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 186f03c6599SAl Viro unsigned int count = 0; 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 220f03c6599SAl Viro this_cpu_add(mnt->mnt_pcp->mnt_count, 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 6147e3d0eb0SAl Viro static inline void __mnt_make_longterm(struct vfsmount *mnt) 6157e3d0eb0SAl Viro { 6167e3d0eb0SAl Viro #ifdef CONFIG_SMP 6177e3d0eb0SAl Viro atomic_inc(&mnt->mnt_longterm); 6187e3d0eb0SAl Viro #endif 6197e3d0eb0SAl Viro } 6207e3d0eb0SAl Viro 6217e3d0eb0SAl Viro /* needs vfsmount lock for write */ 6227e3d0eb0SAl Viro static inline void __mnt_make_shortterm(struct vfsmount *mnt) 6237e3d0eb0SAl Viro { 6247e3d0eb0SAl Viro #ifdef CONFIG_SMP 6257e3d0eb0SAl Viro atomic_dec(&mnt->mnt_longterm); 6267e3d0eb0SAl Viro #endif 6277e3d0eb0SAl Viro } 6287e3d0eb0SAl Viro 629b90fa9aeSRam Pai /* 63099b7db7bSNick Piggin * vfsmount lock must be held for write 631b90fa9aeSRam Pai */ 632b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 633b90fa9aeSRam Pai { 634b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 635b90fa9aeSRam Pai struct vfsmount *m; 636b90fa9aeSRam Pai LIST_HEAD(head); 6376b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 638b90fa9aeSRam Pai 639b90fa9aeSRam Pai BUG_ON(parent == mnt); 640b90fa9aeSRam Pai 641b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 642f03c6599SAl Viro list_for_each_entry(m, &head, mnt_list) { 6436b3286edSKirill Korotaev m->mnt_ns = n; 6447e3d0eb0SAl Viro __mnt_make_longterm(m); 645f03c6599SAl Viro } 646f03c6599SAl Viro 647b90fa9aeSRam Pai list_splice(&head, n->list.prev); 648b90fa9aeSRam Pai 649b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 650b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 651b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 6526b3286edSKirill Korotaev touch_mnt_namespace(n); 6531da177e4SLinus Torvalds } 6541da177e4SLinus Torvalds 6551da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 6561da177e4SLinus Torvalds { 6571da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 6581da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 6591da177e4SLinus Torvalds while (1) { 6601da177e4SLinus Torvalds if (p == root) 6611da177e4SLinus Torvalds return NULL; 6621da177e4SLinus Torvalds next = p->mnt_child.next; 6631da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 6641da177e4SLinus Torvalds break; 6651da177e4SLinus Torvalds p = p->mnt_parent; 6661da177e4SLinus Torvalds } 6671da177e4SLinus Torvalds } 6681da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 6691da177e4SLinus Torvalds } 6701da177e4SLinus Torvalds 6719676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 6729676f0c6SRam Pai { 6739676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 6749676f0c6SRam Pai while (prev != &p->mnt_mounts) { 6759676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 6769676f0c6SRam Pai prev = p->mnt_mounts.prev; 6779676f0c6SRam Pai } 6789676f0c6SRam Pai return p; 6799676f0c6SRam Pai } 6809676f0c6SRam Pai 68136341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 68236341f64SRam Pai int flag) 6831da177e4SLinus Torvalds { 6841da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 6851da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 6861da177e4SLinus Torvalds 6871da177e4SLinus Torvalds if (mnt) { 688719f5d7fSMiklos Szeredi if (flag & (CL_SLAVE | CL_PRIVATE)) 689719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; /* not a peer of original */ 690719f5d7fSMiklos Szeredi else 691719f5d7fSMiklos Szeredi mnt->mnt_group_id = old->mnt_group_id; 692719f5d7fSMiklos Szeredi 693719f5d7fSMiklos Szeredi if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 694719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(mnt); 695719f5d7fSMiklos Szeredi if (err) 696719f5d7fSMiklos Szeredi goto out_free; 697719f5d7fSMiklos Szeredi } 698719f5d7fSMiklos Szeredi 699be1a16a0SMiklos Szeredi mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD; 7001da177e4SLinus Torvalds atomic_inc(&sb->s_active); 7011da177e4SLinus Torvalds mnt->mnt_sb = sb; 7021da177e4SLinus Torvalds mnt->mnt_root = dget(root); 7031da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 7041da177e4SLinus Torvalds mnt->mnt_parent = mnt; 705b90fa9aeSRam Pai 7065afe0022SRam Pai if (flag & CL_SLAVE) { 7075afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 7085afe0022SRam Pai mnt->mnt_master = old; 7095afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 7108aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 711796a6b52SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 712b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 7135afe0022SRam Pai if (IS_MNT_SLAVE(old)) 7145afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 7155afe0022SRam Pai mnt->mnt_master = old->mnt_master; 7165afe0022SRam Pai } 717b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 718b90fa9aeSRam Pai set_mnt_shared(mnt); 7191da177e4SLinus Torvalds 7201da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 7211da177e4SLinus Torvalds * as the original if that was on one */ 72236341f64SRam Pai if (flag & CL_EXPIRE) { 72355e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 72455e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 7251da177e4SLinus Torvalds } 72636341f64SRam Pai } 7271da177e4SLinus Torvalds return mnt; 728719f5d7fSMiklos Szeredi 729719f5d7fSMiklos Szeredi out_free: 730719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 731719f5d7fSMiklos Szeredi return NULL; 7321da177e4SLinus Torvalds } 7331da177e4SLinus Torvalds 734b3e19d92SNick Piggin static inline void mntfree(struct vfsmount *mnt) 7351da177e4SLinus Torvalds { 7361da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 737b3e19d92SNick Piggin 7383d733633SDave Hansen /* 7393d733633SDave Hansen * This probably indicates that somebody messed 7403d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 7413d733633SDave Hansen * happens, the filesystem was probably unable 7423d733633SDave Hansen * to make r/w->r/o transitions. 7433d733633SDave Hansen */ 744d3ef3d73Snpiggin@suse.de /* 745b3e19d92SNick Piggin * The locking used to deal with mnt_count decrement provides barriers, 746b3e19d92SNick Piggin * so mnt_get_writers() below is safe. 747d3ef3d73Snpiggin@suse.de */ 748c6653a83SNick Piggin WARN_ON(mnt_get_writers(mnt)); 749ca9c726eSAndreas Gruenbacher fsnotify_vfsmount_delete(mnt); 7501da177e4SLinus Torvalds dput(mnt->mnt_root); 7511da177e4SLinus Torvalds free_vfsmnt(mnt); 7521da177e4SLinus Torvalds deactivate_super(sb); 7531da177e4SLinus Torvalds } 7541da177e4SLinus Torvalds 755f03c6599SAl Viro static void mntput_no_expire(struct vfsmount *mnt) 7567b7b1aceSAl Viro { 757b3e19d92SNick Piggin put_again: 758f03c6599SAl Viro #ifdef CONFIG_SMP 759b3e19d92SNick Piggin br_read_lock(vfsmount_lock); 760f03c6599SAl Viro if (likely(atomic_read(&mnt->mnt_longterm))) { 761b3e19d92SNick Piggin mnt_dec_count(mnt); 762b3e19d92SNick Piggin br_read_unlock(vfsmount_lock); 76399b7db7bSNick Piggin return; 764b3e19d92SNick Piggin } 765b3e19d92SNick Piggin br_read_unlock(vfsmount_lock); 766b3e19d92SNick Piggin 76799b7db7bSNick Piggin br_write_lock(vfsmount_lock); 768b3e19d92SNick Piggin mnt_dec_count(mnt); 769b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 77099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 77199b7db7bSNick Piggin return; 77299b7db7bSNick Piggin } 773b3e19d92SNick Piggin #else 774b3e19d92SNick Piggin mnt_dec_count(mnt); 775b3e19d92SNick Piggin if (likely(mnt_get_count(mnt))) 776b3e19d92SNick Piggin return; 777b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 778f03c6599SAl Viro #endif 779b3e19d92SNick Piggin if (unlikely(mnt->mnt_pinned)) { 780b3e19d92SNick Piggin mnt_add_count(mnt, mnt->mnt_pinned + 1); 781b3e19d92SNick Piggin mnt->mnt_pinned = 0; 782b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 783b3e19d92SNick Piggin acct_auto_close_mnt(mnt); 784b3e19d92SNick Piggin goto put_again; 785b3e19d92SNick Piggin } 786b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 787b3e19d92SNick Piggin mntfree(mnt); 788b3e19d92SNick Piggin } 789b3e19d92SNick Piggin 790b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 791b3e19d92SNick Piggin { 792b3e19d92SNick Piggin if (mnt) { 793b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 794b3e19d92SNick Piggin if (unlikely(mnt->mnt_expiry_mark)) 795b3e19d92SNick Piggin mnt->mnt_expiry_mark = 0; 796f03c6599SAl Viro mntput_no_expire(mnt); 797b3e19d92SNick Piggin } 798b3e19d92SNick Piggin } 799b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 800b3e19d92SNick Piggin 801b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 802b3e19d92SNick Piggin { 803b3e19d92SNick Piggin if (mnt) 804b3e19d92SNick Piggin mnt_inc_count(mnt); 805b3e19d92SNick Piggin return mnt; 806b3e19d92SNick Piggin } 807b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 808b3e19d92SNick Piggin 8097b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 8107b7b1aceSAl Viro { 81199b7db7bSNick Piggin br_write_lock(vfsmount_lock); 8127b7b1aceSAl Viro mnt->mnt_pinned++; 81399b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 8147b7b1aceSAl Viro } 8157b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 8167b7b1aceSAl Viro 8177b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 8187b7b1aceSAl Viro { 81999b7db7bSNick Piggin br_write_lock(vfsmount_lock); 8207b7b1aceSAl Viro if (mnt->mnt_pinned) { 821b3e19d92SNick Piggin mnt_inc_count(mnt); 8227b7b1aceSAl Viro mnt->mnt_pinned--; 8237b7b1aceSAl Viro } 82499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 8257b7b1aceSAl Viro } 8267b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 8271da177e4SLinus Torvalds 828b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 829b3b304a2SMiklos Szeredi { 830b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 831b3b304a2SMiklos Szeredi } 832b3b304a2SMiklos Szeredi 833b3b304a2SMiklos Szeredi /* 834b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 835b3b304a2SMiklos Szeredi * implement more complex mount option showing. 836b3b304a2SMiklos Szeredi * 837b3b304a2SMiklos Szeredi * See also save_mount_options(). 838b3b304a2SMiklos Szeredi */ 839b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 840b3b304a2SMiklos Szeredi { 8412a32cebdSAl Viro const char *options; 8422a32cebdSAl Viro 8432a32cebdSAl Viro rcu_read_lock(); 8442a32cebdSAl Viro options = rcu_dereference(mnt->mnt_sb->s_options); 845b3b304a2SMiklos Szeredi 846b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 847b3b304a2SMiklos Szeredi seq_putc(m, ','); 848b3b304a2SMiklos Szeredi mangle(m, options); 849b3b304a2SMiklos Szeredi } 8502a32cebdSAl Viro rcu_read_unlock(); 851b3b304a2SMiklos Szeredi 852b3b304a2SMiklos Szeredi return 0; 853b3b304a2SMiklos Szeredi } 854b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 855b3b304a2SMiklos Szeredi 856b3b304a2SMiklos Szeredi /* 857b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 858b3b304a2SMiklos Szeredi * called from the fill_super() callback. 859b3b304a2SMiklos Szeredi * 860b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 861b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 862b3b304a2SMiklos Szeredi * remount fails. 863b3b304a2SMiklos Szeredi * 864b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 865b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 866b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 867b3b304a2SMiklos Szeredi * any more. 868b3b304a2SMiklos Szeredi */ 869b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 870b3b304a2SMiklos Szeredi { 8712a32cebdSAl Viro BUG_ON(sb->s_options); 8722a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 873b3b304a2SMiklos Szeredi } 874b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 875b3b304a2SMiklos Szeredi 8762a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 8772a32cebdSAl Viro { 8782a32cebdSAl Viro char *old = sb->s_options; 8792a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 8802a32cebdSAl Viro if (old) { 8812a32cebdSAl Viro synchronize_rcu(); 8822a32cebdSAl Viro kfree(old); 8832a32cebdSAl Viro } 8842a32cebdSAl Viro } 8852a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 8862a32cebdSAl Viro 887a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 8881da177e4SLinus Torvalds /* iterator */ 8891da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 8901da177e4SLinus Torvalds { 891a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 8921da177e4SLinus Torvalds 893390c6843SRam Pai down_read(&namespace_sem); 894a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 8951da177e4SLinus Torvalds } 8961da177e4SLinus Torvalds 8971da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 8981da177e4SLinus Torvalds { 899a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 900b0765fb8SPavel Emelianov 901a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 9021da177e4SLinus Torvalds } 9031da177e4SLinus Torvalds 9041da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 9051da177e4SLinus Torvalds { 906390c6843SRam Pai up_read(&namespace_sem); 9071da177e4SLinus Torvalds } 9081da177e4SLinus Torvalds 9099f5596afSAl Viro int mnt_had_events(struct proc_mounts *p) 9109f5596afSAl Viro { 9119f5596afSAl Viro struct mnt_namespace *ns = p->ns; 9129f5596afSAl Viro int res = 0; 9139f5596afSAl Viro 91499b7db7bSNick Piggin br_read_lock(vfsmount_lock); 9159f5596afSAl Viro if (p->event != ns->event) { 9169f5596afSAl Viro p->event = ns->event; 9179f5596afSAl Viro res = 1; 9189f5596afSAl Viro } 91999b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 9209f5596afSAl Viro 9219f5596afSAl Viro return res; 9229f5596afSAl Viro } 9239f5596afSAl Viro 9242d4d4864SRam Pai struct proc_fs_info { 9251da177e4SLinus Torvalds int flag; 9262d4d4864SRam Pai const char *str; 9272d4d4864SRam Pai }; 9282d4d4864SRam Pai 9292069f457SEric Paris static int show_sb_opts(struct seq_file *m, struct super_block *sb) 9302d4d4864SRam Pai { 9312d4d4864SRam Pai static const struct proc_fs_info fs_info[] = { 9321da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 9331da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 9341da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 9351da177e4SLinus Torvalds { 0, NULL } 9361da177e4SLinus Torvalds }; 9372d4d4864SRam Pai const struct proc_fs_info *fs_infop; 9382d4d4864SRam Pai 9392d4d4864SRam Pai for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 9402d4d4864SRam Pai if (sb->s_flags & fs_infop->flag) 9412d4d4864SRam Pai seq_puts(m, fs_infop->str); 9422d4d4864SRam Pai } 9432069f457SEric Paris 9442069f457SEric Paris return security_sb_show_options(m, sb); 9452d4d4864SRam Pai } 9462d4d4864SRam Pai 9472d4d4864SRam Pai static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) 9482d4d4864SRam Pai { 9492d4d4864SRam Pai static const struct proc_fs_info mnt_info[] = { 9501da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 9511da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 9521da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 953fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 954fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 95547ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 9561da177e4SLinus Torvalds { 0, NULL } 9571da177e4SLinus Torvalds }; 9582d4d4864SRam Pai const struct proc_fs_info *fs_infop; 9592d4d4864SRam Pai 9602d4d4864SRam Pai for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 9612d4d4864SRam Pai if (mnt->mnt_flags & fs_infop->flag) 9622d4d4864SRam Pai seq_puts(m, fs_infop->str); 9632d4d4864SRam Pai } 9642d4d4864SRam Pai } 9652d4d4864SRam Pai 9662d4d4864SRam Pai static void show_type(struct seq_file *m, struct super_block *sb) 9672d4d4864SRam Pai { 9682d4d4864SRam Pai mangle(m, sb->s_type->name); 9692d4d4864SRam Pai if (sb->s_subtype && sb->s_subtype[0]) { 9702d4d4864SRam Pai seq_putc(m, '.'); 9712d4d4864SRam Pai mangle(m, sb->s_subtype); 9722d4d4864SRam Pai } 9732d4d4864SRam Pai } 9742d4d4864SRam Pai 9752d4d4864SRam Pai static int show_vfsmnt(struct seq_file *m, void *v) 9762d4d4864SRam Pai { 9772d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 9782d4d4864SRam Pai int err = 0; 979c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 9801da177e4SLinus Torvalds 9811da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 9821da177e4SLinus Torvalds seq_putc(m, ' '); 983c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 9841da177e4SLinus Torvalds seq_putc(m, ' '); 9852d4d4864SRam Pai show_type(m, mnt->mnt_sb); 9862e4b7fcdSDave Hansen seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 9872069f457SEric Paris err = show_sb_opts(m, mnt->mnt_sb); 9882069f457SEric Paris if (err) 9892069f457SEric Paris goto out; 9902d4d4864SRam Pai show_mnt_opts(m, mnt); 9911da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 9921da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 9931da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 9942069f457SEric Paris out: 9951da177e4SLinus Torvalds return err; 9961da177e4SLinus Torvalds } 9971da177e4SLinus Torvalds 998a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 9991da177e4SLinus Torvalds .start = m_start, 10001da177e4SLinus Torvalds .next = m_next, 10011da177e4SLinus Torvalds .stop = m_stop, 10021da177e4SLinus Torvalds .show = show_vfsmnt 10031da177e4SLinus Torvalds }; 10041da177e4SLinus Torvalds 10052d4d4864SRam Pai static int show_mountinfo(struct seq_file *m, void *v) 10062d4d4864SRam Pai { 10072d4d4864SRam Pai struct proc_mounts *p = m->private; 10082d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 10092d4d4864SRam Pai struct super_block *sb = mnt->mnt_sb; 10102d4d4864SRam Pai struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 10112d4d4864SRam Pai struct path root = p->root; 10122d4d4864SRam Pai int err = 0; 10132d4d4864SRam Pai 10142d4d4864SRam Pai seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, 10152d4d4864SRam Pai MAJOR(sb->s_dev), MINOR(sb->s_dev)); 10162d4d4864SRam Pai seq_dentry(m, mnt->mnt_root, " \t\n\\"); 10172d4d4864SRam Pai seq_putc(m, ' '); 10182d4d4864SRam Pai seq_path_root(m, &mnt_path, &root, " \t\n\\"); 10192d4d4864SRam Pai if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { 10202d4d4864SRam Pai /* 10212d4d4864SRam Pai * Mountpoint is outside root, discard that one. Ugly, 10222d4d4864SRam Pai * but less so than trying to do that in iterator in a 10232d4d4864SRam Pai * race-free way (due to renames). 10242d4d4864SRam Pai */ 10252d4d4864SRam Pai return SEQ_SKIP; 10262d4d4864SRam Pai } 10272d4d4864SRam Pai seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); 10282d4d4864SRam Pai show_mnt_opts(m, mnt); 10292d4d4864SRam Pai 10302d4d4864SRam Pai /* Tagged fields ("foo:X" or "bar") */ 10312d4d4864SRam Pai if (IS_MNT_SHARED(mnt)) 10322d4d4864SRam Pai seq_printf(m, " shared:%i", mnt->mnt_group_id); 103397e7e0f7SMiklos Szeredi if (IS_MNT_SLAVE(mnt)) { 103497e7e0f7SMiklos Szeredi int master = mnt->mnt_master->mnt_group_id; 103597e7e0f7SMiklos Szeredi int dom = get_dominating_id(mnt, &p->root); 103697e7e0f7SMiklos Szeredi seq_printf(m, " master:%i", master); 103797e7e0f7SMiklos Szeredi if (dom && dom != master) 103897e7e0f7SMiklos Szeredi seq_printf(m, " propagate_from:%i", dom); 103997e7e0f7SMiklos Szeredi } 10402d4d4864SRam Pai if (IS_MNT_UNBINDABLE(mnt)) 10412d4d4864SRam Pai seq_puts(m, " unbindable"); 10422d4d4864SRam Pai 10432d4d4864SRam Pai /* Filesystem specific data */ 10442d4d4864SRam Pai seq_puts(m, " - "); 10452d4d4864SRam Pai show_type(m, sb); 10462d4d4864SRam Pai seq_putc(m, ' '); 10472d4d4864SRam Pai mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 10482d4d4864SRam Pai seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); 10492069f457SEric Paris err = show_sb_opts(m, sb); 10502069f457SEric Paris if (err) 10512069f457SEric Paris goto out; 10522d4d4864SRam Pai if (sb->s_op->show_options) 10532d4d4864SRam Pai err = sb->s_op->show_options(m, mnt); 10542d4d4864SRam Pai seq_putc(m, '\n'); 10552069f457SEric Paris out: 10562d4d4864SRam Pai return err; 10572d4d4864SRam Pai } 10582d4d4864SRam Pai 10592d4d4864SRam Pai const struct seq_operations mountinfo_op = { 10602d4d4864SRam Pai .start = m_start, 10612d4d4864SRam Pai .next = m_next, 10622d4d4864SRam Pai .stop = m_stop, 10632d4d4864SRam Pai .show = show_mountinfo, 10642d4d4864SRam Pai }; 10652d4d4864SRam Pai 1066b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 1067b4629fe2SChuck Lever { 1068b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 1069c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 1070b4629fe2SChuck Lever int err = 0; 1071b4629fe2SChuck Lever 1072b4629fe2SChuck Lever /* device */ 1073b4629fe2SChuck Lever if (mnt->mnt_devname) { 1074b4629fe2SChuck Lever seq_puts(m, "device "); 1075b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 1076b4629fe2SChuck Lever } else 1077b4629fe2SChuck Lever seq_puts(m, "no device"); 1078b4629fe2SChuck Lever 1079b4629fe2SChuck Lever /* mount point */ 1080b4629fe2SChuck Lever seq_puts(m, " mounted on "); 1081c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 1082b4629fe2SChuck Lever seq_putc(m, ' '); 1083b4629fe2SChuck Lever 1084b4629fe2SChuck Lever /* file system type */ 1085b4629fe2SChuck Lever seq_puts(m, "with fstype "); 10862d4d4864SRam Pai show_type(m, mnt->mnt_sb); 1087b4629fe2SChuck Lever 1088b4629fe2SChuck Lever /* optional statistics */ 1089b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 1090b4629fe2SChuck Lever seq_putc(m, ' '); 1091b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 1092b4629fe2SChuck Lever } 1093b4629fe2SChuck Lever 1094b4629fe2SChuck Lever seq_putc(m, '\n'); 1095b4629fe2SChuck Lever return err; 1096b4629fe2SChuck Lever } 1097b4629fe2SChuck Lever 1098a1a2c409SMiklos Szeredi const struct seq_operations mountstats_op = { 1099b4629fe2SChuck Lever .start = m_start, 1100b4629fe2SChuck Lever .next = m_next, 1101b4629fe2SChuck Lever .stop = m_stop, 1102b4629fe2SChuck Lever .show = show_vfsstat, 1103b4629fe2SChuck Lever }; 1104a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1105b4629fe2SChuck Lever 11061da177e4SLinus Torvalds /** 11071da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 11081da177e4SLinus Torvalds * @mnt: root of mount tree 11091da177e4SLinus Torvalds * 11101da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 11111da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 11121da177e4SLinus Torvalds * busy. 11131da177e4SLinus Torvalds */ 11141da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 11151da177e4SLinus Torvalds { 111636341f64SRam Pai int actual_refs = 0; 111736341f64SRam Pai int minimum_refs = 0; 111836341f64SRam Pai struct vfsmount *p; 11191da177e4SLinus Torvalds 1120b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1121b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 112236341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 1123b3e19d92SNick Piggin actual_refs += mnt_get_count(p); 11241da177e4SLinus Torvalds minimum_refs += 2; 11251da177e4SLinus Torvalds } 1126b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 11271da177e4SLinus Torvalds 11281da177e4SLinus Torvalds if (actual_refs > minimum_refs) 11291da177e4SLinus Torvalds return 0; 1130e3474a8eSIan Kent 1131e3474a8eSIan Kent return 1; 11321da177e4SLinus Torvalds } 11331da177e4SLinus Torvalds 11341da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 11351da177e4SLinus Torvalds 11361da177e4SLinus Torvalds /** 11371da177e4SLinus Torvalds * may_umount - check if a mount point is busy 11381da177e4SLinus Torvalds * @mnt: root of mount 11391da177e4SLinus Torvalds * 11401da177e4SLinus Torvalds * This is called to check if a mount point has any 11411da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 11421da177e4SLinus Torvalds * mount has sub mounts this will return busy 11431da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 11441da177e4SLinus Torvalds * 11451da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 11461da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 11471da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 11481da177e4SLinus Torvalds */ 11491da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 11501da177e4SLinus Torvalds { 1151e3474a8eSIan Kent int ret = 1; 11528ad08d8aSAl Viro down_read(&namespace_sem); 1153b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 1154a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 1155e3474a8eSIan Kent ret = 0; 1156b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 11578ad08d8aSAl Viro up_read(&namespace_sem); 1158a05964f3SRam Pai return ret; 11591da177e4SLinus Torvalds } 11601da177e4SLinus Torvalds 11611da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 11621da177e4SLinus Torvalds 1163b90fa9aeSRam Pai void release_mounts(struct list_head *head) 11641da177e4SLinus Torvalds { 116570fbcdf4SRam Pai struct vfsmount *mnt; 116670fbcdf4SRam Pai while (!list_empty(head)) { 1167b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 116870fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 116970fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 117070fbcdf4SRam Pai struct dentry *dentry; 117170fbcdf4SRam Pai struct vfsmount *m; 117299b7db7bSNick Piggin 117399b7db7bSNick Piggin br_write_lock(vfsmount_lock); 117470fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 117570fbcdf4SRam Pai m = mnt->mnt_parent; 117670fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 117770fbcdf4SRam Pai mnt->mnt_parent = mnt; 11787c4b93d8SAl Viro m->mnt_ghosts--; 117999b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 118070fbcdf4SRam Pai dput(dentry); 118170fbcdf4SRam Pai mntput(m); 11821da177e4SLinus Torvalds } 1183f03c6599SAl Viro mntput(mnt); 118470fbcdf4SRam Pai } 118570fbcdf4SRam Pai } 118670fbcdf4SRam Pai 118799b7db7bSNick Piggin /* 118899b7db7bSNick Piggin * vfsmount lock must be held for write 118999b7db7bSNick Piggin * namespace_sem must be held for write 119099b7db7bSNick Piggin */ 1191a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 119270fbcdf4SRam Pai { 11937b8a53fdSAl Viro LIST_HEAD(tmp_list); 119470fbcdf4SRam Pai struct vfsmount *p; 119570fbcdf4SRam Pai 11961bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 11977b8a53fdSAl Viro list_move(&p->mnt_hash, &tmp_list); 119870fbcdf4SRam Pai 1199a05964f3SRam Pai if (propagate) 12007b8a53fdSAl Viro propagate_umount(&tmp_list); 1201a05964f3SRam Pai 12027b8a53fdSAl Viro list_for_each_entry(p, &tmp_list, mnt_hash) { 120370fbcdf4SRam Pai list_del_init(&p->mnt_expire); 120470fbcdf4SRam Pai list_del_init(&p->mnt_list); 12056b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 12066b3286edSKirill Korotaev p->mnt_ns = NULL; 12077e3d0eb0SAl Viro __mnt_make_shortterm(p); 120870fbcdf4SRam Pai list_del_init(&p->mnt_child); 12097c4b93d8SAl Viro if (p->mnt_parent != p) { 12107c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 12115f57cbccSNick Piggin dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint); 12127c4b93d8SAl Viro } 1213a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 12141da177e4SLinus Torvalds } 12157b8a53fdSAl Viro list_splice(&tmp_list, kill); 12161da177e4SLinus Torvalds } 12171da177e4SLinus Torvalds 1218c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 1219c35038beSAl Viro 12201da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 12211da177e4SLinus Torvalds { 12221da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 12231da177e4SLinus Torvalds int retval; 122470fbcdf4SRam Pai LIST_HEAD(umount_list); 12251da177e4SLinus Torvalds 12261da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 12271da177e4SLinus Torvalds if (retval) 12281da177e4SLinus Torvalds return retval; 12291da177e4SLinus Torvalds 12301da177e4SLinus Torvalds /* 12311da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 12321da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 12331da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 12341da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 12351da177e4SLinus Torvalds */ 12361da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 12376ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 12381da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 12391da177e4SLinus Torvalds return -EINVAL; 12401da177e4SLinus Torvalds 1241b3e19d92SNick Piggin /* 1242b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1243b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1244b3e19d92SNick Piggin */ 1245b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 1246b3e19d92SNick Piggin if (mnt_get_count(mnt) != 2) { 1247b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 12481da177e4SLinus Torvalds return -EBUSY; 1249b3e19d92SNick Piggin } 1250b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 12531da177e4SLinus Torvalds return -EAGAIN; 12541da177e4SLinus Torvalds } 12551da177e4SLinus Torvalds 12561da177e4SLinus Torvalds /* 12571da177e4SLinus Torvalds * If we may have to abort operations to get out of this 12581da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 12591da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 12601da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 12611da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 12621da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 12631da177e4SLinus Torvalds * about for the moment. 12641da177e4SLinus Torvalds */ 12651da177e4SLinus Torvalds 126642faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 126742faad99SAl Viro sb->s_op->umount_begin(sb); 126842faad99SAl Viro } 12691da177e4SLinus Torvalds 12701da177e4SLinus Torvalds /* 12711da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 12721da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 12731da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 12741da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 12751da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 12761da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 12771da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 12781da177e4SLinus Torvalds */ 12796ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 12801da177e4SLinus Torvalds /* 12811da177e4SLinus Torvalds * Special case for "unmounting" root ... 12821da177e4SLinus Torvalds * we just try to remount it readonly. 12831da177e4SLinus Torvalds */ 12841da177e4SLinus Torvalds down_write(&sb->s_umount); 12854aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 12861da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 12871da177e4SLinus Torvalds up_write(&sb->s_umount); 12881da177e4SLinus Torvalds return retval; 12891da177e4SLinus Torvalds } 12901da177e4SLinus Torvalds 1291390c6843SRam Pai down_write(&namespace_sem); 129299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 12935addc5ddSAl Viro event++; 12941da177e4SLinus Torvalds 1295c35038beSAl Viro if (!(flags & MNT_DETACH)) 1296c35038beSAl Viro shrink_submounts(mnt, &umount_list); 1297c35038beSAl Viro 12981da177e4SLinus Torvalds retval = -EBUSY; 1299a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 13001da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 1301a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 13021da177e4SLinus Torvalds retval = 0; 13031da177e4SLinus Torvalds } 130499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1305390c6843SRam Pai up_write(&namespace_sem); 130670fbcdf4SRam Pai release_mounts(&umount_list); 13071da177e4SLinus Torvalds return retval; 13081da177e4SLinus Torvalds } 13091da177e4SLinus Torvalds 13101da177e4SLinus Torvalds /* 13111da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 13121da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 13131da177e4SLinus Torvalds * 13141da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 13151da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 13161da177e4SLinus Torvalds */ 13171da177e4SLinus Torvalds 1318bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 13191da177e4SLinus Torvalds { 13202d8f3038SAl Viro struct path path; 13211da177e4SLinus Torvalds int retval; 1322db1f05bbSMiklos Szeredi int lookup_flags = 0; 13231da177e4SLinus Torvalds 1324db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1325db1f05bbSMiklos Szeredi return -EINVAL; 1326db1f05bbSMiklos Szeredi 1327db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1328db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1329db1f05bbSMiklos Szeredi 1330db1f05bbSMiklos Szeredi retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); 13311da177e4SLinus Torvalds if (retval) 13321da177e4SLinus Torvalds goto out; 13331da177e4SLinus Torvalds retval = -EINVAL; 13342d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 13351da177e4SLinus Torvalds goto dput_and_out; 13362d8f3038SAl Viro if (!check_mnt(path.mnt)) 13371da177e4SLinus Torvalds goto dput_and_out; 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds retval = -EPERM; 13401da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 13411da177e4SLinus Torvalds goto dput_and_out; 13421da177e4SLinus Torvalds 13432d8f3038SAl Viro retval = do_umount(path.mnt, flags); 13441da177e4SLinus Torvalds dput_and_out: 1345429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 13462d8f3038SAl Viro dput(path.dentry); 13472d8f3038SAl Viro mntput_no_expire(path.mnt); 13481da177e4SLinus Torvalds out: 13491da177e4SLinus Torvalds return retval; 13501da177e4SLinus Torvalds } 13511da177e4SLinus Torvalds 13521da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 13531da177e4SLinus Torvalds 13541da177e4SLinus Torvalds /* 13551da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 13561da177e4SLinus Torvalds */ 1357bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 13581da177e4SLinus Torvalds { 13591da177e4SLinus Torvalds return sys_umount(name, 0); 13601da177e4SLinus Torvalds } 13611da177e4SLinus Torvalds 13621da177e4SLinus Torvalds #endif 13631da177e4SLinus Torvalds 13642d92ab3cSAl Viro static int mount_is_safe(struct path *path) 13651da177e4SLinus Torvalds { 13661da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 13671da177e4SLinus Torvalds return 0; 13681da177e4SLinus Torvalds return -EPERM; 13691da177e4SLinus Torvalds #ifdef notyet 13702d92ab3cSAl Viro if (S_ISLNK(path->dentry->d_inode->i_mode)) 13711da177e4SLinus Torvalds return -EPERM; 13722d92ab3cSAl Viro if (path->dentry->d_inode->i_mode & S_ISVTX) { 1373da9592edSDavid Howells if (current_uid() != path->dentry->d_inode->i_uid) 13741da177e4SLinus Torvalds return -EPERM; 13751da177e4SLinus Torvalds } 13762d92ab3cSAl Viro if (inode_permission(path->dentry->d_inode, MAY_WRITE)) 13771da177e4SLinus Torvalds return -EPERM; 13781da177e4SLinus Torvalds return 0; 13791da177e4SLinus Torvalds #endif 13801da177e4SLinus Torvalds } 13811da177e4SLinus Torvalds 1382b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 138336341f64SRam Pai int flag) 13841da177e4SLinus Torvalds { 13851da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 13861a390689SAl Viro struct path path; 13871da177e4SLinus Torvalds 13889676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 13899676f0c6SRam Pai return NULL; 13909676f0c6SRam Pai 139136341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 13921da177e4SLinus Torvalds if (!q) 13931da177e4SLinus Torvalds goto Enomem; 13941da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 13951da177e4SLinus Torvalds 13961da177e4SLinus Torvalds p = mnt; 1397fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 13987ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 13991da177e4SLinus Torvalds continue; 14001da177e4SLinus Torvalds 14011da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 14029676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 14039676f0c6SRam Pai s = skip_mnt_tree(s); 14049676f0c6SRam Pai continue; 14059676f0c6SRam Pai } 14061da177e4SLinus Torvalds while (p != s->mnt_parent) { 14071da177e4SLinus Torvalds p = p->mnt_parent; 14081da177e4SLinus Torvalds q = q->mnt_parent; 14091da177e4SLinus Torvalds } 14101da177e4SLinus Torvalds p = s; 14111a390689SAl Viro path.mnt = q; 14121a390689SAl Viro path.dentry = p->mnt_mountpoint; 141336341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 14141da177e4SLinus Torvalds if (!q) 14151da177e4SLinus Torvalds goto Enomem; 141699b7db7bSNick Piggin br_write_lock(vfsmount_lock); 14171da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 14181a390689SAl Viro attach_mnt(q, &path); 141999b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 14201da177e4SLinus Torvalds } 14211da177e4SLinus Torvalds } 14221da177e4SLinus Torvalds return res; 14231da177e4SLinus Torvalds Enomem: 14241da177e4SLinus Torvalds if (res) { 142570fbcdf4SRam Pai LIST_HEAD(umount_list); 142699b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1427a05964f3SRam Pai umount_tree(res, 0, &umount_list); 142899b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 142970fbcdf4SRam Pai release_mounts(&umount_list); 14301da177e4SLinus Torvalds } 14311da177e4SLinus Torvalds return NULL; 14321da177e4SLinus Torvalds } 14331da177e4SLinus Torvalds 1434589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 14358aec0809SAl Viro { 14368aec0809SAl Viro struct vfsmount *tree; 14371a60a280SAl Viro down_write(&namespace_sem); 1438589ff870SAl Viro tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); 14391a60a280SAl Viro up_write(&namespace_sem); 14408aec0809SAl Viro return tree; 14418aec0809SAl Viro } 14428aec0809SAl Viro 14438aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 14448aec0809SAl Viro { 14458aec0809SAl Viro LIST_HEAD(umount_list); 14461a60a280SAl Viro down_write(&namespace_sem); 144799b7db7bSNick Piggin br_write_lock(vfsmount_lock); 14488aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 144999b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 14501a60a280SAl Viro up_write(&namespace_sem); 14518aec0809SAl Viro release_mounts(&umount_list); 14528aec0809SAl Viro } 14538aec0809SAl Viro 14541f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 14551f707137SAl Viro struct vfsmount *root) 14561f707137SAl Viro { 14571f707137SAl Viro struct vfsmount *mnt; 14581f707137SAl Viro int res = f(root, arg); 14591f707137SAl Viro if (res) 14601f707137SAl Viro return res; 14611f707137SAl Viro list_for_each_entry(mnt, &root->mnt_list, mnt_list) { 14621f707137SAl Viro res = f(mnt, arg); 14631f707137SAl Viro if (res) 14641f707137SAl Viro return res; 14651f707137SAl Viro } 14661f707137SAl Viro return 0; 14671f707137SAl Viro } 14681f707137SAl Viro 1469719f5d7fSMiklos Szeredi static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) 1470719f5d7fSMiklos Szeredi { 1471719f5d7fSMiklos Szeredi struct vfsmount *p; 1472719f5d7fSMiklos Szeredi 1473719f5d7fSMiklos Szeredi for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1474719f5d7fSMiklos Szeredi if (p->mnt_group_id && !IS_MNT_SHARED(p)) 1475719f5d7fSMiklos Szeredi mnt_release_group_id(p); 1476719f5d7fSMiklos Szeredi } 1477719f5d7fSMiklos Szeredi } 1478719f5d7fSMiklos Szeredi 1479719f5d7fSMiklos Szeredi static int invent_group_ids(struct vfsmount *mnt, bool recurse) 1480719f5d7fSMiklos Szeredi { 1481719f5d7fSMiklos Szeredi struct vfsmount *p; 1482719f5d7fSMiklos Szeredi 1483719f5d7fSMiklos Szeredi for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1484719f5d7fSMiklos Szeredi if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 1485719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(p); 1486719f5d7fSMiklos Szeredi if (err) { 1487719f5d7fSMiklos Szeredi cleanup_group_ids(mnt, p); 1488719f5d7fSMiklos Szeredi return err; 1489719f5d7fSMiklos Szeredi } 1490719f5d7fSMiklos Szeredi } 1491719f5d7fSMiklos Szeredi } 1492719f5d7fSMiklos Szeredi 1493719f5d7fSMiklos Szeredi return 0; 1494719f5d7fSMiklos Szeredi } 1495719f5d7fSMiklos Szeredi 1496b90fa9aeSRam Pai /* 1497b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1498b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 149921444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 150021444403SRam Pai * store the parent mount and mountpoint dentry. 150121444403SRam Pai * (done when source_mnt is moved) 1502b90fa9aeSRam Pai * 1503b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1504b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 15059676f0c6SRam Pai * --------------------------------------------------------------------------- 1506b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 15079676f0c6SRam Pai * |************************************************************************** 15089676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15099676f0c6SRam Pai * | dest | | | | | 15109676f0c6SRam Pai * | | | | | | | 15119676f0c6SRam Pai * | v | | | | | 15129676f0c6SRam Pai * |************************************************************************** 15139676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 15145afe0022SRam Pai * | | | | | | 15159676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 15169676f0c6SRam Pai * *************************************************************************** 1517b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1518b90fa9aeSRam Pai * destination mount. 1519b90fa9aeSRam Pai * 1520b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1521b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1522b90fa9aeSRam Pai * the peer group of the source mount. 1523b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1524b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1525b90fa9aeSRam Pai * mount. 15265afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 15275afe0022SRam Pai * of the destination mount and the cloned mount is made slave 15285afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 15295afe0022SRam Pai * is marked as 'shared and slave'. 15305afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 15315afe0022SRam Pai * source mount. 15325afe0022SRam Pai * 15339676f0c6SRam Pai * --------------------------------------------------------------------------- 153421444403SRam Pai * | MOVE MOUNT OPERATION | 15359676f0c6SRam Pai * |************************************************************************** 15369676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15379676f0c6SRam Pai * | dest | | | | | 15389676f0c6SRam Pai * | | | | | | | 15399676f0c6SRam Pai * | v | | | | | 15409676f0c6SRam Pai * |************************************************************************** 15419676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 15425afe0022SRam Pai * | | | | | | 15439676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 15449676f0c6SRam Pai * *************************************************************************** 15455afe0022SRam Pai * 15465afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 15475afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 154821444403SRam Pai * (+*) the mount is moved to the destination. 15495afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 15505afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 15515afe0022SRam Pai * the mount is marked as 'shared and slave'. 15525afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1553b90fa9aeSRam Pai * 1554b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1555b90fa9aeSRam Pai * applied to each mount in the tree. 1556b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1557b90fa9aeSRam Pai * in allocations. 1558b90fa9aeSRam Pai */ 1559b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 15601a390689SAl Viro struct path *path, struct path *parent_path) 1561b90fa9aeSRam Pai { 1562b90fa9aeSRam Pai LIST_HEAD(tree_list); 15631a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 15641a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1565b90fa9aeSRam Pai struct vfsmount *child, *p; 1566719f5d7fSMiklos Szeredi int err; 1567b90fa9aeSRam Pai 1568719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) { 1569719f5d7fSMiklos Szeredi err = invent_group_ids(source_mnt, true); 1570719f5d7fSMiklos Szeredi if (err) 1571719f5d7fSMiklos Szeredi goto out; 1572719f5d7fSMiklos Szeredi } 1573719f5d7fSMiklos Szeredi err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1574719f5d7fSMiklos Szeredi if (err) 1575719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1576b90fa9aeSRam Pai 157799b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1578df1a1ad2SAl Viro 1579b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1580b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1581b90fa9aeSRam Pai set_mnt_shared(p); 1582b90fa9aeSRam Pai } 15831a390689SAl Viro if (parent_path) { 15841a390689SAl Viro detach_mnt(source_mnt, parent_path); 15851a390689SAl Viro attach_mnt(source_mnt, path); 1586e5d67f07SAl Viro touch_mnt_namespace(parent_path->mnt->mnt_ns); 158721444403SRam Pai } else { 1588b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1589b90fa9aeSRam Pai commit_tree(source_mnt); 159021444403SRam Pai } 1591b90fa9aeSRam Pai 1592b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1593b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1594b90fa9aeSRam Pai commit_tree(child); 1595b90fa9aeSRam Pai } 159699b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 159799b7db7bSNick Piggin 1598b90fa9aeSRam Pai return 0; 1599719f5d7fSMiklos Szeredi 1600719f5d7fSMiklos Szeredi out_cleanup_ids: 1601719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) 1602719f5d7fSMiklos Szeredi cleanup_group_ids(source_mnt, NULL); 1603719f5d7fSMiklos Szeredi out: 1604719f5d7fSMiklos Szeredi return err; 1605b90fa9aeSRam Pai } 1606b90fa9aeSRam Pai 16078c3ee42eSAl Viro static int graft_tree(struct vfsmount *mnt, struct path *path) 16081da177e4SLinus Torvalds { 16091da177e4SLinus Torvalds int err; 16101da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 16111da177e4SLinus Torvalds return -EINVAL; 16121da177e4SLinus Torvalds 16138c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 16141da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 16151da177e4SLinus Torvalds return -ENOTDIR; 16161da177e4SLinus Torvalds 16171da177e4SLinus Torvalds err = -ENOENT; 16188c3ee42eSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 1619d83c49f3SAl Viro if (cant_mount(path->dentry)) 16201da177e4SLinus Torvalds goto out_unlock; 16211da177e4SLinus Torvalds 1622f3da392eSAlexey Dobriyan if (!d_unlinked(path->dentry)) 16238c3ee42eSAl Viro err = attach_recursive_mnt(mnt, path, NULL); 16241da177e4SLinus Torvalds out_unlock: 16258c3ee42eSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 16261da177e4SLinus Torvalds return err; 16271da177e4SLinus Torvalds } 16281da177e4SLinus Torvalds 16291da177e4SLinus Torvalds /* 16307a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 16317a2e8a8fSValerie Aurora */ 16327a2e8a8fSValerie Aurora 16337a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 16347a2e8a8fSValerie Aurora { 16357a2e8a8fSValerie Aurora int type = flags & ~MS_REC; 16367a2e8a8fSValerie Aurora 16377a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 16387a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 16397a2e8a8fSValerie Aurora return 0; 16407a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 16417a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 16427a2e8a8fSValerie Aurora return 0; 16437a2e8a8fSValerie Aurora return type; 16447a2e8a8fSValerie Aurora } 16457a2e8a8fSValerie Aurora 16467a2e8a8fSValerie Aurora /* 164707b20889SRam Pai * recursively change the type of the mountpoint. 164807b20889SRam Pai */ 16490a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 165007b20889SRam Pai { 16512d92ab3cSAl Viro struct vfsmount *m, *mnt = path->mnt; 165207b20889SRam Pai int recurse = flag & MS_REC; 16537a2e8a8fSValerie Aurora int type; 1654719f5d7fSMiklos Szeredi int err = 0; 165507b20889SRam Pai 1656ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1657ee6f9582SMiklos Szeredi return -EPERM; 1658ee6f9582SMiklos Szeredi 16592d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 166007b20889SRam Pai return -EINVAL; 166107b20889SRam Pai 16627a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 16637a2e8a8fSValerie Aurora if (!type) 16647a2e8a8fSValerie Aurora return -EINVAL; 16657a2e8a8fSValerie Aurora 166607b20889SRam Pai down_write(&namespace_sem); 1667719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1668719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1669719f5d7fSMiklos Szeredi if (err) 1670719f5d7fSMiklos Szeredi goto out_unlock; 1671719f5d7fSMiklos Szeredi } 1672719f5d7fSMiklos Szeredi 167399b7db7bSNick Piggin br_write_lock(vfsmount_lock); 167407b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 167507b20889SRam Pai change_mnt_propagation(m, type); 167699b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1677719f5d7fSMiklos Szeredi 1678719f5d7fSMiklos Szeredi out_unlock: 167907b20889SRam Pai up_write(&namespace_sem); 1680719f5d7fSMiklos Szeredi return err; 168107b20889SRam Pai } 168207b20889SRam Pai 168307b20889SRam Pai /* 16841da177e4SLinus Torvalds * do loopback mount. 16851da177e4SLinus Torvalds */ 16860a0d8a46SAl Viro static int do_loopback(struct path *path, char *old_name, 16872dafe1c4SEric Sandeen int recurse) 16881da177e4SLinus Torvalds { 16892d92ab3cSAl Viro struct path old_path; 16901da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 16912d92ab3cSAl Viro int err = mount_is_safe(path); 16921da177e4SLinus Torvalds if (err) 16931da177e4SLinus Torvalds return err; 16941da177e4SLinus Torvalds if (!old_name || !*old_name) 16951da177e4SLinus Torvalds return -EINVAL; 16962d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 16971da177e4SLinus Torvalds if (err) 16981da177e4SLinus Torvalds return err; 16991da177e4SLinus Torvalds 1700390c6843SRam Pai down_write(&namespace_sem); 17011da177e4SLinus Torvalds err = -EINVAL; 17022d92ab3cSAl Viro if (IS_MNT_UNBINDABLE(old_path.mnt)) 17039676f0c6SRam Pai goto out; 17049676f0c6SRam Pai 17052d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 1706ccd48bc7SAl Viro goto out; 1707ccd48bc7SAl Viro 17081da177e4SLinus Torvalds err = -ENOMEM; 17091da177e4SLinus Torvalds if (recurse) 17102d92ab3cSAl Viro mnt = copy_tree(old_path.mnt, old_path.dentry, 0); 17111da177e4SLinus Torvalds else 17122d92ab3cSAl Viro mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); 17131da177e4SLinus Torvalds 1714ccd48bc7SAl Viro if (!mnt) 1715ccd48bc7SAl Viro goto out; 1716ccd48bc7SAl Viro 17172d92ab3cSAl Viro err = graft_tree(mnt, path); 17181da177e4SLinus Torvalds if (err) { 171970fbcdf4SRam Pai LIST_HEAD(umount_list); 172099b7db7bSNick Piggin 172199b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1722a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 172399b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 172470fbcdf4SRam Pai release_mounts(&umount_list); 17255b83d2c5SRam Pai } 17261da177e4SLinus Torvalds 1727ccd48bc7SAl Viro out: 1728390c6843SRam Pai up_write(&namespace_sem); 17292d92ab3cSAl Viro path_put(&old_path); 17301da177e4SLinus Torvalds return err; 17311da177e4SLinus Torvalds } 17321da177e4SLinus Torvalds 17332e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 17342e4b7fcdSDave Hansen { 17352e4b7fcdSDave Hansen int error = 0; 17362e4b7fcdSDave Hansen int readonly_request = 0; 17372e4b7fcdSDave Hansen 17382e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 17392e4b7fcdSDave Hansen readonly_request = 1; 17402e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 17412e4b7fcdSDave Hansen return 0; 17422e4b7fcdSDave Hansen 17432e4b7fcdSDave Hansen if (readonly_request) 17442e4b7fcdSDave Hansen error = mnt_make_readonly(mnt); 17452e4b7fcdSDave Hansen else 17462e4b7fcdSDave Hansen __mnt_unmake_readonly(mnt); 17472e4b7fcdSDave Hansen return error; 17482e4b7fcdSDave Hansen } 17492e4b7fcdSDave Hansen 17501da177e4SLinus Torvalds /* 17511da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 17521da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 17531da177e4SLinus Torvalds * on it - tough luck. 17541da177e4SLinus Torvalds */ 17550a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 17561da177e4SLinus Torvalds void *data) 17571da177e4SLinus Torvalds { 17581da177e4SLinus Torvalds int err; 17592d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 17601da177e4SLinus Torvalds 17611da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 17621da177e4SLinus Torvalds return -EPERM; 17631da177e4SLinus Torvalds 17642d92ab3cSAl Viro if (!check_mnt(path->mnt)) 17651da177e4SLinus Torvalds return -EINVAL; 17661da177e4SLinus Torvalds 17672d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 17681da177e4SLinus Torvalds return -EINVAL; 17691da177e4SLinus Torvalds 17701da177e4SLinus Torvalds down_write(&sb->s_umount); 17712e4b7fcdSDave Hansen if (flags & MS_BIND) 17722d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 17734aa98cf7SAl Viro else 17741da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 17757b43a79fSAl Viro if (!err) { 177699b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1777495d6c9cSValerie Aurora mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK; 17782d92ab3cSAl Viro path->mnt->mnt_flags = mnt_flags; 177999b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 17807b43a79fSAl Viro } 17811da177e4SLinus Torvalds up_write(&sb->s_umount); 17820e55a7ccSDan Williams if (!err) { 178399b7db7bSNick Piggin br_write_lock(vfsmount_lock); 17840e55a7ccSDan Williams touch_mnt_namespace(path->mnt->mnt_ns); 178599b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 17860e55a7ccSDan Williams } 17871da177e4SLinus Torvalds return err; 17881da177e4SLinus Torvalds } 17891da177e4SLinus Torvalds 17909676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 17919676f0c6SRam Pai { 17929676f0c6SRam Pai struct vfsmount *p; 17939676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 17949676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 17959676f0c6SRam Pai return 1; 17969676f0c6SRam Pai } 17979676f0c6SRam Pai return 0; 17989676f0c6SRam Pai } 17999676f0c6SRam Pai 18000a0d8a46SAl Viro static int do_move_mount(struct path *path, char *old_name) 18011da177e4SLinus Torvalds { 18022d92ab3cSAl Viro struct path old_path, parent_path; 18031da177e4SLinus Torvalds struct vfsmount *p; 18041da177e4SLinus Torvalds int err = 0; 18051da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 18061da177e4SLinus Torvalds return -EPERM; 18071da177e4SLinus Torvalds if (!old_name || !*old_name) 18081da177e4SLinus Torvalds return -EINVAL; 18092d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 18101da177e4SLinus Torvalds if (err) 18111da177e4SLinus Torvalds return err; 18121da177e4SLinus Torvalds 1813390c6843SRam Pai down_write(&namespace_sem); 1814cc53ce53SDavid Howells err = follow_down(path, true); 1815cc53ce53SDavid Howells if (err < 0) 1816cc53ce53SDavid Howells goto out; 1817cc53ce53SDavid Howells 18181da177e4SLinus Torvalds err = -EINVAL; 18192d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 18201da177e4SLinus Torvalds goto out; 18211da177e4SLinus Torvalds 18221da177e4SLinus Torvalds err = -ENOENT; 18232d92ab3cSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 1824d83c49f3SAl Viro if (cant_mount(path->dentry)) 18251da177e4SLinus Torvalds goto out1; 18261da177e4SLinus Torvalds 1827f3da392eSAlexey Dobriyan if (d_unlinked(path->dentry)) 182821444403SRam Pai goto out1; 18291da177e4SLinus Torvalds 18301da177e4SLinus Torvalds err = -EINVAL; 18312d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 183221444403SRam Pai goto out1; 18331da177e4SLinus Torvalds 18342d92ab3cSAl Viro if (old_path.mnt == old_path.mnt->mnt_parent) 183521444403SRam Pai goto out1; 18361da177e4SLinus Torvalds 18372d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 18382d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 183921444403SRam Pai goto out1; 184021444403SRam Pai /* 184121444403SRam Pai * Don't move a mount residing in a shared parent. 184221444403SRam Pai */ 18432d92ab3cSAl Viro if (old_path.mnt->mnt_parent && 18442d92ab3cSAl Viro IS_MNT_SHARED(old_path.mnt->mnt_parent)) 184521444403SRam Pai goto out1; 18469676f0c6SRam Pai /* 18479676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 18489676f0c6SRam Pai * mount which is shared. 18499676f0c6SRam Pai */ 18502d92ab3cSAl Viro if (IS_MNT_SHARED(path->mnt) && 18512d92ab3cSAl Viro tree_contains_unbindable(old_path.mnt)) 18529676f0c6SRam Pai goto out1; 18531da177e4SLinus Torvalds err = -ELOOP; 18542d92ab3cSAl Viro for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) 18552d92ab3cSAl Viro if (p == old_path.mnt) 185621444403SRam Pai goto out1; 18571da177e4SLinus Torvalds 18582d92ab3cSAl Viro err = attach_recursive_mnt(old_path.mnt, path, &parent_path); 18594ac91378SJan Blunck if (err) 186021444403SRam Pai goto out1; 18611da177e4SLinus Torvalds 18621da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 18631da177e4SLinus Torvalds * automatically */ 18642d92ab3cSAl Viro list_del_init(&old_path.mnt->mnt_expire); 18651da177e4SLinus Torvalds out1: 18662d92ab3cSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 18671da177e4SLinus Torvalds out: 1868390c6843SRam Pai up_write(&namespace_sem); 18691da177e4SLinus Torvalds if (!err) 18701a390689SAl Viro path_put(&parent_path); 18712d92ab3cSAl Viro path_put(&old_path); 18721da177e4SLinus Torvalds return err; 18731da177e4SLinus Torvalds } 18741da177e4SLinus Torvalds 1875b1e75df4SAl Viro static int do_add_mount(struct vfsmount *, struct path *, int); 1876b1e75df4SAl Viro 18771da177e4SLinus Torvalds /* 18781da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 18791da177e4SLinus Torvalds * namespace's tree 18801da177e4SLinus Torvalds */ 18810a0d8a46SAl Viro static int do_new_mount(struct path *path, char *type, int flags, 18821da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 18831da177e4SLinus Torvalds { 18841da177e4SLinus Torvalds struct vfsmount *mnt; 188515f9a3f3SAl Viro int err; 18861da177e4SLinus Torvalds 1887eca6f534SVegard Nossum if (!type) 18881da177e4SLinus Torvalds return -EINVAL; 18891da177e4SLinus Torvalds 18901da177e4SLinus Torvalds /* we need capabilities... */ 18911da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 18921da177e4SLinus Torvalds return -EPERM; 18931da177e4SLinus Torvalds 18941da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 18951da177e4SLinus Torvalds if (IS_ERR(mnt)) 18961da177e4SLinus Torvalds return PTR_ERR(mnt); 18971da177e4SLinus Torvalds 189815f9a3f3SAl Viro err = do_add_mount(mnt, path, mnt_flags); 189915f9a3f3SAl Viro if (err) 190015f9a3f3SAl Viro mntput(mnt); 190115f9a3f3SAl Viro return err; 19021da177e4SLinus Torvalds } 19031da177e4SLinus Torvalds 190419a167afSAl Viro int finish_automount(struct vfsmount *m, struct path *path) 190519a167afSAl Viro { 190619a167afSAl Viro int err; 190719a167afSAl Viro /* The new mount record should have at least 2 refs to prevent it being 190819a167afSAl Viro * expired before we get a chance to add it 190919a167afSAl Viro */ 191019a167afSAl Viro BUG_ON(mnt_get_count(m) < 2); 191119a167afSAl Viro 191219a167afSAl Viro if (m->mnt_sb == path->mnt->mnt_sb && 191319a167afSAl Viro m->mnt_root == path->dentry) { 1914b1e75df4SAl Viro err = -ELOOP; 1915b1e75df4SAl Viro goto fail; 191619a167afSAl Viro } 191719a167afSAl Viro 191819a167afSAl Viro err = do_add_mount(m, path, path->mnt->mnt_flags | MNT_SHRINKABLE); 1919b1e75df4SAl Viro if (!err) 1920b1e75df4SAl Viro return 0; 1921b1e75df4SAl Viro fail: 1922b1e75df4SAl Viro /* remove m from any expiration list it may be on */ 1923b1e75df4SAl Viro if (!list_empty(&m->mnt_expire)) { 1924b1e75df4SAl Viro down_write(&namespace_sem); 1925b1e75df4SAl Viro br_write_lock(vfsmount_lock); 1926b1e75df4SAl Viro list_del_init(&m->mnt_expire); 1927b1e75df4SAl Viro br_write_unlock(vfsmount_lock); 1928b1e75df4SAl Viro up_write(&namespace_sem); 192919a167afSAl Viro } 1930b1e75df4SAl Viro mntput(m); 1931b1e75df4SAl Viro mntput(m); 193219a167afSAl Viro return err; 193319a167afSAl Viro } 193419a167afSAl Viro 19351da177e4SLinus Torvalds /* 19361da177e4SLinus Torvalds * add a mount into a namespace's mount tree 19371da177e4SLinus Torvalds */ 1938b1e75df4SAl Viro static int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags) 19391da177e4SLinus Torvalds { 19401da177e4SLinus Torvalds int err; 19411da177e4SLinus Torvalds 19428089352aSAl Viro mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); 194327d55f1fSAl Viro 1944390c6843SRam Pai down_write(&namespace_sem); 19451da177e4SLinus Torvalds /* Something was mounted here while we slept */ 1946cc53ce53SDavid Howells err = follow_down(path, true); 1947cc53ce53SDavid Howells if (err < 0) 1948cc53ce53SDavid Howells goto unlock; 1949cc53ce53SDavid Howells 19501da177e4SLinus Torvalds err = -EINVAL; 1951dd5cae6eSAl Viro if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) 19521da177e4SLinus Torvalds goto unlock; 19531da177e4SLinus Torvalds 19541da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 19551da177e4SLinus Torvalds err = -EBUSY; 19568d66bf54SAl Viro if (path->mnt->mnt_sb == newmnt->mnt_sb && 19578d66bf54SAl Viro path->mnt->mnt_root == path->dentry) 19581da177e4SLinus Torvalds goto unlock; 19591da177e4SLinus Torvalds 19601da177e4SLinus Torvalds err = -EINVAL; 19611da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 19621da177e4SLinus Torvalds goto unlock; 19631da177e4SLinus Torvalds 19641da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 1965b1e75df4SAl Viro err = graft_tree(newmnt, path); 19661da177e4SLinus Torvalds 19671da177e4SLinus Torvalds unlock: 1968390c6843SRam Pai up_write(&namespace_sem); 19691da177e4SLinus Torvalds return err; 19701da177e4SLinus Torvalds } 19711da177e4SLinus Torvalds 1972ea5b778aSDavid Howells /** 1973ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 1974ea5b778aSDavid Howells * @mnt: The mount to list. 1975ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 1976ea5b778aSDavid Howells */ 1977ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 1978ea5b778aSDavid Howells { 1979ea5b778aSDavid Howells down_write(&namespace_sem); 1980ea5b778aSDavid Howells br_write_lock(vfsmount_lock); 1981ea5b778aSDavid Howells 1982ea5b778aSDavid Howells list_add_tail(&mnt->mnt_expire, expiry_list); 1983ea5b778aSDavid Howells 1984ea5b778aSDavid Howells br_write_unlock(vfsmount_lock); 1985ea5b778aSDavid Howells up_write(&namespace_sem); 1986ea5b778aSDavid Howells } 1987ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 1988ea5b778aSDavid Howells 1989ea5b778aSDavid Howells /* 19901da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 19911da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 19921da177e4SLinus Torvalds * here 19931da177e4SLinus Torvalds */ 19941da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 19951da177e4SLinus Torvalds { 19961da177e4SLinus Torvalds struct vfsmount *mnt, *next; 19971da177e4SLinus Torvalds LIST_HEAD(graveyard); 1998bcc5c7d2SAl Viro LIST_HEAD(umounts); 19991da177e4SLinus Torvalds 20001da177e4SLinus Torvalds if (list_empty(mounts)) 20011da177e4SLinus Torvalds return; 20021da177e4SLinus Torvalds 2003bcc5c7d2SAl Viro down_write(&namespace_sem); 200499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 20051da177e4SLinus Torvalds 20061da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 20071da177e4SLinus Torvalds * following criteria: 20081da177e4SLinus Torvalds * - only referenced by its parent vfsmount 20091da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 20101da177e4SLinus Torvalds * cleared by mntput()) 20111da177e4SLinus Torvalds */ 201255e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 20131da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 2014bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 20151da177e4SLinus Torvalds continue; 201655e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 20171da177e4SLinus Torvalds } 2018bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2019bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 2020bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 2021bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 2022bcc5c7d2SAl Viro } 202399b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 2024bcc5c7d2SAl Viro up_write(&namespace_sem); 2025bcc5c7d2SAl Viro 2026bcc5c7d2SAl Viro release_mounts(&umounts); 20271da177e4SLinus Torvalds } 20281da177e4SLinus Torvalds 20291da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 20301da177e4SLinus Torvalds 20311da177e4SLinus Torvalds /* 20325528f911STrond Myklebust * Ripoff of 'select_parent()' 20335528f911STrond Myklebust * 20345528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 20355528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 20365528f911STrond Myklebust */ 20375528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 20385528f911STrond Myklebust { 20395528f911STrond Myklebust struct vfsmount *this_parent = parent; 20405528f911STrond Myklebust struct list_head *next; 20415528f911STrond Myklebust int found = 0; 20425528f911STrond Myklebust 20435528f911STrond Myklebust repeat: 20445528f911STrond Myklebust next = this_parent->mnt_mounts.next; 20455528f911STrond Myklebust resume: 20465528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 20475528f911STrond Myklebust struct list_head *tmp = next; 20485528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 20495528f911STrond Myklebust 20505528f911STrond Myklebust next = tmp->next; 20515528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 20525528f911STrond Myklebust continue; 20535528f911STrond Myklebust /* 20545528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 20555528f911STrond Myklebust */ 20565528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 20575528f911STrond Myklebust this_parent = mnt; 20585528f911STrond Myklebust goto repeat; 20595528f911STrond Myklebust } 20605528f911STrond Myklebust 20615528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 20625528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 20635528f911STrond Myklebust found++; 20645528f911STrond Myklebust } 20655528f911STrond Myklebust } 20665528f911STrond Myklebust /* 20675528f911STrond Myklebust * All done at this level ... ascend and resume the search 20685528f911STrond Myklebust */ 20695528f911STrond Myklebust if (this_parent != parent) { 20705528f911STrond Myklebust next = this_parent->mnt_child.next; 20715528f911STrond Myklebust this_parent = this_parent->mnt_parent; 20725528f911STrond Myklebust goto resume; 20735528f911STrond Myklebust } 20745528f911STrond Myklebust return found; 20755528f911STrond Myklebust } 20765528f911STrond Myklebust 20775528f911STrond Myklebust /* 20785528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 20795528f911STrond Myklebust * submounts of a specific parent mountpoint 208099b7db7bSNick Piggin * 208199b7db7bSNick Piggin * vfsmount_lock must be held for write 20825528f911STrond Myklebust */ 2083c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 20845528f911STrond Myklebust { 20855528f911STrond Myklebust LIST_HEAD(graveyard); 2086c35038beSAl Viro struct vfsmount *m; 20875528f911STrond Myklebust 20885528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2089c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2090bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2091c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 2092bcc5c7d2SAl Viro mnt_expire); 2093afef80b3SEric W. Biederman touch_mnt_namespace(m->mnt_ns); 2094afef80b3SEric W. Biederman umount_tree(m, 1, umounts); 2095bcc5c7d2SAl Viro } 2096bcc5c7d2SAl Viro } 20975528f911STrond Myklebust } 20985528f911STrond Myklebust 20995528f911STrond Myklebust /* 21001da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 21011da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 21021da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 21031da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 21041da177e4SLinus Torvalds */ 2105b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2106b58fed8bSRam Pai unsigned long n) 21071da177e4SLinus Torvalds { 21081da177e4SLinus Torvalds char *t = to; 21091da177e4SLinus Torvalds const char __user *f = from; 21101da177e4SLinus Torvalds char c; 21111da177e4SLinus Torvalds 21121da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 21131da177e4SLinus Torvalds return n; 21141da177e4SLinus Torvalds 21151da177e4SLinus Torvalds while (n) { 21161da177e4SLinus Torvalds if (__get_user(c, f)) { 21171da177e4SLinus Torvalds memset(t, 0, n); 21181da177e4SLinus Torvalds break; 21191da177e4SLinus Torvalds } 21201da177e4SLinus Torvalds *t++ = c; 21211da177e4SLinus Torvalds f++; 21221da177e4SLinus Torvalds n--; 21231da177e4SLinus Torvalds } 21241da177e4SLinus Torvalds return n; 21251da177e4SLinus Torvalds } 21261da177e4SLinus Torvalds 21271da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 21281da177e4SLinus Torvalds { 21291da177e4SLinus Torvalds int i; 21301da177e4SLinus Torvalds unsigned long page; 21311da177e4SLinus Torvalds unsigned long size; 21321da177e4SLinus Torvalds 21331da177e4SLinus Torvalds *where = 0; 21341da177e4SLinus Torvalds if (!data) 21351da177e4SLinus Torvalds return 0; 21361da177e4SLinus Torvalds 21371da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 21381da177e4SLinus Torvalds return -ENOMEM; 21391da177e4SLinus Torvalds 21401da177e4SLinus Torvalds /* We only care that *some* data at the address the user 21411da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 21421da177e4SLinus Torvalds * the remainder of the page. 21431da177e4SLinus Torvalds */ 21441da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 21451da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 21461da177e4SLinus Torvalds if (size > PAGE_SIZE) 21471da177e4SLinus Torvalds size = PAGE_SIZE; 21481da177e4SLinus Torvalds 21491da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 21501da177e4SLinus Torvalds if (!i) { 21511da177e4SLinus Torvalds free_page(page); 21521da177e4SLinus Torvalds return -EFAULT; 21531da177e4SLinus Torvalds } 21541da177e4SLinus Torvalds if (i != PAGE_SIZE) 21551da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 21561da177e4SLinus Torvalds *where = page; 21571da177e4SLinus Torvalds return 0; 21581da177e4SLinus Torvalds } 21591da177e4SLinus Torvalds 2160eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 2161eca6f534SVegard Nossum { 2162eca6f534SVegard Nossum char *tmp; 2163eca6f534SVegard Nossum 2164eca6f534SVegard Nossum if (!data) { 2165eca6f534SVegard Nossum *where = NULL; 2166eca6f534SVegard Nossum return 0; 2167eca6f534SVegard Nossum } 2168eca6f534SVegard Nossum 2169eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 2170eca6f534SVegard Nossum if (IS_ERR(tmp)) 2171eca6f534SVegard Nossum return PTR_ERR(tmp); 2172eca6f534SVegard Nossum 2173eca6f534SVegard Nossum *where = tmp; 2174eca6f534SVegard Nossum return 0; 2175eca6f534SVegard Nossum } 2176eca6f534SVegard Nossum 21771da177e4SLinus Torvalds /* 21781da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 21791da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 21801da177e4SLinus Torvalds * 21811da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 21821da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 21831da177e4SLinus Torvalds * information (or be NULL). 21841da177e4SLinus Torvalds * 21851da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 21861da177e4SLinus Torvalds * When the flags word was introduced its top half was required 21871da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 21881da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 21891da177e4SLinus Torvalds * and must be discarded. 21901da177e4SLinus Torvalds */ 21911da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 21921da177e4SLinus Torvalds unsigned long flags, void *data_page) 21931da177e4SLinus Torvalds { 21942d92ab3cSAl Viro struct path path; 21951da177e4SLinus Torvalds int retval = 0; 21961da177e4SLinus Torvalds int mnt_flags = 0; 21971da177e4SLinus Torvalds 21981da177e4SLinus Torvalds /* Discard magic */ 21991da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 22001da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 22011da177e4SLinus Torvalds 22021da177e4SLinus Torvalds /* Basic sanity checks */ 22031da177e4SLinus Torvalds 22041da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 22051da177e4SLinus Torvalds return -EINVAL; 22061da177e4SLinus Torvalds 22071da177e4SLinus Torvalds if (data_page) 22081da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 22091da177e4SLinus Torvalds 2210a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 2211a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2212a27ab9f2STetsuo Handa if (retval) 2213a27ab9f2STetsuo Handa return retval; 2214a27ab9f2STetsuo Handa 2215a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2216a27ab9f2STetsuo Handa type_page, flags, data_page); 2217a27ab9f2STetsuo Handa if (retval) 2218a27ab9f2STetsuo Handa goto dput_out; 2219a27ab9f2STetsuo Handa 2220613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2221613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 22220a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 22230a1c01c9SMatthew Garrett 22241da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 22251da177e4SLinus Torvalds if (flags & MS_NOSUID) 22261da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 22271da177e4SLinus Torvalds if (flags & MS_NODEV) 22281da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 22291da177e4SLinus Torvalds if (flags & MS_NOEXEC) 22301da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2231fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2232fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2233fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2234fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2235d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2236d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 22372e4b7fcdSDave Hansen if (flags & MS_RDONLY) 22382e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2239fc33a7bbSChristoph Hellwig 22407a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2241d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2242d0adde57SMatthew Garrett MS_STRICTATIME); 22431da177e4SLinus Torvalds 22441da177e4SLinus Torvalds if (flags & MS_REMOUNT) 22452d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 22461da177e4SLinus Torvalds data_page); 22471da177e4SLinus Torvalds else if (flags & MS_BIND) 22482d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 22499676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 22502d92ab3cSAl Viro retval = do_change_type(&path, flags); 22511da177e4SLinus Torvalds else if (flags & MS_MOVE) 22522d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 22531da177e4SLinus Torvalds else 22542d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 22551da177e4SLinus Torvalds dev_name, data_page); 22561da177e4SLinus Torvalds dput_out: 22572d92ab3cSAl Viro path_put(&path); 22581da177e4SLinus Torvalds return retval; 22591da177e4SLinus Torvalds } 22601da177e4SLinus Torvalds 2261cf8d2c11STrond Myklebust static struct mnt_namespace *alloc_mnt_ns(void) 2262cf8d2c11STrond Myklebust { 2263cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2264cf8d2c11STrond Myklebust 2265cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2266cf8d2c11STrond Myklebust if (!new_ns) 2267cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2268cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2269cf8d2c11STrond Myklebust new_ns->root = NULL; 2270cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2271cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2272cf8d2c11STrond Myklebust new_ns->event = 0; 2273cf8d2c11STrond Myklebust return new_ns; 2274cf8d2c11STrond Myklebust } 2275cf8d2c11STrond Myklebust 2276f03c6599SAl Viro void mnt_make_longterm(struct vfsmount *mnt) 2277f03c6599SAl Viro { 22787e3d0eb0SAl Viro __mnt_make_longterm(mnt); 2279f03c6599SAl Viro } 2280f03c6599SAl Viro 2281f03c6599SAl Viro void mnt_make_shortterm(struct vfsmount *mnt) 2282f03c6599SAl Viro { 22837e3d0eb0SAl Viro #ifdef CONFIG_SMP 2284f03c6599SAl Viro if (atomic_add_unless(&mnt->mnt_longterm, -1, 1)) 2285f03c6599SAl Viro return; 2286f03c6599SAl Viro br_write_lock(vfsmount_lock); 2287f03c6599SAl Viro atomic_dec(&mnt->mnt_longterm); 2288f03c6599SAl Viro br_write_unlock(vfsmount_lock); 22897e3d0eb0SAl Viro #endif 2290f03c6599SAl Viro } 2291f03c6599SAl Viro 2292741a2951SJANAK DESAI /* 2293741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 2294741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 2295741a2951SJANAK DESAI */ 2296e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 22976b3286edSKirill Korotaev struct fs_struct *fs) 22981da177e4SLinus Torvalds { 22996b3286edSKirill Korotaev struct mnt_namespace *new_ns; 23007f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 23011da177e4SLinus Torvalds struct vfsmount *p, *q; 23021da177e4SLinus Torvalds 2303cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2304cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2305cf8d2c11STrond Myklebust return new_ns; 23061da177e4SLinus Torvalds 2307390c6843SRam Pai down_write(&namespace_sem); 23081da177e4SLinus Torvalds /* First pass: copy the tree topology */ 23096b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 23109676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 23111da177e4SLinus Torvalds if (!new_ns->root) { 2312390c6843SRam Pai up_write(&namespace_sem); 23131da177e4SLinus Torvalds kfree(new_ns); 23145cc4a034SJulia Lawall return ERR_PTR(-ENOMEM); 23151da177e4SLinus Torvalds } 231699b7db7bSNick Piggin br_write_lock(vfsmount_lock); 23171da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 231899b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 23191da177e4SLinus Torvalds 23201da177e4SLinus Torvalds /* 23211da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 23221da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 23231da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 23241da177e4SLinus Torvalds */ 23256b3286edSKirill Korotaev p = mnt_ns->root; 23261da177e4SLinus Torvalds q = new_ns->root; 23271da177e4SLinus Torvalds while (p) { 23286b3286edSKirill Korotaev q->mnt_ns = new_ns; 23297e3d0eb0SAl Viro __mnt_make_longterm(q); 23301da177e4SLinus Torvalds if (fs) { 23316ac08c39SJan Blunck if (p == fs->root.mnt) { 2332f03c6599SAl Viro fs->root.mnt = mntget(q); 23337e3d0eb0SAl Viro __mnt_make_longterm(q); 2334f03c6599SAl Viro mnt_make_shortterm(p); 23351da177e4SLinus Torvalds rootmnt = p; 23361da177e4SLinus Torvalds } 23376ac08c39SJan Blunck if (p == fs->pwd.mnt) { 2338f03c6599SAl Viro fs->pwd.mnt = mntget(q); 23397e3d0eb0SAl Viro __mnt_make_longterm(q); 2340f03c6599SAl Viro mnt_make_shortterm(p); 23411da177e4SLinus Torvalds pwdmnt = p; 23421da177e4SLinus Torvalds } 23431da177e4SLinus Torvalds } 23446b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 23451da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 23461da177e4SLinus Torvalds } 2347390c6843SRam Pai up_write(&namespace_sem); 23481da177e4SLinus Torvalds 23491da177e4SLinus Torvalds if (rootmnt) 2350f03c6599SAl Viro mntput(rootmnt); 23511da177e4SLinus Torvalds if (pwdmnt) 2352f03c6599SAl Viro mntput(pwdmnt); 23531da177e4SLinus Torvalds 2354741a2951SJANAK DESAI return new_ns; 2355741a2951SJANAK DESAI } 2356741a2951SJANAK DESAI 2357213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2358e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 2359741a2951SJANAK DESAI { 23606b3286edSKirill Korotaev struct mnt_namespace *new_ns; 2361741a2951SJANAK DESAI 2362e3222c4eSBadari Pulavarty BUG_ON(!ns); 23636b3286edSKirill Korotaev get_mnt_ns(ns); 2364741a2951SJANAK DESAI 2365741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 2366e3222c4eSBadari Pulavarty return ns; 2367741a2951SJANAK DESAI 2368e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 2369741a2951SJANAK DESAI 23706b3286edSKirill Korotaev put_mnt_ns(ns); 2371e3222c4eSBadari Pulavarty return new_ns; 23721da177e4SLinus Torvalds } 23731da177e4SLinus Torvalds 2374cf8d2c11STrond Myklebust /** 2375cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2376cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2377cf8d2c11STrond Myklebust */ 2378a2770d86SLinus Torvalds struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt) 2379cf8d2c11STrond Myklebust { 2380cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2381cf8d2c11STrond Myklebust 2382cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2383cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 2384cf8d2c11STrond Myklebust mnt->mnt_ns = new_ns; 23857e3d0eb0SAl Viro __mnt_make_longterm(mnt); 2386cf8d2c11STrond Myklebust new_ns->root = mnt; 2387cf8d2c11STrond Myklebust list_add(&new_ns->list, &new_ns->root->mnt_list); 2388cf8d2c11STrond Myklebust } 2389cf8d2c11STrond Myklebust return new_ns; 2390cf8d2c11STrond Myklebust } 2391a2770d86SLinus Torvalds EXPORT_SYMBOL(create_mnt_ns); 2392cf8d2c11STrond Myklebust 2393bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2394bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 23951da177e4SLinus Torvalds { 2396eca6f534SVegard Nossum int ret; 2397eca6f534SVegard Nossum char *kernel_type; 2398eca6f534SVegard Nossum char *kernel_dir; 2399eca6f534SVegard Nossum char *kernel_dev; 24001da177e4SLinus Torvalds unsigned long data_page; 24011da177e4SLinus Torvalds 2402eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2403eca6f534SVegard Nossum if (ret < 0) 2404eca6f534SVegard Nossum goto out_type; 24051da177e4SLinus Torvalds 2406eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2407eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2408eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2409eca6f534SVegard Nossum goto out_dir; 2410eca6f534SVegard Nossum } 24111da177e4SLinus Torvalds 2412eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2413eca6f534SVegard Nossum if (ret < 0) 2414eca6f534SVegard Nossum goto out_dev; 24151da177e4SLinus Torvalds 2416eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2417eca6f534SVegard Nossum if (ret < 0) 2418eca6f534SVegard Nossum goto out_data; 24191da177e4SLinus Torvalds 2420eca6f534SVegard Nossum ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags, 2421eca6f534SVegard Nossum (void *) data_page); 2422eca6f534SVegard Nossum 24231da177e4SLinus Torvalds free_page(data_page); 2424eca6f534SVegard Nossum out_data: 2425eca6f534SVegard Nossum kfree(kernel_dev); 2426eca6f534SVegard Nossum out_dev: 2427eca6f534SVegard Nossum putname(kernel_dir); 2428eca6f534SVegard Nossum out_dir: 2429eca6f534SVegard Nossum kfree(kernel_type); 2430eca6f534SVegard Nossum out_type: 2431eca6f534SVegard Nossum return ret; 24321da177e4SLinus Torvalds } 24331da177e4SLinus Torvalds 24341da177e4SLinus Torvalds /* 24351da177e4SLinus Torvalds * pivot_root Semantics: 24361da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 24371da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 24381da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 24391da177e4SLinus Torvalds * 24401da177e4SLinus Torvalds * Restrictions: 24411da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 24421da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 24431da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 24441da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 24451da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 24461da177e4SLinus Torvalds * 24474a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 24484a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 24494a0d11faSNeil Brown * in this situation. 24504a0d11faSNeil Brown * 24511da177e4SLinus Torvalds * Notes: 24521da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 24531da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 24541da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 24551da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 24561da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 24571da177e4SLinus Torvalds * first. 24581da177e4SLinus Torvalds */ 24593480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 24603480b257SHeiko Carstens const char __user *, put_old) 24611da177e4SLinus Torvalds { 24621da177e4SLinus Torvalds struct vfsmount *tmp; 24632d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 24641da177e4SLinus Torvalds int error; 24651da177e4SLinus Torvalds 24661da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 24671da177e4SLinus Torvalds return -EPERM; 24681da177e4SLinus Torvalds 24692d8f3038SAl Viro error = user_path_dir(new_root, &new); 24701da177e4SLinus Torvalds if (error) 24711da177e4SLinus Torvalds goto out0; 24721da177e4SLinus Torvalds error = -EINVAL; 24732d8f3038SAl Viro if (!check_mnt(new.mnt)) 24741da177e4SLinus Torvalds goto out1; 24751da177e4SLinus Torvalds 24762d8f3038SAl Viro error = user_path_dir(put_old, &old); 24771da177e4SLinus Torvalds if (error) 24781da177e4SLinus Torvalds goto out1; 24791da177e4SLinus Torvalds 24802d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 24811da177e4SLinus Torvalds if (error) { 24822d8f3038SAl Viro path_put(&old); 24831da177e4SLinus Torvalds goto out1; 24841da177e4SLinus Torvalds } 24851da177e4SLinus Torvalds 2486f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 2487390c6843SRam Pai down_write(&namespace_sem); 24882d8f3038SAl Viro mutex_lock(&old.dentry->d_inode->i_mutex); 24891da177e4SLinus Torvalds error = -EINVAL; 24902d8f3038SAl Viro if (IS_MNT_SHARED(old.mnt) || 24912d8f3038SAl Viro IS_MNT_SHARED(new.mnt->mnt_parent) || 24928c3ee42eSAl Viro IS_MNT_SHARED(root.mnt->mnt_parent)) 249321444403SRam Pai goto out2; 24948c3ee42eSAl Viro if (!check_mnt(root.mnt)) 24951da177e4SLinus Torvalds goto out2; 24961da177e4SLinus Torvalds error = -ENOENT; 2497d83c49f3SAl Viro if (cant_mount(old.dentry)) 24981da177e4SLinus Torvalds goto out2; 2499f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 25001da177e4SLinus Torvalds goto out2; 2501f3da392eSAlexey Dobriyan if (d_unlinked(old.dentry)) 25021da177e4SLinus Torvalds goto out2; 25031da177e4SLinus Torvalds error = -EBUSY; 25042d8f3038SAl Viro if (new.mnt == root.mnt || 25052d8f3038SAl Viro old.mnt == root.mnt) 25061da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 25071da177e4SLinus Torvalds error = -EINVAL; 25088c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 25091da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 25108c3ee42eSAl Viro if (root.mnt->mnt_parent == root.mnt) 25110bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 25122d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 25131da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 25142d8f3038SAl Viro if (new.mnt->mnt_parent == new.mnt) 25150bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 25164ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 25172d8f3038SAl Viro tmp = old.mnt; 251899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 25192d8f3038SAl Viro if (tmp != new.mnt) { 25201da177e4SLinus Torvalds for (;;) { 25211da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 25221da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 25232d8f3038SAl Viro if (tmp->mnt_parent == new.mnt) 25241da177e4SLinus Torvalds break; 25251da177e4SLinus Torvalds tmp = tmp->mnt_parent; 25261da177e4SLinus Torvalds } 25272d8f3038SAl Viro if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) 25281da177e4SLinus Torvalds goto out3; 25292d8f3038SAl Viro } else if (!is_subdir(old.dentry, new.dentry)) 25301da177e4SLinus Torvalds goto out3; 25312d8f3038SAl Viro detach_mnt(new.mnt, &parent_path); 25328c3ee42eSAl Viro detach_mnt(root.mnt, &root_parent); 25334ac91378SJan Blunck /* mount old root on put_old */ 25342d8f3038SAl Viro attach_mnt(root.mnt, &old); 25354ac91378SJan Blunck /* mount new_root on / */ 25362d8f3038SAl Viro attach_mnt(new.mnt, &root_parent); 25376b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 253899b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 25392d8f3038SAl Viro chroot_fs_refs(&root, &new); 2540b3e19d92SNick Piggin 25411da177e4SLinus Torvalds error = 0; 25421a390689SAl Viro path_put(&root_parent); 25431a390689SAl Viro path_put(&parent_path); 25441da177e4SLinus Torvalds out2: 25452d8f3038SAl Viro mutex_unlock(&old.dentry->d_inode->i_mutex); 2546390c6843SRam Pai up_write(&namespace_sem); 25478c3ee42eSAl Viro path_put(&root); 25482d8f3038SAl Viro path_put(&old); 25491da177e4SLinus Torvalds out1: 25502d8f3038SAl Viro path_put(&new); 25511da177e4SLinus Torvalds out0: 25521da177e4SLinus Torvalds return error; 25531da177e4SLinus Torvalds out3: 255499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 25551da177e4SLinus Torvalds goto out2; 25561da177e4SLinus Torvalds } 25571da177e4SLinus Torvalds 25581da177e4SLinus Torvalds static void __init init_mount_tree(void) 25591da177e4SLinus Torvalds { 25601da177e4SLinus Torvalds struct vfsmount *mnt; 25616b3286edSKirill Korotaev struct mnt_namespace *ns; 2562ac748a09SJan Blunck struct path root; 25631da177e4SLinus Torvalds 25641da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 25651da177e4SLinus Torvalds if (IS_ERR(mnt)) 25661da177e4SLinus Torvalds panic("Can't create rootfs"); 2567b3e19d92SNick Piggin 25683b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 25693b22edc5STrond Myklebust if (IS_ERR(ns)) 25701da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 25711da177e4SLinus Torvalds 25726b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 25736b3286edSKirill Korotaev get_mnt_ns(ns); 25741da177e4SLinus Torvalds 2575ac748a09SJan Blunck root.mnt = ns->root; 2576ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2577ac748a09SJan Blunck 2578ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2579ac748a09SJan Blunck set_fs_root(current->fs, &root); 25801da177e4SLinus Torvalds } 25811da177e4SLinus Torvalds 258274bf17cfSDenis Cheng void __init mnt_init(void) 25831da177e4SLinus Torvalds { 258413f14b4dSEric Dumazet unsigned u; 258515a67dd8SRandy Dunlap int err; 25861da177e4SLinus Torvalds 2587390c6843SRam Pai init_rwsem(&namespace_sem); 2588390c6843SRam Pai 25891da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 259020c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 25911da177e4SLinus Torvalds 2592b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 25931da177e4SLinus Torvalds 25941da177e4SLinus Torvalds if (!mount_hashtable) 25951da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 25961da177e4SLinus Torvalds 259713f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 25981da177e4SLinus Torvalds 259913f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 260013f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 26011da177e4SLinus Torvalds 260299b7db7bSNick Piggin br_lock_init(vfsmount_lock); 260399b7db7bSNick Piggin 260415a67dd8SRandy Dunlap err = sysfs_init(); 260515a67dd8SRandy Dunlap if (err) 260615a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 26078e24eea7SHarvey Harrison __func__, err); 260800d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 260900d26666SGreg Kroah-Hartman if (!fs_kobj) 26108e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 26111da177e4SLinus Torvalds init_rootfs(); 26121da177e4SLinus Torvalds init_mount_tree(); 26131da177e4SLinus Torvalds } 26141da177e4SLinus Torvalds 2615616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 26161da177e4SLinus Torvalds { 261770fbcdf4SRam Pai LIST_HEAD(umount_list); 2618616511d0STrond Myklebust 2619d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 2620616511d0STrond Myklebust return; 2621390c6843SRam Pai down_write(&namespace_sem); 262299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 2623d498b25aSAl Viro umount_tree(ns->root, 0, &umount_list); 262499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 2625390c6843SRam Pai up_write(&namespace_sem); 262670fbcdf4SRam Pai release_mounts(&umount_list); 26276b3286edSKirill Korotaev kfree(ns); 26281da177e4SLinus Torvalds } 2629cf8d2c11STrond Myklebust EXPORT_SYMBOL(put_mnt_ns); 2630