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> 20*3d733633SDave 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> 301da177e4SLinus Torvalds #include <asm/uaccess.h> 311da177e4SLinus Torvalds #include <asm/unistd.h> 3207b20889SRam Pai #include "pnode.h" 33948730b0SAdrian Bunk #include "internal.h" 341da177e4SLinus Torvalds 3513f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 3613f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 3713f14b4dSEric Dumazet 381da177e4SLinus Torvalds /* spinlock for vfsmount related operations, inplace of dcache_lock */ 391da177e4SLinus Torvalds __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock); 401da177e4SLinus Torvalds 415addc5ddSAl Viro static int event; 425addc5ddSAl Viro 43fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 44e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 45390c6843SRam Pai static struct rw_semaphore namespace_sem; 461da177e4SLinus Torvalds 47f87fd4c2SMiklos Szeredi /* /sys/fs */ 4800d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 4900d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 50f87fd4c2SMiklos Szeredi 511da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 521da177e4SLinus Torvalds { 531da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 541da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 5513f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 5613f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 571da177e4SLinus Torvalds } 581da177e4SLinus Torvalds 59*3d733633SDave Hansen #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 60*3d733633SDave Hansen 611da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 621da177e4SLinus Torvalds { 63c3762229SRobert P. J. Day struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 641da177e4SLinus Torvalds if (mnt) { 651da177e4SLinus Torvalds atomic_set(&mnt->mnt_count, 1); 661da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 671da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 681da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 691da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 7055e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 7103e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 72a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 73a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 74*3d733633SDave Hansen atomic_set(&mnt->__mnt_writers, 0); 751da177e4SLinus Torvalds if (name) { 761da177e4SLinus Torvalds int size = strlen(name) + 1; 771da177e4SLinus Torvalds char *newname = kmalloc(size, GFP_KERNEL); 781da177e4SLinus Torvalds if (newname) { 791da177e4SLinus Torvalds memcpy(newname, name, size); 801da177e4SLinus Torvalds mnt->mnt_devname = newname; 811da177e4SLinus Torvalds } 821da177e4SLinus Torvalds } 831da177e4SLinus Torvalds } 841da177e4SLinus Torvalds return mnt; 851da177e4SLinus Torvalds } 861da177e4SLinus Torvalds 878366025eSDave Hansen /* 888366025eSDave Hansen * Most r/o checks on a fs are for operations that take 898366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 908366025eSDave Hansen * We must keep track of when those operations start 918366025eSDave Hansen * (for permission checks) and when they end, so that 928366025eSDave Hansen * we can determine when writes are able to occur to 938366025eSDave Hansen * a filesystem. 948366025eSDave Hansen */ 95*3d733633SDave Hansen /* 96*3d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 97*3d733633SDave Hansen * @mnt: the mount to check for its write status 98*3d733633SDave Hansen * 99*3d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 100*3d733633SDave Hansen * It does not guarantee that the filesystem will stay 101*3d733633SDave Hansen * r/w, just that it is right *now*. This can not and 102*3d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 103*3d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 104*3d733633SDave Hansen * r/w. 105*3d733633SDave Hansen */ 106*3d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 107*3d733633SDave Hansen { 108*3d733633SDave Hansen return (mnt->mnt_sb->s_flags & MS_RDONLY); 109*3d733633SDave Hansen } 110*3d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 111*3d733633SDave Hansen 112*3d733633SDave Hansen struct mnt_writer { 113*3d733633SDave Hansen /* 114*3d733633SDave Hansen * If holding multiple instances of this lock, they 115*3d733633SDave Hansen * must be ordered by cpu number. 116*3d733633SDave Hansen */ 117*3d733633SDave Hansen spinlock_t lock; 118*3d733633SDave Hansen struct lock_class_key lock_class; /* compiles out with !lockdep */ 119*3d733633SDave Hansen unsigned long count; 120*3d733633SDave Hansen struct vfsmount *mnt; 121*3d733633SDave Hansen } ____cacheline_aligned_in_smp; 122*3d733633SDave Hansen static DEFINE_PER_CPU(struct mnt_writer, mnt_writers); 123*3d733633SDave Hansen 124*3d733633SDave Hansen static int __init init_mnt_writers(void) 125*3d733633SDave Hansen { 126*3d733633SDave Hansen int cpu; 127*3d733633SDave Hansen for_each_possible_cpu(cpu) { 128*3d733633SDave Hansen struct mnt_writer *writer = &per_cpu(mnt_writers, cpu); 129*3d733633SDave Hansen spin_lock_init(&writer->lock); 130*3d733633SDave Hansen lockdep_set_class(&writer->lock, &writer->lock_class); 131*3d733633SDave Hansen writer->count = 0; 132*3d733633SDave Hansen } 133*3d733633SDave Hansen return 0; 134*3d733633SDave Hansen } 135*3d733633SDave Hansen fs_initcall(init_mnt_writers); 136*3d733633SDave Hansen 137*3d733633SDave Hansen static void unlock_mnt_writers(void) 138*3d733633SDave Hansen { 139*3d733633SDave Hansen int cpu; 140*3d733633SDave Hansen struct mnt_writer *cpu_writer; 141*3d733633SDave Hansen 142*3d733633SDave Hansen for_each_possible_cpu(cpu) { 143*3d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 144*3d733633SDave Hansen spin_unlock(&cpu_writer->lock); 145*3d733633SDave Hansen } 146*3d733633SDave Hansen } 147*3d733633SDave Hansen 148*3d733633SDave Hansen static inline void __clear_mnt_count(struct mnt_writer *cpu_writer) 149*3d733633SDave Hansen { 150*3d733633SDave Hansen if (!cpu_writer->mnt) 151*3d733633SDave Hansen return; 152*3d733633SDave Hansen /* 153*3d733633SDave Hansen * This is in case anyone ever leaves an invalid, 154*3d733633SDave Hansen * old ->mnt and a count of 0. 155*3d733633SDave Hansen */ 156*3d733633SDave Hansen if (!cpu_writer->count) 157*3d733633SDave Hansen return; 158*3d733633SDave Hansen atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers); 159*3d733633SDave Hansen cpu_writer->count = 0; 160*3d733633SDave Hansen } 161*3d733633SDave Hansen /* 162*3d733633SDave Hansen * must hold cpu_writer->lock 163*3d733633SDave Hansen */ 164*3d733633SDave Hansen static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer, 165*3d733633SDave Hansen struct vfsmount *mnt) 166*3d733633SDave Hansen { 167*3d733633SDave Hansen if (cpu_writer->mnt == mnt) 168*3d733633SDave Hansen return; 169*3d733633SDave Hansen __clear_mnt_count(cpu_writer); 170*3d733633SDave Hansen cpu_writer->mnt = mnt; 171*3d733633SDave Hansen } 172*3d733633SDave Hansen 173*3d733633SDave Hansen /* 174*3d733633SDave Hansen * Most r/o checks on a fs are for operations that take 175*3d733633SDave Hansen * discrete amounts of time, like a write() or unlink(). 176*3d733633SDave Hansen * We must keep track of when those operations start 177*3d733633SDave Hansen * (for permission checks) and when they end, so that 178*3d733633SDave Hansen * we can determine when writes are able to occur to 179*3d733633SDave Hansen * a filesystem. 180*3d733633SDave Hansen */ 1818366025eSDave Hansen /** 1828366025eSDave Hansen * mnt_want_write - get write access to a mount 1838366025eSDave Hansen * @mnt: the mount on which to take a write 1848366025eSDave Hansen * 1858366025eSDave Hansen * This tells the low-level filesystem that a write is 1868366025eSDave Hansen * about to be performed to it, and makes sure that 1878366025eSDave Hansen * writes are allowed before returning success. When 1888366025eSDave Hansen * the write operation is finished, mnt_drop_write() 1898366025eSDave Hansen * must be called. This is effectively a refcount. 1908366025eSDave Hansen */ 1918366025eSDave Hansen int mnt_want_write(struct vfsmount *mnt) 1928366025eSDave Hansen { 193*3d733633SDave Hansen int ret = 0; 194*3d733633SDave Hansen struct mnt_writer *cpu_writer; 195*3d733633SDave Hansen 196*3d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 197*3d733633SDave Hansen spin_lock(&cpu_writer->lock); 198*3d733633SDave Hansen if (__mnt_is_readonly(mnt)) { 199*3d733633SDave Hansen ret = -EROFS; 200*3d733633SDave Hansen goto out; 201*3d733633SDave Hansen } 202*3d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 203*3d733633SDave Hansen cpu_writer->count++; 204*3d733633SDave Hansen out: 205*3d733633SDave Hansen spin_unlock(&cpu_writer->lock); 206*3d733633SDave Hansen put_cpu_var(mnt_writers); 207*3d733633SDave Hansen return ret; 2088366025eSDave Hansen } 2098366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 2108366025eSDave Hansen 211*3d733633SDave Hansen static void lock_mnt_writers(void) 212*3d733633SDave Hansen { 213*3d733633SDave Hansen int cpu; 214*3d733633SDave Hansen struct mnt_writer *cpu_writer; 215*3d733633SDave Hansen 216*3d733633SDave Hansen for_each_possible_cpu(cpu) { 217*3d733633SDave Hansen cpu_writer = &per_cpu(mnt_writers, cpu); 218*3d733633SDave Hansen spin_lock(&cpu_writer->lock); 219*3d733633SDave Hansen __clear_mnt_count(cpu_writer); 220*3d733633SDave Hansen cpu_writer->mnt = NULL; 221*3d733633SDave Hansen } 222*3d733633SDave Hansen } 223*3d733633SDave Hansen 224*3d733633SDave Hansen /* 225*3d733633SDave Hansen * These per-cpu write counts are not guaranteed to have 226*3d733633SDave Hansen * matched increments and decrements on any given cpu. 227*3d733633SDave Hansen * A file open()ed for write on one cpu and close()d on 228*3d733633SDave Hansen * another cpu will imbalance this count. Make sure it 229*3d733633SDave Hansen * does not get too far out of whack. 230*3d733633SDave Hansen */ 231*3d733633SDave Hansen static void handle_write_count_underflow(struct vfsmount *mnt) 232*3d733633SDave Hansen { 233*3d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) >= 234*3d733633SDave Hansen MNT_WRITER_UNDERFLOW_LIMIT) 235*3d733633SDave Hansen return; 236*3d733633SDave Hansen /* 237*3d733633SDave Hansen * It isn't necessary to hold all of the locks 238*3d733633SDave Hansen * at the same time, but doing it this way makes 239*3d733633SDave Hansen * us share a lot more code. 240*3d733633SDave Hansen */ 241*3d733633SDave Hansen lock_mnt_writers(); 242*3d733633SDave Hansen /* 243*3d733633SDave Hansen * vfsmount_lock is for mnt_flags. 244*3d733633SDave Hansen */ 245*3d733633SDave Hansen spin_lock(&vfsmount_lock); 246*3d733633SDave Hansen /* 247*3d733633SDave Hansen * If coalescing the per-cpu writer counts did not 248*3d733633SDave Hansen * get us back to a positive writer count, we have 249*3d733633SDave Hansen * a bug. 250*3d733633SDave Hansen */ 251*3d733633SDave Hansen if ((atomic_read(&mnt->__mnt_writers) < 0) && 252*3d733633SDave Hansen !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) { 253*3d733633SDave Hansen printk(KERN_DEBUG "leak detected on mount(%p) writers " 254*3d733633SDave Hansen "count: %d\n", 255*3d733633SDave Hansen mnt, atomic_read(&mnt->__mnt_writers)); 256*3d733633SDave Hansen WARN_ON(1); 257*3d733633SDave Hansen /* use the flag to keep the dmesg spam down */ 258*3d733633SDave Hansen mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT; 259*3d733633SDave Hansen } 260*3d733633SDave Hansen spin_unlock(&vfsmount_lock); 261*3d733633SDave Hansen unlock_mnt_writers(); 262*3d733633SDave Hansen } 263*3d733633SDave Hansen 2648366025eSDave Hansen /** 2658366025eSDave Hansen * mnt_drop_write - give up write access to a mount 2668366025eSDave Hansen * @mnt: the mount on which to give up write access 2678366025eSDave Hansen * 2688366025eSDave Hansen * Tells the low-level filesystem that we are done 2698366025eSDave Hansen * performing writes to it. Must be matched with 2708366025eSDave Hansen * mnt_want_write() call above. 2718366025eSDave Hansen */ 2728366025eSDave Hansen void mnt_drop_write(struct vfsmount *mnt) 2738366025eSDave Hansen { 274*3d733633SDave Hansen int must_check_underflow = 0; 275*3d733633SDave Hansen struct mnt_writer *cpu_writer; 276*3d733633SDave Hansen 277*3d733633SDave Hansen cpu_writer = &get_cpu_var(mnt_writers); 278*3d733633SDave Hansen spin_lock(&cpu_writer->lock); 279*3d733633SDave Hansen 280*3d733633SDave Hansen use_cpu_writer_for_mount(cpu_writer, mnt); 281*3d733633SDave Hansen if (cpu_writer->count > 0) { 282*3d733633SDave Hansen cpu_writer->count--; 283*3d733633SDave Hansen } else { 284*3d733633SDave Hansen must_check_underflow = 1; 285*3d733633SDave Hansen atomic_dec(&mnt->__mnt_writers); 286*3d733633SDave Hansen } 287*3d733633SDave Hansen 288*3d733633SDave Hansen spin_unlock(&cpu_writer->lock); 289*3d733633SDave Hansen /* 290*3d733633SDave Hansen * Logically, we could call this each time, 291*3d733633SDave Hansen * but the __mnt_writers cacheline tends to 292*3d733633SDave Hansen * be cold, and makes this expensive. 293*3d733633SDave Hansen */ 294*3d733633SDave Hansen if (must_check_underflow) 295*3d733633SDave Hansen handle_write_count_underflow(mnt); 296*3d733633SDave Hansen /* 297*3d733633SDave Hansen * This could be done right after the spinlock 298*3d733633SDave Hansen * is taken because the spinlock keeps us on 299*3d733633SDave Hansen * the cpu, and disables preemption. However, 300*3d733633SDave Hansen * putting it here bounds the amount that 301*3d733633SDave Hansen * __mnt_writers can underflow. Without it, 302*3d733633SDave Hansen * we could theoretically wrap __mnt_writers. 303*3d733633SDave Hansen */ 304*3d733633SDave Hansen put_cpu_var(mnt_writers); 3058366025eSDave Hansen } 3068366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 3078366025eSDave Hansen 308*3d733633SDave Hansen int mnt_make_readonly(struct vfsmount *mnt) 3098366025eSDave Hansen { 310*3d733633SDave Hansen int ret = 0; 311*3d733633SDave Hansen 312*3d733633SDave Hansen lock_mnt_writers(); 313*3d733633SDave Hansen /* 314*3d733633SDave Hansen * With all the locks held, this value is stable 315*3d733633SDave Hansen */ 316*3d733633SDave Hansen if (atomic_read(&mnt->__mnt_writers) > 0) { 317*3d733633SDave Hansen ret = -EBUSY; 318*3d733633SDave Hansen goto out; 3198366025eSDave Hansen } 320*3d733633SDave Hansen /* 321*3d733633SDave Hansen * actually set mount's r/o flag here to make 322*3d733633SDave Hansen * __mnt_is_readonly() true, which keeps anyone 323*3d733633SDave Hansen * from doing a successful mnt_want_write(). 324*3d733633SDave Hansen */ 325*3d733633SDave Hansen out: 326*3d733633SDave Hansen unlock_mnt_writers(); 327*3d733633SDave Hansen return ret; 328*3d733633SDave Hansen } 3298366025eSDave Hansen 330454e2398SDavid Howells int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 331454e2398SDavid Howells { 332454e2398SDavid Howells mnt->mnt_sb = sb; 333454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 334454e2398SDavid Howells return 0; 335454e2398SDavid Howells } 336454e2398SDavid Howells 337454e2398SDavid Howells EXPORT_SYMBOL(simple_set_mnt); 338454e2398SDavid Howells 3391da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 3401da177e4SLinus Torvalds { 3411da177e4SLinus Torvalds kfree(mnt->mnt_devname); 3421da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 3431da177e4SLinus Torvalds } 3441da177e4SLinus Torvalds 3451da177e4SLinus Torvalds /* 346a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 347a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 3481da177e4SLinus Torvalds */ 349a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 350a05964f3SRam Pai int dir) 3511da177e4SLinus Torvalds { 3521da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 3531da177e4SLinus Torvalds struct list_head *tmp = head; 3541da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 3551da177e4SLinus Torvalds 3561da177e4SLinus Torvalds for (;;) { 357a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 3581da177e4SLinus Torvalds p = NULL; 3591da177e4SLinus Torvalds if (tmp == head) 3601da177e4SLinus Torvalds break; 3611da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 3621da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 363a05964f3SRam Pai found = p; 3641da177e4SLinus Torvalds break; 3651da177e4SLinus Torvalds } 3661da177e4SLinus Torvalds } 3671da177e4SLinus Torvalds return found; 3681da177e4SLinus Torvalds } 3691da177e4SLinus Torvalds 370a05964f3SRam Pai /* 371a05964f3SRam Pai * lookup_mnt increments the ref count before returning 372a05964f3SRam Pai * the vfsmount struct. 373a05964f3SRam Pai */ 374a05964f3SRam Pai struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry) 375a05964f3SRam Pai { 376a05964f3SRam Pai struct vfsmount *child_mnt; 377a05964f3SRam Pai spin_lock(&vfsmount_lock); 378a05964f3SRam Pai if ((child_mnt = __lookup_mnt(mnt, dentry, 1))) 379a05964f3SRam Pai mntget(child_mnt); 380a05964f3SRam Pai spin_unlock(&vfsmount_lock); 381a05964f3SRam Pai return child_mnt; 382a05964f3SRam Pai } 383a05964f3SRam Pai 3841da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 3851da177e4SLinus Torvalds { 3866b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 3871da177e4SLinus Torvalds } 3881da177e4SLinus Torvalds 3896b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 3905addc5ddSAl Viro { 3915addc5ddSAl Viro if (ns) { 3925addc5ddSAl Viro ns->event = ++event; 3935addc5ddSAl Viro wake_up_interruptible(&ns->poll); 3945addc5ddSAl Viro } 3955addc5ddSAl Viro } 3965addc5ddSAl Viro 3976b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 3985addc5ddSAl Viro { 3995addc5ddSAl Viro if (ns && ns->event != event) { 4005addc5ddSAl Viro ns->event = event; 4015addc5ddSAl Viro wake_up_interruptible(&ns->poll); 4025addc5ddSAl Viro } 4035addc5ddSAl Viro } 4045addc5ddSAl Viro 4051a390689SAl Viro static void detach_mnt(struct vfsmount *mnt, struct path *old_path) 4061da177e4SLinus Torvalds { 4071a390689SAl Viro old_path->dentry = mnt->mnt_mountpoint; 4081a390689SAl Viro old_path->mnt = mnt->mnt_parent; 4091da177e4SLinus Torvalds mnt->mnt_parent = mnt; 4101da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 4111da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 4121da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 4131a390689SAl Viro old_path->dentry->d_mounted--; 4141da177e4SLinus Torvalds } 4151da177e4SLinus Torvalds 416b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 417b90fa9aeSRam Pai struct vfsmount *child_mnt) 418b90fa9aeSRam Pai { 419b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 420b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 421b90fa9aeSRam Pai dentry->d_mounted++; 422b90fa9aeSRam Pai } 423b90fa9aeSRam Pai 4241a390689SAl Viro static void attach_mnt(struct vfsmount *mnt, struct path *path) 4251da177e4SLinus Torvalds { 4261a390689SAl Viro mnt_set_mountpoint(path->mnt, path->dentry, mnt); 427b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 4281a390689SAl Viro hash(path->mnt, path->dentry)); 4291a390689SAl Viro list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); 430b90fa9aeSRam Pai } 431b90fa9aeSRam Pai 432b90fa9aeSRam Pai /* 433b90fa9aeSRam Pai * the caller must hold vfsmount_lock 434b90fa9aeSRam Pai */ 435b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 436b90fa9aeSRam Pai { 437b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 438b90fa9aeSRam Pai struct vfsmount *m; 439b90fa9aeSRam Pai LIST_HEAD(head); 4406b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 441b90fa9aeSRam Pai 442b90fa9aeSRam Pai BUG_ON(parent == mnt); 443b90fa9aeSRam Pai 444b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 445b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 4466b3286edSKirill Korotaev m->mnt_ns = n; 447b90fa9aeSRam Pai list_splice(&head, n->list.prev); 448b90fa9aeSRam Pai 449b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 450b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 451b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 4526b3286edSKirill Korotaev touch_mnt_namespace(n); 4531da177e4SLinus Torvalds } 4541da177e4SLinus Torvalds 4551da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 4561da177e4SLinus Torvalds { 4571da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 4581da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 4591da177e4SLinus Torvalds while (1) { 4601da177e4SLinus Torvalds if (p == root) 4611da177e4SLinus Torvalds return NULL; 4621da177e4SLinus Torvalds next = p->mnt_child.next; 4631da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 4641da177e4SLinus Torvalds break; 4651da177e4SLinus Torvalds p = p->mnt_parent; 4661da177e4SLinus Torvalds } 4671da177e4SLinus Torvalds } 4681da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 4691da177e4SLinus Torvalds } 4701da177e4SLinus Torvalds 4719676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 4729676f0c6SRam Pai { 4739676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 4749676f0c6SRam Pai while (prev != &p->mnt_mounts) { 4759676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 4769676f0c6SRam Pai prev = p->mnt_mounts.prev; 4779676f0c6SRam Pai } 4789676f0c6SRam Pai return p; 4799676f0c6SRam Pai } 4809676f0c6SRam Pai 48136341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 48236341f64SRam Pai int flag) 4831da177e4SLinus Torvalds { 4841da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 4851da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 4861da177e4SLinus Torvalds 4871da177e4SLinus Torvalds if (mnt) { 4881da177e4SLinus Torvalds mnt->mnt_flags = old->mnt_flags; 4891da177e4SLinus Torvalds atomic_inc(&sb->s_active); 4901da177e4SLinus Torvalds mnt->mnt_sb = sb; 4911da177e4SLinus Torvalds mnt->mnt_root = dget(root); 4921da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 4931da177e4SLinus Torvalds mnt->mnt_parent = mnt; 494b90fa9aeSRam Pai 4955afe0022SRam Pai if (flag & CL_SLAVE) { 4965afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 4975afe0022SRam Pai mnt->mnt_master = old; 4985afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 4998aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 500b90fa9aeSRam Pai if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old)) 501b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 5025afe0022SRam Pai if (IS_MNT_SLAVE(old)) 5035afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 5045afe0022SRam Pai mnt->mnt_master = old->mnt_master; 5055afe0022SRam Pai } 506b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 507b90fa9aeSRam Pai set_mnt_shared(mnt); 5081da177e4SLinus Torvalds 5091da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 5101da177e4SLinus Torvalds * as the original if that was on one */ 51136341f64SRam Pai if (flag & CL_EXPIRE) { 51255e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 51355e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 5141da177e4SLinus Torvalds } 51536341f64SRam Pai } 5161da177e4SLinus Torvalds return mnt; 5171da177e4SLinus Torvalds } 5181da177e4SLinus Torvalds 5197b7b1aceSAl Viro static inline void __mntput(struct vfsmount *mnt) 5201da177e4SLinus Torvalds { 521*3d733633SDave Hansen int cpu; 5221da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 523*3d733633SDave Hansen /* 524*3d733633SDave Hansen * We don't have to hold all of the locks at the 525*3d733633SDave Hansen * same time here because we know that we're the 526*3d733633SDave Hansen * last reference to mnt and that no new writers 527*3d733633SDave Hansen * can come in. 528*3d733633SDave Hansen */ 529*3d733633SDave Hansen for_each_possible_cpu(cpu) { 530*3d733633SDave Hansen struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu); 531*3d733633SDave Hansen if (cpu_writer->mnt != mnt) 532*3d733633SDave Hansen continue; 533*3d733633SDave Hansen spin_lock(&cpu_writer->lock); 534*3d733633SDave Hansen atomic_add(cpu_writer->count, &mnt->__mnt_writers); 535*3d733633SDave Hansen cpu_writer->count = 0; 536*3d733633SDave Hansen /* 537*3d733633SDave Hansen * Might as well do this so that no one 538*3d733633SDave Hansen * ever sees the pointer and expects 539*3d733633SDave Hansen * it to be valid. 540*3d733633SDave Hansen */ 541*3d733633SDave Hansen cpu_writer->mnt = NULL; 542*3d733633SDave Hansen spin_unlock(&cpu_writer->lock); 543*3d733633SDave Hansen } 544*3d733633SDave Hansen /* 545*3d733633SDave Hansen * This probably indicates that somebody messed 546*3d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 547*3d733633SDave Hansen * happens, the filesystem was probably unable 548*3d733633SDave Hansen * to make r/w->r/o transitions. 549*3d733633SDave Hansen */ 550*3d733633SDave Hansen WARN_ON(atomic_read(&mnt->__mnt_writers)); 5511da177e4SLinus Torvalds dput(mnt->mnt_root); 5521da177e4SLinus Torvalds free_vfsmnt(mnt); 5531da177e4SLinus Torvalds deactivate_super(sb); 5541da177e4SLinus Torvalds } 5551da177e4SLinus Torvalds 5567b7b1aceSAl Viro void mntput_no_expire(struct vfsmount *mnt) 5577b7b1aceSAl Viro { 5587b7b1aceSAl Viro repeat: 5597b7b1aceSAl Viro if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) { 5607b7b1aceSAl Viro if (likely(!mnt->mnt_pinned)) { 5617b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 5627b7b1aceSAl Viro __mntput(mnt); 5637b7b1aceSAl Viro return; 5647b7b1aceSAl Viro } 5657b7b1aceSAl Viro atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count); 5667b7b1aceSAl Viro mnt->mnt_pinned = 0; 5677b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 5687b7b1aceSAl Viro acct_auto_close_mnt(mnt); 5697b7b1aceSAl Viro security_sb_umount_close(mnt); 5707b7b1aceSAl Viro goto repeat; 5717b7b1aceSAl Viro } 5727b7b1aceSAl Viro } 5737b7b1aceSAl Viro 5747b7b1aceSAl Viro EXPORT_SYMBOL(mntput_no_expire); 5757b7b1aceSAl Viro 5767b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 5777b7b1aceSAl Viro { 5787b7b1aceSAl Viro spin_lock(&vfsmount_lock); 5797b7b1aceSAl Viro mnt->mnt_pinned++; 5807b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 5817b7b1aceSAl Viro } 5827b7b1aceSAl Viro 5837b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 5847b7b1aceSAl Viro 5857b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 5867b7b1aceSAl Viro { 5877b7b1aceSAl Viro spin_lock(&vfsmount_lock); 5887b7b1aceSAl Viro if (mnt->mnt_pinned) { 5897b7b1aceSAl Viro atomic_inc(&mnt->mnt_count); 5907b7b1aceSAl Viro mnt->mnt_pinned--; 5917b7b1aceSAl Viro } 5927b7b1aceSAl Viro spin_unlock(&vfsmount_lock); 5937b7b1aceSAl Viro } 5947b7b1aceSAl Viro 5957b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 5961da177e4SLinus Torvalds 597b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 598b3b304a2SMiklos Szeredi { 599b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 600b3b304a2SMiklos Szeredi } 601b3b304a2SMiklos Szeredi 602b3b304a2SMiklos Szeredi /* 603b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 604b3b304a2SMiklos Szeredi * implement more complex mount option showing. 605b3b304a2SMiklos Szeredi * 606b3b304a2SMiklos Szeredi * See also save_mount_options(). 607b3b304a2SMiklos Szeredi */ 608b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 609b3b304a2SMiklos Szeredi { 610b3b304a2SMiklos Szeredi const char *options = mnt->mnt_sb->s_options; 611b3b304a2SMiklos Szeredi 612b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 613b3b304a2SMiklos Szeredi seq_putc(m, ','); 614b3b304a2SMiklos Szeredi mangle(m, options); 615b3b304a2SMiklos Szeredi } 616b3b304a2SMiklos Szeredi 617b3b304a2SMiklos Szeredi return 0; 618b3b304a2SMiklos Szeredi } 619b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 620b3b304a2SMiklos Szeredi 621b3b304a2SMiklos Szeredi /* 622b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 623b3b304a2SMiklos Szeredi * called from the fill_super() callback. 624b3b304a2SMiklos Szeredi * 625b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 626b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 627b3b304a2SMiklos Szeredi * remount fails. 628b3b304a2SMiklos Szeredi * 629b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 630b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 631b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 632b3b304a2SMiklos Szeredi * any more. 633b3b304a2SMiklos Szeredi */ 634b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 635b3b304a2SMiklos Szeredi { 636b3b304a2SMiklos Szeredi kfree(sb->s_options); 637b3b304a2SMiklos Szeredi sb->s_options = kstrdup(options, GFP_KERNEL); 638b3b304a2SMiklos Szeredi } 639b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 640b3b304a2SMiklos Szeredi 6411da177e4SLinus Torvalds /* iterator */ 6421da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 6431da177e4SLinus Torvalds { 6446b3286edSKirill Korotaev struct mnt_namespace *n = m->private; 6451da177e4SLinus Torvalds 646390c6843SRam Pai down_read(&namespace_sem); 647b0765fb8SPavel Emelianov return seq_list_start(&n->list, *pos); 6481da177e4SLinus Torvalds } 6491da177e4SLinus Torvalds 6501da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 6511da177e4SLinus Torvalds { 6526b3286edSKirill Korotaev struct mnt_namespace *n = m->private; 653b0765fb8SPavel Emelianov 654b0765fb8SPavel Emelianov return seq_list_next(v, &n->list, pos); 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds 6571da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 6581da177e4SLinus Torvalds { 659390c6843SRam Pai up_read(&namespace_sem); 6601da177e4SLinus Torvalds } 6611da177e4SLinus Torvalds 6621da177e4SLinus Torvalds static int show_vfsmnt(struct seq_file *m, void *v) 6631da177e4SLinus Torvalds { 664b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 6651da177e4SLinus Torvalds int err = 0; 6661da177e4SLinus Torvalds static struct proc_fs_info { 6671da177e4SLinus Torvalds int flag; 6681da177e4SLinus Torvalds char *str; 6691da177e4SLinus Torvalds } fs_info[] = { 6701da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 6711da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 6721da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 6731da177e4SLinus Torvalds { 0, NULL } 6741da177e4SLinus Torvalds }; 6751da177e4SLinus Torvalds static struct proc_fs_info mnt_info[] = { 6761da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 6771da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 6781da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 679fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 680fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 68147ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 6821da177e4SLinus Torvalds { 0, NULL } 6831da177e4SLinus Torvalds }; 6841da177e4SLinus Torvalds struct proc_fs_info *fs_infop; 685c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 6861da177e4SLinus Torvalds 6871da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 6881da177e4SLinus Torvalds seq_putc(m, ' '); 689c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 6901da177e4SLinus Torvalds seq_putc(m, ' '); 6911da177e4SLinus Torvalds mangle(m, mnt->mnt_sb->s_type->name); 69279c0b2dfSMiklos Szeredi if (mnt->mnt_sb->s_subtype && mnt->mnt_sb->s_subtype[0]) { 69379c0b2dfSMiklos Szeredi seq_putc(m, '.'); 69479c0b2dfSMiklos Szeredi mangle(m, mnt->mnt_sb->s_subtype); 69579c0b2dfSMiklos Szeredi } 6961da177e4SLinus Torvalds seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw"); 6971da177e4SLinus Torvalds for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 6981da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & fs_infop->flag) 6991da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 7001da177e4SLinus Torvalds } 7011da177e4SLinus Torvalds for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 7021da177e4SLinus Torvalds if (mnt->mnt_flags & fs_infop->flag) 7031da177e4SLinus Torvalds seq_puts(m, fs_infop->str); 7041da177e4SLinus Torvalds } 7051da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 7061da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 7071da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 7081da177e4SLinus Torvalds return err; 7091da177e4SLinus Torvalds } 7101da177e4SLinus Torvalds 7111da177e4SLinus Torvalds struct seq_operations mounts_op = { 7121da177e4SLinus Torvalds .start = m_start, 7131da177e4SLinus Torvalds .next = m_next, 7141da177e4SLinus Torvalds .stop = m_stop, 7151da177e4SLinus Torvalds .show = show_vfsmnt 7161da177e4SLinus Torvalds }; 7171da177e4SLinus Torvalds 718b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 719b4629fe2SChuck Lever { 720b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 721c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 722b4629fe2SChuck Lever int err = 0; 723b4629fe2SChuck Lever 724b4629fe2SChuck Lever /* device */ 725b4629fe2SChuck Lever if (mnt->mnt_devname) { 726b4629fe2SChuck Lever seq_puts(m, "device "); 727b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 728b4629fe2SChuck Lever } else 729b4629fe2SChuck Lever seq_puts(m, "no device"); 730b4629fe2SChuck Lever 731b4629fe2SChuck Lever /* mount point */ 732b4629fe2SChuck Lever seq_puts(m, " mounted on "); 733c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 734b4629fe2SChuck Lever seq_putc(m, ' '); 735b4629fe2SChuck Lever 736b4629fe2SChuck Lever /* file system type */ 737b4629fe2SChuck Lever seq_puts(m, "with fstype "); 738b4629fe2SChuck Lever mangle(m, mnt->mnt_sb->s_type->name); 739b4629fe2SChuck Lever 740b4629fe2SChuck Lever /* optional statistics */ 741b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 742b4629fe2SChuck Lever seq_putc(m, ' '); 743b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 744b4629fe2SChuck Lever } 745b4629fe2SChuck Lever 746b4629fe2SChuck Lever seq_putc(m, '\n'); 747b4629fe2SChuck Lever return err; 748b4629fe2SChuck Lever } 749b4629fe2SChuck Lever 750b4629fe2SChuck Lever struct seq_operations mountstats_op = { 751b4629fe2SChuck Lever .start = m_start, 752b4629fe2SChuck Lever .next = m_next, 753b4629fe2SChuck Lever .stop = m_stop, 754b4629fe2SChuck Lever .show = show_vfsstat, 755b4629fe2SChuck Lever }; 756b4629fe2SChuck Lever 7571da177e4SLinus Torvalds /** 7581da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 7591da177e4SLinus Torvalds * @mnt: root of mount tree 7601da177e4SLinus Torvalds * 7611da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 7621da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 7631da177e4SLinus Torvalds * busy. 7641da177e4SLinus Torvalds */ 7651da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 7661da177e4SLinus Torvalds { 76736341f64SRam Pai int actual_refs = 0; 76836341f64SRam Pai int minimum_refs = 0; 76936341f64SRam Pai struct vfsmount *p; 7701da177e4SLinus Torvalds 7711da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 77236341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 7731da177e4SLinus Torvalds actual_refs += atomic_read(&p->mnt_count); 7741da177e4SLinus Torvalds minimum_refs += 2; 7751da177e4SLinus Torvalds } 7761da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 7771da177e4SLinus Torvalds 7781da177e4SLinus Torvalds if (actual_refs > minimum_refs) 7791da177e4SLinus Torvalds return 0; 780e3474a8eSIan Kent 781e3474a8eSIan Kent return 1; 7821da177e4SLinus Torvalds } 7831da177e4SLinus Torvalds 7841da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 7851da177e4SLinus Torvalds 7861da177e4SLinus Torvalds /** 7871da177e4SLinus Torvalds * may_umount - check if a mount point is busy 7881da177e4SLinus Torvalds * @mnt: root of mount 7891da177e4SLinus Torvalds * 7901da177e4SLinus Torvalds * This is called to check if a mount point has any 7911da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 7921da177e4SLinus Torvalds * mount has sub mounts this will return busy 7931da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 7941da177e4SLinus Torvalds * 7951da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 7961da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 7971da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 7981da177e4SLinus Torvalds */ 7991da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 8001da177e4SLinus Torvalds { 801e3474a8eSIan Kent int ret = 1; 802a05964f3SRam Pai spin_lock(&vfsmount_lock); 803a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 804e3474a8eSIan Kent ret = 0; 805a05964f3SRam Pai spin_unlock(&vfsmount_lock); 806a05964f3SRam Pai return ret; 8071da177e4SLinus Torvalds } 8081da177e4SLinus Torvalds 8091da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 8101da177e4SLinus Torvalds 811b90fa9aeSRam Pai void release_mounts(struct list_head *head) 8121da177e4SLinus Torvalds { 81370fbcdf4SRam Pai struct vfsmount *mnt; 81470fbcdf4SRam Pai while (!list_empty(head)) { 815b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 81670fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 81770fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 81870fbcdf4SRam Pai struct dentry *dentry; 81970fbcdf4SRam Pai struct vfsmount *m; 82070fbcdf4SRam Pai spin_lock(&vfsmount_lock); 82170fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 82270fbcdf4SRam Pai m = mnt->mnt_parent; 82370fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 82470fbcdf4SRam Pai mnt->mnt_parent = mnt; 8257c4b93d8SAl Viro m->mnt_ghosts--; 8261da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 82770fbcdf4SRam Pai dput(dentry); 82870fbcdf4SRam Pai mntput(m); 8291da177e4SLinus Torvalds } 8301da177e4SLinus Torvalds mntput(mnt); 83170fbcdf4SRam Pai } 83270fbcdf4SRam Pai } 83370fbcdf4SRam Pai 834a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 83570fbcdf4SRam Pai { 83670fbcdf4SRam Pai struct vfsmount *p; 83770fbcdf4SRam Pai 8381bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 8391bfba4e8SAkinobu Mita list_move(&p->mnt_hash, kill); 84070fbcdf4SRam Pai 841a05964f3SRam Pai if (propagate) 842a05964f3SRam Pai propagate_umount(kill); 843a05964f3SRam Pai 84470fbcdf4SRam Pai list_for_each_entry(p, kill, mnt_hash) { 84570fbcdf4SRam Pai list_del_init(&p->mnt_expire); 84670fbcdf4SRam Pai list_del_init(&p->mnt_list); 8476b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 8486b3286edSKirill Korotaev p->mnt_ns = NULL; 84970fbcdf4SRam Pai list_del_init(&p->mnt_child); 8507c4b93d8SAl Viro if (p->mnt_parent != p) { 8517c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 852f30ac319SAl Viro p->mnt_mountpoint->d_mounted--; 8537c4b93d8SAl Viro } 854a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 8551da177e4SLinus Torvalds } 8561da177e4SLinus Torvalds } 8571da177e4SLinus Torvalds 858c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 859c35038beSAl Viro 8601da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 8611da177e4SLinus Torvalds { 8621da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 8631da177e4SLinus Torvalds int retval; 86470fbcdf4SRam Pai LIST_HEAD(umount_list); 8651da177e4SLinus Torvalds 8661da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 8671da177e4SLinus Torvalds if (retval) 8681da177e4SLinus Torvalds return retval; 8691da177e4SLinus Torvalds 8701da177e4SLinus Torvalds /* 8711da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 8721da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 8731da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 8741da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 8751da177e4SLinus Torvalds */ 8761da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 8776ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 8781da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 8791da177e4SLinus Torvalds return -EINVAL; 8801da177e4SLinus Torvalds 8811da177e4SLinus Torvalds if (atomic_read(&mnt->mnt_count) != 2) 8821da177e4SLinus Torvalds return -EBUSY; 8831da177e4SLinus Torvalds 8841da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 8851da177e4SLinus Torvalds return -EAGAIN; 8861da177e4SLinus Torvalds } 8871da177e4SLinus Torvalds 8881da177e4SLinus Torvalds /* 8891da177e4SLinus Torvalds * If we may have to abort operations to get out of this 8901da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 8911da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 8921da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 8931da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 8941da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 8951da177e4SLinus Torvalds * about for the moment. 8961da177e4SLinus Torvalds */ 8971da177e4SLinus Torvalds 8981da177e4SLinus Torvalds lock_kernel(); 8998b512d9aSTrond Myklebust if (sb->s_op->umount_begin) 9008b512d9aSTrond Myklebust sb->s_op->umount_begin(mnt, flags); 9011da177e4SLinus Torvalds unlock_kernel(); 9021da177e4SLinus Torvalds 9031da177e4SLinus Torvalds /* 9041da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 9051da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 9061da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 9071da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 9081da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 9091da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 9101da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 9111da177e4SLinus Torvalds */ 9126ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 9131da177e4SLinus Torvalds /* 9141da177e4SLinus Torvalds * Special case for "unmounting" root ... 9151da177e4SLinus Torvalds * we just try to remount it readonly. 9161da177e4SLinus Torvalds */ 9171da177e4SLinus Torvalds down_write(&sb->s_umount); 9181da177e4SLinus Torvalds if (!(sb->s_flags & MS_RDONLY)) { 9191da177e4SLinus Torvalds lock_kernel(); 9201da177e4SLinus Torvalds DQUOT_OFF(sb); 9211da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 9221da177e4SLinus Torvalds unlock_kernel(); 9231da177e4SLinus Torvalds } 9241da177e4SLinus Torvalds up_write(&sb->s_umount); 9251da177e4SLinus Torvalds return retval; 9261da177e4SLinus Torvalds } 9271da177e4SLinus Torvalds 928390c6843SRam Pai down_write(&namespace_sem); 9291da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 9305addc5ddSAl Viro event++; 9311da177e4SLinus Torvalds 932c35038beSAl Viro if (!(flags & MNT_DETACH)) 933c35038beSAl Viro shrink_submounts(mnt, &umount_list); 934c35038beSAl Viro 9351da177e4SLinus Torvalds retval = -EBUSY; 936a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 9371da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 938a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 9391da177e4SLinus Torvalds retval = 0; 9401da177e4SLinus Torvalds } 9411da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 9421da177e4SLinus Torvalds if (retval) 9431da177e4SLinus Torvalds security_sb_umount_busy(mnt); 944390c6843SRam Pai up_write(&namespace_sem); 94570fbcdf4SRam Pai release_mounts(&umount_list); 9461da177e4SLinus Torvalds return retval; 9471da177e4SLinus Torvalds } 9481da177e4SLinus Torvalds 9491da177e4SLinus Torvalds /* 9501da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 9511da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 9521da177e4SLinus Torvalds * 9531da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 9541da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 9551da177e4SLinus Torvalds */ 9561da177e4SLinus Torvalds 9571da177e4SLinus Torvalds asmlinkage long sys_umount(char __user * name, int flags) 9581da177e4SLinus Torvalds { 9591da177e4SLinus Torvalds struct nameidata nd; 9601da177e4SLinus Torvalds int retval; 9611da177e4SLinus Torvalds 9621da177e4SLinus Torvalds retval = __user_walk(name, LOOKUP_FOLLOW, &nd); 9631da177e4SLinus Torvalds if (retval) 9641da177e4SLinus Torvalds goto out; 9651da177e4SLinus Torvalds retval = -EINVAL; 9664ac91378SJan Blunck if (nd.path.dentry != nd.path.mnt->mnt_root) 9671da177e4SLinus Torvalds goto dput_and_out; 9684ac91378SJan Blunck if (!check_mnt(nd.path.mnt)) 9691da177e4SLinus Torvalds goto dput_and_out; 9701da177e4SLinus Torvalds 9711da177e4SLinus Torvalds retval = -EPERM; 9721da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 9731da177e4SLinus Torvalds goto dput_and_out; 9741da177e4SLinus Torvalds 9754ac91378SJan Blunck retval = do_umount(nd.path.mnt, flags); 9761da177e4SLinus Torvalds dput_and_out: 977429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 9784ac91378SJan Blunck dput(nd.path.dentry); 9794ac91378SJan Blunck mntput_no_expire(nd.path.mnt); 9801da177e4SLinus Torvalds out: 9811da177e4SLinus Torvalds return retval; 9821da177e4SLinus Torvalds } 9831da177e4SLinus Torvalds 9841da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 9851da177e4SLinus Torvalds 9861da177e4SLinus Torvalds /* 9871da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 9881da177e4SLinus Torvalds */ 9891da177e4SLinus Torvalds asmlinkage long sys_oldumount(char __user * name) 9901da177e4SLinus Torvalds { 9911da177e4SLinus Torvalds return sys_umount(name, 0); 9921da177e4SLinus Torvalds } 9931da177e4SLinus Torvalds 9941da177e4SLinus Torvalds #endif 9951da177e4SLinus Torvalds 9961da177e4SLinus Torvalds static int mount_is_safe(struct nameidata *nd) 9971da177e4SLinus Torvalds { 9981da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 9991da177e4SLinus Torvalds return 0; 10001da177e4SLinus Torvalds return -EPERM; 10011da177e4SLinus Torvalds #ifdef notyet 10024ac91378SJan Blunck if (S_ISLNK(nd->path.dentry->d_inode->i_mode)) 10031da177e4SLinus Torvalds return -EPERM; 10044ac91378SJan Blunck if (nd->path.dentry->d_inode->i_mode & S_ISVTX) { 10054ac91378SJan Blunck if (current->uid != nd->path.dentry->d_inode->i_uid) 10061da177e4SLinus Torvalds return -EPERM; 10071da177e4SLinus Torvalds } 1008e4543eddSChristoph Hellwig if (vfs_permission(nd, MAY_WRITE)) 10091da177e4SLinus Torvalds return -EPERM; 10101da177e4SLinus Torvalds return 0; 10111da177e4SLinus Torvalds #endif 10121da177e4SLinus Torvalds } 10131da177e4SLinus Torvalds 1014b58fed8bSRam Pai static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry) 10151da177e4SLinus Torvalds { 10161da177e4SLinus Torvalds while (1) { 10171da177e4SLinus Torvalds if (d == dentry) 10181da177e4SLinus Torvalds return 1; 10191da177e4SLinus Torvalds if (d == NULL || d == d->d_parent) 10201da177e4SLinus Torvalds return 0; 10211da177e4SLinus Torvalds d = d->d_parent; 10221da177e4SLinus Torvalds } 10231da177e4SLinus Torvalds } 10241da177e4SLinus Torvalds 1025b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 102636341f64SRam Pai int flag) 10271da177e4SLinus Torvalds { 10281da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 10291a390689SAl Viro struct path path; 10301da177e4SLinus Torvalds 10319676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 10329676f0c6SRam Pai return NULL; 10339676f0c6SRam Pai 103436341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 10351da177e4SLinus Torvalds if (!q) 10361da177e4SLinus Torvalds goto Enomem; 10371da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 10381da177e4SLinus Torvalds 10391da177e4SLinus Torvalds p = mnt; 1040fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 10411da177e4SLinus Torvalds if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry)) 10421da177e4SLinus Torvalds continue; 10431da177e4SLinus Torvalds 10441da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 10459676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 10469676f0c6SRam Pai s = skip_mnt_tree(s); 10479676f0c6SRam Pai continue; 10489676f0c6SRam Pai } 10491da177e4SLinus Torvalds while (p != s->mnt_parent) { 10501da177e4SLinus Torvalds p = p->mnt_parent; 10511da177e4SLinus Torvalds q = q->mnt_parent; 10521da177e4SLinus Torvalds } 10531da177e4SLinus Torvalds p = s; 10541a390689SAl Viro path.mnt = q; 10551a390689SAl Viro path.dentry = p->mnt_mountpoint; 105636341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 10571da177e4SLinus Torvalds if (!q) 10581da177e4SLinus Torvalds goto Enomem; 10591da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 10601da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 10611a390689SAl Viro attach_mnt(q, &path); 10621da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 10631da177e4SLinus Torvalds } 10641da177e4SLinus Torvalds } 10651da177e4SLinus Torvalds return res; 10661da177e4SLinus Torvalds Enomem: 10671da177e4SLinus Torvalds if (res) { 106870fbcdf4SRam Pai LIST_HEAD(umount_list); 10691da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1070a05964f3SRam Pai umount_tree(res, 0, &umount_list); 10711da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 107270fbcdf4SRam Pai release_mounts(&umount_list); 10731da177e4SLinus Torvalds } 10741da177e4SLinus Torvalds return NULL; 10751da177e4SLinus Torvalds } 10761da177e4SLinus Torvalds 10778aec0809SAl Viro struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry) 10788aec0809SAl Viro { 10798aec0809SAl Viro struct vfsmount *tree; 10808aec0809SAl Viro down_read(&namespace_sem); 10818aec0809SAl Viro tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE); 10828aec0809SAl Viro up_read(&namespace_sem); 10838aec0809SAl Viro return tree; 10848aec0809SAl Viro } 10858aec0809SAl Viro 10868aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 10878aec0809SAl Viro { 10888aec0809SAl Viro LIST_HEAD(umount_list); 10898aec0809SAl Viro down_read(&namespace_sem); 10908aec0809SAl Viro spin_lock(&vfsmount_lock); 10918aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 10928aec0809SAl Viro spin_unlock(&vfsmount_lock); 10938aec0809SAl Viro up_read(&namespace_sem); 10948aec0809SAl Viro release_mounts(&umount_list); 10958aec0809SAl Viro } 10968aec0809SAl Viro 1097b90fa9aeSRam Pai /* 1098b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1099b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 110021444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 110121444403SRam Pai * store the parent mount and mountpoint dentry. 110221444403SRam Pai * (done when source_mnt is moved) 1103b90fa9aeSRam Pai * 1104b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1105b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 11069676f0c6SRam Pai * --------------------------------------------------------------------------- 1107b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 11089676f0c6SRam Pai * |************************************************************************** 11099676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 11109676f0c6SRam Pai * | dest | | | | | 11119676f0c6SRam Pai * | | | | | | | 11129676f0c6SRam Pai * | v | | | | | 11139676f0c6SRam Pai * |************************************************************************** 11149676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 11155afe0022SRam Pai * | | | | | | 11169676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 11179676f0c6SRam Pai * *************************************************************************** 1118b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1119b90fa9aeSRam Pai * destination mount. 1120b90fa9aeSRam Pai * 1121b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1122b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1123b90fa9aeSRam Pai * the peer group of the source mount. 1124b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1125b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1126b90fa9aeSRam Pai * mount. 11275afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 11285afe0022SRam Pai * of the destination mount and the cloned mount is made slave 11295afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 11305afe0022SRam Pai * is marked as 'shared and slave'. 11315afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 11325afe0022SRam Pai * source mount. 11335afe0022SRam Pai * 11349676f0c6SRam Pai * --------------------------------------------------------------------------- 113521444403SRam Pai * | MOVE MOUNT OPERATION | 11369676f0c6SRam Pai * |************************************************************************** 11379676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 11389676f0c6SRam Pai * | dest | | | | | 11399676f0c6SRam Pai * | | | | | | | 11409676f0c6SRam Pai * | v | | | | | 11419676f0c6SRam Pai * |************************************************************************** 11429676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 11435afe0022SRam Pai * | | | | | | 11449676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 11459676f0c6SRam Pai * *************************************************************************** 11465afe0022SRam Pai * 11475afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 11485afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 114921444403SRam Pai * (+*) the mount is moved to the destination. 11505afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 11515afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 11525afe0022SRam Pai * the mount is marked as 'shared and slave'. 11535afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1154b90fa9aeSRam Pai * 1155b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1156b90fa9aeSRam Pai * applied to each mount in the tree. 1157b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1158b90fa9aeSRam Pai * in allocations. 1159b90fa9aeSRam Pai */ 1160b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 11611a390689SAl Viro struct path *path, struct path *parent_path) 1162b90fa9aeSRam Pai { 1163b90fa9aeSRam Pai LIST_HEAD(tree_list); 11641a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 11651a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1166b90fa9aeSRam Pai struct vfsmount *child, *p; 1167b90fa9aeSRam Pai 1168b90fa9aeSRam Pai if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list)) 1169b90fa9aeSRam Pai return -EINVAL; 1170b90fa9aeSRam Pai 1171b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1172b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1173b90fa9aeSRam Pai set_mnt_shared(p); 1174b90fa9aeSRam Pai } 1175b90fa9aeSRam Pai 1176b90fa9aeSRam Pai spin_lock(&vfsmount_lock); 11771a390689SAl Viro if (parent_path) { 11781a390689SAl Viro detach_mnt(source_mnt, parent_path); 11791a390689SAl Viro attach_mnt(source_mnt, path); 11806b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 118121444403SRam Pai } else { 1182b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1183b90fa9aeSRam Pai commit_tree(source_mnt); 118421444403SRam Pai } 1185b90fa9aeSRam Pai 1186b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1187b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1188b90fa9aeSRam Pai commit_tree(child); 1189b90fa9aeSRam Pai } 1190b90fa9aeSRam Pai spin_unlock(&vfsmount_lock); 1191b90fa9aeSRam Pai return 0; 1192b90fa9aeSRam Pai } 1193b90fa9aeSRam Pai 11941da177e4SLinus Torvalds static int graft_tree(struct vfsmount *mnt, struct nameidata *nd) 11951da177e4SLinus Torvalds { 11961da177e4SLinus Torvalds int err; 11971da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 11981da177e4SLinus Torvalds return -EINVAL; 11991da177e4SLinus Torvalds 12004ac91378SJan Blunck if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != 12011da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 12021da177e4SLinus Torvalds return -ENOTDIR; 12031da177e4SLinus Torvalds 12041da177e4SLinus Torvalds err = -ENOENT; 12054ac91378SJan Blunck mutex_lock(&nd->path.dentry->d_inode->i_mutex); 12064ac91378SJan Blunck if (IS_DEADDIR(nd->path.dentry->d_inode)) 12071da177e4SLinus Torvalds goto out_unlock; 12081da177e4SLinus Torvalds 12091da177e4SLinus Torvalds err = security_sb_check_sb(mnt, nd); 12101da177e4SLinus Torvalds if (err) 12111da177e4SLinus Torvalds goto out_unlock; 12121da177e4SLinus Torvalds 12131da177e4SLinus Torvalds err = -ENOENT; 12144ac91378SJan Blunck if (IS_ROOT(nd->path.dentry) || !d_unhashed(nd->path.dentry)) 12151a390689SAl Viro err = attach_recursive_mnt(mnt, &nd->path, NULL); 12161da177e4SLinus Torvalds out_unlock: 12174ac91378SJan Blunck mutex_unlock(&nd->path.dentry->d_inode->i_mutex); 12181da177e4SLinus Torvalds if (!err) 12191da177e4SLinus Torvalds security_sb_post_addmount(mnt, nd); 12201da177e4SLinus Torvalds return err; 12211da177e4SLinus Torvalds } 12221da177e4SLinus Torvalds 12231da177e4SLinus Torvalds /* 122407b20889SRam Pai * recursively change the type of the mountpoint. 12252dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 122607b20889SRam Pai */ 12272dafe1c4SEric Sandeen static noinline int do_change_type(struct nameidata *nd, int flag) 122807b20889SRam Pai { 12294ac91378SJan Blunck struct vfsmount *m, *mnt = nd->path.mnt; 123007b20889SRam Pai int recurse = flag & MS_REC; 123107b20889SRam Pai int type = flag & ~MS_REC; 123207b20889SRam Pai 1233ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1234ee6f9582SMiklos Szeredi return -EPERM; 1235ee6f9582SMiklos Szeredi 12364ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 123707b20889SRam Pai return -EINVAL; 123807b20889SRam Pai 123907b20889SRam Pai down_write(&namespace_sem); 124007b20889SRam Pai spin_lock(&vfsmount_lock); 124107b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 124207b20889SRam Pai change_mnt_propagation(m, type); 124307b20889SRam Pai spin_unlock(&vfsmount_lock); 124407b20889SRam Pai up_write(&namespace_sem); 124507b20889SRam Pai return 0; 124607b20889SRam Pai } 124707b20889SRam Pai 124807b20889SRam Pai /* 12491da177e4SLinus Torvalds * do loopback mount. 12502dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 12511da177e4SLinus Torvalds */ 12522dafe1c4SEric Sandeen static noinline int do_loopback(struct nameidata *nd, char *old_name, 12532dafe1c4SEric Sandeen int recurse) 12541da177e4SLinus Torvalds { 12551da177e4SLinus Torvalds struct nameidata old_nd; 12561da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 12571da177e4SLinus Torvalds int err = mount_is_safe(nd); 12581da177e4SLinus Torvalds if (err) 12591da177e4SLinus Torvalds return err; 12601da177e4SLinus Torvalds if (!old_name || !*old_name) 12611da177e4SLinus Torvalds return -EINVAL; 12621da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 12631da177e4SLinus Torvalds if (err) 12641da177e4SLinus Torvalds return err; 12651da177e4SLinus Torvalds 1266390c6843SRam Pai down_write(&namespace_sem); 12671da177e4SLinus Torvalds err = -EINVAL; 12684ac91378SJan Blunck if (IS_MNT_UNBINDABLE(old_nd.path.mnt)) 12699676f0c6SRam Pai goto out; 12709676f0c6SRam Pai 12714ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 1272ccd48bc7SAl Viro goto out; 1273ccd48bc7SAl Viro 12741da177e4SLinus Torvalds err = -ENOMEM; 12751da177e4SLinus Torvalds if (recurse) 12764ac91378SJan Blunck mnt = copy_tree(old_nd.path.mnt, old_nd.path.dentry, 0); 12771da177e4SLinus Torvalds else 12784ac91378SJan Blunck mnt = clone_mnt(old_nd.path.mnt, old_nd.path.dentry, 0); 12791da177e4SLinus Torvalds 1280ccd48bc7SAl Viro if (!mnt) 1281ccd48bc7SAl Viro goto out; 1282ccd48bc7SAl Viro 12831da177e4SLinus Torvalds err = graft_tree(mnt, nd); 12841da177e4SLinus Torvalds if (err) { 128570fbcdf4SRam Pai LIST_HEAD(umount_list); 12861da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 1287a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 12881da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 128970fbcdf4SRam Pai release_mounts(&umount_list); 12905b83d2c5SRam Pai } 12911da177e4SLinus Torvalds 1292ccd48bc7SAl Viro out: 1293390c6843SRam Pai up_write(&namespace_sem); 12941d957f9bSJan Blunck path_put(&old_nd.path); 12951da177e4SLinus Torvalds return err; 12961da177e4SLinus Torvalds } 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds /* 12991da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 13001da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 13011da177e4SLinus Torvalds * on it - tough luck. 13022dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 13031da177e4SLinus Torvalds */ 13042dafe1c4SEric Sandeen static noinline int do_remount(struct nameidata *nd, int flags, int mnt_flags, 13051da177e4SLinus Torvalds void *data) 13061da177e4SLinus Torvalds { 13071da177e4SLinus Torvalds int err; 13084ac91378SJan Blunck struct super_block *sb = nd->path.mnt->mnt_sb; 13091da177e4SLinus Torvalds 13101da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 13111da177e4SLinus Torvalds return -EPERM; 13121da177e4SLinus Torvalds 13134ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 13141da177e4SLinus Torvalds return -EINVAL; 13151da177e4SLinus Torvalds 13164ac91378SJan Blunck if (nd->path.dentry != nd->path.mnt->mnt_root) 13171da177e4SLinus Torvalds return -EINVAL; 13181da177e4SLinus Torvalds 13191da177e4SLinus Torvalds down_write(&sb->s_umount); 13201da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 13211da177e4SLinus Torvalds if (!err) 13224ac91378SJan Blunck nd->path.mnt->mnt_flags = mnt_flags; 13231da177e4SLinus Torvalds up_write(&sb->s_umount); 13241da177e4SLinus Torvalds if (!err) 13254ac91378SJan Blunck security_sb_post_remount(nd->path.mnt, flags, data); 13261da177e4SLinus Torvalds return err; 13271da177e4SLinus Torvalds } 13281da177e4SLinus Torvalds 13299676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 13309676f0c6SRam Pai { 13319676f0c6SRam Pai struct vfsmount *p; 13329676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 13339676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 13349676f0c6SRam Pai return 1; 13359676f0c6SRam Pai } 13369676f0c6SRam Pai return 0; 13379676f0c6SRam Pai } 13389676f0c6SRam Pai 13392dafe1c4SEric Sandeen /* 13402dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 13412dafe1c4SEric Sandeen */ 13422dafe1c4SEric Sandeen static noinline int do_move_mount(struct nameidata *nd, char *old_name) 13431da177e4SLinus Torvalds { 13441a390689SAl Viro struct nameidata old_nd; 13451a390689SAl Viro struct path parent_path; 13461da177e4SLinus Torvalds struct vfsmount *p; 13471da177e4SLinus Torvalds int err = 0; 13481da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 13491da177e4SLinus Torvalds return -EPERM; 13501da177e4SLinus Torvalds if (!old_name || !*old_name) 13511da177e4SLinus Torvalds return -EINVAL; 13521da177e4SLinus Torvalds err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd); 13531da177e4SLinus Torvalds if (err) 13541da177e4SLinus Torvalds return err; 13551da177e4SLinus Torvalds 1356390c6843SRam Pai down_write(&namespace_sem); 13574ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 13584ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 13591da177e4SLinus Torvalds ; 13601da177e4SLinus Torvalds err = -EINVAL; 13614ac91378SJan Blunck if (!check_mnt(nd->path.mnt) || !check_mnt(old_nd.path.mnt)) 13621da177e4SLinus Torvalds goto out; 13631da177e4SLinus Torvalds 13641da177e4SLinus Torvalds err = -ENOENT; 13654ac91378SJan Blunck mutex_lock(&nd->path.dentry->d_inode->i_mutex); 13664ac91378SJan Blunck if (IS_DEADDIR(nd->path.dentry->d_inode)) 13671da177e4SLinus Torvalds goto out1; 13681da177e4SLinus Torvalds 13694ac91378SJan Blunck if (!IS_ROOT(nd->path.dentry) && d_unhashed(nd->path.dentry)) 137021444403SRam Pai goto out1; 13711da177e4SLinus Torvalds 13721da177e4SLinus Torvalds err = -EINVAL; 13734ac91378SJan Blunck if (old_nd.path.dentry != old_nd.path.mnt->mnt_root) 137421444403SRam Pai goto out1; 13751da177e4SLinus Torvalds 13764ac91378SJan Blunck if (old_nd.path.mnt == old_nd.path.mnt->mnt_parent) 137721444403SRam Pai goto out1; 13781da177e4SLinus Torvalds 13794ac91378SJan Blunck if (S_ISDIR(nd->path.dentry->d_inode->i_mode) != 13804ac91378SJan Blunck S_ISDIR(old_nd.path.dentry->d_inode->i_mode)) 138121444403SRam Pai goto out1; 138221444403SRam Pai /* 138321444403SRam Pai * Don't move a mount residing in a shared parent. 138421444403SRam Pai */ 13854ac91378SJan Blunck if (old_nd.path.mnt->mnt_parent && 13864ac91378SJan Blunck IS_MNT_SHARED(old_nd.path.mnt->mnt_parent)) 138721444403SRam Pai goto out1; 13889676f0c6SRam Pai /* 13899676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 13909676f0c6SRam Pai * mount which is shared. 13919676f0c6SRam Pai */ 13924ac91378SJan Blunck if (IS_MNT_SHARED(nd->path.mnt) && 13934ac91378SJan Blunck tree_contains_unbindable(old_nd.path.mnt)) 13949676f0c6SRam Pai goto out1; 13951da177e4SLinus Torvalds err = -ELOOP; 13964ac91378SJan Blunck for (p = nd->path.mnt; p->mnt_parent != p; p = p->mnt_parent) 13974ac91378SJan Blunck if (p == old_nd.path.mnt) 139821444403SRam Pai goto out1; 13991da177e4SLinus Torvalds 14001a390689SAl Viro err = attach_recursive_mnt(old_nd.path.mnt, &nd->path, &parent_path); 14014ac91378SJan Blunck if (err) 140221444403SRam Pai goto out1; 14031da177e4SLinus Torvalds 14041da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 14051da177e4SLinus Torvalds * automatically */ 14064ac91378SJan Blunck list_del_init(&old_nd.path.mnt->mnt_expire); 14071da177e4SLinus Torvalds out1: 14084ac91378SJan Blunck mutex_unlock(&nd->path.dentry->d_inode->i_mutex); 14091da177e4SLinus Torvalds out: 1410390c6843SRam Pai up_write(&namespace_sem); 14111da177e4SLinus Torvalds if (!err) 14121a390689SAl Viro path_put(&parent_path); 14131d957f9bSJan Blunck path_put(&old_nd.path); 14141da177e4SLinus Torvalds return err; 14151da177e4SLinus Torvalds } 14161da177e4SLinus Torvalds 14171da177e4SLinus Torvalds /* 14181da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 14191da177e4SLinus Torvalds * namespace's tree 14202dafe1c4SEric Sandeen * noinline this do_mount helper to save do_mount stack space. 14211da177e4SLinus Torvalds */ 14222dafe1c4SEric Sandeen static noinline int do_new_mount(struct nameidata *nd, char *type, int flags, 14231da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 14241da177e4SLinus Torvalds { 14251da177e4SLinus Torvalds struct vfsmount *mnt; 14261da177e4SLinus Torvalds 14271da177e4SLinus Torvalds if (!type || !memchr(type, 0, PAGE_SIZE)) 14281da177e4SLinus Torvalds return -EINVAL; 14291da177e4SLinus Torvalds 14301da177e4SLinus Torvalds /* we need capabilities... */ 14311da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 14321da177e4SLinus Torvalds return -EPERM; 14331da177e4SLinus Torvalds 14341da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 14351da177e4SLinus Torvalds if (IS_ERR(mnt)) 14361da177e4SLinus Torvalds return PTR_ERR(mnt); 14371da177e4SLinus Torvalds 14381da177e4SLinus Torvalds return do_add_mount(mnt, nd, mnt_flags, NULL); 14391da177e4SLinus Torvalds } 14401da177e4SLinus Torvalds 14411da177e4SLinus Torvalds /* 14421da177e4SLinus Torvalds * add a mount into a namespace's mount tree 14431da177e4SLinus Torvalds * - provide the option of adding the new mount to an expiration list 14441da177e4SLinus Torvalds */ 14451da177e4SLinus Torvalds int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd, 14461da177e4SLinus Torvalds int mnt_flags, struct list_head *fslist) 14471da177e4SLinus Torvalds { 14481da177e4SLinus Torvalds int err; 14491da177e4SLinus Torvalds 1450390c6843SRam Pai down_write(&namespace_sem); 14511da177e4SLinus Torvalds /* Something was mounted here while we slept */ 14524ac91378SJan Blunck while (d_mountpoint(nd->path.dentry) && 14534ac91378SJan Blunck follow_down(&nd->path.mnt, &nd->path.dentry)) 14541da177e4SLinus Torvalds ; 14551da177e4SLinus Torvalds err = -EINVAL; 14564ac91378SJan Blunck if (!check_mnt(nd->path.mnt)) 14571da177e4SLinus Torvalds goto unlock; 14581da177e4SLinus Torvalds 14591da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 14601da177e4SLinus Torvalds err = -EBUSY; 14614ac91378SJan Blunck if (nd->path.mnt->mnt_sb == newmnt->mnt_sb && 14624ac91378SJan Blunck nd->path.mnt->mnt_root == nd->path.dentry) 14631da177e4SLinus Torvalds goto unlock; 14641da177e4SLinus Torvalds 14651da177e4SLinus Torvalds err = -EINVAL; 14661da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 14671da177e4SLinus Torvalds goto unlock; 14681da177e4SLinus Torvalds 14691da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 14705b83d2c5SRam Pai if ((err = graft_tree(newmnt, nd))) 14715b83d2c5SRam Pai goto unlock; 14721da177e4SLinus Torvalds 14736758f953SAl Viro if (fslist) /* add to the specified expiration list */ 147455e700b9SMiklos Szeredi list_add_tail(&newmnt->mnt_expire, fslist); 14756758f953SAl Viro 1476390c6843SRam Pai up_write(&namespace_sem); 14775b83d2c5SRam Pai return 0; 14781da177e4SLinus Torvalds 14791da177e4SLinus Torvalds unlock: 1480390c6843SRam Pai up_write(&namespace_sem); 14811da177e4SLinus Torvalds mntput(newmnt); 14821da177e4SLinus Torvalds return err; 14831da177e4SLinus Torvalds } 14841da177e4SLinus Torvalds 14851da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(do_add_mount); 14861da177e4SLinus Torvalds 14875528f911STrond Myklebust /* 14881da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 14891da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 14901da177e4SLinus Torvalds * here 14911da177e4SLinus Torvalds */ 14921da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 14931da177e4SLinus Torvalds { 14941da177e4SLinus Torvalds struct vfsmount *mnt, *next; 14951da177e4SLinus Torvalds LIST_HEAD(graveyard); 1496bcc5c7d2SAl Viro LIST_HEAD(umounts); 14971da177e4SLinus Torvalds 14981da177e4SLinus Torvalds if (list_empty(mounts)) 14991da177e4SLinus Torvalds return; 15001da177e4SLinus Torvalds 1501bcc5c7d2SAl Viro down_write(&namespace_sem); 15021da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 15031da177e4SLinus Torvalds 15041da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 15051da177e4SLinus Torvalds * following criteria: 15061da177e4SLinus Torvalds * - only referenced by its parent vfsmount 15071da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 15081da177e4SLinus Torvalds * cleared by mntput()) 15091da177e4SLinus Torvalds */ 151055e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 15111da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 1512bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 15131da177e4SLinus Torvalds continue; 151455e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 15151da177e4SLinus Torvalds } 1516bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1517bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 1518bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1519bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 1520bcc5c7d2SAl Viro } 15211da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 1522bcc5c7d2SAl Viro up_write(&namespace_sem); 1523bcc5c7d2SAl Viro 1524bcc5c7d2SAl Viro release_mounts(&umounts); 15251da177e4SLinus Torvalds } 15261da177e4SLinus Torvalds 15271da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 15281da177e4SLinus Torvalds 15291da177e4SLinus Torvalds /* 15305528f911STrond Myklebust * Ripoff of 'select_parent()' 15315528f911STrond Myklebust * 15325528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 15335528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 15345528f911STrond Myklebust */ 15355528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 15365528f911STrond Myklebust { 15375528f911STrond Myklebust struct vfsmount *this_parent = parent; 15385528f911STrond Myklebust struct list_head *next; 15395528f911STrond Myklebust int found = 0; 15405528f911STrond Myklebust 15415528f911STrond Myklebust repeat: 15425528f911STrond Myklebust next = this_parent->mnt_mounts.next; 15435528f911STrond Myklebust resume: 15445528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 15455528f911STrond Myklebust struct list_head *tmp = next; 15465528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 15475528f911STrond Myklebust 15485528f911STrond Myklebust next = tmp->next; 15495528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 15505528f911STrond Myklebust continue; 15515528f911STrond Myklebust /* 15525528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 15535528f911STrond Myklebust */ 15545528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 15555528f911STrond Myklebust this_parent = mnt; 15565528f911STrond Myklebust goto repeat; 15575528f911STrond Myklebust } 15585528f911STrond Myklebust 15595528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 15605528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 15615528f911STrond Myklebust found++; 15625528f911STrond Myklebust } 15635528f911STrond Myklebust } 15645528f911STrond Myklebust /* 15655528f911STrond Myklebust * All done at this level ... ascend and resume the search 15665528f911STrond Myklebust */ 15675528f911STrond Myklebust if (this_parent != parent) { 15685528f911STrond Myklebust next = this_parent->mnt_child.next; 15695528f911STrond Myklebust this_parent = this_parent->mnt_parent; 15705528f911STrond Myklebust goto resume; 15715528f911STrond Myklebust } 15725528f911STrond Myklebust return found; 15735528f911STrond Myklebust } 15745528f911STrond Myklebust 15755528f911STrond Myklebust /* 15765528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 15775528f911STrond Myklebust * submounts of a specific parent mountpoint 15785528f911STrond Myklebust */ 1579c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 15805528f911STrond Myklebust { 15815528f911STrond Myklebust LIST_HEAD(graveyard); 1582c35038beSAl Viro struct vfsmount *m; 15835528f911STrond Myklebust 15845528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 1585c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 1586bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1587c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 1588bcc5c7d2SAl Viro mnt_expire); 1589bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1590c35038beSAl Viro umount_tree(mnt, 1, umounts); 1591bcc5c7d2SAl Viro } 1592bcc5c7d2SAl Viro } 15935528f911STrond Myklebust } 15945528f911STrond Myklebust 15955528f911STrond Myklebust /* 15961da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 15971da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 15981da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 15991da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 16001da177e4SLinus Torvalds */ 1601b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 1602b58fed8bSRam Pai unsigned long n) 16031da177e4SLinus Torvalds { 16041da177e4SLinus Torvalds char *t = to; 16051da177e4SLinus Torvalds const char __user *f = from; 16061da177e4SLinus Torvalds char c; 16071da177e4SLinus Torvalds 16081da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 16091da177e4SLinus Torvalds return n; 16101da177e4SLinus Torvalds 16111da177e4SLinus Torvalds while (n) { 16121da177e4SLinus Torvalds if (__get_user(c, f)) { 16131da177e4SLinus Torvalds memset(t, 0, n); 16141da177e4SLinus Torvalds break; 16151da177e4SLinus Torvalds } 16161da177e4SLinus Torvalds *t++ = c; 16171da177e4SLinus Torvalds f++; 16181da177e4SLinus Torvalds n--; 16191da177e4SLinus Torvalds } 16201da177e4SLinus Torvalds return n; 16211da177e4SLinus Torvalds } 16221da177e4SLinus Torvalds 16231da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 16241da177e4SLinus Torvalds { 16251da177e4SLinus Torvalds int i; 16261da177e4SLinus Torvalds unsigned long page; 16271da177e4SLinus Torvalds unsigned long size; 16281da177e4SLinus Torvalds 16291da177e4SLinus Torvalds *where = 0; 16301da177e4SLinus Torvalds if (!data) 16311da177e4SLinus Torvalds return 0; 16321da177e4SLinus Torvalds 16331da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 16341da177e4SLinus Torvalds return -ENOMEM; 16351da177e4SLinus Torvalds 16361da177e4SLinus Torvalds /* We only care that *some* data at the address the user 16371da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 16381da177e4SLinus Torvalds * the remainder of the page. 16391da177e4SLinus Torvalds */ 16401da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 16411da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 16421da177e4SLinus Torvalds if (size > PAGE_SIZE) 16431da177e4SLinus Torvalds size = PAGE_SIZE; 16441da177e4SLinus Torvalds 16451da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 16461da177e4SLinus Torvalds if (!i) { 16471da177e4SLinus Torvalds free_page(page); 16481da177e4SLinus Torvalds return -EFAULT; 16491da177e4SLinus Torvalds } 16501da177e4SLinus Torvalds if (i != PAGE_SIZE) 16511da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 16521da177e4SLinus Torvalds *where = page; 16531da177e4SLinus Torvalds return 0; 16541da177e4SLinus Torvalds } 16551da177e4SLinus Torvalds 16561da177e4SLinus Torvalds /* 16571da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 16581da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 16591da177e4SLinus Torvalds * 16601da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 16611da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 16621da177e4SLinus Torvalds * information (or be NULL). 16631da177e4SLinus Torvalds * 16641da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 16651da177e4SLinus Torvalds * When the flags word was introduced its top half was required 16661da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 16671da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 16681da177e4SLinus Torvalds * and must be discarded. 16691da177e4SLinus Torvalds */ 16701da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 16711da177e4SLinus Torvalds unsigned long flags, void *data_page) 16721da177e4SLinus Torvalds { 16731da177e4SLinus Torvalds struct nameidata nd; 16741da177e4SLinus Torvalds int retval = 0; 16751da177e4SLinus Torvalds int mnt_flags = 0; 16761da177e4SLinus Torvalds 16771da177e4SLinus Torvalds /* Discard magic */ 16781da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 16791da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 16801da177e4SLinus Torvalds 16811da177e4SLinus Torvalds /* Basic sanity checks */ 16821da177e4SLinus Torvalds 16831da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 16841da177e4SLinus Torvalds return -EINVAL; 16851da177e4SLinus Torvalds if (dev_name && !memchr(dev_name, 0, PAGE_SIZE)) 16861da177e4SLinus Torvalds return -EINVAL; 16871da177e4SLinus Torvalds 16881da177e4SLinus Torvalds if (data_page) 16891da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 16901da177e4SLinus Torvalds 16911da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 16921da177e4SLinus Torvalds if (flags & MS_NOSUID) 16931da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 16941da177e4SLinus Torvalds if (flags & MS_NODEV) 16951da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 16961da177e4SLinus Torvalds if (flags & MS_NOEXEC) 16971da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 1698fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 1699fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 1700fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 1701fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 170247ae32d6SValerie Henson if (flags & MS_RELATIME) 170347ae32d6SValerie Henson mnt_flags |= MNT_RELATIME; 1704fc33a7bbSChristoph Hellwig 1705fc33a7bbSChristoph Hellwig flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | 17068bf9725cSPavel Emelyanov MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT); 17071da177e4SLinus Torvalds 17081da177e4SLinus Torvalds /* ... and get the mountpoint */ 17091da177e4SLinus Torvalds retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd); 17101da177e4SLinus Torvalds if (retval) 17111da177e4SLinus Torvalds return retval; 17121da177e4SLinus Torvalds 17131da177e4SLinus Torvalds retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page); 17141da177e4SLinus Torvalds if (retval) 17151da177e4SLinus Torvalds goto dput_out; 17161da177e4SLinus Torvalds 17171da177e4SLinus Torvalds if (flags & MS_REMOUNT) 17181da177e4SLinus Torvalds retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags, 17191da177e4SLinus Torvalds data_page); 17201da177e4SLinus Torvalds else if (flags & MS_BIND) 1721eee391a6SAndrew Morton retval = do_loopback(&nd, dev_name, flags & MS_REC); 17229676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 172307b20889SRam Pai retval = do_change_type(&nd, flags); 17241da177e4SLinus Torvalds else if (flags & MS_MOVE) 17251da177e4SLinus Torvalds retval = do_move_mount(&nd, dev_name); 17261da177e4SLinus Torvalds else 17271da177e4SLinus Torvalds retval = do_new_mount(&nd, type_page, flags, mnt_flags, 17281da177e4SLinus Torvalds dev_name, data_page); 17291da177e4SLinus Torvalds dput_out: 17301d957f9bSJan Blunck path_put(&nd.path); 17311da177e4SLinus Torvalds return retval; 17321da177e4SLinus Torvalds } 17331da177e4SLinus Torvalds 1734741a2951SJANAK DESAI /* 1735741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 1736741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 1737741a2951SJANAK DESAI */ 1738e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 17396b3286edSKirill Korotaev struct fs_struct *fs) 17401da177e4SLinus Torvalds { 17416b3286edSKirill Korotaev struct mnt_namespace *new_ns; 17421da177e4SLinus Torvalds struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL; 17431da177e4SLinus Torvalds struct vfsmount *p, *q; 17441da177e4SLinus Torvalds 17456b3286edSKirill Korotaev new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 17461da177e4SLinus Torvalds if (!new_ns) 1747467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM); 17481da177e4SLinus Torvalds 17491da177e4SLinus Torvalds atomic_set(&new_ns->count, 1); 17501da177e4SLinus Torvalds INIT_LIST_HEAD(&new_ns->list); 17515addc5ddSAl Viro init_waitqueue_head(&new_ns->poll); 17525addc5ddSAl Viro new_ns->event = 0; 17531da177e4SLinus Torvalds 1754390c6843SRam Pai down_write(&namespace_sem); 17551da177e4SLinus Torvalds /* First pass: copy the tree topology */ 17566b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 17579676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 17581da177e4SLinus Torvalds if (!new_ns->root) { 1759390c6843SRam Pai up_write(&namespace_sem); 17601da177e4SLinus Torvalds kfree(new_ns); 1761467e9f4bSCedric Le Goater return ERR_PTR(-ENOMEM);; 17621da177e4SLinus Torvalds } 17631da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 17641da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 17651da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 17661da177e4SLinus Torvalds 17671da177e4SLinus Torvalds /* 17681da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 17691da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 17701da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 17711da177e4SLinus Torvalds */ 17726b3286edSKirill Korotaev p = mnt_ns->root; 17731da177e4SLinus Torvalds q = new_ns->root; 17741da177e4SLinus Torvalds while (p) { 17756b3286edSKirill Korotaev q->mnt_ns = new_ns; 17761da177e4SLinus Torvalds if (fs) { 17776ac08c39SJan Blunck if (p == fs->root.mnt) { 17781da177e4SLinus Torvalds rootmnt = p; 17796ac08c39SJan Blunck fs->root.mnt = mntget(q); 17801da177e4SLinus Torvalds } 17816ac08c39SJan Blunck if (p == fs->pwd.mnt) { 17821da177e4SLinus Torvalds pwdmnt = p; 17836ac08c39SJan Blunck fs->pwd.mnt = mntget(q); 17841da177e4SLinus Torvalds } 17856ac08c39SJan Blunck if (p == fs->altroot.mnt) { 17861da177e4SLinus Torvalds altrootmnt = p; 17876ac08c39SJan Blunck fs->altroot.mnt = mntget(q); 17881da177e4SLinus Torvalds } 17891da177e4SLinus Torvalds } 17906b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 17911da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 17921da177e4SLinus Torvalds } 1793390c6843SRam Pai up_write(&namespace_sem); 17941da177e4SLinus Torvalds 17951da177e4SLinus Torvalds if (rootmnt) 17961da177e4SLinus Torvalds mntput(rootmnt); 17971da177e4SLinus Torvalds if (pwdmnt) 17981da177e4SLinus Torvalds mntput(pwdmnt); 17991da177e4SLinus Torvalds if (altrootmnt) 18001da177e4SLinus Torvalds mntput(altrootmnt); 18011da177e4SLinus Torvalds 1802741a2951SJANAK DESAI return new_ns; 1803741a2951SJANAK DESAI } 1804741a2951SJANAK DESAI 1805213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 1806e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 1807741a2951SJANAK DESAI { 18086b3286edSKirill Korotaev struct mnt_namespace *new_ns; 1809741a2951SJANAK DESAI 1810e3222c4eSBadari Pulavarty BUG_ON(!ns); 18116b3286edSKirill Korotaev get_mnt_ns(ns); 1812741a2951SJANAK DESAI 1813741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 1814e3222c4eSBadari Pulavarty return ns; 1815741a2951SJANAK DESAI 1816e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 1817741a2951SJANAK DESAI 18186b3286edSKirill Korotaev put_mnt_ns(ns); 1819e3222c4eSBadari Pulavarty return new_ns; 18201da177e4SLinus Torvalds } 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name, 18231da177e4SLinus Torvalds char __user * type, unsigned long flags, 18241da177e4SLinus Torvalds void __user * data) 18251da177e4SLinus Torvalds { 18261da177e4SLinus Torvalds int retval; 18271da177e4SLinus Torvalds unsigned long data_page; 18281da177e4SLinus Torvalds unsigned long type_page; 18291da177e4SLinus Torvalds unsigned long dev_page; 18301da177e4SLinus Torvalds char *dir_page; 18311da177e4SLinus Torvalds 18321da177e4SLinus Torvalds retval = copy_mount_options(type, &type_page); 18331da177e4SLinus Torvalds if (retval < 0) 18341da177e4SLinus Torvalds return retval; 18351da177e4SLinus Torvalds 18361da177e4SLinus Torvalds dir_page = getname(dir_name); 18371da177e4SLinus Torvalds retval = PTR_ERR(dir_page); 18381da177e4SLinus Torvalds if (IS_ERR(dir_page)) 18391da177e4SLinus Torvalds goto out1; 18401da177e4SLinus Torvalds 18411da177e4SLinus Torvalds retval = copy_mount_options(dev_name, &dev_page); 18421da177e4SLinus Torvalds if (retval < 0) 18431da177e4SLinus Torvalds goto out2; 18441da177e4SLinus Torvalds 18451da177e4SLinus Torvalds retval = copy_mount_options(data, &data_page); 18461da177e4SLinus Torvalds if (retval < 0) 18471da177e4SLinus Torvalds goto out3; 18481da177e4SLinus Torvalds 18491da177e4SLinus Torvalds lock_kernel(); 18501da177e4SLinus Torvalds retval = do_mount((char *)dev_page, dir_page, (char *)type_page, 18511da177e4SLinus Torvalds flags, (void *)data_page); 18521da177e4SLinus Torvalds unlock_kernel(); 18531da177e4SLinus Torvalds free_page(data_page); 18541da177e4SLinus Torvalds 18551da177e4SLinus Torvalds out3: 18561da177e4SLinus Torvalds free_page(dev_page); 18571da177e4SLinus Torvalds out2: 18581da177e4SLinus Torvalds putname(dir_page); 18591da177e4SLinus Torvalds out1: 18601da177e4SLinus Torvalds free_page(type_page); 18611da177e4SLinus Torvalds return retval; 18621da177e4SLinus Torvalds } 18631da177e4SLinus Torvalds 18641da177e4SLinus Torvalds /* 18651da177e4SLinus Torvalds * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values. 18661da177e4SLinus Torvalds * It can block. Requires the big lock held. 18671da177e4SLinus Torvalds */ 1868ac748a09SJan Blunck void set_fs_root(struct fs_struct *fs, struct path *path) 18691da177e4SLinus Torvalds { 18706ac08c39SJan Blunck struct path old_root; 18716ac08c39SJan Blunck 18721da177e4SLinus Torvalds write_lock(&fs->lock); 18731da177e4SLinus Torvalds old_root = fs->root; 1874ac748a09SJan Blunck fs->root = *path; 1875ac748a09SJan Blunck path_get(path); 18761da177e4SLinus Torvalds write_unlock(&fs->lock); 18776ac08c39SJan Blunck if (old_root.dentry) 18786ac08c39SJan Blunck path_put(&old_root); 18791da177e4SLinus Torvalds } 18801da177e4SLinus Torvalds 18811da177e4SLinus Torvalds /* 18821da177e4SLinus Torvalds * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values. 18831da177e4SLinus Torvalds * It can block. Requires the big lock held. 18841da177e4SLinus Torvalds */ 1885ac748a09SJan Blunck void set_fs_pwd(struct fs_struct *fs, struct path *path) 18861da177e4SLinus Torvalds { 18876ac08c39SJan Blunck struct path old_pwd; 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds write_lock(&fs->lock); 18901da177e4SLinus Torvalds old_pwd = fs->pwd; 1891ac748a09SJan Blunck fs->pwd = *path; 1892ac748a09SJan Blunck path_get(path); 18931da177e4SLinus Torvalds write_unlock(&fs->lock); 18941da177e4SLinus Torvalds 18956ac08c39SJan Blunck if (old_pwd.dentry) 18966ac08c39SJan Blunck path_put(&old_pwd); 18971da177e4SLinus Torvalds } 18981da177e4SLinus Torvalds 18991a390689SAl Viro static void chroot_fs_refs(struct path *old_root, struct path *new_root) 19001da177e4SLinus Torvalds { 19011da177e4SLinus Torvalds struct task_struct *g, *p; 19021da177e4SLinus Torvalds struct fs_struct *fs; 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds read_lock(&tasklist_lock); 19051da177e4SLinus Torvalds do_each_thread(g, p) { 19061da177e4SLinus Torvalds task_lock(p); 19071da177e4SLinus Torvalds fs = p->fs; 19081da177e4SLinus Torvalds if (fs) { 19091da177e4SLinus Torvalds atomic_inc(&fs->count); 19101da177e4SLinus Torvalds task_unlock(p); 19111a390689SAl Viro if (fs->root.dentry == old_root->dentry 19121a390689SAl Viro && fs->root.mnt == old_root->mnt) 19131a390689SAl Viro set_fs_root(fs, new_root); 19141a390689SAl Viro if (fs->pwd.dentry == old_root->dentry 19151a390689SAl Viro && fs->pwd.mnt == old_root->mnt) 19161a390689SAl Viro set_fs_pwd(fs, new_root); 19171da177e4SLinus Torvalds put_fs_struct(fs); 19181da177e4SLinus Torvalds } else 19191da177e4SLinus Torvalds task_unlock(p); 19201da177e4SLinus Torvalds } while_each_thread(g, p); 19211da177e4SLinus Torvalds read_unlock(&tasklist_lock); 19221da177e4SLinus Torvalds } 19231da177e4SLinus Torvalds 19241da177e4SLinus Torvalds /* 19251da177e4SLinus Torvalds * pivot_root Semantics: 19261da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 19271da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 19281da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 19291da177e4SLinus Torvalds * 19301da177e4SLinus Torvalds * Restrictions: 19311da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 19321da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 19331da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 19341da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 19351da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 19361da177e4SLinus Torvalds * 19374a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 19384a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 19394a0d11faSNeil Brown * in this situation. 19404a0d11faSNeil Brown * 19411da177e4SLinus Torvalds * Notes: 19421da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 19431da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 19441da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 19451da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 19461da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 19471da177e4SLinus Torvalds * first. 19481da177e4SLinus Torvalds */ 1949b58fed8bSRam Pai asmlinkage long sys_pivot_root(const char __user * new_root, 1950b58fed8bSRam Pai const char __user * put_old) 19511da177e4SLinus Torvalds { 19521da177e4SLinus Torvalds struct vfsmount *tmp; 19531a390689SAl Viro struct nameidata new_nd, old_nd, user_nd; 19541a390689SAl Viro struct path parent_path, root_parent; 19551da177e4SLinus Torvalds int error; 19561da177e4SLinus Torvalds 19571da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 19581da177e4SLinus Torvalds return -EPERM; 19591da177e4SLinus Torvalds 19601da177e4SLinus Torvalds lock_kernel(); 19611da177e4SLinus Torvalds 1962b58fed8bSRam Pai error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, 1963b58fed8bSRam Pai &new_nd); 19641da177e4SLinus Torvalds if (error) 19651da177e4SLinus Torvalds goto out0; 19661da177e4SLinus Torvalds error = -EINVAL; 19674ac91378SJan Blunck if (!check_mnt(new_nd.path.mnt)) 19681da177e4SLinus Torvalds goto out1; 19691da177e4SLinus Torvalds 19701da177e4SLinus Torvalds error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd); 19711da177e4SLinus Torvalds if (error) 19721da177e4SLinus Torvalds goto out1; 19731da177e4SLinus Torvalds 19741da177e4SLinus Torvalds error = security_sb_pivotroot(&old_nd, &new_nd); 19751da177e4SLinus Torvalds if (error) { 19761d957f9bSJan Blunck path_put(&old_nd.path); 19771da177e4SLinus Torvalds goto out1; 19781da177e4SLinus Torvalds } 19791da177e4SLinus Torvalds 19801da177e4SLinus Torvalds read_lock(¤t->fs->lock); 19816ac08c39SJan Blunck user_nd.path = current->fs->root; 19826ac08c39SJan Blunck path_get(¤t->fs->root); 19831da177e4SLinus Torvalds read_unlock(¤t->fs->lock); 1984390c6843SRam Pai down_write(&namespace_sem); 19854ac91378SJan Blunck mutex_lock(&old_nd.path.dentry->d_inode->i_mutex); 19861da177e4SLinus Torvalds error = -EINVAL; 19874ac91378SJan Blunck if (IS_MNT_SHARED(old_nd.path.mnt) || 19884ac91378SJan Blunck IS_MNT_SHARED(new_nd.path.mnt->mnt_parent) || 19894ac91378SJan Blunck IS_MNT_SHARED(user_nd.path.mnt->mnt_parent)) 199021444403SRam Pai goto out2; 19914ac91378SJan Blunck if (!check_mnt(user_nd.path.mnt)) 19921da177e4SLinus Torvalds goto out2; 19931da177e4SLinus Torvalds error = -ENOENT; 19944ac91378SJan Blunck if (IS_DEADDIR(new_nd.path.dentry->d_inode)) 19951da177e4SLinus Torvalds goto out2; 19964ac91378SJan Blunck if (d_unhashed(new_nd.path.dentry) && !IS_ROOT(new_nd.path.dentry)) 19971da177e4SLinus Torvalds goto out2; 19984ac91378SJan Blunck if (d_unhashed(old_nd.path.dentry) && !IS_ROOT(old_nd.path.dentry)) 19991da177e4SLinus Torvalds goto out2; 20001da177e4SLinus Torvalds error = -EBUSY; 20014ac91378SJan Blunck if (new_nd.path.mnt == user_nd.path.mnt || 20024ac91378SJan Blunck old_nd.path.mnt == user_nd.path.mnt) 20031da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 20041da177e4SLinus Torvalds error = -EINVAL; 20054ac91378SJan Blunck if (user_nd.path.mnt->mnt_root != user_nd.path.dentry) 20061da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 20074ac91378SJan Blunck if (user_nd.path.mnt->mnt_parent == user_nd.path.mnt) 20080bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 20094ac91378SJan Blunck if (new_nd.path.mnt->mnt_root != new_nd.path.dentry) 20101da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 20114ac91378SJan Blunck if (new_nd.path.mnt->mnt_parent == new_nd.path.mnt) 20120bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 20134ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 20144ac91378SJan Blunck tmp = old_nd.path.mnt; 20151da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 20164ac91378SJan Blunck if (tmp != new_nd.path.mnt) { 20171da177e4SLinus Torvalds for (;;) { 20181da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 20191da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 20204ac91378SJan Blunck if (tmp->mnt_parent == new_nd.path.mnt) 20211da177e4SLinus Torvalds break; 20221da177e4SLinus Torvalds tmp = tmp->mnt_parent; 20231da177e4SLinus Torvalds } 20244ac91378SJan Blunck if (!is_subdir(tmp->mnt_mountpoint, new_nd.path.dentry)) 20251da177e4SLinus Torvalds goto out3; 20264ac91378SJan Blunck } else if (!is_subdir(old_nd.path.dentry, new_nd.path.dentry)) 20271da177e4SLinus Torvalds goto out3; 20281a390689SAl Viro detach_mnt(new_nd.path.mnt, &parent_path); 20294ac91378SJan Blunck detach_mnt(user_nd.path.mnt, &root_parent); 20304ac91378SJan Blunck /* mount old root on put_old */ 20311a390689SAl Viro attach_mnt(user_nd.path.mnt, &old_nd.path); 20324ac91378SJan Blunck /* mount new_root on / */ 20334ac91378SJan Blunck attach_mnt(new_nd.path.mnt, &root_parent); 20346b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 20351da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 20361a390689SAl Viro chroot_fs_refs(&user_nd.path, &new_nd.path); 20371da177e4SLinus Torvalds security_sb_post_pivotroot(&user_nd, &new_nd); 20381da177e4SLinus Torvalds error = 0; 20391a390689SAl Viro path_put(&root_parent); 20401a390689SAl Viro path_put(&parent_path); 20411da177e4SLinus Torvalds out2: 20424ac91378SJan Blunck mutex_unlock(&old_nd.path.dentry->d_inode->i_mutex); 2043390c6843SRam Pai up_write(&namespace_sem); 20441d957f9bSJan Blunck path_put(&user_nd.path); 20451d957f9bSJan Blunck path_put(&old_nd.path); 20461da177e4SLinus Torvalds out1: 20471d957f9bSJan Blunck path_put(&new_nd.path); 20481da177e4SLinus Torvalds out0: 20491da177e4SLinus Torvalds unlock_kernel(); 20501da177e4SLinus Torvalds return error; 20511da177e4SLinus Torvalds out3: 20521da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 20531da177e4SLinus Torvalds goto out2; 20541da177e4SLinus Torvalds } 20551da177e4SLinus Torvalds 20561da177e4SLinus Torvalds static void __init init_mount_tree(void) 20571da177e4SLinus Torvalds { 20581da177e4SLinus Torvalds struct vfsmount *mnt; 20596b3286edSKirill Korotaev struct mnt_namespace *ns; 2060ac748a09SJan Blunck struct path root; 20611da177e4SLinus Torvalds 20621da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 20631da177e4SLinus Torvalds if (IS_ERR(mnt)) 20641da177e4SLinus Torvalds panic("Can't create rootfs"); 20656b3286edSKirill Korotaev ns = kmalloc(sizeof(*ns), GFP_KERNEL); 20666b3286edSKirill Korotaev if (!ns) 20671da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 20686b3286edSKirill Korotaev atomic_set(&ns->count, 1); 20696b3286edSKirill Korotaev INIT_LIST_HEAD(&ns->list); 20706b3286edSKirill Korotaev init_waitqueue_head(&ns->poll); 20716b3286edSKirill Korotaev ns->event = 0; 20726b3286edSKirill Korotaev list_add(&mnt->mnt_list, &ns->list); 20736b3286edSKirill Korotaev ns->root = mnt; 20746b3286edSKirill Korotaev mnt->mnt_ns = ns; 20751da177e4SLinus Torvalds 20766b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 20776b3286edSKirill Korotaev get_mnt_ns(ns); 20781da177e4SLinus Torvalds 2079ac748a09SJan Blunck root.mnt = ns->root; 2080ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2081ac748a09SJan Blunck 2082ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2083ac748a09SJan Blunck set_fs_root(current->fs, &root); 20841da177e4SLinus Torvalds } 20851da177e4SLinus Torvalds 208674bf17cfSDenis Cheng void __init mnt_init(void) 20871da177e4SLinus Torvalds { 208813f14b4dSEric Dumazet unsigned u; 208915a67dd8SRandy Dunlap int err; 20901da177e4SLinus Torvalds 2091390c6843SRam Pai init_rwsem(&namespace_sem); 2092390c6843SRam Pai 20931da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 209420c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 20951da177e4SLinus Torvalds 2096b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 20971da177e4SLinus Torvalds 20981da177e4SLinus Torvalds if (!mount_hashtable) 20991da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 21001da177e4SLinus Torvalds 210113f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 21021da177e4SLinus Torvalds 210313f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 210413f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 21051da177e4SLinus Torvalds 210615a67dd8SRandy Dunlap err = sysfs_init(); 210715a67dd8SRandy Dunlap if (err) 210815a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 210915a67dd8SRandy Dunlap __FUNCTION__, err); 211000d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 211100d26666SGreg Kroah-Hartman if (!fs_kobj) 211200d26666SGreg Kroah-Hartman printk(KERN_WARNING "%s: kobj create error\n", __FUNCTION__); 21131da177e4SLinus Torvalds init_rootfs(); 21141da177e4SLinus Torvalds init_mount_tree(); 21151da177e4SLinus Torvalds } 21161da177e4SLinus Torvalds 21176b3286edSKirill Korotaev void __put_mnt_ns(struct mnt_namespace *ns) 21181da177e4SLinus Torvalds { 21196b3286edSKirill Korotaev struct vfsmount *root = ns->root; 212070fbcdf4SRam Pai LIST_HEAD(umount_list); 21216b3286edSKirill Korotaev ns->root = NULL; 21221ce88cf4SMiklos Szeredi spin_unlock(&vfsmount_lock); 2123390c6843SRam Pai down_write(&namespace_sem); 21241da177e4SLinus Torvalds spin_lock(&vfsmount_lock); 2125a05964f3SRam Pai umount_tree(root, 0, &umount_list); 21261da177e4SLinus Torvalds spin_unlock(&vfsmount_lock); 2127390c6843SRam Pai up_write(&namespace_sem); 212870fbcdf4SRam Pai release_mounts(&umount_list); 21296b3286edSKirill Korotaev kfree(ns); 21301da177e4SLinus Torvalds } 2131