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> 1499b7db7bSNick Piggin #include <linux/spinlock.h> 1599b7db7bSNick Piggin #include <linux/percpu.h> 161da177e4SLinus Torvalds #include <linux/init.h> 1715a67dd8SRandy Dunlap #include <linux/kernel.h> 181da177e4SLinus Torvalds #include <linux/acct.h> 1916f7e0feSRandy Dunlap #include <linux/capability.h> 203d733633SDave Hansen #include <linux/cpumask.h> 211da177e4SLinus Torvalds #include <linux/module.h> 22f20a9eadSAndrew Morton #include <linux/sysfs.h> 231da177e4SLinus Torvalds #include <linux/seq_file.h> 246b3286edSKirill Korotaev #include <linux/mnt_namespace.h> 251da177e4SLinus Torvalds #include <linux/namei.h> 26b43f3cbdSAlexey Dobriyan #include <linux/nsproxy.h> 271da177e4SLinus Torvalds #include <linux/security.h> 281da177e4SLinus Torvalds #include <linux/mount.h> 2907f3f05cSDavid Howells #include <linux/ramfs.h> 3013f14b4dSEric Dumazet #include <linux/log2.h> 3173cd49ecSMiklos Szeredi #include <linux/idr.h> 325ad4e53bSAl Viro #include <linux/fs_struct.h> 332504c5d6SAndreas Gruenbacher #include <linux/fsnotify.h> 341da177e4SLinus Torvalds #include <asm/uaccess.h> 351da177e4SLinus Torvalds #include <asm/unistd.h> 3607b20889SRam Pai #include "pnode.h" 37948730b0SAdrian Bunk #include "internal.h" 381da177e4SLinus Torvalds 3913f14b4dSEric Dumazet #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head)) 4013f14b4dSEric Dumazet #define HASH_SIZE (1UL << HASH_SHIFT) 4113f14b4dSEric Dumazet 425addc5ddSAl Viro static int event; 4373cd49ecSMiklos Szeredi static DEFINE_IDA(mnt_id_ida); 44719f5d7fSMiklos Szeredi static DEFINE_IDA(mnt_group_ida); 4599b7db7bSNick Piggin static DEFINE_SPINLOCK(mnt_id_lock); 46f21f6220SAl Viro static int mnt_id_start = 0; 47f21f6220SAl Viro static int mnt_group_start = 1; 485addc5ddSAl Viro 49fa3536ccSEric Dumazet static struct list_head *mount_hashtable __read_mostly; 50e18b890bSChristoph Lameter static struct kmem_cache *mnt_cache __read_mostly; 51390c6843SRam Pai static struct rw_semaphore namespace_sem; 521da177e4SLinus Torvalds 53f87fd4c2SMiklos Szeredi /* /sys/fs */ 5400d26666SGreg Kroah-Hartman struct kobject *fs_kobj; 5500d26666SGreg Kroah-Hartman EXPORT_SYMBOL_GPL(fs_kobj); 56f87fd4c2SMiklos Szeredi 5799b7db7bSNick Piggin /* 5899b7db7bSNick Piggin * vfsmount lock may be taken for read to prevent changes to the 5999b7db7bSNick Piggin * vfsmount hash, ie. during mountpoint lookups or walking back 6099b7db7bSNick Piggin * up the tree. 6199b7db7bSNick Piggin * 6299b7db7bSNick Piggin * It should be taken for write in all cases where the vfsmount 6399b7db7bSNick Piggin * tree or hash is modified or when a vfsmount structure is modified. 6499b7db7bSNick Piggin */ 6599b7db7bSNick Piggin DEFINE_BRLOCK(vfsmount_lock); 6699b7db7bSNick Piggin 671da177e4SLinus Torvalds static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry) 681da177e4SLinus Torvalds { 691da177e4SLinus Torvalds unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES); 701da177e4SLinus Torvalds tmp += ((unsigned long)dentry / L1_CACHE_BYTES); 7113f14b4dSEric Dumazet tmp = tmp + (tmp >> HASH_SHIFT); 7213f14b4dSEric Dumazet return tmp & (HASH_SIZE - 1); 731da177e4SLinus Torvalds } 741da177e4SLinus Torvalds 753d733633SDave Hansen #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16) 763d733633SDave Hansen 7799b7db7bSNick Piggin /* 7899b7db7bSNick Piggin * allocation is serialized by namespace_sem, but we need the spinlock to 7999b7db7bSNick Piggin * serialize with freeing. 8099b7db7bSNick Piggin */ 8173cd49ecSMiklos Szeredi static int mnt_alloc_id(struct vfsmount *mnt) 8273cd49ecSMiklos Szeredi { 8373cd49ecSMiklos Szeredi int res; 8473cd49ecSMiklos Szeredi 8573cd49ecSMiklos Szeredi retry: 8673cd49ecSMiklos Szeredi ida_pre_get(&mnt_id_ida, GFP_KERNEL); 8799b7db7bSNick Piggin spin_lock(&mnt_id_lock); 88f21f6220SAl Viro res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id); 89f21f6220SAl Viro if (!res) 90f21f6220SAl Viro mnt_id_start = mnt->mnt_id + 1; 9199b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 9273cd49ecSMiklos Szeredi if (res == -EAGAIN) 9373cd49ecSMiklos Szeredi goto retry; 9473cd49ecSMiklos Szeredi 9573cd49ecSMiklos Szeredi return res; 9673cd49ecSMiklos Szeredi } 9773cd49ecSMiklos Szeredi 9873cd49ecSMiklos Szeredi static void mnt_free_id(struct vfsmount *mnt) 9973cd49ecSMiklos Szeredi { 100f21f6220SAl Viro int id = mnt->mnt_id; 10199b7db7bSNick Piggin spin_lock(&mnt_id_lock); 102f21f6220SAl Viro ida_remove(&mnt_id_ida, id); 103f21f6220SAl Viro if (mnt_id_start > id) 104f21f6220SAl Viro mnt_id_start = id; 10599b7db7bSNick Piggin spin_unlock(&mnt_id_lock); 10673cd49ecSMiklos Szeredi } 10773cd49ecSMiklos Szeredi 108719f5d7fSMiklos Szeredi /* 109719f5d7fSMiklos Szeredi * Allocate a new peer group ID 110719f5d7fSMiklos Szeredi * 111719f5d7fSMiklos Szeredi * mnt_group_ida is protected by namespace_sem 112719f5d7fSMiklos Szeredi */ 113719f5d7fSMiklos Szeredi static int mnt_alloc_group_id(struct vfsmount *mnt) 114719f5d7fSMiklos Szeredi { 115f21f6220SAl Viro int res; 116f21f6220SAl Viro 117719f5d7fSMiklos Szeredi if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL)) 118719f5d7fSMiklos Szeredi return -ENOMEM; 119719f5d7fSMiklos Szeredi 120f21f6220SAl Viro res = ida_get_new_above(&mnt_group_ida, 121f21f6220SAl Viro mnt_group_start, 122f21f6220SAl Viro &mnt->mnt_group_id); 123f21f6220SAl Viro if (!res) 124f21f6220SAl Viro mnt_group_start = mnt->mnt_group_id + 1; 125f21f6220SAl Viro 126f21f6220SAl Viro return res; 127719f5d7fSMiklos Szeredi } 128719f5d7fSMiklos Szeredi 129719f5d7fSMiklos Szeredi /* 130719f5d7fSMiklos Szeredi * Release a peer group ID 131719f5d7fSMiklos Szeredi */ 132719f5d7fSMiklos Szeredi void mnt_release_group_id(struct vfsmount *mnt) 133719f5d7fSMiklos Szeredi { 134f21f6220SAl Viro int id = mnt->mnt_group_id; 135f21f6220SAl Viro ida_remove(&mnt_group_ida, id); 136f21f6220SAl Viro if (mnt_group_start > id) 137f21f6220SAl Viro mnt_group_start = id; 138719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; 139719f5d7fSMiklos Szeredi } 140719f5d7fSMiklos Szeredi 141b3e19d92SNick Piggin /* 142b3e19d92SNick Piggin * vfsmount lock must be held for read 143b3e19d92SNick Piggin */ 144b3e19d92SNick Piggin static inline void mnt_add_count(struct vfsmount *mnt, int n) 145b3e19d92SNick Piggin { 146b3e19d92SNick Piggin #ifdef CONFIG_SMP 147b3e19d92SNick Piggin this_cpu_add(mnt->mnt_pcp->mnt_count, n); 148b3e19d92SNick Piggin #else 149b3e19d92SNick Piggin preempt_disable(); 150b3e19d92SNick Piggin mnt->mnt_count += n; 151b3e19d92SNick Piggin preempt_enable(); 152b3e19d92SNick Piggin #endif 153b3e19d92SNick Piggin } 154b3e19d92SNick Piggin 155b3e19d92SNick Piggin static inline void mnt_set_count(struct vfsmount *mnt, int n) 156b3e19d92SNick Piggin { 157b3e19d92SNick Piggin #ifdef CONFIG_SMP 158b3e19d92SNick Piggin this_cpu_write(mnt->mnt_pcp->mnt_count, n); 159b3e19d92SNick Piggin #else 160b3e19d92SNick Piggin mnt->mnt_count = n; 161b3e19d92SNick Piggin #endif 162b3e19d92SNick Piggin } 163b3e19d92SNick Piggin 164b3e19d92SNick Piggin /* 165b3e19d92SNick Piggin * vfsmount lock must be held for read 166b3e19d92SNick Piggin */ 167b3e19d92SNick Piggin static inline void mnt_inc_count(struct vfsmount *mnt) 168b3e19d92SNick Piggin { 169b3e19d92SNick Piggin mnt_add_count(mnt, 1); 170b3e19d92SNick Piggin } 171b3e19d92SNick Piggin 172b3e19d92SNick Piggin /* 173b3e19d92SNick Piggin * vfsmount lock must be held for read 174b3e19d92SNick Piggin */ 175b3e19d92SNick Piggin static inline void mnt_dec_count(struct vfsmount *mnt) 176b3e19d92SNick Piggin { 177b3e19d92SNick Piggin mnt_add_count(mnt, -1); 178b3e19d92SNick Piggin } 179b3e19d92SNick Piggin 180b3e19d92SNick Piggin /* 181b3e19d92SNick Piggin * vfsmount lock must be held for write 182b3e19d92SNick Piggin */ 183b3e19d92SNick Piggin unsigned int mnt_get_count(struct vfsmount *mnt) 184b3e19d92SNick Piggin { 185b3e19d92SNick Piggin #ifdef CONFIG_SMP 186b3e19d92SNick Piggin unsigned int count = atomic_read(&mnt->mnt_longrefs); 187b3e19d92SNick Piggin int cpu; 188b3e19d92SNick Piggin 189b3e19d92SNick Piggin for_each_possible_cpu(cpu) { 190b3e19d92SNick Piggin count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count; 191b3e19d92SNick Piggin } 192b3e19d92SNick Piggin 193b3e19d92SNick Piggin return count; 194b3e19d92SNick Piggin #else 195b3e19d92SNick Piggin return mnt->mnt_count; 196b3e19d92SNick Piggin #endif 197b3e19d92SNick Piggin } 198b3e19d92SNick Piggin 1991da177e4SLinus Torvalds struct vfsmount *alloc_vfsmnt(const char *name) 2001da177e4SLinus Torvalds { 201c3762229SRobert P. J. Day struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL); 2021da177e4SLinus Torvalds if (mnt) { 20373cd49ecSMiklos Szeredi int err; 20473cd49ecSMiklos Szeredi 20573cd49ecSMiklos Szeredi err = mnt_alloc_id(mnt); 20688b38782SLi Zefan if (err) 20788b38782SLi Zefan goto out_free_cache; 20888b38782SLi Zefan 20988b38782SLi Zefan if (name) { 21088b38782SLi Zefan mnt->mnt_devname = kstrdup(name, GFP_KERNEL); 21188b38782SLi Zefan if (!mnt->mnt_devname) 21288b38782SLi Zefan goto out_free_id; 21373cd49ecSMiklos Szeredi } 21473cd49ecSMiklos Szeredi 215b3e19d92SNick Piggin #ifdef CONFIG_SMP 216b3e19d92SNick Piggin mnt->mnt_pcp = alloc_percpu(struct mnt_pcp); 217b3e19d92SNick Piggin if (!mnt->mnt_pcp) 218b3e19d92SNick Piggin goto out_free_devname; 219b3e19d92SNick Piggin 220b3e19d92SNick Piggin atomic_set(&mnt->mnt_longrefs, 1); 221b3e19d92SNick Piggin #else 222b3e19d92SNick Piggin mnt->mnt_count = 1; 223b3e19d92SNick Piggin mnt->mnt_writers = 0; 224b3e19d92SNick Piggin #endif 225b3e19d92SNick Piggin 2261da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_hash); 2271da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_child); 2281da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_mounts); 2291da177e4SLinus Torvalds INIT_LIST_HEAD(&mnt->mnt_list); 23055e700b9SMiklos Szeredi INIT_LIST_HEAD(&mnt->mnt_expire); 23103e06e68SRam Pai INIT_LIST_HEAD(&mnt->mnt_share); 232a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave_list); 233a58b0eb8SRam Pai INIT_LIST_HEAD(&mnt->mnt_slave); 2342504c5d6SAndreas Gruenbacher #ifdef CONFIG_FSNOTIFY 2352504c5d6SAndreas Gruenbacher INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks); 2362504c5d6SAndreas Gruenbacher #endif 2371da177e4SLinus Torvalds } 2381da177e4SLinus Torvalds return mnt; 23988b38782SLi Zefan 240d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 241d3ef3d73Snpiggin@suse.de out_free_devname: 242d3ef3d73Snpiggin@suse.de kfree(mnt->mnt_devname); 243d3ef3d73Snpiggin@suse.de #endif 24488b38782SLi Zefan out_free_id: 24588b38782SLi Zefan mnt_free_id(mnt); 24688b38782SLi Zefan out_free_cache: 24788b38782SLi Zefan kmem_cache_free(mnt_cache, mnt); 24888b38782SLi Zefan return NULL; 2491da177e4SLinus Torvalds } 2501da177e4SLinus Torvalds 2518366025eSDave Hansen /* 2528366025eSDave Hansen * Most r/o checks on a fs are for operations that take 2538366025eSDave Hansen * discrete amounts of time, like a write() or unlink(). 2548366025eSDave Hansen * We must keep track of when those operations start 2558366025eSDave Hansen * (for permission checks) and when they end, so that 2568366025eSDave Hansen * we can determine when writes are able to occur to 2578366025eSDave Hansen * a filesystem. 2588366025eSDave Hansen */ 2593d733633SDave Hansen /* 2603d733633SDave Hansen * __mnt_is_readonly: check whether a mount is read-only 2613d733633SDave Hansen * @mnt: the mount to check for its write status 2623d733633SDave Hansen * 2633d733633SDave Hansen * This shouldn't be used directly ouside of the VFS. 2643d733633SDave Hansen * It does not guarantee that the filesystem will stay 2653d733633SDave Hansen * r/w, just that it is right *now*. This can not and 2663d733633SDave Hansen * should not be used in place of IS_RDONLY(inode). 2673d733633SDave Hansen * mnt_want/drop_write() will _keep_ the filesystem 2683d733633SDave Hansen * r/w. 2693d733633SDave Hansen */ 2703d733633SDave Hansen int __mnt_is_readonly(struct vfsmount *mnt) 2713d733633SDave Hansen { 2722e4b7fcdSDave Hansen if (mnt->mnt_flags & MNT_READONLY) 2732e4b7fcdSDave Hansen return 1; 2742e4b7fcdSDave Hansen if (mnt->mnt_sb->s_flags & MS_RDONLY) 2752e4b7fcdSDave Hansen return 1; 2762e4b7fcdSDave Hansen return 0; 2773d733633SDave Hansen } 2783d733633SDave Hansen EXPORT_SYMBOL_GPL(__mnt_is_readonly); 2793d733633SDave Hansen 280c6653a83SNick Piggin static inline void mnt_inc_writers(struct vfsmount *mnt) 2813d733633SDave Hansen { 282d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 283b3e19d92SNick Piggin this_cpu_inc(mnt->mnt_pcp->mnt_writers); 284d3ef3d73Snpiggin@suse.de #else 285d3ef3d73Snpiggin@suse.de mnt->mnt_writers++; 286d3ef3d73Snpiggin@suse.de #endif 2873d733633SDave Hansen } 2883d733633SDave Hansen 289c6653a83SNick Piggin static inline void mnt_dec_writers(struct vfsmount *mnt) 2903d733633SDave Hansen { 291d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 292b3e19d92SNick Piggin this_cpu_dec(mnt->mnt_pcp->mnt_writers); 293d3ef3d73Snpiggin@suse.de #else 294d3ef3d73Snpiggin@suse.de mnt->mnt_writers--; 295d3ef3d73Snpiggin@suse.de #endif 296d3ef3d73Snpiggin@suse.de } 297d3ef3d73Snpiggin@suse.de 298c6653a83SNick Piggin static unsigned int mnt_get_writers(struct vfsmount *mnt) 299d3ef3d73Snpiggin@suse.de { 300d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 301d3ef3d73Snpiggin@suse.de unsigned int count = 0; 3023d733633SDave Hansen int cpu; 3033d733633SDave Hansen 3043d733633SDave Hansen for_each_possible_cpu(cpu) { 305b3e19d92SNick Piggin count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers; 3063d733633SDave Hansen } 3073d733633SDave Hansen 308d3ef3d73Snpiggin@suse.de return count; 309d3ef3d73Snpiggin@suse.de #else 310d3ef3d73Snpiggin@suse.de return mnt->mnt_writers; 311d3ef3d73Snpiggin@suse.de #endif 3123d733633SDave Hansen } 3133d733633SDave Hansen 3143d733633SDave Hansen /* 3153d733633SDave Hansen * Most r/o checks on a fs are for operations that take 3163d733633SDave Hansen * discrete amounts of time, like a write() or unlink(). 3173d733633SDave Hansen * We must keep track of when those operations start 3183d733633SDave Hansen * (for permission checks) and when they end, so that 3193d733633SDave Hansen * we can determine when writes are able to occur to 3203d733633SDave Hansen * a filesystem. 3213d733633SDave Hansen */ 3228366025eSDave Hansen /** 3238366025eSDave Hansen * mnt_want_write - get write access to a mount 3248366025eSDave Hansen * @mnt: the mount on which to take a write 3258366025eSDave Hansen * 3268366025eSDave Hansen * This tells the low-level filesystem that a write is 3278366025eSDave Hansen * about to be performed to it, and makes sure that 3288366025eSDave Hansen * writes are allowed before returning success. When 3298366025eSDave Hansen * the write operation is finished, mnt_drop_write() 3308366025eSDave Hansen * must be called. This is effectively a refcount. 3318366025eSDave Hansen */ 3328366025eSDave Hansen int mnt_want_write(struct vfsmount *mnt) 3338366025eSDave Hansen { 3343d733633SDave Hansen int ret = 0; 3353d733633SDave Hansen 336d3ef3d73Snpiggin@suse.de preempt_disable(); 337c6653a83SNick Piggin mnt_inc_writers(mnt); 338d3ef3d73Snpiggin@suse.de /* 339c6653a83SNick Piggin * The store to mnt_inc_writers must be visible before we pass 340d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD loop below, so that the slowpath can see our 341d3ef3d73Snpiggin@suse.de * incremented count after it has set MNT_WRITE_HOLD. 342d3ef3d73Snpiggin@suse.de */ 343d3ef3d73Snpiggin@suse.de smp_mb(); 344d3ef3d73Snpiggin@suse.de while (mnt->mnt_flags & MNT_WRITE_HOLD) 345d3ef3d73Snpiggin@suse.de cpu_relax(); 346d3ef3d73Snpiggin@suse.de /* 347d3ef3d73Snpiggin@suse.de * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will 348d3ef3d73Snpiggin@suse.de * be set to match its requirements. So we must not load that until 349d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD is cleared. 350d3ef3d73Snpiggin@suse.de */ 351d3ef3d73Snpiggin@suse.de smp_rmb(); 3523d733633SDave Hansen if (__mnt_is_readonly(mnt)) { 353c6653a83SNick Piggin mnt_dec_writers(mnt); 3543d733633SDave Hansen ret = -EROFS; 3553d733633SDave Hansen goto out; 3563d733633SDave Hansen } 3573d733633SDave Hansen out: 358d3ef3d73Snpiggin@suse.de preempt_enable(); 3593d733633SDave Hansen return ret; 3608366025eSDave Hansen } 3618366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_want_write); 3628366025eSDave Hansen 3638366025eSDave Hansen /** 36496029c4eSnpiggin@suse.de * mnt_clone_write - get write access to a mount 36596029c4eSnpiggin@suse.de * @mnt: the mount on which to take a write 36696029c4eSnpiggin@suse.de * 36796029c4eSnpiggin@suse.de * This is effectively like mnt_want_write, except 36896029c4eSnpiggin@suse.de * it must only be used to take an extra write reference 36996029c4eSnpiggin@suse.de * on a mountpoint that we already know has a write reference 37096029c4eSnpiggin@suse.de * on it. This allows some optimisation. 37196029c4eSnpiggin@suse.de * 37296029c4eSnpiggin@suse.de * After finished, mnt_drop_write must be called as usual to 37396029c4eSnpiggin@suse.de * drop the reference. 37496029c4eSnpiggin@suse.de */ 37596029c4eSnpiggin@suse.de int mnt_clone_write(struct vfsmount *mnt) 37696029c4eSnpiggin@suse.de { 37796029c4eSnpiggin@suse.de /* superblock may be r/o */ 37896029c4eSnpiggin@suse.de if (__mnt_is_readonly(mnt)) 37996029c4eSnpiggin@suse.de return -EROFS; 38096029c4eSnpiggin@suse.de preempt_disable(); 381c6653a83SNick Piggin mnt_inc_writers(mnt); 38296029c4eSnpiggin@suse.de preempt_enable(); 38396029c4eSnpiggin@suse.de return 0; 38496029c4eSnpiggin@suse.de } 38596029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_clone_write); 38696029c4eSnpiggin@suse.de 38796029c4eSnpiggin@suse.de /** 38896029c4eSnpiggin@suse.de * mnt_want_write_file - get write access to a file's mount 38996029c4eSnpiggin@suse.de * @file: the file who's mount on which to take a write 39096029c4eSnpiggin@suse.de * 39196029c4eSnpiggin@suse.de * This is like mnt_want_write, but it takes a file and can 39296029c4eSnpiggin@suse.de * do some optimisations if the file is open for write already 39396029c4eSnpiggin@suse.de */ 39496029c4eSnpiggin@suse.de int mnt_want_write_file(struct file *file) 39596029c4eSnpiggin@suse.de { 3962d8dd38aSOGAWA Hirofumi struct inode *inode = file->f_dentry->d_inode; 3972d8dd38aSOGAWA Hirofumi if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode)) 39896029c4eSnpiggin@suse.de return mnt_want_write(file->f_path.mnt); 39996029c4eSnpiggin@suse.de else 40096029c4eSnpiggin@suse.de return mnt_clone_write(file->f_path.mnt); 40196029c4eSnpiggin@suse.de } 40296029c4eSnpiggin@suse.de EXPORT_SYMBOL_GPL(mnt_want_write_file); 40396029c4eSnpiggin@suse.de 40496029c4eSnpiggin@suse.de /** 4058366025eSDave Hansen * mnt_drop_write - give up write access to a mount 4068366025eSDave Hansen * @mnt: the mount on which to give up write access 4078366025eSDave Hansen * 4088366025eSDave Hansen * Tells the low-level filesystem that we are done 4098366025eSDave Hansen * performing writes to it. Must be matched with 4108366025eSDave Hansen * mnt_want_write() call above. 4118366025eSDave Hansen */ 4128366025eSDave Hansen void mnt_drop_write(struct vfsmount *mnt) 4138366025eSDave Hansen { 414d3ef3d73Snpiggin@suse.de preempt_disable(); 415c6653a83SNick Piggin mnt_dec_writers(mnt); 416d3ef3d73Snpiggin@suse.de preempt_enable(); 4178366025eSDave Hansen } 4188366025eSDave Hansen EXPORT_SYMBOL_GPL(mnt_drop_write); 4198366025eSDave Hansen 4202e4b7fcdSDave Hansen static int mnt_make_readonly(struct vfsmount *mnt) 4218366025eSDave Hansen { 4223d733633SDave Hansen int ret = 0; 4233d733633SDave Hansen 42499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 425d3ef3d73Snpiggin@suse.de mnt->mnt_flags |= MNT_WRITE_HOLD; 426d3ef3d73Snpiggin@suse.de /* 427d3ef3d73Snpiggin@suse.de * After storing MNT_WRITE_HOLD, we'll read the counters. This store 428d3ef3d73Snpiggin@suse.de * should be visible before we do. 429d3ef3d73Snpiggin@suse.de */ 430d3ef3d73Snpiggin@suse.de smp_mb(); 431d3ef3d73Snpiggin@suse.de 432d3ef3d73Snpiggin@suse.de /* 433d3ef3d73Snpiggin@suse.de * With writers on hold, if this value is zero, then there are 434d3ef3d73Snpiggin@suse.de * definitely no active writers (although held writers may subsequently 435d3ef3d73Snpiggin@suse.de * increment the count, they'll have to wait, and decrement it after 436d3ef3d73Snpiggin@suse.de * seeing MNT_READONLY). 437d3ef3d73Snpiggin@suse.de * 438d3ef3d73Snpiggin@suse.de * It is OK to have counter incremented on one CPU and decremented on 439d3ef3d73Snpiggin@suse.de * another: the sum will add up correctly. The danger would be when we 440d3ef3d73Snpiggin@suse.de * sum up each counter, if we read a counter before it is incremented, 441d3ef3d73Snpiggin@suse.de * but then read another CPU's count which it has been subsequently 442d3ef3d73Snpiggin@suse.de * decremented from -- we would see more decrements than we should. 443d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD protects against this scenario, because 444d3ef3d73Snpiggin@suse.de * mnt_want_write first increments count, then smp_mb, then spins on 445d3ef3d73Snpiggin@suse.de * MNT_WRITE_HOLD, so it can't be decremented by another CPU while 446d3ef3d73Snpiggin@suse.de * we're counting up here. 447d3ef3d73Snpiggin@suse.de */ 448c6653a83SNick Piggin if (mnt_get_writers(mnt) > 0) 449d3ef3d73Snpiggin@suse.de ret = -EBUSY; 450d3ef3d73Snpiggin@suse.de else 4512e4b7fcdSDave Hansen mnt->mnt_flags |= MNT_READONLY; 452d3ef3d73Snpiggin@suse.de /* 453d3ef3d73Snpiggin@suse.de * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers 454d3ef3d73Snpiggin@suse.de * that become unheld will see MNT_READONLY. 455d3ef3d73Snpiggin@suse.de */ 456d3ef3d73Snpiggin@suse.de smp_wmb(); 457d3ef3d73Snpiggin@suse.de mnt->mnt_flags &= ~MNT_WRITE_HOLD; 45899b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 4593d733633SDave Hansen return ret; 4603d733633SDave Hansen } 4618366025eSDave Hansen 4622e4b7fcdSDave Hansen static void __mnt_unmake_readonly(struct vfsmount *mnt) 4632e4b7fcdSDave Hansen { 46499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 4652e4b7fcdSDave Hansen mnt->mnt_flags &= ~MNT_READONLY; 46699b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 4672e4b7fcdSDave Hansen } 4682e4b7fcdSDave Hansen 469a3ec947cSSukadev Bhattiprolu void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb) 470454e2398SDavid Howells { 471454e2398SDavid Howells mnt->mnt_sb = sb; 472454e2398SDavid Howells mnt->mnt_root = dget(sb->s_root); 473454e2398SDavid Howells } 474454e2398SDavid Howells 475454e2398SDavid Howells EXPORT_SYMBOL(simple_set_mnt); 476454e2398SDavid Howells 4771da177e4SLinus Torvalds void free_vfsmnt(struct vfsmount *mnt) 4781da177e4SLinus Torvalds { 4791da177e4SLinus Torvalds kfree(mnt->mnt_devname); 48073cd49ecSMiklos Szeredi mnt_free_id(mnt); 481d3ef3d73Snpiggin@suse.de #ifdef CONFIG_SMP 482b3e19d92SNick Piggin free_percpu(mnt->mnt_pcp); 483d3ef3d73Snpiggin@suse.de #endif 4841da177e4SLinus Torvalds kmem_cache_free(mnt_cache, mnt); 4851da177e4SLinus Torvalds } 4861da177e4SLinus Torvalds 4871da177e4SLinus Torvalds /* 488a05964f3SRam Pai * find the first or last mount at @dentry on vfsmount @mnt depending on 489a05964f3SRam Pai * @dir. If @dir is set return the first mount else return the last mount. 49099b7db7bSNick Piggin * vfsmount_lock must be held for read or write. 4911da177e4SLinus Torvalds */ 492a05964f3SRam Pai struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry, 493a05964f3SRam Pai int dir) 4941da177e4SLinus Torvalds { 4951da177e4SLinus Torvalds struct list_head *head = mount_hashtable + hash(mnt, dentry); 4961da177e4SLinus Torvalds struct list_head *tmp = head; 4971da177e4SLinus Torvalds struct vfsmount *p, *found = NULL; 4981da177e4SLinus Torvalds 4991da177e4SLinus Torvalds for (;;) { 500a05964f3SRam Pai tmp = dir ? tmp->next : tmp->prev; 5011da177e4SLinus Torvalds p = NULL; 5021da177e4SLinus Torvalds if (tmp == head) 5031da177e4SLinus Torvalds break; 5041da177e4SLinus Torvalds p = list_entry(tmp, struct vfsmount, mnt_hash); 5051da177e4SLinus Torvalds if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) { 506a05964f3SRam Pai found = p; 5071da177e4SLinus Torvalds break; 5081da177e4SLinus Torvalds } 5091da177e4SLinus Torvalds } 5101da177e4SLinus Torvalds return found; 5111da177e4SLinus Torvalds } 5121da177e4SLinus Torvalds 513a05964f3SRam Pai /* 514a05964f3SRam Pai * lookup_mnt increments the ref count before returning 515a05964f3SRam Pai * the vfsmount struct. 516a05964f3SRam Pai */ 5171c755af4SAl Viro struct vfsmount *lookup_mnt(struct path *path) 518a05964f3SRam Pai { 519a05964f3SRam Pai struct vfsmount *child_mnt; 52099b7db7bSNick Piggin 52199b7db7bSNick Piggin br_read_lock(vfsmount_lock); 5221c755af4SAl Viro if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1))) 523a05964f3SRam Pai mntget(child_mnt); 52499b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 525a05964f3SRam Pai return child_mnt; 526a05964f3SRam Pai } 527a05964f3SRam Pai 5281da177e4SLinus Torvalds static inline int check_mnt(struct vfsmount *mnt) 5291da177e4SLinus Torvalds { 5306b3286edSKirill Korotaev return mnt->mnt_ns == current->nsproxy->mnt_ns; 5311da177e4SLinus Torvalds } 5321da177e4SLinus Torvalds 53399b7db7bSNick Piggin /* 53499b7db7bSNick Piggin * vfsmount lock must be held for write 53599b7db7bSNick Piggin */ 5366b3286edSKirill Korotaev static void touch_mnt_namespace(struct mnt_namespace *ns) 5375addc5ddSAl Viro { 5385addc5ddSAl Viro if (ns) { 5395addc5ddSAl Viro ns->event = ++event; 5405addc5ddSAl Viro wake_up_interruptible(&ns->poll); 5415addc5ddSAl Viro } 5425addc5ddSAl Viro } 5435addc5ddSAl Viro 54499b7db7bSNick Piggin /* 54599b7db7bSNick Piggin * vfsmount lock must be held for write 54699b7db7bSNick Piggin */ 5476b3286edSKirill Korotaev static void __touch_mnt_namespace(struct mnt_namespace *ns) 5485addc5ddSAl Viro { 5495addc5ddSAl Viro if (ns && ns->event != event) { 5505addc5ddSAl Viro ns->event = event; 5515addc5ddSAl Viro wake_up_interruptible(&ns->poll); 5525addc5ddSAl Viro } 5535addc5ddSAl Viro } 5545addc5ddSAl Viro 55599b7db7bSNick Piggin /* 5565f57cbccSNick Piggin * Clear dentry's mounted state if it has no remaining mounts. 5575f57cbccSNick Piggin * vfsmount_lock must be held for write. 5585f57cbccSNick Piggin */ 5595f57cbccSNick Piggin static void dentry_reset_mounted(struct vfsmount *mnt, struct dentry *dentry) 5605f57cbccSNick Piggin { 5615f57cbccSNick Piggin unsigned u; 5625f57cbccSNick Piggin 5635f57cbccSNick Piggin for (u = 0; u < HASH_SIZE; u++) { 5645f57cbccSNick Piggin struct vfsmount *p; 5655f57cbccSNick Piggin 5665f57cbccSNick Piggin list_for_each_entry(p, &mount_hashtable[u], mnt_hash) { 5675f57cbccSNick Piggin if (p->mnt_mountpoint == dentry) 5685f57cbccSNick Piggin return; 5695f57cbccSNick Piggin } 5705f57cbccSNick Piggin } 5715f57cbccSNick Piggin spin_lock(&dentry->d_lock); 5725f57cbccSNick Piggin dentry->d_flags &= ~DCACHE_MOUNTED; 5735f57cbccSNick Piggin spin_unlock(&dentry->d_lock); 5745f57cbccSNick Piggin } 5755f57cbccSNick Piggin 5765f57cbccSNick Piggin /* 57799b7db7bSNick Piggin * vfsmount lock must be held for write 57899b7db7bSNick Piggin */ 5791a390689SAl Viro static void detach_mnt(struct vfsmount *mnt, struct path *old_path) 5801da177e4SLinus Torvalds { 5811a390689SAl Viro old_path->dentry = mnt->mnt_mountpoint; 5821a390689SAl Viro old_path->mnt = mnt->mnt_parent; 5831da177e4SLinus Torvalds mnt->mnt_parent = mnt; 5841da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 5851da177e4SLinus Torvalds list_del_init(&mnt->mnt_child); 5861da177e4SLinus Torvalds list_del_init(&mnt->mnt_hash); 5875f57cbccSNick Piggin dentry_reset_mounted(old_path->mnt, old_path->dentry); 5881da177e4SLinus Torvalds } 5891da177e4SLinus Torvalds 59099b7db7bSNick Piggin /* 59199b7db7bSNick Piggin * vfsmount lock must be held for write 59299b7db7bSNick Piggin */ 593b90fa9aeSRam Pai void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry, 594b90fa9aeSRam Pai struct vfsmount *child_mnt) 595b90fa9aeSRam Pai { 596b90fa9aeSRam Pai child_mnt->mnt_parent = mntget(mnt); 597b90fa9aeSRam Pai child_mnt->mnt_mountpoint = dget(dentry); 5985f57cbccSNick Piggin spin_lock(&dentry->d_lock); 5995f57cbccSNick Piggin dentry->d_flags |= DCACHE_MOUNTED; 6005f57cbccSNick Piggin spin_unlock(&dentry->d_lock); 601b90fa9aeSRam Pai } 602b90fa9aeSRam Pai 60399b7db7bSNick Piggin /* 60499b7db7bSNick Piggin * vfsmount lock must be held for write 60599b7db7bSNick Piggin */ 6061a390689SAl Viro static void attach_mnt(struct vfsmount *mnt, struct path *path) 6071da177e4SLinus Torvalds { 6081a390689SAl Viro mnt_set_mountpoint(path->mnt, path->dentry, mnt); 609b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 6101a390689SAl Viro hash(path->mnt, path->dentry)); 6111a390689SAl Viro list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts); 612b90fa9aeSRam Pai } 613b90fa9aeSRam Pai 614b90fa9aeSRam Pai /* 61599b7db7bSNick Piggin * vfsmount lock must be held for write 616b90fa9aeSRam Pai */ 617b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 618b90fa9aeSRam Pai { 619b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 620b90fa9aeSRam Pai struct vfsmount *m; 621b90fa9aeSRam Pai LIST_HEAD(head); 6226b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 623b90fa9aeSRam Pai 624b90fa9aeSRam Pai BUG_ON(parent == mnt); 625b90fa9aeSRam Pai 626b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 627b90fa9aeSRam Pai list_for_each_entry(m, &head, mnt_list) 6286b3286edSKirill Korotaev m->mnt_ns = n; 629b90fa9aeSRam Pai list_splice(&head, n->list.prev); 630b90fa9aeSRam Pai 631b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 632b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 633b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 6346b3286edSKirill Korotaev touch_mnt_namespace(n); 6351da177e4SLinus Torvalds } 6361da177e4SLinus Torvalds 6371da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 6381da177e4SLinus Torvalds { 6391da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 6401da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 6411da177e4SLinus Torvalds while (1) { 6421da177e4SLinus Torvalds if (p == root) 6431da177e4SLinus Torvalds return NULL; 6441da177e4SLinus Torvalds next = p->mnt_child.next; 6451da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 6461da177e4SLinus Torvalds break; 6471da177e4SLinus Torvalds p = p->mnt_parent; 6481da177e4SLinus Torvalds } 6491da177e4SLinus Torvalds } 6501da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 6511da177e4SLinus Torvalds } 6521da177e4SLinus Torvalds 6539676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 6549676f0c6SRam Pai { 6559676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 6569676f0c6SRam Pai while (prev != &p->mnt_mounts) { 6579676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 6589676f0c6SRam Pai prev = p->mnt_mounts.prev; 6599676f0c6SRam Pai } 6609676f0c6SRam Pai return p; 6619676f0c6SRam Pai } 6629676f0c6SRam Pai 66336341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 66436341f64SRam Pai int flag) 6651da177e4SLinus Torvalds { 6661da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 6671da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds if (mnt) { 670719f5d7fSMiklos Szeredi if (flag & (CL_SLAVE | CL_PRIVATE)) 671719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; /* not a peer of original */ 672719f5d7fSMiklos Szeredi else 673719f5d7fSMiklos Szeredi mnt->mnt_group_id = old->mnt_group_id; 674719f5d7fSMiklos Szeredi 675719f5d7fSMiklos Szeredi if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 676719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(mnt); 677719f5d7fSMiklos Szeredi if (err) 678719f5d7fSMiklos Szeredi goto out_free; 679719f5d7fSMiklos Szeredi } 680719f5d7fSMiklos Szeredi 681be1a16a0SMiklos Szeredi mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD; 6821da177e4SLinus Torvalds atomic_inc(&sb->s_active); 6831da177e4SLinus Torvalds mnt->mnt_sb = sb; 6841da177e4SLinus Torvalds mnt->mnt_root = dget(root); 6851da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 6861da177e4SLinus Torvalds mnt->mnt_parent = mnt; 687b90fa9aeSRam Pai 6885afe0022SRam Pai if (flag & CL_SLAVE) { 6895afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 6905afe0022SRam Pai mnt->mnt_master = old; 6915afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 6928aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 693796a6b52SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 694b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 6955afe0022SRam Pai if (IS_MNT_SLAVE(old)) 6965afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 6975afe0022SRam Pai mnt->mnt_master = old->mnt_master; 6985afe0022SRam Pai } 699b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 700b90fa9aeSRam Pai set_mnt_shared(mnt); 7011da177e4SLinus Torvalds 7021da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 7031da177e4SLinus Torvalds * as the original if that was on one */ 70436341f64SRam Pai if (flag & CL_EXPIRE) { 70555e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 70655e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 7071da177e4SLinus Torvalds } 70836341f64SRam Pai } 7091da177e4SLinus Torvalds return mnt; 710719f5d7fSMiklos Szeredi 711719f5d7fSMiklos Szeredi out_free: 712719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 713719f5d7fSMiklos Szeredi return NULL; 7141da177e4SLinus Torvalds } 7151da177e4SLinus Torvalds 716b3e19d92SNick Piggin static inline void mntfree(struct vfsmount *mnt) 7171da177e4SLinus Torvalds { 7181da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 719b3e19d92SNick Piggin 7203d733633SDave Hansen /* 7213d733633SDave Hansen * This probably indicates that somebody messed 7223d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 7233d733633SDave Hansen * happens, the filesystem was probably unable 7243d733633SDave Hansen * to make r/w->r/o transitions. 7253d733633SDave Hansen */ 726d3ef3d73Snpiggin@suse.de /* 727b3e19d92SNick Piggin * The locking used to deal with mnt_count decrement provides barriers, 728b3e19d92SNick Piggin * so mnt_get_writers() below is safe. 729d3ef3d73Snpiggin@suse.de */ 730c6653a83SNick Piggin WARN_ON(mnt_get_writers(mnt)); 731ca9c726eSAndreas Gruenbacher fsnotify_vfsmount_delete(mnt); 7321da177e4SLinus Torvalds dput(mnt->mnt_root); 7331da177e4SLinus Torvalds free_vfsmnt(mnt); 7341da177e4SLinus Torvalds deactivate_super(sb); 7351da177e4SLinus Torvalds } 7361da177e4SLinus Torvalds 737b3e19d92SNick Piggin #ifdef CONFIG_SMP 738b3e19d92SNick Piggin static inline void __mntput(struct vfsmount *mnt, int longrefs) 7397b7b1aceSAl Viro { 740b3e19d92SNick Piggin if (!longrefs) { 741b3e19d92SNick Piggin put_again: 742b3e19d92SNick Piggin br_read_lock(vfsmount_lock); 743b3e19d92SNick Piggin if (likely(atomic_read(&mnt->mnt_longrefs))) { 744b3e19d92SNick Piggin mnt_dec_count(mnt); 745b3e19d92SNick Piggin br_read_unlock(vfsmount_lock); 74699b7db7bSNick Piggin return; 747b3e19d92SNick Piggin } 748b3e19d92SNick Piggin br_read_unlock(vfsmount_lock); 749b3e19d92SNick Piggin } else { 750b3e19d92SNick Piggin BUG_ON(!atomic_read(&mnt->mnt_longrefs)); 751b3e19d92SNick Piggin if (atomic_add_unless(&mnt->mnt_longrefs, -1, 1)) 752b3e19d92SNick Piggin return; 753b3e19d92SNick Piggin } 754b3e19d92SNick Piggin 75599b7db7bSNick Piggin br_write_lock(vfsmount_lock); 756b3e19d92SNick Piggin if (!longrefs) 757b3e19d92SNick Piggin mnt_dec_count(mnt); 758b3e19d92SNick Piggin else 759b3e19d92SNick Piggin atomic_dec(&mnt->mnt_longrefs); 760b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 76199b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 76299b7db7bSNick Piggin return; 76399b7db7bSNick Piggin } 764b3e19d92SNick Piggin if (unlikely(mnt->mnt_pinned)) { 765b3e19d92SNick Piggin mnt_add_count(mnt, mnt->mnt_pinned + 1); 7667b7b1aceSAl Viro mnt->mnt_pinned = 0; 76799b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 7687b7b1aceSAl Viro acct_auto_close_mnt(mnt); 769b3e19d92SNick Piggin goto put_again; 7707b7b1aceSAl Viro } 771b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 772b3e19d92SNick Piggin mntfree(mnt); 773b3e19d92SNick Piggin } 774b3e19d92SNick Piggin #else 775b3e19d92SNick Piggin static inline void __mntput(struct vfsmount *mnt, int longrefs) 776b3e19d92SNick Piggin { 777b3e19d92SNick Piggin put_again: 778b3e19d92SNick Piggin mnt_dec_count(mnt); 779b3e19d92SNick Piggin if (likely(mnt_get_count(mnt))) 780b3e19d92SNick Piggin return; 781b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 782b3e19d92SNick Piggin if (unlikely(mnt->mnt_pinned)) { 783b3e19d92SNick Piggin mnt_add_count(mnt, mnt->mnt_pinned + 1); 784b3e19d92SNick Piggin mnt->mnt_pinned = 0; 785b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 786b3e19d92SNick Piggin acct_auto_close_mnt(mnt); 787b3e19d92SNick Piggin goto put_again; 788b3e19d92SNick Piggin } 789b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 790b3e19d92SNick Piggin mntfree(mnt); 791b3e19d92SNick Piggin } 792b3e19d92SNick Piggin #endif 793b3e19d92SNick Piggin 794b3e19d92SNick Piggin static void mntput_no_expire(struct vfsmount *mnt) 795b3e19d92SNick Piggin { 796b3e19d92SNick Piggin __mntput(mnt, 0); 797b3e19d92SNick Piggin } 798b3e19d92SNick Piggin 799b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 800b3e19d92SNick Piggin { 801b3e19d92SNick Piggin if (mnt) { 802b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 803b3e19d92SNick Piggin if (unlikely(mnt->mnt_expiry_mark)) 804b3e19d92SNick Piggin mnt->mnt_expiry_mark = 0; 805b3e19d92SNick Piggin __mntput(mnt, 0); 806b3e19d92SNick Piggin } 807b3e19d92SNick Piggin } 808b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 809b3e19d92SNick Piggin 810b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 811b3e19d92SNick Piggin { 812b3e19d92SNick Piggin if (mnt) 813b3e19d92SNick Piggin mnt_inc_count(mnt); 814b3e19d92SNick Piggin return mnt; 815b3e19d92SNick Piggin } 816b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 817b3e19d92SNick Piggin 818b3e19d92SNick Piggin void mntput_long(struct vfsmount *mnt) 819b3e19d92SNick Piggin { 820b3e19d92SNick Piggin #ifdef CONFIG_SMP 821b3e19d92SNick Piggin if (mnt) { 822b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 823b3e19d92SNick Piggin if (unlikely(mnt->mnt_expiry_mark)) 824b3e19d92SNick Piggin mnt->mnt_expiry_mark = 0; 825b3e19d92SNick Piggin __mntput(mnt, 1); 826b3e19d92SNick Piggin } 827b3e19d92SNick Piggin #else 828b3e19d92SNick Piggin mntput(mnt); 829b3e19d92SNick Piggin #endif 830b3e19d92SNick Piggin } 831b3e19d92SNick Piggin EXPORT_SYMBOL(mntput_long); 832b3e19d92SNick Piggin 833b3e19d92SNick Piggin struct vfsmount *mntget_long(struct vfsmount *mnt) 834b3e19d92SNick Piggin { 835b3e19d92SNick Piggin #ifdef CONFIG_SMP 836b3e19d92SNick Piggin if (mnt) 837b3e19d92SNick Piggin atomic_inc(&mnt->mnt_longrefs); 838b3e19d92SNick Piggin return mnt; 839b3e19d92SNick Piggin #else 840b3e19d92SNick Piggin return mntget(mnt); 841b3e19d92SNick Piggin #endif 842b3e19d92SNick Piggin } 843b3e19d92SNick Piggin EXPORT_SYMBOL(mntget_long); 8447b7b1aceSAl Viro 8457b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 8467b7b1aceSAl Viro { 84799b7db7bSNick Piggin br_write_lock(vfsmount_lock); 8487b7b1aceSAl Viro mnt->mnt_pinned++; 84999b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 8507b7b1aceSAl Viro } 8517b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 8527b7b1aceSAl Viro 8537b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 8547b7b1aceSAl Viro { 85599b7db7bSNick Piggin br_write_lock(vfsmount_lock); 8567b7b1aceSAl Viro if (mnt->mnt_pinned) { 857b3e19d92SNick Piggin mnt_inc_count(mnt); 8587b7b1aceSAl Viro mnt->mnt_pinned--; 8597b7b1aceSAl Viro } 86099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 8617b7b1aceSAl Viro } 8627b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 8631da177e4SLinus Torvalds 864b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 865b3b304a2SMiklos Szeredi { 866b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 867b3b304a2SMiklos Szeredi } 868b3b304a2SMiklos Szeredi 869b3b304a2SMiklos Szeredi /* 870b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 871b3b304a2SMiklos Szeredi * implement more complex mount option showing. 872b3b304a2SMiklos Szeredi * 873b3b304a2SMiklos Szeredi * See also save_mount_options(). 874b3b304a2SMiklos Szeredi */ 875b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 876b3b304a2SMiklos Szeredi { 8772a32cebdSAl Viro const char *options; 8782a32cebdSAl Viro 8792a32cebdSAl Viro rcu_read_lock(); 8802a32cebdSAl Viro options = rcu_dereference(mnt->mnt_sb->s_options); 881b3b304a2SMiklos Szeredi 882b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 883b3b304a2SMiklos Szeredi seq_putc(m, ','); 884b3b304a2SMiklos Szeredi mangle(m, options); 885b3b304a2SMiklos Szeredi } 8862a32cebdSAl Viro rcu_read_unlock(); 887b3b304a2SMiklos Szeredi 888b3b304a2SMiklos Szeredi return 0; 889b3b304a2SMiklos Szeredi } 890b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 891b3b304a2SMiklos Szeredi 892b3b304a2SMiklos Szeredi /* 893b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 894b3b304a2SMiklos Szeredi * called from the fill_super() callback. 895b3b304a2SMiklos Szeredi * 896b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 897b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 898b3b304a2SMiklos Szeredi * remount fails. 899b3b304a2SMiklos Szeredi * 900b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 901b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 902b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 903b3b304a2SMiklos Szeredi * any more. 904b3b304a2SMiklos Szeredi */ 905b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 906b3b304a2SMiklos Szeredi { 9072a32cebdSAl Viro BUG_ON(sb->s_options); 9082a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 909b3b304a2SMiklos Szeredi } 910b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 911b3b304a2SMiklos Szeredi 9122a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 9132a32cebdSAl Viro { 9142a32cebdSAl Viro char *old = sb->s_options; 9152a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 9162a32cebdSAl Viro if (old) { 9172a32cebdSAl Viro synchronize_rcu(); 9182a32cebdSAl Viro kfree(old); 9192a32cebdSAl Viro } 9202a32cebdSAl Viro } 9212a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 9222a32cebdSAl Viro 923a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 9241da177e4SLinus Torvalds /* iterator */ 9251da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 9261da177e4SLinus Torvalds { 927a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 9281da177e4SLinus Torvalds 929390c6843SRam Pai down_read(&namespace_sem); 930a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 9311da177e4SLinus Torvalds } 9321da177e4SLinus Torvalds 9331da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 9341da177e4SLinus Torvalds { 935a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 936b0765fb8SPavel Emelianov 937a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 9381da177e4SLinus Torvalds } 9391da177e4SLinus Torvalds 9401da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 9411da177e4SLinus Torvalds { 942390c6843SRam Pai up_read(&namespace_sem); 9431da177e4SLinus Torvalds } 9441da177e4SLinus Torvalds 9459f5596afSAl Viro int mnt_had_events(struct proc_mounts *p) 9469f5596afSAl Viro { 9479f5596afSAl Viro struct mnt_namespace *ns = p->ns; 9489f5596afSAl Viro int res = 0; 9499f5596afSAl Viro 95099b7db7bSNick Piggin br_read_lock(vfsmount_lock); 9519f5596afSAl Viro if (p->event != ns->event) { 9529f5596afSAl Viro p->event = ns->event; 9539f5596afSAl Viro res = 1; 9549f5596afSAl Viro } 95599b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 9569f5596afSAl Viro 9579f5596afSAl Viro return res; 9589f5596afSAl Viro } 9599f5596afSAl Viro 9602d4d4864SRam Pai struct proc_fs_info { 9611da177e4SLinus Torvalds int flag; 9622d4d4864SRam Pai const char *str; 9632d4d4864SRam Pai }; 9642d4d4864SRam Pai 9652069f457SEric Paris static int show_sb_opts(struct seq_file *m, struct super_block *sb) 9662d4d4864SRam Pai { 9672d4d4864SRam Pai static const struct proc_fs_info fs_info[] = { 9681da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 9691da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 9701da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 9711da177e4SLinus Torvalds { 0, NULL } 9721da177e4SLinus Torvalds }; 9732d4d4864SRam Pai const struct proc_fs_info *fs_infop; 9742d4d4864SRam Pai 9752d4d4864SRam Pai for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 9762d4d4864SRam Pai if (sb->s_flags & fs_infop->flag) 9772d4d4864SRam Pai seq_puts(m, fs_infop->str); 9782d4d4864SRam Pai } 9792069f457SEric Paris 9802069f457SEric Paris return security_sb_show_options(m, sb); 9812d4d4864SRam Pai } 9822d4d4864SRam Pai 9832d4d4864SRam Pai static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) 9842d4d4864SRam Pai { 9852d4d4864SRam Pai static const struct proc_fs_info mnt_info[] = { 9861da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 9871da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 9881da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 989fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 990fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 99147ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 9921da177e4SLinus Torvalds { 0, NULL } 9931da177e4SLinus Torvalds }; 9942d4d4864SRam Pai const struct proc_fs_info *fs_infop; 9952d4d4864SRam Pai 9962d4d4864SRam Pai for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 9972d4d4864SRam Pai if (mnt->mnt_flags & fs_infop->flag) 9982d4d4864SRam Pai seq_puts(m, fs_infop->str); 9992d4d4864SRam Pai } 10002d4d4864SRam Pai } 10012d4d4864SRam Pai 10022d4d4864SRam Pai static void show_type(struct seq_file *m, struct super_block *sb) 10032d4d4864SRam Pai { 10042d4d4864SRam Pai mangle(m, sb->s_type->name); 10052d4d4864SRam Pai if (sb->s_subtype && sb->s_subtype[0]) { 10062d4d4864SRam Pai seq_putc(m, '.'); 10072d4d4864SRam Pai mangle(m, sb->s_subtype); 10082d4d4864SRam Pai } 10092d4d4864SRam Pai } 10102d4d4864SRam Pai 10112d4d4864SRam Pai static int show_vfsmnt(struct seq_file *m, void *v) 10122d4d4864SRam Pai { 10132d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 10142d4d4864SRam Pai int err = 0; 1015c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 10161da177e4SLinus Torvalds 10171da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 10181da177e4SLinus Torvalds seq_putc(m, ' '); 1019c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 10201da177e4SLinus Torvalds seq_putc(m, ' '); 10212d4d4864SRam Pai show_type(m, mnt->mnt_sb); 10222e4b7fcdSDave Hansen seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 10232069f457SEric Paris err = show_sb_opts(m, mnt->mnt_sb); 10242069f457SEric Paris if (err) 10252069f457SEric Paris goto out; 10262d4d4864SRam Pai show_mnt_opts(m, mnt); 10271da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 10281da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 10291da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 10302069f457SEric Paris out: 10311da177e4SLinus Torvalds return err; 10321da177e4SLinus Torvalds } 10331da177e4SLinus Torvalds 1034a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 10351da177e4SLinus Torvalds .start = m_start, 10361da177e4SLinus Torvalds .next = m_next, 10371da177e4SLinus Torvalds .stop = m_stop, 10381da177e4SLinus Torvalds .show = show_vfsmnt 10391da177e4SLinus Torvalds }; 10401da177e4SLinus Torvalds 10412d4d4864SRam Pai static int show_mountinfo(struct seq_file *m, void *v) 10422d4d4864SRam Pai { 10432d4d4864SRam Pai struct proc_mounts *p = m->private; 10442d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 10452d4d4864SRam Pai struct super_block *sb = mnt->mnt_sb; 10462d4d4864SRam Pai struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 10472d4d4864SRam Pai struct path root = p->root; 10482d4d4864SRam Pai int err = 0; 10492d4d4864SRam Pai 10502d4d4864SRam Pai seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, 10512d4d4864SRam Pai MAJOR(sb->s_dev), MINOR(sb->s_dev)); 10522d4d4864SRam Pai seq_dentry(m, mnt->mnt_root, " \t\n\\"); 10532d4d4864SRam Pai seq_putc(m, ' '); 10542d4d4864SRam Pai seq_path_root(m, &mnt_path, &root, " \t\n\\"); 10552d4d4864SRam Pai if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { 10562d4d4864SRam Pai /* 10572d4d4864SRam Pai * Mountpoint is outside root, discard that one. Ugly, 10582d4d4864SRam Pai * but less so than trying to do that in iterator in a 10592d4d4864SRam Pai * race-free way (due to renames). 10602d4d4864SRam Pai */ 10612d4d4864SRam Pai return SEQ_SKIP; 10622d4d4864SRam Pai } 10632d4d4864SRam Pai seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); 10642d4d4864SRam Pai show_mnt_opts(m, mnt); 10652d4d4864SRam Pai 10662d4d4864SRam Pai /* Tagged fields ("foo:X" or "bar") */ 10672d4d4864SRam Pai if (IS_MNT_SHARED(mnt)) 10682d4d4864SRam Pai seq_printf(m, " shared:%i", mnt->mnt_group_id); 106997e7e0f7SMiklos Szeredi if (IS_MNT_SLAVE(mnt)) { 107097e7e0f7SMiklos Szeredi int master = mnt->mnt_master->mnt_group_id; 107197e7e0f7SMiklos Szeredi int dom = get_dominating_id(mnt, &p->root); 107297e7e0f7SMiklos Szeredi seq_printf(m, " master:%i", master); 107397e7e0f7SMiklos Szeredi if (dom && dom != master) 107497e7e0f7SMiklos Szeredi seq_printf(m, " propagate_from:%i", dom); 107597e7e0f7SMiklos Szeredi } 10762d4d4864SRam Pai if (IS_MNT_UNBINDABLE(mnt)) 10772d4d4864SRam Pai seq_puts(m, " unbindable"); 10782d4d4864SRam Pai 10792d4d4864SRam Pai /* Filesystem specific data */ 10802d4d4864SRam Pai seq_puts(m, " - "); 10812d4d4864SRam Pai show_type(m, sb); 10822d4d4864SRam Pai seq_putc(m, ' '); 10832d4d4864SRam Pai mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 10842d4d4864SRam Pai seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); 10852069f457SEric Paris err = show_sb_opts(m, sb); 10862069f457SEric Paris if (err) 10872069f457SEric Paris goto out; 10882d4d4864SRam Pai if (sb->s_op->show_options) 10892d4d4864SRam Pai err = sb->s_op->show_options(m, mnt); 10902d4d4864SRam Pai seq_putc(m, '\n'); 10912069f457SEric Paris out: 10922d4d4864SRam Pai return err; 10932d4d4864SRam Pai } 10942d4d4864SRam Pai 10952d4d4864SRam Pai const struct seq_operations mountinfo_op = { 10962d4d4864SRam Pai .start = m_start, 10972d4d4864SRam Pai .next = m_next, 10982d4d4864SRam Pai .stop = m_stop, 10992d4d4864SRam Pai .show = show_mountinfo, 11002d4d4864SRam Pai }; 11012d4d4864SRam Pai 1102b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 1103b4629fe2SChuck Lever { 1104b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 1105c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 1106b4629fe2SChuck Lever int err = 0; 1107b4629fe2SChuck Lever 1108b4629fe2SChuck Lever /* device */ 1109b4629fe2SChuck Lever if (mnt->mnt_devname) { 1110b4629fe2SChuck Lever seq_puts(m, "device "); 1111b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 1112b4629fe2SChuck Lever } else 1113b4629fe2SChuck Lever seq_puts(m, "no device"); 1114b4629fe2SChuck Lever 1115b4629fe2SChuck Lever /* mount point */ 1116b4629fe2SChuck Lever seq_puts(m, " mounted on "); 1117c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 1118b4629fe2SChuck Lever seq_putc(m, ' '); 1119b4629fe2SChuck Lever 1120b4629fe2SChuck Lever /* file system type */ 1121b4629fe2SChuck Lever seq_puts(m, "with fstype "); 11222d4d4864SRam Pai show_type(m, mnt->mnt_sb); 1123b4629fe2SChuck Lever 1124b4629fe2SChuck Lever /* optional statistics */ 1125b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 1126b4629fe2SChuck Lever seq_putc(m, ' '); 1127b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 1128b4629fe2SChuck Lever } 1129b4629fe2SChuck Lever 1130b4629fe2SChuck Lever seq_putc(m, '\n'); 1131b4629fe2SChuck Lever return err; 1132b4629fe2SChuck Lever } 1133b4629fe2SChuck Lever 1134a1a2c409SMiklos Szeredi const struct seq_operations mountstats_op = { 1135b4629fe2SChuck Lever .start = m_start, 1136b4629fe2SChuck Lever .next = m_next, 1137b4629fe2SChuck Lever .stop = m_stop, 1138b4629fe2SChuck Lever .show = show_vfsstat, 1139b4629fe2SChuck Lever }; 1140a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1141b4629fe2SChuck Lever 11421da177e4SLinus Torvalds /** 11431da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 11441da177e4SLinus Torvalds * @mnt: root of mount tree 11451da177e4SLinus Torvalds * 11461da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 11471da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 11481da177e4SLinus Torvalds * busy. 11491da177e4SLinus Torvalds */ 11501da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 11511da177e4SLinus Torvalds { 115236341f64SRam Pai int actual_refs = 0; 115336341f64SRam Pai int minimum_refs = 0; 115436341f64SRam Pai struct vfsmount *p; 11551da177e4SLinus Torvalds 1156b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1157b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 115836341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 1159b3e19d92SNick Piggin actual_refs += mnt_get_count(p); 11601da177e4SLinus Torvalds minimum_refs += 2; 11611da177e4SLinus Torvalds } 1162b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 11631da177e4SLinus Torvalds 11641da177e4SLinus Torvalds if (actual_refs > minimum_refs) 11651da177e4SLinus Torvalds return 0; 1166e3474a8eSIan Kent 1167e3474a8eSIan Kent return 1; 11681da177e4SLinus Torvalds } 11691da177e4SLinus Torvalds 11701da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 11711da177e4SLinus Torvalds 11721da177e4SLinus Torvalds /** 11731da177e4SLinus Torvalds * may_umount - check if a mount point is busy 11741da177e4SLinus Torvalds * @mnt: root of mount 11751da177e4SLinus Torvalds * 11761da177e4SLinus Torvalds * This is called to check if a mount point has any 11771da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 11781da177e4SLinus Torvalds * mount has sub mounts this will return busy 11791da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 11801da177e4SLinus Torvalds * 11811da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 11821da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 11831da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 11841da177e4SLinus Torvalds */ 11851da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 11861da177e4SLinus Torvalds { 1187e3474a8eSIan Kent int ret = 1; 11888ad08d8aSAl Viro down_read(&namespace_sem); 1189b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 1190a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 1191e3474a8eSIan Kent ret = 0; 1192b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 11938ad08d8aSAl Viro up_read(&namespace_sem); 1194a05964f3SRam Pai return ret; 11951da177e4SLinus Torvalds } 11961da177e4SLinus Torvalds 11971da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 11981da177e4SLinus Torvalds 1199b90fa9aeSRam Pai void release_mounts(struct list_head *head) 12001da177e4SLinus Torvalds { 120170fbcdf4SRam Pai struct vfsmount *mnt; 120270fbcdf4SRam Pai while (!list_empty(head)) { 1203b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 120470fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 120570fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 120670fbcdf4SRam Pai struct dentry *dentry; 120770fbcdf4SRam Pai struct vfsmount *m; 120899b7db7bSNick Piggin 120999b7db7bSNick Piggin br_write_lock(vfsmount_lock); 121070fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 121170fbcdf4SRam Pai m = mnt->mnt_parent; 121270fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 121370fbcdf4SRam Pai mnt->mnt_parent = mnt; 12147c4b93d8SAl Viro m->mnt_ghosts--; 121599b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 121670fbcdf4SRam Pai dput(dentry); 121770fbcdf4SRam Pai mntput(m); 12181da177e4SLinus Torvalds } 1219b3e19d92SNick Piggin mntput_long(mnt); 122070fbcdf4SRam Pai } 122170fbcdf4SRam Pai } 122270fbcdf4SRam Pai 122399b7db7bSNick Piggin /* 122499b7db7bSNick Piggin * vfsmount lock must be held for write 122599b7db7bSNick Piggin * namespace_sem must be held for write 122699b7db7bSNick Piggin */ 1227a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 122870fbcdf4SRam Pai { 12297b8a53fdSAl Viro LIST_HEAD(tmp_list); 123070fbcdf4SRam Pai struct vfsmount *p; 123170fbcdf4SRam Pai 12321bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 12337b8a53fdSAl Viro list_move(&p->mnt_hash, &tmp_list); 123470fbcdf4SRam Pai 1235a05964f3SRam Pai if (propagate) 12367b8a53fdSAl Viro propagate_umount(&tmp_list); 1237a05964f3SRam Pai 12387b8a53fdSAl Viro list_for_each_entry(p, &tmp_list, mnt_hash) { 123970fbcdf4SRam Pai list_del_init(&p->mnt_expire); 124070fbcdf4SRam Pai list_del_init(&p->mnt_list); 12416b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 12426b3286edSKirill Korotaev p->mnt_ns = NULL; 124370fbcdf4SRam Pai list_del_init(&p->mnt_child); 12447c4b93d8SAl Viro if (p->mnt_parent != p) { 12457c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 12465f57cbccSNick Piggin dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint); 12477c4b93d8SAl Viro } 1248a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 12491da177e4SLinus Torvalds } 12507b8a53fdSAl Viro list_splice(&tmp_list, kill); 12511da177e4SLinus Torvalds } 12521da177e4SLinus Torvalds 1253c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 1254c35038beSAl Viro 12551da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 12561da177e4SLinus Torvalds { 12571da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 12581da177e4SLinus Torvalds int retval; 125970fbcdf4SRam Pai LIST_HEAD(umount_list); 12601da177e4SLinus Torvalds 12611da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 12621da177e4SLinus Torvalds if (retval) 12631da177e4SLinus Torvalds return retval; 12641da177e4SLinus Torvalds 12651da177e4SLinus Torvalds /* 12661da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 12671da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 12681da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 12691da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 12701da177e4SLinus Torvalds */ 12711da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 12726ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 12731da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 12741da177e4SLinus Torvalds return -EINVAL; 12751da177e4SLinus Torvalds 1276b3e19d92SNick Piggin /* 1277b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1278b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1279b3e19d92SNick Piggin */ 1280b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 1281b3e19d92SNick Piggin if (mnt_get_count(mnt) != 2) { 1282b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 12831da177e4SLinus Torvalds return -EBUSY; 1284b3e19d92SNick Piggin } 1285b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 12861da177e4SLinus Torvalds 12871da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 12881da177e4SLinus Torvalds return -EAGAIN; 12891da177e4SLinus Torvalds } 12901da177e4SLinus Torvalds 12911da177e4SLinus Torvalds /* 12921da177e4SLinus Torvalds * If we may have to abort operations to get out of this 12931da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 12941da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 12951da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 12961da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 12971da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 12981da177e4SLinus Torvalds * about for the moment. 12991da177e4SLinus Torvalds */ 13001da177e4SLinus Torvalds 130142faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 130242faad99SAl Viro sb->s_op->umount_begin(sb); 130342faad99SAl Viro } 13041da177e4SLinus Torvalds 13051da177e4SLinus Torvalds /* 13061da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 13071da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 13081da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 13091da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 13101da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 13111da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 13121da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 13131da177e4SLinus Torvalds */ 13146ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 13151da177e4SLinus Torvalds /* 13161da177e4SLinus Torvalds * Special case for "unmounting" root ... 13171da177e4SLinus Torvalds * we just try to remount it readonly. 13181da177e4SLinus Torvalds */ 13191da177e4SLinus Torvalds down_write(&sb->s_umount); 13204aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 13211da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 13221da177e4SLinus Torvalds up_write(&sb->s_umount); 13231da177e4SLinus Torvalds return retval; 13241da177e4SLinus Torvalds } 13251da177e4SLinus Torvalds 1326390c6843SRam Pai down_write(&namespace_sem); 132799b7db7bSNick Piggin br_write_lock(vfsmount_lock); 13285addc5ddSAl Viro event++; 13291da177e4SLinus Torvalds 1330c35038beSAl Viro if (!(flags & MNT_DETACH)) 1331c35038beSAl Viro shrink_submounts(mnt, &umount_list); 1332c35038beSAl Viro 13331da177e4SLinus Torvalds retval = -EBUSY; 1334a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 13351da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 1336a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 13371da177e4SLinus Torvalds retval = 0; 13381da177e4SLinus Torvalds } 133999b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1340390c6843SRam Pai up_write(&namespace_sem); 134170fbcdf4SRam Pai release_mounts(&umount_list); 13421da177e4SLinus Torvalds return retval; 13431da177e4SLinus Torvalds } 13441da177e4SLinus Torvalds 13451da177e4SLinus Torvalds /* 13461da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 13471da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 13481da177e4SLinus Torvalds * 13491da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 13501da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 13511da177e4SLinus Torvalds */ 13521da177e4SLinus Torvalds 1353bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 13541da177e4SLinus Torvalds { 13552d8f3038SAl Viro struct path path; 13561da177e4SLinus Torvalds int retval; 1357db1f05bbSMiklos Szeredi int lookup_flags = 0; 13581da177e4SLinus Torvalds 1359db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1360db1f05bbSMiklos Szeredi return -EINVAL; 1361db1f05bbSMiklos Szeredi 1362db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1363db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1364db1f05bbSMiklos Szeredi 1365db1f05bbSMiklos Szeredi retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); 13661da177e4SLinus Torvalds if (retval) 13671da177e4SLinus Torvalds goto out; 13681da177e4SLinus Torvalds retval = -EINVAL; 13692d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 13701da177e4SLinus Torvalds goto dput_and_out; 13712d8f3038SAl Viro if (!check_mnt(path.mnt)) 13721da177e4SLinus Torvalds goto dput_and_out; 13731da177e4SLinus Torvalds 13741da177e4SLinus Torvalds retval = -EPERM; 13751da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 13761da177e4SLinus Torvalds goto dput_and_out; 13771da177e4SLinus Torvalds 13782d8f3038SAl Viro retval = do_umount(path.mnt, flags); 13791da177e4SLinus Torvalds dput_and_out: 1380429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 13812d8f3038SAl Viro dput(path.dentry); 13822d8f3038SAl Viro mntput_no_expire(path.mnt); 13831da177e4SLinus Torvalds out: 13841da177e4SLinus Torvalds return retval; 13851da177e4SLinus Torvalds } 13861da177e4SLinus Torvalds 13871da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 13881da177e4SLinus Torvalds 13891da177e4SLinus Torvalds /* 13901da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 13911da177e4SLinus Torvalds */ 1392bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 13931da177e4SLinus Torvalds { 13941da177e4SLinus Torvalds return sys_umount(name, 0); 13951da177e4SLinus Torvalds } 13961da177e4SLinus Torvalds 13971da177e4SLinus Torvalds #endif 13981da177e4SLinus Torvalds 13992d92ab3cSAl Viro static int mount_is_safe(struct path *path) 14001da177e4SLinus Torvalds { 14011da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 14021da177e4SLinus Torvalds return 0; 14031da177e4SLinus Torvalds return -EPERM; 14041da177e4SLinus Torvalds #ifdef notyet 14052d92ab3cSAl Viro if (S_ISLNK(path->dentry->d_inode->i_mode)) 14061da177e4SLinus Torvalds return -EPERM; 14072d92ab3cSAl Viro if (path->dentry->d_inode->i_mode & S_ISVTX) { 1408da9592edSDavid Howells if (current_uid() != path->dentry->d_inode->i_uid) 14091da177e4SLinus Torvalds return -EPERM; 14101da177e4SLinus Torvalds } 14112d92ab3cSAl Viro if (inode_permission(path->dentry->d_inode, MAY_WRITE)) 14121da177e4SLinus Torvalds return -EPERM; 14131da177e4SLinus Torvalds return 0; 14141da177e4SLinus Torvalds #endif 14151da177e4SLinus Torvalds } 14161da177e4SLinus Torvalds 1417b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 141836341f64SRam Pai int flag) 14191da177e4SLinus Torvalds { 14201da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 14211a390689SAl Viro struct path path; 14221da177e4SLinus Torvalds 14239676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 14249676f0c6SRam Pai return NULL; 14259676f0c6SRam Pai 142636341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 14271da177e4SLinus Torvalds if (!q) 14281da177e4SLinus Torvalds goto Enomem; 14291da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 14301da177e4SLinus Torvalds 14311da177e4SLinus Torvalds p = mnt; 1432fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 14337ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 14341da177e4SLinus Torvalds continue; 14351da177e4SLinus Torvalds 14361da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 14379676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 14389676f0c6SRam Pai s = skip_mnt_tree(s); 14399676f0c6SRam Pai continue; 14409676f0c6SRam Pai } 14411da177e4SLinus Torvalds while (p != s->mnt_parent) { 14421da177e4SLinus Torvalds p = p->mnt_parent; 14431da177e4SLinus Torvalds q = q->mnt_parent; 14441da177e4SLinus Torvalds } 14451da177e4SLinus Torvalds p = s; 14461a390689SAl Viro path.mnt = q; 14471a390689SAl Viro path.dentry = p->mnt_mountpoint; 144836341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 14491da177e4SLinus Torvalds if (!q) 14501da177e4SLinus Torvalds goto Enomem; 145199b7db7bSNick Piggin br_write_lock(vfsmount_lock); 14521da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 14531a390689SAl Viro attach_mnt(q, &path); 145499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 14551da177e4SLinus Torvalds } 14561da177e4SLinus Torvalds } 14571da177e4SLinus Torvalds return res; 14581da177e4SLinus Torvalds Enomem: 14591da177e4SLinus Torvalds if (res) { 146070fbcdf4SRam Pai LIST_HEAD(umount_list); 146199b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1462a05964f3SRam Pai umount_tree(res, 0, &umount_list); 146399b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 146470fbcdf4SRam Pai release_mounts(&umount_list); 14651da177e4SLinus Torvalds } 14661da177e4SLinus Torvalds return NULL; 14671da177e4SLinus Torvalds } 14681da177e4SLinus Torvalds 1469589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 14708aec0809SAl Viro { 14718aec0809SAl Viro struct vfsmount *tree; 14721a60a280SAl Viro down_write(&namespace_sem); 1473589ff870SAl Viro tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); 14741a60a280SAl Viro up_write(&namespace_sem); 14758aec0809SAl Viro return tree; 14768aec0809SAl Viro } 14778aec0809SAl Viro 14788aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 14798aec0809SAl Viro { 14808aec0809SAl Viro LIST_HEAD(umount_list); 14811a60a280SAl Viro down_write(&namespace_sem); 148299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 14838aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 148499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 14851a60a280SAl Viro up_write(&namespace_sem); 14868aec0809SAl Viro release_mounts(&umount_list); 14878aec0809SAl Viro } 14888aec0809SAl Viro 14891f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 14901f707137SAl Viro struct vfsmount *root) 14911f707137SAl Viro { 14921f707137SAl Viro struct vfsmount *mnt; 14931f707137SAl Viro int res = f(root, arg); 14941f707137SAl Viro if (res) 14951f707137SAl Viro return res; 14961f707137SAl Viro list_for_each_entry(mnt, &root->mnt_list, mnt_list) { 14971f707137SAl Viro res = f(mnt, arg); 14981f707137SAl Viro if (res) 14991f707137SAl Viro return res; 15001f707137SAl Viro } 15011f707137SAl Viro return 0; 15021f707137SAl Viro } 15031f707137SAl Viro 1504719f5d7fSMiklos Szeredi static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) 1505719f5d7fSMiklos Szeredi { 1506719f5d7fSMiklos Szeredi struct vfsmount *p; 1507719f5d7fSMiklos Szeredi 1508719f5d7fSMiklos Szeredi for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1509719f5d7fSMiklos Szeredi if (p->mnt_group_id && !IS_MNT_SHARED(p)) 1510719f5d7fSMiklos Szeredi mnt_release_group_id(p); 1511719f5d7fSMiklos Szeredi } 1512719f5d7fSMiklos Szeredi } 1513719f5d7fSMiklos Szeredi 1514719f5d7fSMiklos Szeredi static int invent_group_ids(struct vfsmount *mnt, bool recurse) 1515719f5d7fSMiklos Szeredi { 1516719f5d7fSMiklos Szeredi struct vfsmount *p; 1517719f5d7fSMiklos Szeredi 1518719f5d7fSMiklos Szeredi for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1519719f5d7fSMiklos Szeredi if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 1520719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(p); 1521719f5d7fSMiklos Szeredi if (err) { 1522719f5d7fSMiklos Szeredi cleanup_group_ids(mnt, p); 1523719f5d7fSMiklos Szeredi return err; 1524719f5d7fSMiklos Szeredi } 1525719f5d7fSMiklos Szeredi } 1526719f5d7fSMiklos Szeredi } 1527719f5d7fSMiklos Szeredi 1528719f5d7fSMiklos Szeredi return 0; 1529719f5d7fSMiklos Szeredi } 1530719f5d7fSMiklos Szeredi 1531b90fa9aeSRam Pai /* 1532b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1533b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 153421444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 153521444403SRam Pai * store the parent mount and mountpoint dentry. 153621444403SRam Pai * (done when source_mnt is moved) 1537b90fa9aeSRam Pai * 1538b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1539b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 15409676f0c6SRam Pai * --------------------------------------------------------------------------- 1541b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 15429676f0c6SRam Pai * |************************************************************************** 15439676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15449676f0c6SRam Pai * | dest | | | | | 15459676f0c6SRam Pai * | | | | | | | 15469676f0c6SRam Pai * | v | | | | | 15479676f0c6SRam Pai * |************************************************************************** 15489676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 15495afe0022SRam Pai * | | | | | | 15509676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 15519676f0c6SRam Pai * *************************************************************************** 1552b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1553b90fa9aeSRam Pai * destination mount. 1554b90fa9aeSRam Pai * 1555b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1556b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1557b90fa9aeSRam Pai * the peer group of the source mount. 1558b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1559b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1560b90fa9aeSRam Pai * mount. 15615afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 15625afe0022SRam Pai * of the destination mount and the cloned mount is made slave 15635afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 15645afe0022SRam Pai * is marked as 'shared and slave'. 15655afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 15665afe0022SRam Pai * source mount. 15675afe0022SRam Pai * 15689676f0c6SRam Pai * --------------------------------------------------------------------------- 156921444403SRam Pai * | MOVE MOUNT OPERATION | 15709676f0c6SRam Pai * |************************************************************************** 15719676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15729676f0c6SRam Pai * | dest | | | | | 15739676f0c6SRam Pai * | | | | | | | 15749676f0c6SRam Pai * | v | | | | | 15759676f0c6SRam Pai * |************************************************************************** 15769676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 15775afe0022SRam Pai * | | | | | | 15789676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 15799676f0c6SRam Pai * *************************************************************************** 15805afe0022SRam Pai * 15815afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 15825afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 158321444403SRam Pai * (+*) the mount is moved to the destination. 15845afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 15855afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 15865afe0022SRam Pai * the mount is marked as 'shared and slave'. 15875afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1588b90fa9aeSRam Pai * 1589b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1590b90fa9aeSRam Pai * applied to each mount in the tree. 1591b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1592b90fa9aeSRam Pai * in allocations. 1593b90fa9aeSRam Pai */ 1594b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 15951a390689SAl Viro struct path *path, struct path *parent_path) 1596b90fa9aeSRam Pai { 1597b90fa9aeSRam Pai LIST_HEAD(tree_list); 15981a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 15991a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1600b90fa9aeSRam Pai struct vfsmount *child, *p; 1601719f5d7fSMiklos Szeredi int err; 1602b90fa9aeSRam Pai 1603719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) { 1604719f5d7fSMiklos Szeredi err = invent_group_ids(source_mnt, true); 1605719f5d7fSMiklos Szeredi if (err) 1606719f5d7fSMiklos Szeredi goto out; 1607719f5d7fSMiklos Szeredi } 1608719f5d7fSMiklos Szeredi err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1609719f5d7fSMiklos Szeredi if (err) 1610719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1611b90fa9aeSRam Pai 161299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1613df1a1ad2SAl Viro 1614b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1615b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1616b90fa9aeSRam Pai set_mnt_shared(p); 1617b90fa9aeSRam Pai } 16181a390689SAl Viro if (parent_path) { 16191a390689SAl Viro detach_mnt(source_mnt, parent_path); 16201a390689SAl Viro attach_mnt(source_mnt, path); 1621e5d67f07SAl Viro touch_mnt_namespace(parent_path->mnt->mnt_ns); 162221444403SRam Pai } else { 1623b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1624b90fa9aeSRam Pai commit_tree(source_mnt); 162521444403SRam Pai } 1626b90fa9aeSRam Pai 1627b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1628b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1629b90fa9aeSRam Pai commit_tree(child); 1630b90fa9aeSRam Pai } 163199b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 163299b7db7bSNick Piggin 1633b90fa9aeSRam Pai return 0; 1634719f5d7fSMiklos Szeredi 1635719f5d7fSMiklos Szeredi out_cleanup_ids: 1636719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) 1637719f5d7fSMiklos Szeredi cleanup_group_ids(source_mnt, NULL); 1638719f5d7fSMiklos Szeredi out: 1639719f5d7fSMiklos Szeredi return err; 1640b90fa9aeSRam Pai } 1641b90fa9aeSRam Pai 16428c3ee42eSAl Viro static int graft_tree(struct vfsmount *mnt, struct path *path) 16431da177e4SLinus Torvalds { 16441da177e4SLinus Torvalds int err; 16451da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 16461da177e4SLinus Torvalds return -EINVAL; 16471da177e4SLinus Torvalds 16488c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 16491da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 16501da177e4SLinus Torvalds return -ENOTDIR; 16511da177e4SLinus Torvalds 16521da177e4SLinus Torvalds err = -ENOENT; 16538c3ee42eSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 1654d83c49f3SAl Viro if (cant_mount(path->dentry)) 16551da177e4SLinus Torvalds goto out_unlock; 16561da177e4SLinus Torvalds 1657f3da392eSAlexey Dobriyan if (!d_unlinked(path->dentry)) 16588c3ee42eSAl Viro err = attach_recursive_mnt(mnt, path, NULL); 16591da177e4SLinus Torvalds out_unlock: 16608c3ee42eSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 16611da177e4SLinus Torvalds return err; 16621da177e4SLinus Torvalds } 16631da177e4SLinus Torvalds 16641da177e4SLinus Torvalds /* 16657a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 16667a2e8a8fSValerie Aurora */ 16677a2e8a8fSValerie Aurora 16687a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 16697a2e8a8fSValerie Aurora { 16707a2e8a8fSValerie Aurora int type = flags & ~MS_REC; 16717a2e8a8fSValerie Aurora 16727a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 16737a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 16747a2e8a8fSValerie Aurora return 0; 16757a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 16767a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 16777a2e8a8fSValerie Aurora return 0; 16787a2e8a8fSValerie Aurora return type; 16797a2e8a8fSValerie Aurora } 16807a2e8a8fSValerie Aurora 16817a2e8a8fSValerie Aurora /* 168207b20889SRam Pai * recursively change the type of the mountpoint. 168307b20889SRam Pai */ 16840a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 168507b20889SRam Pai { 16862d92ab3cSAl Viro struct vfsmount *m, *mnt = path->mnt; 168707b20889SRam Pai int recurse = flag & MS_REC; 16887a2e8a8fSValerie Aurora int type; 1689719f5d7fSMiklos Szeredi int err = 0; 169007b20889SRam Pai 1691ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1692ee6f9582SMiklos Szeredi return -EPERM; 1693ee6f9582SMiklos Szeredi 16942d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 169507b20889SRam Pai return -EINVAL; 169607b20889SRam Pai 16977a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 16987a2e8a8fSValerie Aurora if (!type) 16997a2e8a8fSValerie Aurora return -EINVAL; 17007a2e8a8fSValerie Aurora 170107b20889SRam Pai down_write(&namespace_sem); 1702719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1703719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1704719f5d7fSMiklos Szeredi if (err) 1705719f5d7fSMiklos Szeredi goto out_unlock; 1706719f5d7fSMiklos Szeredi } 1707719f5d7fSMiklos Szeredi 170899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 170907b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 171007b20889SRam Pai change_mnt_propagation(m, type); 171199b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1712719f5d7fSMiklos Szeredi 1713719f5d7fSMiklos Szeredi out_unlock: 171407b20889SRam Pai up_write(&namespace_sem); 1715719f5d7fSMiklos Szeredi return err; 171607b20889SRam Pai } 171707b20889SRam Pai 171807b20889SRam Pai /* 17191da177e4SLinus Torvalds * do loopback mount. 17201da177e4SLinus Torvalds */ 17210a0d8a46SAl Viro static int do_loopback(struct path *path, char *old_name, 17222dafe1c4SEric Sandeen int recurse) 17231da177e4SLinus Torvalds { 17242d92ab3cSAl Viro struct path old_path; 17251da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 17262d92ab3cSAl Viro int err = mount_is_safe(path); 17271da177e4SLinus Torvalds if (err) 17281da177e4SLinus Torvalds return err; 17291da177e4SLinus Torvalds if (!old_name || !*old_name) 17301da177e4SLinus Torvalds return -EINVAL; 17312d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 17321da177e4SLinus Torvalds if (err) 17331da177e4SLinus Torvalds return err; 17341da177e4SLinus Torvalds 1735390c6843SRam Pai down_write(&namespace_sem); 17361da177e4SLinus Torvalds err = -EINVAL; 17372d92ab3cSAl Viro if (IS_MNT_UNBINDABLE(old_path.mnt)) 17389676f0c6SRam Pai goto out; 17399676f0c6SRam Pai 17402d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 1741ccd48bc7SAl Viro goto out; 1742ccd48bc7SAl Viro 17431da177e4SLinus Torvalds err = -ENOMEM; 17441da177e4SLinus Torvalds if (recurse) 17452d92ab3cSAl Viro mnt = copy_tree(old_path.mnt, old_path.dentry, 0); 17461da177e4SLinus Torvalds else 17472d92ab3cSAl Viro mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); 17481da177e4SLinus Torvalds 1749ccd48bc7SAl Viro if (!mnt) 1750ccd48bc7SAl Viro goto out; 1751ccd48bc7SAl Viro 17522d92ab3cSAl Viro err = graft_tree(mnt, path); 17531da177e4SLinus Torvalds if (err) { 175470fbcdf4SRam Pai LIST_HEAD(umount_list); 175599b7db7bSNick Piggin 175699b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1757a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 175899b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 175970fbcdf4SRam Pai release_mounts(&umount_list); 17605b83d2c5SRam Pai } 17611da177e4SLinus Torvalds 1762ccd48bc7SAl Viro out: 1763390c6843SRam Pai up_write(&namespace_sem); 17642d92ab3cSAl Viro path_put(&old_path); 17651da177e4SLinus Torvalds return err; 17661da177e4SLinus Torvalds } 17671da177e4SLinus Torvalds 17682e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 17692e4b7fcdSDave Hansen { 17702e4b7fcdSDave Hansen int error = 0; 17712e4b7fcdSDave Hansen int readonly_request = 0; 17722e4b7fcdSDave Hansen 17732e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 17742e4b7fcdSDave Hansen readonly_request = 1; 17752e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 17762e4b7fcdSDave Hansen return 0; 17772e4b7fcdSDave Hansen 17782e4b7fcdSDave Hansen if (readonly_request) 17792e4b7fcdSDave Hansen error = mnt_make_readonly(mnt); 17802e4b7fcdSDave Hansen else 17812e4b7fcdSDave Hansen __mnt_unmake_readonly(mnt); 17822e4b7fcdSDave Hansen return error; 17832e4b7fcdSDave Hansen } 17842e4b7fcdSDave Hansen 17851da177e4SLinus Torvalds /* 17861da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 17871da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 17881da177e4SLinus Torvalds * on it - tough luck. 17891da177e4SLinus Torvalds */ 17900a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 17911da177e4SLinus Torvalds void *data) 17921da177e4SLinus Torvalds { 17931da177e4SLinus Torvalds int err; 17942d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 17951da177e4SLinus Torvalds 17961da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 17971da177e4SLinus Torvalds return -EPERM; 17981da177e4SLinus Torvalds 17992d92ab3cSAl Viro if (!check_mnt(path->mnt)) 18001da177e4SLinus Torvalds return -EINVAL; 18011da177e4SLinus Torvalds 18022d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 18031da177e4SLinus Torvalds return -EINVAL; 18041da177e4SLinus Torvalds 18051da177e4SLinus Torvalds down_write(&sb->s_umount); 18062e4b7fcdSDave Hansen if (flags & MS_BIND) 18072d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 18084aa98cf7SAl Viro else 18091da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 18107b43a79fSAl Viro if (!err) { 181199b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1812495d6c9cSValerie Aurora mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK; 18132d92ab3cSAl Viro path->mnt->mnt_flags = mnt_flags; 181499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 18157b43a79fSAl Viro } 18161da177e4SLinus Torvalds up_write(&sb->s_umount); 18170e55a7ccSDan Williams if (!err) { 181899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 18190e55a7ccSDan Williams touch_mnt_namespace(path->mnt->mnt_ns); 182099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 18210e55a7ccSDan Williams } 18221da177e4SLinus Torvalds return err; 18231da177e4SLinus Torvalds } 18241da177e4SLinus Torvalds 18259676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 18269676f0c6SRam Pai { 18279676f0c6SRam Pai struct vfsmount *p; 18289676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 18299676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 18309676f0c6SRam Pai return 1; 18319676f0c6SRam Pai } 18329676f0c6SRam Pai return 0; 18339676f0c6SRam Pai } 18349676f0c6SRam Pai 18350a0d8a46SAl Viro static int do_move_mount(struct path *path, char *old_name) 18361da177e4SLinus Torvalds { 18372d92ab3cSAl Viro struct path old_path, parent_path; 18381da177e4SLinus Torvalds struct vfsmount *p; 18391da177e4SLinus Torvalds int err = 0; 18401da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 18411da177e4SLinus Torvalds return -EPERM; 18421da177e4SLinus Torvalds if (!old_name || !*old_name) 18431da177e4SLinus Torvalds return -EINVAL; 18442d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 18451da177e4SLinus Torvalds if (err) 18461da177e4SLinus Torvalds return err; 18471da177e4SLinus Torvalds 1848390c6843SRam Pai down_write(&namespace_sem); 1849cc53ce53SDavid Howells err = follow_down(path, true); 1850cc53ce53SDavid Howells if (err < 0) 1851cc53ce53SDavid Howells goto out; 1852cc53ce53SDavid Howells 18531da177e4SLinus Torvalds err = -EINVAL; 18542d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 18551da177e4SLinus Torvalds goto out; 18561da177e4SLinus Torvalds 18571da177e4SLinus Torvalds err = -ENOENT; 18582d92ab3cSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 1859d83c49f3SAl Viro if (cant_mount(path->dentry)) 18601da177e4SLinus Torvalds goto out1; 18611da177e4SLinus Torvalds 1862f3da392eSAlexey Dobriyan if (d_unlinked(path->dentry)) 186321444403SRam Pai goto out1; 18641da177e4SLinus Torvalds 18651da177e4SLinus Torvalds err = -EINVAL; 18662d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 186721444403SRam Pai goto out1; 18681da177e4SLinus Torvalds 18692d92ab3cSAl Viro if (old_path.mnt == old_path.mnt->mnt_parent) 187021444403SRam Pai goto out1; 18711da177e4SLinus Torvalds 18722d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 18732d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 187421444403SRam Pai goto out1; 187521444403SRam Pai /* 187621444403SRam Pai * Don't move a mount residing in a shared parent. 187721444403SRam Pai */ 18782d92ab3cSAl Viro if (old_path.mnt->mnt_parent && 18792d92ab3cSAl Viro IS_MNT_SHARED(old_path.mnt->mnt_parent)) 188021444403SRam Pai goto out1; 18819676f0c6SRam Pai /* 18829676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 18839676f0c6SRam Pai * mount which is shared. 18849676f0c6SRam Pai */ 18852d92ab3cSAl Viro if (IS_MNT_SHARED(path->mnt) && 18862d92ab3cSAl Viro tree_contains_unbindable(old_path.mnt)) 18879676f0c6SRam Pai goto out1; 18881da177e4SLinus Torvalds err = -ELOOP; 18892d92ab3cSAl Viro for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) 18902d92ab3cSAl Viro if (p == old_path.mnt) 189121444403SRam Pai goto out1; 18921da177e4SLinus Torvalds 18932d92ab3cSAl Viro err = attach_recursive_mnt(old_path.mnt, path, &parent_path); 18944ac91378SJan Blunck if (err) 189521444403SRam Pai goto out1; 18961da177e4SLinus Torvalds 18971da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 18981da177e4SLinus Torvalds * automatically */ 18992d92ab3cSAl Viro list_del_init(&old_path.mnt->mnt_expire); 19001da177e4SLinus Torvalds out1: 19012d92ab3cSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 19021da177e4SLinus Torvalds out: 1903390c6843SRam Pai up_write(&namespace_sem); 19041da177e4SLinus Torvalds if (!err) 19051a390689SAl Viro path_put(&parent_path); 19062d92ab3cSAl Viro path_put(&old_path); 19071da177e4SLinus Torvalds return err; 19081da177e4SLinus Torvalds } 19091da177e4SLinus Torvalds 19101da177e4SLinus Torvalds /* 19111da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 19121da177e4SLinus Torvalds * namespace's tree 19131da177e4SLinus Torvalds */ 19140a0d8a46SAl Viro static int do_new_mount(struct path *path, char *type, int flags, 19151da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 19161da177e4SLinus Torvalds { 19171da177e4SLinus Torvalds struct vfsmount *mnt; 19181da177e4SLinus Torvalds 1919eca6f534SVegard Nossum if (!type) 19201da177e4SLinus Torvalds return -EINVAL; 19211da177e4SLinus Torvalds 19221da177e4SLinus Torvalds /* we need capabilities... */ 19231da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 19241da177e4SLinus Torvalds return -EPERM; 19251da177e4SLinus Torvalds 19261da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 19271da177e4SLinus Torvalds if (IS_ERR(mnt)) 19281da177e4SLinus Torvalds return PTR_ERR(mnt); 19291da177e4SLinus Torvalds 1930ea5b778aSDavid Howells return do_add_mount(mnt, path, mnt_flags); 19311da177e4SLinus Torvalds } 19321da177e4SLinus Torvalds 19331da177e4SLinus Torvalds /* 19341da177e4SLinus Torvalds * add a mount into a namespace's mount tree 1935ea5b778aSDavid Howells * - this unconditionally eats one of the caller's references to newmnt. 19361da177e4SLinus Torvalds */ 1937ea5b778aSDavid Howells int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags) 19381da177e4SLinus Torvalds { 19391da177e4SLinus Torvalds int err; 19401da177e4SLinus Torvalds 19418089352aSAl Viro mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); 194227d55f1fSAl Viro 1943390c6843SRam Pai down_write(&namespace_sem); 19441da177e4SLinus Torvalds /* Something was mounted here while we slept */ 1945cc53ce53SDavid Howells err = follow_down(path, true); 1946cc53ce53SDavid Howells if (err < 0) 1947cc53ce53SDavid Howells goto unlock; 1948cc53ce53SDavid Howells 19491da177e4SLinus Torvalds err = -EINVAL; 1950dd5cae6eSAl Viro if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) 19511da177e4SLinus Torvalds goto unlock; 19521da177e4SLinus Torvalds 19531da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 19541da177e4SLinus Torvalds err = -EBUSY; 19558d66bf54SAl Viro if (path->mnt->mnt_sb == newmnt->mnt_sb && 19568d66bf54SAl Viro path->mnt->mnt_root == path->dentry) 19571da177e4SLinus Torvalds goto unlock; 19581da177e4SLinus Torvalds 19591da177e4SLinus Torvalds err = -EINVAL; 19601da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 19611da177e4SLinus Torvalds goto unlock; 19621da177e4SLinus Torvalds 19631da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 19648d66bf54SAl Viro if ((err = graft_tree(newmnt, path))) 19655b83d2c5SRam Pai goto unlock; 19661da177e4SLinus Torvalds 1967390c6843SRam Pai up_write(&namespace_sem); 19685b83d2c5SRam Pai return 0; 19691da177e4SLinus Torvalds 19701da177e4SLinus Torvalds unlock: 1971390c6843SRam Pai up_write(&namespace_sem); 1972b3e19d92SNick Piggin mntput_long(newmnt); 19731da177e4SLinus Torvalds return err; 19741da177e4SLinus Torvalds } 19751da177e4SLinus Torvalds 1976ea5b778aSDavid Howells /** 1977ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 1978ea5b778aSDavid Howells * @mnt: The mount to list. 1979ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 1980ea5b778aSDavid Howells */ 1981ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 1982ea5b778aSDavid Howells { 1983ea5b778aSDavid Howells down_write(&namespace_sem); 1984ea5b778aSDavid Howells br_write_lock(vfsmount_lock); 1985ea5b778aSDavid Howells 1986ea5b778aSDavid Howells list_add_tail(&mnt->mnt_expire, expiry_list); 1987ea5b778aSDavid Howells 1988ea5b778aSDavid Howells br_write_unlock(vfsmount_lock); 1989ea5b778aSDavid Howells up_write(&namespace_sem); 1990ea5b778aSDavid Howells } 1991ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 1992ea5b778aSDavid Howells 1993ea5b778aSDavid Howells /* 1994ea5b778aSDavid Howells * Remove a vfsmount from any expiration list it may be on 1995ea5b778aSDavid Howells */ 1996ea5b778aSDavid Howells void mnt_clear_expiry(struct vfsmount *mnt) 1997ea5b778aSDavid Howells { 1998ea5b778aSDavid Howells if (!list_empty(&mnt->mnt_expire)) { 1999ea5b778aSDavid Howells down_write(&namespace_sem); 2000ea5b778aSDavid Howells br_write_lock(vfsmount_lock); 2001ea5b778aSDavid Howells list_del_init(&mnt->mnt_expire); 2002ea5b778aSDavid Howells br_write_unlock(vfsmount_lock); 2003ea5b778aSDavid Howells up_write(&namespace_sem); 2004ea5b778aSDavid Howells } 2005ea5b778aSDavid Howells } 20061da177e4SLinus Torvalds 20075528f911STrond Myklebust /* 20081da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 20091da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 20101da177e4SLinus Torvalds * here 20111da177e4SLinus Torvalds */ 20121da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 20131da177e4SLinus Torvalds { 20141da177e4SLinus Torvalds struct vfsmount *mnt, *next; 20151da177e4SLinus Torvalds LIST_HEAD(graveyard); 2016bcc5c7d2SAl Viro LIST_HEAD(umounts); 20171da177e4SLinus Torvalds 20181da177e4SLinus Torvalds if (list_empty(mounts)) 20191da177e4SLinus Torvalds return; 20201da177e4SLinus Torvalds 2021bcc5c7d2SAl Viro down_write(&namespace_sem); 202299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 20231da177e4SLinus Torvalds 20241da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 20251da177e4SLinus Torvalds * following criteria: 20261da177e4SLinus Torvalds * - only referenced by its parent vfsmount 20271da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 20281da177e4SLinus Torvalds * cleared by mntput()) 20291da177e4SLinus Torvalds */ 203055e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 20311da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 2032bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 20331da177e4SLinus Torvalds continue; 203455e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 20351da177e4SLinus Torvalds } 2036bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2037bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 2038bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2039bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 2040bcc5c7d2SAl Viro } 204199b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 2042bcc5c7d2SAl Viro up_write(&namespace_sem); 2043bcc5c7d2SAl Viro 2044bcc5c7d2SAl Viro release_mounts(&umounts); 20451da177e4SLinus Torvalds } 20461da177e4SLinus Torvalds 20471da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 20481da177e4SLinus Torvalds 20491da177e4SLinus Torvalds /* 20505528f911STrond Myklebust * Ripoff of 'select_parent()' 20515528f911STrond Myklebust * 20525528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 20535528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 20545528f911STrond Myklebust */ 20555528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 20565528f911STrond Myklebust { 20575528f911STrond Myklebust struct vfsmount *this_parent = parent; 20585528f911STrond Myklebust struct list_head *next; 20595528f911STrond Myklebust int found = 0; 20605528f911STrond Myklebust 20615528f911STrond Myklebust repeat: 20625528f911STrond Myklebust next = this_parent->mnt_mounts.next; 20635528f911STrond Myklebust resume: 20645528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 20655528f911STrond Myklebust struct list_head *tmp = next; 20665528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 20675528f911STrond Myklebust 20685528f911STrond Myklebust next = tmp->next; 20695528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 20705528f911STrond Myklebust continue; 20715528f911STrond Myklebust /* 20725528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 20735528f911STrond Myklebust */ 20745528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 20755528f911STrond Myklebust this_parent = mnt; 20765528f911STrond Myklebust goto repeat; 20775528f911STrond Myklebust } 20785528f911STrond Myklebust 20795528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 20805528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 20815528f911STrond Myklebust found++; 20825528f911STrond Myklebust } 20835528f911STrond Myklebust } 20845528f911STrond Myklebust /* 20855528f911STrond Myklebust * All done at this level ... ascend and resume the search 20865528f911STrond Myklebust */ 20875528f911STrond Myklebust if (this_parent != parent) { 20885528f911STrond Myklebust next = this_parent->mnt_child.next; 20895528f911STrond Myklebust this_parent = this_parent->mnt_parent; 20905528f911STrond Myklebust goto resume; 20915528f911STrond Myklebust } 20925528f911STrond Myklebust return found; 20935528f911STrond Myklebust } 20945528f911STrond Myklebust 20955528f911STrond Myklebust /* 20965528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 20975528f911STrond Myklebust * submounts of a specific parent mountpoint 209899b7db7bSNick Piggin * 209999b7db7bSNick Piggin * vfsmount_lock must be held for write 21005528f911STrond Myklebust */ 2101c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 21025528f911STrond Myklebust { 21035528f911STrond Myklebust LIST_HEAD(graveyard); 2104c35038beSAl Viro struct vfsmount *m; 21055528f911STrond Myklebust 21065528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2107c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2108bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2109c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 2110bcc5c7d2SAl Viro mnt_expire); 2111afef80b3SEric W. Biederman touch_mnt_namespace(m->mnt_ns); 2112afef80b3SEric W. Biederman umount_tree(m, 1, umounts); 2113bcc5c7d2SAl Viro } 2114bcc5c7d2SAl Viro } 21155528f911STrond Myklebust } 21165528f911STrond Myklebust 21175528f911STrond Myklebust /* 21181da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 21191da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 21201da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 21211da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 21221da177e4SLinus Torvalds */ 2123b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2124b58fed8bSRam Pai unsigned long n) 21251da177e4SLinus Torvalds { 21261da177e4SLinus Torvalds char *t = to; 21271da177e4SLinus Torvalds const char __user *f = from; 21281da177e4SLinus Torvalds char c; 21291da177e4SLinus Torvalds 21301da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 21311da177e4SLinus Torvalds return n; 21321da177e4SLinus Torvalds 21331da177e4SLinus Torvalds while (n) { 21341da177e4SLinus Torvalds if (__get_user(c, f)) { 21351da177e4SLinus Torvalds memset(t, 0, n); 21361da177e4SLinus Torvalds break; 21371da177e4SLinus Torvalds } 21381da177e4SLinus Torvalds *t++ = c; 21391da177e4SLinus Torvalds f++; 21401da177e4SLinus Torvalds n--; 21411da177e4SLinus Torvalds } 21421da177e4SLinus Torvalds return n; 21431da177e4SLinus Torvalds } 21441da177e4SLinus Torvalds 21451da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 21461da177e4SLinus Torvalds { 21471da177e4SLinus Torvalds int i; 21481da177e4SLinus Torvalds unsigned long page; 21491da177e4SLinus Torvalds unsigned long size; 21501da177e4SLinus Torvalds 21511da177e4SLinus Torvalds *where = 0; 21521da177e4SLinus Torvalds if (!data) 21531da177e4SLinus Torvalds return 0; 21541da177e4SLinus Torvalds 21551da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 21561da177e4SLinus Torvalds return -ENOMEM; 21571da177e4SLinus Torvalds 21581da177e4SLinus Torvalds /* We only care that *some* data at the address the user 21591da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 21601da177e4SLinus Torvalds * the remainder of the page. 21611da177e4SLinus Torvalds */ 21621da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 21631da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 21641da177e4SLinus Torvalds if (size > PAGE_SIZE) 21651da177e4SLinus Torvalds size = PAGE_SIZE; 21661da177e4SLinus Torvalds 21671da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 21681da177e4SLinus Torvalds if (!i) { 21691da177e4SLinus Torvalds free_page(page); 21701da177e4SLinus Torvalds return -EFAULT; 21711da177e4SLinus Torvalds } 21721da177e4SLinus Torvalds if (i != PAGE_SIZE) 21731da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 21741da177e4SLinus Torvalds *where = page; 21751da177e4SLinus Torvalds return 0; 21761da177e4SLinus Torvalds } 21771da177e4SLinus Torvalds 2178eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 2179eca6f534SVegard Nossum { 2180eca6f534SVegard Nossum char *tmp; 2181eca6f534SVegard Nossum 2182eca6f534SVegard Nossum if (!data) { 2183eca6f534SVegard Nossum *where = NULL; 2184eca6f534SVegard Nossum return 0; 2185eca6f534SVegard Nossum } 2186eca6f534SVegard Nossum 2187eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 2188eca6f534SVegard Nossum if (IS_ERR(tmp)) 2189eca6f534SVegard Nossum return PTR_ERR(tmp); 2190eca6f534SVegard Nossum 2191eca6f534SVegard Nossum *where = tmp; 2192eca6f534SVegard Nossum return 0; 2193eca6f534SVegard Nossum } 2194eca6f534SVegard Nossum 21951da177e4SLinus Torvalds /* 21961da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 21971da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 21981da177e4SLinus Torvalds * 21991da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 22001da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 22011da177e4SLinus Torvalds * information (or be NULL). 22021da177e4SLinus Torvalds * 22031da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 22041da177e4SLinus Torvalds * When the flags word was introduced its top half was required 22051da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 22061da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 22071da177e4SLinus Torvalds * and must be discarded. 22081da177e4SLinus Torvalds */ 22091da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 22101da177e4SLinus Torvalds unsigned long flags, void *data_page) 22111da177e4SLinus Torvalds { 22122d92ab3cSAl Viro struct path path; 22131da177e4SLinus Torvalds int retval = 0; 22141da177e4SLinus Torvalds int mnt_flags = 0; 22151da177e4SLinus Torvalds 22161da177e4SLinus Torvalds /* Discard magic */ 22171da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 22181da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 22191da177e4SLinus Torvalds 22201da177e4SLinus Torvalds /* Basic sanity checks */ 22211da177e4SLinus Torvalds 22221da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 22231da177e4SLinus Torvalds return -EINVAL; 22241da177e4SLinus Torvalds 22251da177e4SLinus Torvalds if (data_page) 22261da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 22271da177e4SLinus Torvalds 2228a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 2229a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2230a27ab9f2STetsuo Handa if (retval) 2231a27ab9f2STetsuo Handa return retval; 2232a27ab9f2STetsuo Handa 2233a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2234a27ab9f2STetsuo Handa type_page, flags, data_page); 2235a27ab9f2STetsuo Handa if (retval) 2236a27ab9f2STetsuo Handa goto dput_out; 2237a27ab9f2STetsuo Handa 2238613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2239613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 22400a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 22410a1c01c9SMatthew Garrett 22421da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 22431da177e4SLinus Torvalds if (flags & MS_NOSUID) 22441da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 22451da177e4SLinus Torvalds if (flags & MS_NODEV) 22461da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 22471da177e4SLinus Torvalds if (flags & MS_NOEXEC) 22481da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2249fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2250fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2251fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2252fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2253d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2254d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 22552e4b7fcdSDave Hansen if (flags & MS_RDONLY) 22562e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2257fc33a7bbSChristoph Hellwig 22587a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2259d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2260d0adde57SMatthew Garrett MS_STRICTATIME); 22611da177e4SLinus Torvalds 22621da177e4SLinus Torvalds if (flags & MS_REMOUNT) 22632d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 22641da177e4SLinus Torvalds data_page); 22651da177e4SLinus Torvalds else if (flags & MS_BIND) 22662d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 22679676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 22682d92ab3cSAl Viro retval = do_change_type(&path, flags); 22691da177e4SLinus Torvalds else if (flags & MS_MOVE) 22702d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 22711da177e4SLinus Torvalds else 22722d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 22731da177e4SLinus Torvalds dev_name, data_page); 22741da177e4SLinus Torvalds dput_out: 22752d92ab3cSAl Viro path_put(&path); 22761da177e4SLinus Torvalds return retval; 22771da177e4SLinus Torvalds } 22781da177e4SLinus Torvalds 2279cf8d2c11STrond Myklebust static struct mnt_namespace *alloc_mnt_ns(void) 2280cf8d2c11STrond Myklebust { 2281cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2282cf8d2c11STrond Myklebust 2283cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2284cf8d2c11STrond Myklebust if (!new_ns) 2285cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2286cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2287cf8d2c11STrond Myklebust new_ns->root = NULL; 2288cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2289cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2290cf8d2c11STrond Myklebust new_ns->event = 0; 2291cf8d2c11STrond Myklebust return new_ns; 2292cf8d2c11STrond Myklebust } 2293cf8d2c11STrond Myklebust 2294741a2951SJANAK DESAI /* 2295741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 2296741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 2297741a2951SJANAK DESAI */ 2298e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 22996b3286edSKirill Korotaev struct fs_struct *fs) 23001da177e4SLinus Torvalds { 23016b3286edSKirill Korotaev struct mnt_namespace *new_ns; 23027f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 23031da177e4SLinus Torvalds struct vfsmount *p, *q; 23041da177e4SLinus Torvalds 2305cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2306cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2307cf8d2c11STrond Myklebust return new_ns; 23081da177e4SLinus Torvalds 2309390c6843SRam Pai down_write(&namespace_sem); 23101da177e4SLinus Torvalds /* First pass: copy the tree topology */ 23116b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 23129676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 23131da177e4SLinus Torvalds if (!new_ns->root) { 2314390c6843SRam Pai up_write(&namespace_sem); 23151da177e4SLinus Torvalds kfree(new_ns); 23165cc4a034SJulia Lawall return ERR_PTR(-ENOMEM); 23171da177e4SLinus Torvalds } 231899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 23191da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 232099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 23211da177e4SLinus Torvalds 23221da177e4SLinus Torvalds /* 23231da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 23241da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 23251da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 23261da177e4SLinus Torvalds */ 23276b3286edSKirill Korotaev p = mnt_ns->root; 23281da177e4SLinus Torvalds q = new_ns->root; 23291da177e4SLinus Torvalds while (p) { 23306b3286edSKirill Korotaev q->mnt_ns = new_ns; 23311da177e4SLinus Torvalds if (fs) { 23326ac08c39SJan Blunck if (p == fs->root.mnt) { 23331da177e4SLinus Torvalds rootmnt = p; 2334b3e19d92SNick Piggin fs->root.mnt = mntget_long(q); 23351da177e4SLinus Torvalds } 23366ac08c39SJan Blunck if (p == fs->pwd.mnt) { 23371da177e4SLinus Torvalds pwdmnt = p; 2338b3e19d92SNick Piggin fs->pwd.mnt = mntget_long(q); 23391da177e4SLinus Torvalds } 23401da177e4SLinus Torvalds } 23416b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 23421da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 23431da177e4SLinus Torvalds } 2344390c6843SRam Pai up_write(&namespace_sem); 23451da177e4SLinus Torvalds 23461da177e4SLinus Torvalds if (rootmnt) 2347b3e19d92SNick Piggin mntput_long(rootmnt); 23481da177e4SLinus Torvalds if (pwdmnt) 2349b3e19d92SNick Piggin mntput_long(pwdmnt); 23501da177e4SLinus Torvalds 2351741a2951SJANAK DESAI return new_ns; 2352741a2951SJANAK DESAI } 2353741a2951SJANAK DESAI 2354213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2355e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 2356741a2951SJANAK DESAI { 23576b3286edSKirill Korotaev struct mnt_namespace *new_ns; 2358741a2951SJANAK DESAI 2359e3222c4eSBadari Pulavarty BUG_ON(!ns); 23606b3286edSKirill Korotaev get_mnt_ns(ns); 2361741a2951SJANAK DESAI 2362741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 2363e3222c4eSBadari Pulavarty return ns; 2364741a2951SJANAK DESAI 2365e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 2366741a2951SJANAK DESAI 23676b3286edSKirill Korotaev put_mnt_ns(ns); 2368e3222c4eSBadari Pulavarty return new_ns; 23691da177e4SLinus Torvalds } 23701da177e4SLinus Torvalds 2371cf8d2c11STrond Myklebust /** 2372cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2373cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2374cf8d2c11STrond Myklebust */ 2375a2770d86SLinus Torvalds struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt) 2376cf8d2c11STrond Myklebust { 2377cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2378cf8d2c11STrond Myklebust 2379cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2380cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 2381cf8d2c11STrond Myklebust mnt->mnt_ns = new_ns; 2382cf8d2c11STrond Myklebust new_ns->root = mnt; 2383cf8d2c11STrond Myklebust list_add(&new_ns->list, &new_ns->root->mnt_list); 2384cf8d2c11STrond Myklebust } 2385cf8d2c11STrond Myklebust return new_ns; 2386cf8d2c11STrond Myklebust } 2387a2770d86SLinus Torvalds EXPORT_SYMBOL(create_mnt_ns); 2388cf8d2c11STrond Myklebust 2389bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2390bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 23911da177e4SLinus Torvalds { 2392eca6f534SVegard Nossum int ret; 2393eca6f534SVegard Nossum char *kernel_type; 2394eca6f534SVegard Nossum char *kernel_dir; 2395eca6f534SVegard Nossum char *kernel_dev; 23961da177e4SLinus Torvalds unsigned long data_page; 23971da177e4SLinus Torvalds 2398eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2399eca6f534SVegard Nossum if (ret < 0) 2400eca6f534SVegard Nossum goto out_type; 24011da177e4SLinus Torvalds 2402eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2403eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2404eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2405eca6f534SVegard Nossum goto out_dir; 2406eca6f534SVegard Nossum } 24071da177e4SLinus Torvalds 2408eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2409eca6f534SVegard Nossum if (ret < 0) 2410eca6f534SVegard Nossum goto out_dev; 24111da177e4SLinus Torvalds 2412eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2413eca6f534SVegard Nossum if (ret < 0) 2414eca6f534SVegard Nossum goto out_data; 24151da177e4SLinus Torvalds 2416eca6f534SVegard Nossum ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags, 2417eca6f534SVegard Nossum (void *) data_page); 2418eca6f534SVegard Nossum 24191da177e4SLinus Torvalds free_page(data_page); 2420eca6f534SVegard Nossum out_data: 2421eca6f534SVegard Nossum kfree(kernel_dev); 2422eca6f534SVegard Nossum out_dev: 2423eca6f534SVegard Nossum putname(kernel_dir); 2424eca6f534SVegard Nossum out_dir: 2425eca6f534SVegard Nossum kfree(kernel_type); 2426eca6f534SVegard Nossum out_type: 2427eca6f534SVegard Nossum return ret; 24281da177e4SLinus Torvalds } 24291da177e4SLinus Torvalds 24301da177e4SLinus Torvalds /* 24311da177e4SLinus Torvalds * pivot_root Semantics: 24321da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 24331da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 24341da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 24351da177e4SLinus Torvalds * 24361da177e4SLinus Torvalds * Restrictions: 24371da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 24381da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 24391da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 24401da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 24411da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 24421da177e4SLinus Torvalds * 24434a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 24444a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 24454a0d11faSNeil Brown * in this situation. 24464a0d11faSNeil Brown * 24471da177e4SLinus Torvalds * Notes: 24481da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 24491da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 24501da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 24511da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 24521da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 24531da177e4SLinus Torvalds * first. 24541da177e4SLinus Torvalds */ 24553480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 24563480b257SHeiko Carstens const char __user *, put_old) 24571da177e4SLinus Torvalds { 24581da177e4SLinus Torvalds struct vfsmount *tmp; 24592d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 24601da177e4SLinus Torvalds int error; 24611da177e4SLinus Torvalds 24621da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 24631da177e4SLinus Torvalds return -EPERM; 24641da177e4SLinus Torvalds 24652d8f3038SAl Viro error = user_path_dir(new_root, &new); 24661da177e4SLinus Torvalds if (error) 24671da177e4SLinus Torvalds goto out0; 24681da177e4SLinus Torvalds error = -EINVAL; 24692d8f3038SAl Viro if (!check_mnt(new.mnt)) 24701da177e4SLinus Torvalds goto out1; 24711da177e4SLinus Torvalds 24722d8f3038SAl Viro error = user_path_dir(put_old, &old); 24731da177e4SLinus Torvalds if (error) 24741da177e4SLinus Torvalds goto out1; 24751da177e4SLinus Torvalds 24762d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 24771da177e4SLinus Torvalds if (error) { 24782d8f3038SAl Viro path_put(&old); 24791da177e4SLinus Torvalds goto out1; 24801da177e4SLinus Torvalds } 24811da177e4SLinus Torvalds 2482f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 2483390c6843SRam Pai down_write(&namespace_sem); 24842d8f3038SAl Viro mutex_lock(&old.dentry->d_inode->i_mutex); 24851da177e4SLinus Torvalds error = -EINVAL; 24862d8f3038SAl Viro if (IS_MNT_SHARED(old.mnt) || 24872d8f3038SAl Viro IS_MNT_SHARED(new.mnt->mnt_parent) || 24888c3ee42eSAl Viro IS_MNT_SHARED(root.mnt->mnt_parent)) 248921444403SRam Pai goto out2; 24908c3ee42eSAl Viro if (!check_mnt(root.mnt)) 24911da177e4SLinus Torvalds goto out2; 24921da177e4SLinus Torvalds error = -ENOENT; 2493d83c49f3SAl Viro if (cant_mount(old.dentry)) 24941da177e4SLinus Torvalds goto out2; 2495f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 24961da177e4SLinus Torvalds goto out2; 2497f3da392eSAlexey Dobriyan if (d_unlinked(old.dentry)) 24981da177e4SLinus Torvalds goto out2; 24991da177e4SLinus Torvalds error = -EBUSY; 25002d8f3038SAl Viro if (new.mnt == root.mnt || 25012d8f3038SAl Viro old.mnt == root.mnt) 25021da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 25031da177e4SLinus Torvalds error = -EINVAL; 25048c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 25051da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 25068c3ee42eSAl Viro if (root.mnt->mnt_parent == root.mnt) 25070bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 25082d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 25091da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 25102d8f3038SAl Viro if (new.mnt->mnt_parent == new.mnt) 25110bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 25124ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 25132d8f3038SAl Viro tmp = old.mnt; 251499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 25152d8f3038SAl Viro if (tmp != new.mnt) { 25161da177e4SLinus Torvalds for (;;) { 25171da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 25181da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 25192d8f3038SAl Viro if (tmp->mnt_parent == new.mnt) 25201da177e4SLinus Torvalds break; 25211da177e4SLinus Torvalds tmp = tmp->mnt_parent; 25221da177e4SLinus Torvalds } 25232d8f3038SAl Viro if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) 25241da177e4SLinus Torvalds goto out3; 25252d8f3038SAl Viro } else if (!is_subdir(old.dentry, new.dentry)) 25261da177e4SLinus Torvalds goto out3; 25272d8f3038SAl Viro detach_mnt(new.mnt, &parent_path); 25288c3ee42eSAl Viro detach_mnt(root.mnt, &root_parent); 25294ac91378SJan Blunck /* mount old root on put_old */ 25302d8f3038SAl Viro attach_mnt(root.mnt, &old); 25314ac91378SJan Blunck /* mount new_root on / */ 25322d8f3038SAl Viro attach_mnt(new.mnt, &root_parent); 25336b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 253499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 25352d8f3038SAl Viro chroot_fs_refs(&root, &new); 2536b3e19d92SNick Piggin 25371da177e4SLinus Torvalds error = 0; 25381a390689SAl Viro path_put(&root_parent); 25391a390689SAl Viro path_put(&parent_path); 25401da177e4SLinus Torvalds out2: 25412d8f3038SAl Viro mutex_unlock(&old.dentry->d_inode->i_mutex); 2542390c6843SRam Pai up_write(&namespace_sem); 25438c3ee42eSAl Viro path_put(&root); 25442d8f3038SAl Viro path_put(&old); 25451da177e4SLinus Torvalds out1: 25462d8f3038SAl Viro path_put(&new); 25471da177e4SLinus Torvalds out0: 25481da177e4SLinus Torvalds return error; 25491da177e4SLinus Torvalds out3: 255099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 25511da177e4SLinus Torvalds goto out2; 25521da177e4SLinus Torvalds } 25531da177e4SLinus Torvalds 25541da177e4SLinus Torvalds static void __init init_mount_tree(void) 25551da177e4SLinus Torvalds { 25561da177e4SLinus Torvalds struct vfsmount *mnt; 25576b3286edSKirill Korotaev struct mnt_namespace *ns; 2558ac748a09SJan Blunck struct path root; 25591da177e4SLinus Torvalds 25601da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 25611da177e4SLinus Torvalds if (IS_ERR(mnt)) 25621da177e4SLinus Torvalds panic("Can't create rootfs"); 2563b3e19d92SNick Piggin 25643b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 25653b22edc5STrond Myklebust if (IS_ERR(ns)) 25661da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 25671da177e4SLinus Torvalds 25686b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 25696b3286edSKirill Korotaev get_mnt_ns(ns); 25701da177e4SLinus Torvalds 2571ac748a09SJan Blunck root.mnt = ns->root; 2572ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2573ac748a09SJan Blunck 2574ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2575ac748a09SJan Blunck set_fs_root(current->fs, &root); 25761da177e4SLinus Torvalds } 25771da177e4SLinus Torvalds 257874bf17cfSDenis Cheng void __init mnt_init(void) 25791da177e4SLinus Torvalds { 258013f14b4dSEric Dumazet unsigned u; 258115a67dd8SRandy Dunlap int err; 25821da177e4SLinus Torvalds 2583390c6843SRam Pai init_rwsem(&namespace_sem); 2584390c6843SRam Pai 25851da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 258620c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 25871da177e4SLinus Torvalds 2588b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 25891da177e4SLinus Torvalds 25901da177e4SLinus Torvalds if (!mount_hashtable) 25911da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 25921da177e4SLinus Torvalds 259313f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 25941da177e4SLinus Torvalds 259513f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 259613f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 25971da177e4SLinus Torvalds 259899b7db7bSNick Piggin br_lock_init(vfsmount_lock); 259999b7db7bSNick Piggin 260015a67dd8SRandy Dunlap err = sysfs_init(); 260115a67dd8SRandy Dunlap if (err) 260215a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 26038e24eea7SHarvey Harrison __func__, err); 260400d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 260500d26666SGreg Kroah-Hartman if (!fs_kobj) 26068e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 26071da177e4SLinus Torvalds init_rootfs(); 26081da177e4SLinus Torvalds init_mount_tree(); 26091da177e4SLinus Torvalds } 26101da177e4SLinus Torvalds 2611616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 26121da177e4SLinus Torvalds { 261370fbcdf4SRam Pai LIST_HEAD(umount_list); 2614616511d0STrond Myklebust 2615d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 2616616511d0STrond Myklebust return; 2617390c6843SRam Pai down_write(&namespace_sem); 261899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 2619d498b25aSAl Viro umount_tree(ns->root, 0, &umount_list); 262099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 2621390c6843SRam Pai up_write(&namespace_sem); 262270fbcdf4SRam Pai release_mounts(&umount_list); 26236b3286edSKirill Korotaev kfree(ns); 26241da177e4SLinus Torvalds } 2625cf8d2c11STrond Myklebust EXPORT_SYMBOL(put_mnt_ns); 2626