607b20889SRam Pai { 12774ac91378SJan Blunck struct vfsmount *m, *mnt = nd->path.mnt; 127807b20889SRam Pai int recurse = flag & MS_REC; 127907b20889SRam Pai int type = flag & ~MS_REC; 128007b20889SRam Pai 1281ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1282ee6f9582SMiklos Szeredi return -EPERM; 1283ee6f9582SMiklos Szeredi 12844ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 128507b20889SRam Pai return -EINVAL; 128607b20889SRam Pai 128707b20889SRam Pai down_write(&namespace_sem); 128807b20889SRam Pai spin_lock(&vfsmount_lock); 128907b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 129007b20889SRam Pai change_mnt_propagation(m, type); 129107b20889SRam Pai spin_unlock(&vfsmount_lock); 129207b20889SRam Pai up_write(&namespace_sem); 129307b20889SRam Pai return 0; 129407b20889SRam Pai } 129507b20889SRam Pai 129607b20889SRam Pai /* 12971da177e4SLinus Torvalds * do loopback mount. 12982dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 12991da177e4SLinus Torvalds */ 13002dafe1c4SEric Sandeen static noinline int do_loopback(struct nameidata *nd, char *old_name, 13012dafe1c4SEric Sandeen int recurse) 13021da177e4SLinus Torvalds { 13031da177e4SLinus Torvalds struct nameidata old_nd; 13041da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 13051da177e4SLinus Torvalds int err = mount_is_safe(nd); 13061da177e4SLinus Torvalds if (err) 13071da177e4SLinus Torvalds return err; 13081da177e4SLinus Torvalds if (!old_name || !*old_name) 13091da177e4SLinus Torvalds return -EINVAL; 13101da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 13111da177e4SLinus Torvalds if (err) 13121da177e4SLinus Torvalds return err; 13131da177e4SLinus Torvalds 1314390c6843SRam Pai down_write(&namespace_sem); 13151da177e4SLinus Torvalds err = -EINVAL; 13164ac91378SJan Blunck if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) 13179676f0c6SRam Pai goto out; 13189676f0c6SRam Pai 13194ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 1320ccd48bc7SAl Viro goto out; 1321ccd48bc7SAl Viro 13221da177e4SLinus Torvalds err = -ENOMEM; 13231da177e4SLinus Torvalds if (recurse) 13244ac91378SJan Blunck mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); 13251da177e4SLinus Torvalds else 13264ac91378SJan Blunck mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); 13271da177e4SLinus Torvalds 1328ccd48bc7SAl Viro if (!mnt) 1329ccd48bc7SAl Viro goto out; 1330ccd48bc7SAl Viro 13318c3ee42eSAl Viro err = graft_tree(mnt, &nd->path); 13321da177e4SLinus Torvalds if (err) { 133370fbcdf4SRam Pai LIST_HEAD(umount_list); 13341da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1335a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 13361da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 133770fbcdf4SRam Pai release_mounts(&umount_list); 13385b83d2c5SRam Pai } 13391da177e4SLinus Torvalds 1340ccd48bc7SAl Viro out: 1341390c6843SRam Pai up_write(&namespace_sem); 13421d957f9bSJan Blunck path_put(&old_nd.path); 13431da177e4SLinus Torvalds return err; 13441da177e4SLinus Torvalds } 13451da177e4SLinus Torvalds 13462e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 13472e4b7fcdSDave Hansen { 13482e4b7fcdSDave Hansen int error = 0; 13492e4b7fcdSDave Hansen int readonly_request = 0; 13502e4b7fcdSDave Hansen 13512e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 13522e4b7fcdSDave Hansen readonly_request = 1; 13532e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 13542e4b7fcdSDave Hansen return 0; 13552e4b7fcdSDave Hansen 13562e4b7fcdSDave Hansen if (readonly_request) 13572e4b7fcdSDave Hansen error = mnt_make_readonly(mnt); 13582e4b7fcdSDave Hansen else 13592e4b7fcdSDave Hansen __mnt_unmake_readonly(mnt); 13602e4b7fcdSDave Hansen return error; 13612e4b7fcdSDave Hansen } 13622e4b7fcdSDave Hansen 13631da177e4SLinus Torvalds /* 13641da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 13651da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 13661da177e4SLinus Torvalds * on it - tough luck. 13672dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 13681da177e4SLinus Torvalds */ 13692dafe1c4SEric Sandeen static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, 13701da177e4SLinus Torvalds void *data) 13711da177e4SLinus Torvalds { 13721da177e4SLinus Torvalds int err; 13734ac91378SJan Blunck struct super_block *sb = nd->path.mnt->mnt_sb; 13741da177e4SLinus Torvalds 13751da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 13761da177e4SLinus Torvalds return -EPERM; 13771da177e4SLinus Torvalds 13784ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 13791da177e4SLinus Torvalds return -EINVAL; 13801da177e4SLinus Torvalds 13814ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 13821da177e4SLinus Torvalds return -EINVAL; 13831da177e4SLinus Torvalds 13841da177e4SLinus Torvalds down_write(&sb->s_umount); 13852e4b7fcdSDave Hansen if (flags & MS_BIND) 13862e4b7fcdSDave Hansen err = change_mount_flags(nd->path.mnt, flags); 13872e4b7fcdSDave Hansen else 13881da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 13891da177e4SLinus Torvalds if (!err) 13904ac91378SJan Blunck nd->path.mnt->mnt_flags = mnt_flags; 13911da177e4SLinus Torvalds up_write(&sb->s_umount); 13921da177e4SLinus Torvalds if (!err) 13934ac91378SJan Blunck security_sb_post_remount(nd->path.mnt, flags, data); 13941da177e4SLinus Torvalds return err; 13951da177e4SLinus Torvalds } 13961da177e4SLinus Torvalds 13979676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 13989676f0c6SRam Pai { 13999676f0c6SRam Pai struct vfsmount *p; 14009676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 14019676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 14029676f0c6SRam Pai return 1; 14039676f0c6SRam Pai } 14049676f0c6SRam Pai return 0; 14059676f0c6SRam Pai } 14069676f0c6SRam Pai 14072dafe1c4SEric Sandeen /* 14082dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 14092dafe1c4SEric Sandeen */ 14102dafe1c4SEric Sandeen static noinline int do_move_mount(struct nameidata *nd, char *old_name) 14111da177e4SLinus Torvalds { 14121a390689SAl Viro struct nameidata old_nd; 14131a390689SAl Viro struct path parent_path; 14141da177e4SLinus Torvalds struct vfsmount *p; 14151da177e4SLinus Torvalds int err = 0; 14161da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 14171da177e4SLinus Torvalds return -EPERM; 14181da177e4SLinus Torvalds if (!old_name || !*old_name) 14191da177e4SLinus Torvalds return -EINVAL; 14201da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 14211da177e4SLinus Torvalds if (err) 14221da177e4SLinus Torvalds return err; 14231da177e4SLinus Torvalds 1424390c6843SRam Pai down_write(&namespace_sem); 14254ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 14264ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 14271da177e4SLinus Torvalds ; 14281da177e4SLinus Torvalds err = -EINVAL; 14294ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 14301da177e4SLinus Torvalds goto out; 14311da177e4SLinus Torvalds 14321da177e4SLinus Torvalds err = -ENOENT; 14334ac91378SJan Blunck mutex_lock(&nd->path.dentry->d_inode->i_mutex); 14344ac91378SJan Blunck if (IS_DEADDIR(nd->path.dentry->d_inode)) 14351da177e4SLinus Torvalds goto out1; 14361da177e4SLinus Torvalds 14374ac91378SJan Blunck if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) 143821444403SRam Pai goto out1; 14391da177e4SLinus Torvalds 14401da177e4SLinus Torvalds err = -EINVAL; 14414ac91378SJan Blunck if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) 144221444403SRam Pai goto out1; 14431da177e4SLinus Torvalds 14444ac91378SJan Blunck if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) 144521444403SRam Pai goto out1; 14461da177e4SLinus Torvalds 14474ac91378SJan Blunck if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != 14484ac91378SJan Blunck S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) 144921444403SRam Pai goto out1; 145021444403SRam Pai /* 145121444403SRam Pai * Don't move a mount residing in a shared parent. 145221444403SRam Pai */ 14534ac91378SJan Blunck if (old_nd.path.mnt->mnt_parent && 14544ac91378SJan Blunck IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) 145521444403SRam Pai goto out1; 14569676f0c6SRam Pai /* 14579676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 14589676f0c6SRam Pai * mount which is shared. 14599676f0c6SRam Pai */ 14604ac91378SJan Blunck if (IS_MNT_SHARED(nd->path.mnt) && 14614ac91378SJan Blunck tree_contains_unbindable(old_nd.path.mnt)) 14629676f0c6SRam Pai goto out1; 14631da177e4SLinus Torvalds err = -ELOOP; 14644ac91378SJan Blunck for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) 14654ac91378SJan Blunck if (p == old_nd.path.mnt) 146621444403SRam Pai goto out1; 14671da177e4SLinus Torvalds 14681a390689SAl Viro err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); 14694ac91378SJan Blunck if (err) 147021444403SRam Pai goto out1; 14711da177e4SLinus Torvalds 14721da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 14731da177e4SLinus Torvalds * automatically */ 14744ac91378SJan Blunck list_del_init(&old_nd.path.mnt->mnt_expire); 14751da177e4SLinus Torvalds out1: 14764ac91378SJan Blunck mutex_unlock(&nd->path.dentry->d_inode->i_mutex); 14771da177e4SLinus Torvalds out: 1478390c6843SRam Pai up_write(&namespace_sem); 14791da177e4SLinus Torvalds if (!err) 14801a390689SAl Viro path_put(&parent_path); 14811d957f9bSJan Blunck path_put(&old_nd.path); 14821da177e4SLinus Torvalds return err; 14831da177e4SLinus Torvalds } 14841da177e4SLinus Torvalds 14851da177e4SLinus Torvalds /* 14861da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 14871da177e4SLinus Torvalds * namespace's tree 14882dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 14891da177e4SLinus Torvalds */ 14902dafe1c4SEric Sandeen static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, 14911da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 14921da177e4SLinus Torvalds { 14931da177e4SLinus Torvalds struct vfsmount *mnt; 14941da177e4SLinus Torvalds 14951da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 14961da177e4SLinus Torvalds return -EINVAL; 14971da177e4SLinus Torvalds 14981da177e4SLinus Torvalds /* we need capabilities... */ 14991da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 15001da177e4SLinus Torvalds return -EPERM; 15011da177e4SLinus Torvalds 15021da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 15031da177e4SLinus Torvalds if (IS_ERR(mnt)) 15041da177e4SLinus Torvalds return PTR_ERR(mnt); 15051da177e4SLinus Torvalds 15061da177e4SLinus Torvalds return do_add_mount(mnt, nd, mnt_flags, NULL); 15071da177e4SLinus Torvalds } 15081da177e4SLinus Torvalds 15091da177e4SLinus Torvalds /* 15101da177e4SLinus Torvalds * add a mount into a namespace's mount tree 15111da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 15121da177e4SLinus Torvalds */ 15131da177e4SLinus Torvalds int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 15141da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 15151da177e4SLinus Torvalds { 15161da177e4SLinus Torvalds int err; 15171da177e4SLinus Torvalds 1518390c6843SRam Pai down_write(&namespace_sem); 15191da177e4SLinus Torvalds /* Something was mounted here while we slept */ 15204ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 15214ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 15221da177e4SLinus Torvalds ; 15231da177e4SLinus Torvalds err = -EINVAL; 15244ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 15251da177e4SLinus Torvalds goto unlock; 15261da177e4SLinus Torvalds 15271da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 15281da177e4SLinus Torvalds err = -EBUSY; 15294ac91378SJan Blunck if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && 15304ac91378SJan Blunck nd->path.mnt->mnt_root == nd->path.dentry) 15311da177e4SLinus Torvalds goto unlock; 15321da177e4SLinus Torvalds 15331da177e4SLinus Torvalds err = -EINVAL; 15341da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 15351da177e4SLinus Torvalds goto unlock; 15361da177e4SLinus Torvalds 15371da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 15388c3ee42eSAl Viro if ((err = graft_tree(newmnt, &nd->path))) 15395b83d2c5SRam Pai goto unlock; 15401da177e4SLinus Torvalds 15416758f953SAl Viro if (fslist) /* add to the specified expiration list */ 154255e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 15436758f953SAl Viro 1544390c6843SRam Pai up_write(&namespace_sem); 15455b83d2c5SRam Pai return 0; 15461da177e4SLinus Torvalds 15471da177e4SLinus Torvalds unlock: 1548390c6843SRam Pai up_write(&namespace_sem); 15491da177e4SLinus Torvalds mntput(newmnt); 15501da177e4SLinus Torvalds return err; 15511da177e4SLinus Torvalds } 15521da177e4SLinus Torvalds 15531da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 15541da177e4SLinus Torvalds 15555528f911STrond Myklebust /* 15561da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 15571da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 15581da177e4SLinus Torvalds * here 15591da177e4SLinus Torvalds */ 15601da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 15611da177e4SLinus Torvalds { 15621da177e4SLinus Torvalds struct vfsmount *mnt, *next; 15631da177e4SLinus Torvalds LIST_HEAD(graveyard); 1564bcc5c7d2SAl Viro LIST_HEAD(umounts); 15651da177e4SLinus Torvalds 15661da177e4SLinus Torvalds if (list_empty(mounts)) 15671da177e4SLinus Torvalds return; 15681da177e4SLinus Torvalds 1569bcc5c7d2SAl Viro down_write(&namespace_sem); 15701da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 15711da177e4SLinus Torvalds 15721da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 15731da177e4SLinus Torvalds * following criteria: 15741da177e4SLinus Torvalds * - only referenced by its parent vfsmount 15751da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 15761da177e4SLinus Torvalds * cleared by mntput()) 15771da177e4SLinus Torvalds */ 157855e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 15791da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 1580bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 15811da177e4SLinus Torvalds continue; 158255e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 15831da177e4SLinus Torvalds } 1584bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1585bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 1586bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1587bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 1588bcc5c7d2SAl Viro } 15891da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1590bcc5c7d2SAl Viro up_write(&namespace_sem); 1591bcc5c7d2SAl Viro 1592bcc5c7d2SAl Viro release_mounts(&umounts); 15931da177e4SLinus Torvalds } 15941da177e4SLinus Torvalds 15951da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 15961da177e4SLinus Torvalds 15971da177e4SLinus Torvalds /* 15985528f911STrond Myklebust * Ripoff of 'select_parent()' 15995528f911STrond Myklebust * 16005528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 16015528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 16025528f911STrond Myklebust */ 16035528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 16045528f911STrond Myklebust { 16055528f911STrond Myklebust struct vfsmount *this_parent = parent; 16065528f911STrond Myklebust struct list_head *next; 16075528f911STrond Myklebust int found = 0; 16085528f911STrond Myklebust 16095528f911STrond Myklebust repeat: 16105528f911STrond Myklebust next = this_parent->mnt_mounts.next; 16115528f911STrond Myklebust resume: 16125528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 16135528f911STrond Myklebust struct list_head *tmp = next; 16145528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 16155528f911STrond Myklebust 16165528f911STrond Myklebust next = tmp->next; 16175528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 16185528f911STrond Myklebust continue; 16195528f911STrond Myklebust /* 16205528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 16215528f911STrond Myklebust */ 16225528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 16235528f911STrond Myklebust this_parent = mnt; 16245528f911STrond Myklebust goto repeat; 16255528f911STrond Myklebust } 16265528f911STrond Myklebust 16275528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 16285528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 16295528f911STrond Myklebust found++; 16305528f911STrond Myklebust } 16315528f911STrond Myklebust } 16325528f911STrond Myklebust /* 16335528f911STrond Myklebust * All done at this level ... ascend and resume the search 16345528f911STrond Myklebust */ 16355528f911STrond Myklebust if (this_parent != parent) { 16365528f911STrond Myklebust next = this_parent->mnt_child.next; 16375528f911STrond Myklebust this_parent = this_parent->mnt_parent; 16385528f911STrond Myklebust goto resume; 16395528f911STrond Myklebust } 16405528f911STrond Myklebust return found; 16415528f911STrond Myklebust } 16425528f911STrond Myklebust 16435528f911STrond Myklebust /* 16445528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 16455528f911STrond Myklebust * submounts of a specific parent mountpoint 16465528f911STrond Myklebust */ 1647c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 16485528f911STrond Myklebust { 16495528f911STrond Myklebust LIST_HEAD(graveyard); 1650c35038beSAl Viro struct vfsmount *m; 16515528f911STrond Myklebust 16525528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 1653c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 1654bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1655c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 1656bcc5c7d2SAl Viro mnt_expire); 1657bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1658c35038beSAl Viro umount_tree(mnt, 1, umounts); 1659bcc5c7d2SAl Viro } 1660bcc5c7d2SAl Viro } 16615528f911STrond Myklebust } 16625528f911STrond Myklebust 16635528f911STrond Myklebust /* 16641da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 16651da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 16661da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 16671da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 16681da177e4SLinus Torvalds */ 1669b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1670b58fed8bSRam Pai unsigned long n) 16711da177e4SLinus Torvalds { 16721da177e4SLinus Torvalds char *t = to; 16731da177e4SLinus Torvalds const char __user *f = from; 16741da177e4SLinus Torvalds char c; 16751da177e4SLinus Torvalds 16761da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 16771da177e4SLinus Torvalds return n; 16781da177e4SLinus Torvalds 16791da177e4SLinus Torvalds while (n) { 16801da177e4SLinus Torvalds if (__get_user(c, f)) { 16811da177e4SLinus Torvalds memset(t, 0, n); 16821da177e4SLinus Torvalds break; 16831da177e4SLinus Torvalds } 16841da177e4SLinus Torvalds *t++ = c; 16851da177e4SLinus Torvalds f++; 16861da177e4SLinus Torvalds n--; 16871da177e4SLinus Torvalds } 16881da177e4SLinus Torvalds return n; 16891da177e4SLinus Torvalds } 16901da177e4SLinus Torvalds 16911da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 16921da177e4SLinus Torvalds { 16931da177e4SLinus Torvalds int i; 16941da177e4SLinus Torvalds unsigned long page; 16951da177e4SLinus Torvalds unsigned long size; 16961da177e4SLinus Torvalds 16971da177e4SLinus Torvalds *where = 0; 16981da177e4SLinus Torvalds if (!data) 16991da177e4SLinus Torvalds return 0; 17001da177e4SLinus Torvalds 17011da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 17021da177e4SLinus Torvalds return -ENOMEM; 17031da177e4SLinus Torvalds 17041da177e4SLinus Torvalds /* We only care that *some* data at the address the user 17051da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 17061da177e4SLinus Torvalds * the remainder of the page. 17071da177e4SLinus Torvalds */ 17081da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 17091da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 17101da177e4SLinus Torvalds if (size > PAGE_SIZE) 17111da177e4SLinus Torvalds size = PAGE_SIZE; 17121da177e4SLinus Torvalds 17131da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 17141da177e4SLinus Torvalds if (!i) { 17151da177e4SLinus Torvalds free_page(page); 17161da177e4SLinus Torvalds return -EFAULT; 17171da177e4SLinus Torvalds } 17181da177e4SLinus Torvalds if (i != PAGE_SIZE) 17191da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 17201da177e4SLinus Torvalds *where = page; 17211da177e4SLinus Torvalds return 0; 17221da177e4SLinus Torvalds } 17231da177e4SLinus Torvalds 17241da177e4SLinus Torvalds /* 17251da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 17261da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 17271da177e4SLinus Torvalds * 17281da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 17291da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 17301da177e4SLinus Torvalds * information (or be NULL). 17311da177e4SLinus Torvalds * 17321da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 17331da177e4SLinus Torvalds * When the flags word was introduced its top half was required 17341da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 17351da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 17361da177e4SLinus Torvalds * and must be discarded. 17371da177e4SLinus Torvalds */ 17381da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 17391da177e4SLinus Torvalds unsigned long flags, void *data_page) 17401da177e4SLinus Torvalds { 17411da177e4SLinus Torvalds struct nameidata nd; 17421da177e4SLinus Torvalds int retval = 0; 17431da177e4SLinus Torvalds int mnt_flags = 0; 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds /* Discard magic */ 17461da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 17471da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 17481da177e4SLinus Torvalds 17491da177e4SLinus Torvalds /* Basic sanity checks */ 17501da177e4SLinus Torvalds 17511da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 17521da177e4SLinus Torvalds return -EINVAL; 17531da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 17541da177e4SLinus Torvalds return -EINVAL; 17551da177e4SLinus Torvalds 17561da177e4SLinus Torvalds if (data_page) 17571da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 17581da177e4SLinus Torvalds 17591da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 17601da177e4SLinus Torvalds if (flags & MS_NOSUID) 17611da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 17621da177e4SLinus Torvalds if (flags & MS_NODEV) 17631da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 17641da177e4SLinus Torvalds if (flags & MS_NOEXEC) 17651da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 1766fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 1767fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 1768fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 1769fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 177047ae32d6SValerie Henson if (flags & MS_RELATIME) 177147ae32d6SValerie Henson mnt_flags |= MNT_RELATIME; 17722e4b7fcdSDave Hansen if (flags & MS_RDONLY) 17732e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 1774fc33a7bbSChristoph Hellwig 1775fc33a7bbSChristoph Hellwig flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 17768bf9725cSPavel Emelyanov MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); 17771da177e4SLinus Torvalds 17781da177e4SLinus Torvalds /* ... and get the mountpoint */ 17791da177e4SLinus Torvalds retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 17801da177e4SLinus Torvalds if (retval) 17811da177e4SLinus Torvalds return retval; 17821da177e4SLinus Torvalds 1783b5266eb4SAl Viro retval = security_sb_mount(dev_name, &nd.path, 1784b5266eb4SAl Viro type_page, flags, data_page); 17851da177e4SLinus Torvalds if (retval) 17861da177e4SLinus Torvalds goto dput_out; 17871da177e4SLinus Torvalds 17881da177e4SLinus Torvalds if (flags & MS_REMOUNT) 17891da177e4SLinus Torvalds retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 17901da177e4SLinus Torvalds data_page); 17911da177e4SLinus Torvalds else if (flags & MS_BIND) 1792eee391a6SAndrew Morton retval = do_loopback(&nd, dev_name, flags & MS_REC); 17939676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 179407b20889SRam Pai retval = do_change_type(&nd, flags); 17951da177e4SLinus Torvalds else if (flags & MS_MOVE) 17961da177e4SLinus Torvalds retval = do_move_mount(&nd, dev_name); 17971da177e4SLinus Torvalds else 17981da177e4SLinus Torvalds retval = do_new_mount(&nd, type_page, flags, mnt_flags, 17991da177e4SLinus Torvalds dev_name, data_page); 18001da177e4SLinus Torvalds dput_out: 18011d957f9bSJan Blunck path_put(&nd.path); 18021da177e4SLinus Torvalds return retval; 18031da177e4SLinus Torvalds } 18041da177e4SLinus Torvalds 1805741a2951SJANAK DESAI /* 1806741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 1807741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 1808741a2951SJANAK DESAI */ 1809e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 18106b3286edSKirill Korotaev struct fs_struct *fs) 18111da177e4SLinus Torvalds { 18126b3286edSKirill Korotaev struct mnt_namespace *new_ns; 18131da177e4SLinus Torvalds struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 18141da177e4SLinus Torvalds struct vfsmount *p, *q; 18151da177e4SLinus Torvalds 18166b3286edSKirill Korotaev new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 18171da177e4SLinus Torvalds if (!new_ns) 1818467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM); 18191da177e4SLinus Torvalds 18201da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 18211da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 18225addc5ddSAl Viro init_waitqueue_head(&new_ns->poll); 18235addc5ddSAl Viro new_ns->event = 0; 18241da177e4SLinus Torvalds 1825390c6843SRam Pai down_write(&namespace_sem); 18261da177e4SLinus Torvalds /* First pass: copy the tree topology */ 18276b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 18289676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 18291da177e4SLinus Torvalds if (!new_ns->root) { 1830390c6843SRam Pai up_write(&namespace_sem); 18311da177e4SLinus Torvalds kfree(new_ns); 1832467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM);; 18331da177e4SLinus Torvalds } 18341da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 18351da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 18361da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 18371da177e4SLinus Torvalds 18381da177e4SLinus Torvalds /* 18391da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 18401da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 18411da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 18421da177e4SLinus Torvalds */ 18436b3286edSKirill Korotaev p = mnt_ns->root; 18441da177e4SLinus Torvalds q = new_ns->root; 18451da177e4SLinus Torvalds while (p) { 18466b3286edSKirill Korotaev q->mnt_ns = new_ns; 18471da177e4SLinus Torvalds if (fs) { 18486ac08c39SJan Blunck if (p == fs->root.mnt) { 18491da177e4SLinus Torvalds rootmnt = p; 18506ac08c39SJan Blunck fs->root.mnt = mntget(q); 18511da177e4SLinus Torvalds } 18526ac08c39SJan Blunck if (p == fs->pwd.mnt) { 18531da177e4SLinus Torvalds pwdmnt = p; 18546ac08c39SJan Blunck fs->pwd.mnt = mntget(q); 18551da177e4SLinus Torvalds } 18566ac08c39SJan Blunck if (p == fs->altroot.mnt) { 18571da177e4SLinus Torvalds altrootmnt = p; 18586ac08c39SJan Blunck fs->altroot.mnt = mntget(q); 18591da177e4SLinus Torvalds } 18601da177e4SLinus Torvalds } 18616b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 18621da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 18631da177e4SLinus Torvalds } 1864390c6843SRam Pai up_write(&namespace_sem); 18651da177e4SLinus Torvalds 18661da177e4SLinus Torvalds if (rootmnt) 18671da177e4SLinus Torvalds mntput(rootmnt); 18681da177e4SLinus Torvalds if (pwdmnt) 18691da177e4SLinus Torvalds mntput(pwdmnt); 18701da177e4SLinus Torvalds if (altrootmnt) 18711da177e4SLinus Torvalds mntput(altrootmnt); 18721da177e4SLinus Torvalds 1873741a2951SJANAK DESAI return new_ns; 1874741a2951SJANAK DESAI } 1875741a2951SJANAK DESAI 1876213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 1877e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 1878741a2951SJANAK DESAI { 18796b3286edSKirill Korotaev struct mnt_namespace *new_ns; 1880741a2951SJANAK DESAI 1881e3222c4eSBadari Pulavarty BUG_ON(!ns); 18826b3286edSKirill Korotaev get_mnt_ns(ns); 1883741a2951SJANAK DESAI 1884741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 1885e3222c4eSBadari Pulavarty return ns; 1886741a2951SJANAK DESAI 1887e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 1888741a2951SJANAK DESAI 18896b3286edSKirill Korotaev put_mnt_ns(ns); 1890e3222c4eSBadari Pulavarty return new_ns; 18911da177e4SLinus Torvalds } 18921da177e4SLinus Torvalds 18931da177e4SLinus Torvalds asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 18941da177e4SLinus Torvalds char __user * type, unsigned long flags, 18951da177e4SLinus Torvalds void __user * data) 18961da177e4SLinus Torvalds { 18971da177e4SLinus Torvalds int retval; 18981da177e4SLinus Torvalds unsigned long data_page; 18991da177e4SLinus Torvalds unsigned long type_page; 19001da177e4SLinus Torvalds unsigned long dev_page; 19011da177e4SLinus Torvalds char *dir_page; 19021da177e4SLinus Torvalds 19031da177e4SLinus Torvalds retval = copy_mount_options(type, &type_page); 19041da177e4SLinus Torvalds if (retval < 0) 19051da177e4SLinus Torvalds return retval; 19061da177e4SLinus Torvalds 19071da177e4SLinus Torvalds dir_page = getname(dir_name); 19081da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 19091da177e4SLinus Torvalds if (IS_ERR(dir_page)) 19101da177e4SLinus Torvalds goto out1; 19111da177e4SLinus Torvalds 19121da177e4SLinus Torvalds retval = copy_mount_options(dev_name, &dev_page); 19131da177e4SLinus Torvalds if (retval < 0) 19141da177e4SLinus Torvalds goto out2; 19151da177e4SLinus Torvalds 19161da177e4SLinus Torvalds retval = copy_mount_options(data, &data_page); 19171da177e4SLinus Torvalds if (retval < 0) 19181da177e4SLinus Torvalds goto out3; 19191da177e4SLinus Torvalds 19201da177e4SLinus Torvalds lock_kernel(); 19211da177e4SLinus Torvalds retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 19221da177e4SLinus Torvalds flags, (void *)data_page); 19231da177e4SLinus Torvalds unlock_kernel(); 19241da177e4SLinus Torvalds free_page(data_page); 19251da177e4SLinus Torvalds 19261da177e4SLinus Torvalds out3: 19271da177e4SLinus Torvalds free_page(dev_page); 19281da177e4SLinus Torvalds out2: 19291da177e4SLinus Torvalds putname(dir_page); 19301da177e4SLinus Torvalds out1: 19311da177e4SLinus Torvalds free_page(type_page); 19321da177e4SLinus Torvalds return retval; 19331da177e4SLinus Torvalds } 19341da177e4SLinus Torvalds 19351da177e4SLinus Torvalds /* 19361da177e4SLinus Torvalds * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 19371da177e4SLinus Torvalds * It can block. Requires the big lock held. 19381da177e4SLinus Torvalds */ 1939ac748a09SJan Blunck void set_fs_root(struct fs_struct *fs, struct path *path) 19401da177e4SLinus Torvalds { 19416ac08c39SJan Blunck struct path old_root; 19426ac08c39SJan Blunck 19431da177e4SLinus Torvalds write_lock(&fs->lock); 19441da177e4SLinus Torvalds old_root = fs->root; 1945ac748a09SJan Blunck fs->root = *path; 1946ac748a09SJan Blunck path_get(path); 19471da177e4SLinus Torvalds write_unlock(&fs->lock); 19486ac08c39SJan Blunck if (old_root.dentry) 19496ac08c39SJan Blunck path_put(&old_root); 19501da177e4SLinus Torvalds } 19511da177e4SLinus Torvalds 19521da177e4SLinus Torvalds /* 19531da177e4SLinus Torvalds * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 19541da177e4SLinus Torvalds * It can block. Requires the big lock held. 19551da177e4SLinus Torvalds */ 1956ac748a09SJan Blunck void set_fs_pwd(struct fs_struct *fs, struct path *path) 19571da177e4SLinus Torvalds { 19586ac08c39SJan Blunck struct path old_pwd; 19591da177e4SLinus Torvalds 19601da177e4SLinus Torvalds write_lock(&fs->lock); 19611da177e4SLinus Torvalds old_pwd = fs->pwd; 1962ac748a09SJan Blunck fs->pwd = *path; 1963ac748a09SJan Blunck path_get(path); 19641da177e4SLinus Torvalds write_unlock(&fs->lock); 19651da177e4SLinus Torvalds 19666ac08c39SJan Blunck if (old_pwd.dentry) 19676ac08c39SJan Blunck path_put(&old_pwd); 19681da177e4SLinus Torvalds } 19691da177e4SLinus Torvalds 19701a390689SAl Viro static void chroot_fs_refs(struct path *old_root, struct path *new_root) 19711da177e4SLinus Torvalds { 19721da177e4SLinus Torvalds struct task_struct *g, *p; 19731da177e4SLinus Torvalds struct fs_struct *fs; 19741da177e4SLinus Torvalds 19751da177e4SLinus Torvalds read_lock(&tasklist_lock); 19761da177e4SLinus Torvalds do_each_thread(g, p) { 19771da177e4SLinus Torvalds task_lock(p); 19781da177e4SLinus Torvalds fs = p->fs; 19791da177e4SLinus Torvalds if (fs) { 19801da177e4SLinus Torvalds atomic_inc(&fs->count); 19811da177e4SLinus Torvalds task_unlock(p); 19821a390689SAl Viro if (fs->root.dentry == old_root->dentry 19831a390689SAl Viro && fs->root.mnt == old_root->mnt) 19841a390689SAl Viro set_fs_root(fs, new_root); 19851a390689SAl Viro if (fs->pwd.dentry == old_root->dentry 19861a390689SAl Viro && fs->pwd.mnt == old_root->mnt) 19871a390689SAl Viro set_fs_pwd(fs, new_root); 19881da177e4SLinus Torvalds put_fs_struct(fs); 19891da177e4SLinus Torvalds } else 19901da177e4SLinus Torvalds task_unlock(p); 19911da177e4SLinus Torvalds } while_each_thread(g, p); 19921da177e4SLinus Torvalds read_unlock(&tasklist_lock); 19931da177e4SLinus Torvalds } 19941da177e4SLinus Torvalds 19951da177e4SLinus Torvalds /* 19961da177e4SLinus Torvalds * pivot_root Semantics: 19971da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 19981da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 19991da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 20001da177e4SLinus Torvalds * 20011da177e4SLinus Torvalds * Restrictions: 20021da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 20031da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 20041da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 20051da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 20061da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 20071da177e4SLinus Torvalds * 20084a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 20094a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 20104a0d11faSNeil Brown * in this situation. 20114a0d11faSNeil Brown * 20121da177e4SLinus Torvalds * Notes: 20131da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 20141da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 20151da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 20161da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 20171da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 20181da177e4SLinus Torvalds * first. 20191da177e4SLinus Torvalds */ 2020b58fed8bSRam Pai asmlinkage long sys_pivot_root(const char __user * new_root, 2021b58fed8bSRam Pai const char __user * put_old) 20221da177e4SLinus Torvalds { 20231da177e4SLinus Torvalds struct vfsmount *tmp; 20248c3ee42eSAl Viro struct nameidata new_nd, old_nd; 20258c3ee42eSAl Viro struct path parent_path, root_parent, root; 20261da177e4SLinus Torvalds int error; 20271da177e4SLinus Torvalds 20281da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 20291da177e4SLinus Torvalds return -EPERM; 20301da177e4SLinus Torvalds 2031b58fed8bSRam Pai error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, 2032b58fed8bSRam Pai &new_nd); 20331da177e4SLinus Torvalds if (error) 20341da177e4SLinus Torvalds goto out0; 20351da177e4SLinus Torvalds error = -EINVAL; 20364ac91378SJan Blunck if (!check_mnt(new_nd.path.mnt)) 20371da177e4SLinus Torvalds goto out1; 20381da177e4SLinus Torvalds 20391da177e4SLinus Torvalds error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); 20401da177e4SLinus Torvalds if (error) 20411da177e4SLinus Torvalds goto out1; 20421da177e4SLinus Torvalds 2043b5266eb4SAl Viro error = security_sb_pivotroot(&old_nd.path, &new_nd.path); 20441da177e4SLinus Torvalds if (error) { 20451d957f9bSJan Blunck path_put(&old_nd.path); 20461da177e4SLinus Torvalds goto out1; 20471da177e4SLinus Torvalds } 20481da177e4SLinus Torvalds 20491da177e4SLinus Torvalds read_lock(¤t->fs->lock); 20508c3ee42eSAl Viro root = current->fs->root; 20516ac08c39SJan Blunck path_get(¤t->fs->root); 20521da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 2053390c6843SRam Pai down_write(&namespace_sem); 20544ac91378SJan Blunck mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); 20551da177e4SLinus Torvalds error = -EINVAL; 20564ac91378SJan Blunck if (IS_MNT_SHARED(old_nd.path.mnt) || 20574ac91378SJan Blunck IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || 20588c3ee42eSAl Viro IS_MNT_SHARED(root.mnt->mnt_parent)) 205921444403SRam Pai goto out2; 20608c3ee42eSAl Viro if (!check_mnt(root.mnt)) 20611da177e4SLinus Torvalds goto out2; 20621da177e4SLinus Torvalds error = -ENOENT; 20634ac91378SJan Blunck if (IS_DEADDIR(new_nd.path.dentry->d_inode)) 20641da177e4SLinus Torvalds goto out2; 20654ac91378SJan Blunck if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) 20661da177e4SLinus Torvalds goto out2; 20674ac91378SJan Blunck if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) 20681da177e4SLinus Torvalds goto out2; 20691da177e4SLinus Torvalds error = -EBUSY; 20708c3ee42eSAl Viro if (new_nd.path.mnt == root.mnt || 20718c3ee42eSAl Viro old_nd.path.mnt == root.mnt) 20721da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 20731da177e4SLinus Torvalds error = -EINVAL; 20748c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 20751da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 20768c3ee42eSAl Viro if (root.mnt->mnt_parent == root.mnt) 20770bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 20784ac91378SJan Blunck if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) 20791da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 20804ac91378SJan Blunck if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) 20810bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 20824ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 20834ac91378SJan Blunck tmp = old_nd.path.mnt; 20841da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 20854ac91378SJan Blunck if (tmp != new_nd.path.mnt) { 20861da177e4SLinus Torvalds for (;;) { 20871da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 20881da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 20894ac91378SJan Blunck if (tmp->mnt_parent == new_nd.path.mnt) 20901da177e4SLinus Torvalds break; 20911da177e4SLinus Torvalds tmp = tmp->mnt_parent; 20921da177e4SLinus Torvalds } 20934ac91378SJan Blunck if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) 20941da177e4SLinus Torvalds goto out3; 20954ac91378SJan Blunck } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) 20961da177e4SLinus Torvalds goto out3; 20971a390689SAl Viro detach_mnt(new_nd.path.mnt, &parent_path); 20988c3ee42eSAl Viro detach_mnt(root.mnt, &root_parent); 20994ac91378SJan Blunck /* mount old root on put_old */ 21008c3ee42eSAl Viro attach_mnt(root.mnt, &old_nd.path); 21014ac91378SJan Blunck /* mount new_root on / */ 21024ac91378SJan Blunck attach_mnt(new_nd.path.mnt, &root_parent); 21036b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 21041da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 21058c3ee42eSAl Viro chroot_fs_refs(&root, &new_nd.path); 21068c3ee42eSAl Viro security_sb_post_pivotroot(&root, &new_nd.path); 21071da177e4SLinus Torvalds error = 0; 21081a390689SAl Viro path_put(&root_parent); 21091a390689SAl Viro path_put(&parent_path); 21101da177e4SLinus Torvalds out2: 21114ac91378SJan Blunck mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); 2112390c6843SRam Pai up_write(&namespace_sem); 21138c3ee42eSAl Viro path_put(&root); 21141d957f9bSJan Blunck path_put(&old_nd.path); 21151da177e4SLinus Torvalds out1: 21161d957f9bSJan Blunck path_put(&new_nd.path); 21171da177e4SLinus Torvalds out0: 21181da177e4SLinus Torvalds return error; 21191da177e4SLinus Torvalds out3: 21201da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 21211da177e4SLinus Torvalds goto out2; 21221da177e4SLinus Torvalds } 21231da177e4SLinus Torvalds 21241da177e4SLinus Torvalds static void __init init_mount_tree(void) 21251da177e4SLinus Torvalds { 21261da177e4SLinus Torvalds struct vfsmount *mnt; 21276b3286edSKirill Korotaev struct mnt_namespace *ns; 2128ac748a09SJan Blunck struct path root; 21291da177e4SLinus Torvalds 21301da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 21311da177e4SLinus Torvalds if (IS_ERR(mnt)) 21321da177e4SLinus Torvalds panic("Can't create rootfs"); 21336b3286edSKirill Korotaev ns = kmalloc(sizeof(*ns), GFP_KERNEL); 21346b3286edSKirill Korotaev if (!ns) 21351da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 21366b3286edSKirill Korotaev atomic_set(&ns->count, 1); 21376b3286edSKirill Korotaev INIT_LIST_HEAD(&ns->list); 21386b3286edSKirill Korotaev init_waitqueue_head(&ns->poll); 21396b3286edSKirill Korotaev ns->event = 0; 21406b3286edSKirill Korotaev list_add(&mnt->mnt_list, &ns->list); 21416b3286edSKirill Korotaev ns->root = mnt; 21426b3286edSKirill Korotaev mnt->mnt_ns = ns; 21431da177e4SLinus Torvalds 21446b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 21456b3286edSKirill Korotaev get_mnt_ns(ns); 21461da177e4SLinus Torvalds 2147ac748a09SJan Blunck root.mnt = ns->root; 2148ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2149ac748a09SJan Blunck 2150ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2151ac748a09SJan Blunck set_fs_root(current->fs, &root); 21521da177e4SLinus Torvalds } 21531da177e4SLinus Torvalds 215474bf17cfSDenis Cheng void __init mnt_init(void) 21551da177e4SLinus Torvalds { 215613f14b4dSEric Dumazet unsigned u; 215715a67dd8SRandy Dunlap int err; 21581da177e4SLinus Torvalds 2159390c6843SRam Pai init_rwsem(&namespace_sem); 2160390c6843SRam Pai 21611da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 216220c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 21631da177e4SLinus Torvalds 2164b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 21651da177e4SLinus Torvalds 21661da177e4SLinus Torvalds if (!mount_hashtable) 21671da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 21681da177e4SLinus Torvalds 216913f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 21701da177e4SLinus Torvalds 217113f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 217213f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 21731da177e4SLinus Torvalds 217415a67dd8SRandy Dunlap err = sysfs_init(); 217515a67dd8SRandy Dunlap if (err) 217615a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 217715a67dd8SRandy Dunlap __FUNCTION__, err); 217800d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 217900d26666SGreg Kroah-Hartman if (!fs_kobj) 218000d26666SGreg Kroah-Hartman printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); 21811da177e4SLinus Torvalds init_rootfs(); 21821da177e4SLinus Torvalds init_mount_tree(); 21831da177e4SLinus Torvalds } 21841da177e4SLinus Torvalds 21856b3286edSKirill Korotaev void __put_mnt_ns(struct mnt_namespace *ns) 21861da177e4SLinus Torvalds { 21876b3286edSKirill Korotaev struct vfsmount *root = ns->root; 218870fbcdf4SRam Pai LIST_HEAD(umount_list); 21896b3286edSKirill Korotaev ns->root = NULL; 21901ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 2191390c6843SRam Pai down_write(&namespace_sem); 21921da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 2193a05964f3SRam Pai umount_tree(root, 0, &umount_list); 21941da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 2195390c6843SRam Pai up_write(&namespace_sem); 219670fbcdf4SRam Pai release_mounts(&umount_list); 21976b3286edSKirill Korotaev kfree(ns); 21981da177e4SLinus Torvalds } 2199