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 614b90fa9aeSRam Pai /* 61599b7db7bSNick Piggin * vfsmount lock must be held for write 616b90fa9aeSRam Pai */ 617b90fa9aeSRam Pai static void commit_tree(struct vfsmount *mnt) 618b90fa9aeSRam Pai { 619b90fa9aeSRam Pai struct vfsmount *parent = mnt->mnt_parent; 620b90fa9aeSRam Pai struct vfsmount *m; 621b90fa9aeSRam Pai LIST_HEAD(head); 6226b3286edSKirill Korotaev struct mnt_namespace *n = parent->mnt_ns; 623b90fa9aeSRam Pai 624b90fa9aeSRam Pai BUG_ON(parent == mnt); 625b90fa9aeSRam Pai 626b90fa9aeSRam Pai list_add_tail(&head, &mnt->mnt_list); 627f03c6599SAl Viro list_for_each_entry(m, &head, mnt_list) { 6286b3286edSKirill Korotaev m->mnt_ns = n; 629f03c6599SAl Viro atomic_inc(&m->mnt_longterm); 630f03c6599SAl Viro } 631f03c6599SAl Viro 632b90fa9aeSRam Pai list_splice(&head, n->list.prev); 633b90fa9aeSRam Pai 634b90fa9aeSRam Pai list_add_tail(&mnt->mnt_hash, mount_hashtable + 635b90fa9aeSRam Pai hash(parent, mnt->mnt_mountpoint)); 636b90fa9aeSRam Pai list_add_tail(&mnt->mnt_child, &parent->mnt_mounts); 6376b3286edSKirill Korotaev touch_mnt_namespace(n); 6381da177e4SLinus Torvalds } 6391da177e4SLinus Torvalds 6401da177e4SLinus Torvalds static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root) 6411da177e4SLinus Torvalds { 6421da177e4SLinus Torvalds struct list_head *next = p->mnt_mounts.next; 6431da177e4SLinus Torvalds if (next == &p->mnt_mounts) { 6441da177e4SLinus Torvalds while (1) { 6451da177e4SLinus Torvalds if (p == root) 6461da177e4SLinus Torvalds return NULL; 6471da177e4SLinus Torvalds next = p->mnt_child.next; 6481da177e4SLinus Torvalds if (next != &p->mnt_parent->mnt_mounts) 6491da177e4SLinus Torvalds break; 6501da177e4SLinus Torvalds p = p->mnt_parent; 6511da177e4SLinus Torvalds } 6521da177e4SLinus Torvalds } 6531da177e4SLinus Torvalds return list_entry(next, struct vfsmount, mnt_child); 6541da177e4SLinus Torvalds } 6551da177e4SLinus Torvalds 6569676f0c6SRam Pai static struct vfsmount *skip_mnt_tree(struct vfsmount *p) 6579676f0c6SRam Pai { 6589676f0c6SRam Pai struct list_head *prev = p->mnt_mounts.prev; 6599676f0c6SRam Pai while (prev != &p->mnt_mounts) { 6609676f0c6SRam Pai p = list_entry(prev, struct vfsmount, mnt_child); 6619676f0c6SRam Pai prev = p->mnt_mounts.prev; 6629676f0c6SRam Pai } 6639676f0c6SRam Pai return p; 6649676f0c6SRam Pai } 6659676f0c6SRam Pai 66636341f64SRam Pai static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root, 66736341f64SRam Pai int flag) 6681da177e4SLinus Torvalds { 6691da177e4SLinus Torvalds struct super_block *sb = old->mnt_sb; 6701da177e4SLinus Torvalds struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname); 6711da177e4SLinus Torvalds 6721da177e4SLinus Torvalds if (mnt) { 673719f5d7fSMiklos Szeredi if (flag & (CL_SLAVE | CL_PRIVATE)) 674719f5d7fSMiklos Szeredi mnt->mnt_group_id = 0; /* not a peer of original */ 675719f5d7fSMiklos Szeredi else 676719f5d7fSMiklos Szeredi mnt->mnt_group_id = old->mnt_group_id; 677719f5d7fSMiklos Szeredi 678719f5d7fSMiklos Szeredi if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) { 679719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(mnt); 680719f5d7fSMiklos Szeredi if (err) 681719f5d7fSMiklos Szeredi goto out_free; 682719f5d7fSMiklos Szeredi } 683719f5d7fSMiklos Szeredi 684be1a16a0SMiklos Szeredi mnt->mnt_flags = old->mnt_flags & ~MNT_WRITE_HOLD; 6851da177e4SLinus Torvalds atomic_inc(&sb->s_active); 6861da177e4SLinus Torvalds mnt->mnt_sb = sb; 6871da177e4SLinus Torvalds mnt->mnt_root = dget(root); 6881da177e4SLinus Torvalds mnt->mnt_mountpoint = mnt->mnt_root; 6891da177e4SLinus Torvalds mnt->mnt_parent = mnt; 690b90fa9aeSRam Pai 6915afe0022SRam Pai if (flag & CL_SLAVE) { 6925afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave_list); 6935afe0022SRam Pai mnt->mnt_master = old; 6945afe0022SRam Pai CLEAR_MNT_SHARED(mnt); 6958aec0809SAl Viro } else if (!(flag & CL_PRIVATE)) { 696796a6b52SAl Viro if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old)) 697b90fa9aeSRam Pai list_add(&mnt->mnt_share, &old->mnt_share); 6985afe0022SRam Pai if (IS_MNT_SLAVE(old)) 6995afe0022SRam Pai list_add(&mnt->mnt_slave, &old->mnt_slave); 7005afe0022SRam Pai mnt->mnt_master = old->mnt_master; 7015afe0022SRam Pai } 702b90fa9aeSRam Pai if (flag & CL_MAKE_SHARED) 703b90fa9aeSRam Pai set_mnt_shared(mnt); 7041da177e4SLinus Torvalds 7051da177e4SLinus Torvalds /* stick the duplicate mount on the same expiry list 7061da177e4SLinus Torvalds * as the original if that was on one */ 70736341f64SRam Pai if (flag & CL_EXPIRE) { 70855e700b9SMiklos Szeredi if (!list_empty(&old->mnt_expire)) 70955e700b9SMiklos Szeredi list_add(&mnt->mnt_expire, &old->mnt_expire); 7101da177e4SLinus Torvalds } 71136341f64SRam Pai } 7121da177e4SLinus Torvalds return mnt; 713719f5d7fSMiklos Szeredi 714719f5d7fSMiklos Szeredi out_free: 715719f5d7fSMiklos Szeredi free_vfsmnt(mnt); 716719f5d7fSMiklos Szeredi return NULL; 7171da177e4SLinus Torvalds } 7181da177e4SLinus Torvalds 719b3e19d92SNick Piggin static inline void mntfree(struct vfsmount *mnt) 7201da177e4SLinus Torvalds { 7211da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 722b3e19d92SNick Piggin 7233d733633SDave Hansen /* 7243d733633SDave Hansen * This probably indicates that somebody messed 7253d733633SDave Hansen * up a mnt_want/drop_write() pair. If this 7263d733633SDave Hansen * happens, the filesystem was probably unable 7273d733633SDave Hansen * to make r/w->r/o transitions. 7283d733633SDave Hansen */ 729d3ef3d73Snpiggin@suse.de /* 730b3e19d92SNick Piggin * The locking used to deal with mnt_count decrement provides barriers, 731b3e19d92SNick Piggin * so mnt_get_writers() below is safe. 732d3ef3d73Snpiggin@suse.de */ 733c6653a83SNick Piggin WARN_ON(mnt_get_writers(mnt)); 734ca9c726eSAndreas Gruenbacher fsnotify_vfsmount_delete(mnt); 7351da177e4SLinus Torvalds dput(mnt->mnt_root); 7361da177e4SLinus Torvalds free_vfsmnt(mnt); 7371da177e4SLinus Torvalds deactivate_super(sb); 7381da177e4SLinus Torvalds } 7391da177e4SLinus Torvalds 740f03c6599SAl Viro static void mntput_no_expire(struct vfsmount *mnt) 7417b7b1aceSAl Viro { 742b3e19d92SNick Piggin put_again: 743f03c6599SAl Viro #ifdef CONFIG_SMP 744b3e19d92SNick Piggin br_read_lock(vfsmount_lock); 745f03c6599SAl Viro if (likely(atomic_read(&mnt->mnt_longterm))) { 746b3e19d92SNick Piggin mnt_dec_count(mnt); 747b3e19d92SNick Piggin br_read_unlock(vfsmount_lock); 74899b7db7bSNick Piggin return; 749b3e19d92SNick Piggin } 750b3e19d92SNick Piggin br_read_unlock(vfsmount_lock); 751b3e19d92SNick Piggin 75299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 753b3e19d92SNick Piggin mnt_dec_count(mnt); 754b3e19d92SNick Piggin if (mnt_get_count(mnt)) { 75599b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 75699b7db7bSNick Piggin return; 75799b7db7bSNick Piggin } 758b3e19d92SNick Piggin #else 759b3e19d92SNick Piggin mnt_dec_count(mnt); 760b3e19d92SNick Piggin if (likely(mnt_get_count(mnt))) 761b3e19d92SNick Piggin return; 762b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 763f03c6599SAl Viro #endif 764b3e19d92SNick Piggin if (unlikely(mnt->mnt_pinned)) { 765b3e19d92SNick Piggin mnt_add_count(mnt, mnt->mnt_pinned + 1); 766b3e19d92SNick Piggin mnt->mnt_pinned = 0; 767b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 768b3e19d92SNick Piggin acct_auto_close_mnt(mnt); 769b3e19d92SNick Piggin goto put_again; 770b3e19d92SNick Piggin } 771b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 772b3e19d92SNick Piggin mntfree(mnt); 773b3e19d92SNick Piggin } 774b3e19d92SNick Piggin 775b3e19d92SNick Piggin void mntput(struct vfsmount *mnt) 776b3e19d92SNick Piggin { 777b3e19d92SNick Piggin if (mnt) { 778b3e19d92SNick Piggin /* avoid cacheline pingpong, hope gcc doesn't get "smart" */ 779b3e19d92SNick Piggin if (unlikely(mnt->mnt_expiry_mark)) 780b3e19d92SNick Piggin mnt->mnt_expiry_mark = 0; 781f03c6599SAl Viro mntput_no_expire(mnt); 782b3e19d92SNick Piggin } 783b3e19d92SNick Piggin } 784b3e19d92SNick Piggin EXPORT_SYMBOL(mntput); 785b3e19d92SNick Piggin 786b3e19d92SNick Piggin struct vfsmount *mntget(struct vfsmount *mnt) 787b3e19d92SNick Piggin { 788b3e19d92SNick Piggin if (mnt) 789b3e19d92SNick Piggin mnt_inc_count(mnt); 790b3e19d92SNick Piggin return mnt; 791b3e19d92SNick Piggin } 792b3e19d92SNick Piggin EXPORT_SYMBOL(mntget); 793b3e19d92SNick Piggin 7947b7b1aceSAl Viro void mnt_pin(struct vfsmount *mnt) 7957b7b1aceSAl Viro { 79699b7db7bSNick Piggin br_write_lock(vfsmount_lock); 7977b7b1aceSAl Viro mnt->mnt_pinned++; 79899b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 7997b7b1aceSAl Viro } 8007b7b1aceSAl Viro EXPORT_SYMBOL(mnt_pin); 8017b7b1aceSAl Viro 8027b7b1aceSAl Viro void mnt_unpin(struct vfsmount *mnt) 8037b7b1aceSAl Viro { 80499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 8057b7b1aceSAl Viro if (mnt->mnt_pinned) { 806b3e19d92SNick Piggin mnt_inc_count(mnt); 8077b7b1aceSAl Viro mnt->mnt_pinned--; 8087b7b1aceSAl Viro } 80999b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 8107b7b1aceSAl Viro } 8117b7b1aceSAl Viro EXPORT_SYMBOL(mnt_unpin); 8121da177e4SLinus Torvalds 813b3b304a2SMiklos Szeredi static inline void mangle(struct seq_file *m, const char *s) 814b3b304a2SMiklos Szeredi { 815b3b304a2SMiklos Szeredi seq_escape(m, s, " \t\n\\"); 816b3b304a2SMiklos Szeredi } 817b3b304a2SMiklos Szeredi 818b3b304a2SMiklos Szeredi /* 819b3b304a2SMiklos Szeredi * Simple .show_options callback for filesystems which don't want to 820b3b304a2SMiklos Szeredi * implement more complex mount option showing. 821b3b304a2SMiklos Szeredi * 822b3b304a2SMiklos Szeredi * See also save_mount_options(). 823b3b304a2SMiklos Szeredi */ 824b3b304a2SMiklos Szeredi int generic_show_options(struct seq_file *m, struct vfsmount *mnt) 825b3b304a2SMiklos Szeredi { 8262a32cebdSAl Viro const char *options; 8272a32cebdSAl Viro 8282a32cebdSAl Viro rcu_read_lock(); 8292a32cebdSAl Viro options = rcu_dereference(mnt->mnt_sb->s_options); 830b3b304a2SMiklos Szeredi 831b3b304a2SMiklos Szeredi if (options != NULL && options[0]) { 832b3b304a2SMiklos Szeredi seq_putc(m, ','); 833b3b304a2SMiklos Szeredi mangle(m, options); 834b3b304a2SMiklos Szeredi } 8352a32cebdSAl Viro rcu_read_unlock(); 836b3b304a2SMiklos Szeredi 837b3b304a2SMiklos Szeredi return 0; 838b3b304a2SMiklos Szeredi } 839b3b304a2SMiklos Szeredi EXPORT_SYMBOL(generic_show_options); 840b3b304a2SMiklos Szeredi 841b3b304a2SMiklos Szeredi /* 842b3b304a2SMiklos Szeredi * If filesystem uses generic_show_options(), this function should be 843b3b304a2SMiklos Szeredi * called from the fill_super() callback. 844b3b304a2SMiklos Szeredi * 845b3b304a2SMiklos Szeredi * The .remount_fs callback usually needs to be handled in a special 846b3b304a2SMiklos Szeredi * way, to make sure, that previous options are not overwritten if the 847b3b304a2SMiklos Szeredi * remount fails. 848b3b304a2SMiklos Szeredi * 849b3b304a2SMiklos Szeredi * Also note, that if the filesystem's .remount_fs function doesn't 850b3b304a2SMiklos Szeredi * reset all options to their default value, but changes only newly 851b3b304a2SMiklos Szeredi * given options, then the displayed options will not reflect reality 852b3b304a2SMiklos Szeredi * any more. 853b3b304a2SMiklos Szeredi */ 854b3b304a2SMiklos Szeredi void save_mount_options(struct super_block *sb, char *options) 855b3b304a2SMiklos Szeredi { 8562a32cebdSAl Viro BUG_ON(sb->s_options); 8572a32cebdSAl Viro rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL)); 858b3b304a2SMiklos Szeredi } 859b3b304a2SMiklos Szeredi EXPORT_SYMBOL(save_mount_options); 860b3b304a2SMiklos Szeredi 8612a32cebdSAl Viro void replace_mount_options(struct super_block *sb, char *options) 8622a32cebdSAl Viro { 8632a32cebdSAl Viro char *old = sb->s_options; 8642a32cebdSAl Viro rcu_assign_pointer(sb->s_options, options); 8652a32cebdSAl Viro if (old) { 8662a32cebdSAl Viro synchronize_rcu(); 8672a32cebdSAl Viro kfree(old); 8682a32cebdSAl Viro } 8692a32cebdSAl Viro } 8702a32cebdSAl Viro EXPORT_SYMBOL(replace_mount_options); 8712a32cebdSAl Viro 872a1a2c409SMiklos Szeredi #ifdef CONFIG_PROC_FS 8731da177e4SLinus Torvalds /* iterator */ 8741da177e4SLinus Torvalds static void *m_start(struct seq_file *m, loff_t *pos) 8751da177e4SLinus Torvalds { 876a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 8771da177e4SLinus Torvalds 878390c6843SRam Pai down_read(&namespace_sem); 879a1a2c409SMiklos Szeredi return seq_list_start(&p->ns->list, *pos); 8801da177e4SLinus Torvalds } 8811da177e4SLinus Torvalds 8821da177e4SLinus Torvalds static void *m_next(struct seq_file *m, void *v, loff_t *pos) 8831da177e4SLinus Torvalds { 884a1a2c409SMiklos Szeredi struct proc_mounts *p = m->private; 885b0765fb8SPavel Emelianov 886a1a2c409SMiklos Szeredi return seq_list_next(v, &p->ns->list, pos); 8871da177e4SLinus Torvalds } 8881da177e4SLinus Torvalds 8891da177e4SLinus Torvalds static void m_stop(struct seq_file *m, void *v) 8901da177e4SLinus Torvalds { 891390c6843SRam Pai up_read(&namespace_sem); 8921da177e4SLinus Torvalds } 8931da177e4SLinus Torvalds 8949f5596afSAl Viro int mnt_had_events(struct proc_mounts *p) 8959f5596afSAl Viro { 8969f5596afSAl Viro struct mnt_namespace *ns = p->ns; 8979f5596afSAl Viro int res = 0; 8989f5596afSAl Viro 89999b7db7bSNick Piggin br_read_lock(vfsmount_lock); 9009f5596afSAl Viro if (p->event != ns->event) { 9019f5596afSAl Viro p->event = ns->event; 9029f5596afSAl Viro res = 1; 9039f5596afSAl Viro } 90499b7db7bSNick Piggin br_read_unlock(vfsmount_lock); 9059f5596afSAl Viro 9069f5596afSAl Viro return res; 9079f5596afSAl Viro } 9089f5596afSAl Viro 9092d4d4864SRam Pai struct proc_fs_info { 9101da177e4SLinus Torvalds int flag; 9112d4d4864SRam Pai const char *str; 9122d4d4864SRam Pai }; 9132d4d4864SRam Pai 9142069f457SEric Paris static int show_sb_opts(struct seq_file *m, struct super_block *sb) 9152d4d4864SRam Pai { 9162d4d4864SRam Pai static const struct proc_fs_info fs_info[] = { 9171da177e4SLinus Torvalds { MS_SYNCHRONOUS, ",sync" }, 9181da177e4SLinus Torvalds { MS_DIRSYNC, ",dirsync" }, 9191da177e4SLinus Torvalds { MS_MANDLOCK, ",mand" }, 9201da177e4SLinus Torvalds { 0, NULL } 9211da177e4SLinus Torvalds }; 9222d4d4864SRam Pai const struct proc_fs_info *fs_infop; 9232d4d4864SRam Pai 9242d4d4864SRam Pai for (fs_infop = fs_info; fs_infop->flag; fs_infop++) { 9252d4d4864SRam Pai if (sb->s_flags & fs_infop->flag) 9262d4d4864SRam Pai seq_puts(m, fs_infop->str); 9272d4d4864SRam Pai } 9282069f457SEric Paris 9292069f457SEric Paris return security_sb_show_options(m, sb); 9302d4d4864SRam Pai } 9312d4d4864SRam Pai 9322d4d4864SRam Pai static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt) 9332d4d4864SRam Pai { 9342d4d4864SRam Pai static const struct proc_fs_info mnt_info[] = { 9351da177e4SLinus Torvalds { MNT_NOSUID, ",nosuid" }, 9361da177e4SLinus Torvalds { MNT_NODEV, ",nodev" }, 9371da177e4SLinus Torvalds { MNT_NOEXEC, ",noexec" }, 938fc33a7bbSChristoph Hellwig { MNT_NOATIME, ",noatime" }, 939fc33a7bbSChristoph Hellwig { MNT_NODIRATIME, ",nodiratime" }, 94047ae32d6SValerie Henson { MNT_RELATIME, ",relatime" }, 9411da177e4SLinus Torvalds { 0, NULL } 9421da177e4SLinus Torvalds }; 9432d4d4864SRam Pai const struct proc_fs_info *fs_infop; 9442d4d4864SRam Pai 9452d4d4864SRam Pai for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) { 9462d4d4864SRam Pai if (mnt->mnt_flags & fs_infop->flag) 9472d4d4864SRam Pai seq_puts(m, fs_infop->str); 9482d4d4864SRam Pai } 9492d4d4864SRam Pai } 9502d4d4864SRam Pai 9512d4d4864SRam Pai static void show_type(struct seq_file *m, struct super_block *sb) 9522d4d4864SRam Pai { 9532d4d4864SRam Pai mangle(m, sb->s_type->name); 9542d4d4864SRam Pai if (sb->s_subtype && sb->s_subtype[0]) { 9552d4d4864SRam Pai seq_putc(m, '.'); 9562d4d4864SRam Pai mangle(m, sb->s_subtype); 9572d4d4864SRam Pai } 9582d4d4864SRam Pai } 9592d4d4864SRam Pai 9602d4d4864SRam Pai static int show_vfsmnt(struct seq_file *m, void *v) 9612d4d4864SRam Pai { 9622d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 9632d4d4864SRam Pai int err = 0; 964c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 9651da177e4SLinus Torvalds 9661da177e4SLinus Torvalds mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 9671da177e4SLinus Torvalds seq_putc(m, ' '); 968c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 9691da177e4SLinus Torvalds seq_putc(m, ' '); 9702d4d4864SRam Pai show_type(m, mnt->mnt_sb); 9712e4b7fcdSDave Hansen seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw"); 9722069f457SEric Paris err = show_sb_opts(m, mnt->mnt_sb); 9732069f457SEric Paris if (err) 9742069f457SEric Paris goto out; 9752d4d4864SRam Pai show_mnt_opts(m, mnt); 9761da177e4SLinus Torvalds if (mnt->mnt_sb->s_op->show_options) 9771da177e4SLinus Torvalds err = mnt->mnt_sb->s_op->show_options(m, mnt); 9781da177e4SLinus Torvalds seq_puts(m, " 0 0\n"); 9792069f457SEric Paris out: 9801da177e4SLinus Torvalds return err; 9811da177e4SLinus Torvalds } 9821da177e4SLinus Torvalds 983a1a2c409SMiklos Szeredi const struct seq_operations mounts_op = { 9841da177e4SLinus Torvalds .start = m_start, 9851da177e4SLinus Torvalds .next = m_next, 9861da177e4SLinus Torvalds .stop = m_stop, 9871da177e4SLinus Torvalds .show = show_vfsmnt 9881da177e4SLinus Torvalds }; 9891da177e4SLinus Torvalds 9902d4d4864SRam Pai static int show_mountinfo(struct seq_file *m, void *v) 9912d4d4864SRam Pai { 9922d4d4864SRam Pai struct proc_mounts *p = m->private; 9932d4d4864SRam Pai struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 9942d4d4864SRam Pai struct super_block *sb = mnt->mnt_sb; 9952d4d4864SRam Pai struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 9962d4d4864SRam Pai struct path root = p->root; 9972d4d4864SRam Pai int err = 0; 9982d4d4864SRam Pai 9992d4d4864SRam Pai seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id, 10002d4d4864SRam Pai MAJOR(sb->s_dev), MINOR(sb->s_dev)); 10012d4d4864SRam Pai seq_dentry(m, mnt->mnt_root, " \t\n\\"); 10022d4d4864SRam Pai seq_putc(m, ' '); 10032d4d4864SRam Pai seq_path_root(m, &mnt_path, &root, " \t\n\\"); 10042d4d4864SRam Pai if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) { 10052d4d4864SRam Pai /* 10062d4d4864SRam Pai * Mountpoint is outside root, discard that one. Ugly, 10072d4d4864SRam Pai * but less so than trying to do that in iterator in a 10082d4d4864SRam Pai * race-free way (due to renames). 10092d4d4864SRam Pai */ 10102d4d4864SRam Pai return SEQ_SKIP; 10112d4d4864SRam Pai } 10122d4d4864SRam Pai seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw"); 10132d4d4864SRam Pai show_mnt_opts(m, mnt); 10142d4d4864SRam Pai 10152d4d4864SRam Pai /* Tagged fields ("foo:X" or "bar") */ 10162d4d4864SRam Pai if (IS_MNT_SHARED(mnt)) 10172d4d4864SRam Pai seq_printf(m, " shared:%i", mnt->mnt_group_id); 101897e7e0f7SMiklos Szeredi if (IS_MNT_SLAVE(mnt)) { 101997e7e0f7SMiklos Szeredi int master = mnt->mnt_master->mnt_group_id; 102097e7e0f7SMiklos Szeredi int dom = get_dominating_id(mnt, &p->root); 102197e7e0f7SMiklos Szeredi seq_printf(m, " master:%i", master); 102297e7e0f7SMiklos Szeredi if (dom && dom != master) 102397e7e0f7SMiklos Szeredi seq_printf(m, " propagate_from:%i", dom); 102497e7e0f7SMiklos Szeredi } 10252d4d4864SRam Pai if (IS_MNT_UNBINDABLE(mnt)) 10262d4d4864SRam Pai seq_puts(m, " unbindable"); 10272d4d4864SRam Pai 10282d4d4864SRam Pai /* Filesystem specific data */ 10292d4d4864SRam Pai seq_puts(m, " - "); 10302d4d4864SRam Pai show_type(m, sb); 10312d4d4864SRam Pai seq_putc(m, ' '); 10322d4d4864SRam Pai mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none"); 10332d4d4864SRam Pai seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw"); 10342069f457SEric Paris err = show_sb_opts(m, sb); 10352069f457SEric Paris if (err) 10362069f457SEric Paris goto out; 10372d4d4864SRam Pai if (sb->s_op->show_options) 10382d4d4864SRam Pai err = sb->s_op->show_options(m, mnt); 10392d4d4864SRam Pai seq_putc(m, '\n'); 10402069f457SEric Paris out: 10412d4d4864SRam Pai return err; 10422d4d4864SRam Pai } 10432d4d4864SRam Pai 10442d4d4864SRam Pai const struct seq_operations mountinfo_op = { 10452d4d4864SRam Pai .start = m_start, 10462d4d4864SRam Pai .next = m_next, 10472d4d4864SRam Pai .stop = m_stop, 10482d4d4864SRam Pai .show = show_mountinfo, 10492d4d4864SRam Pai }; 10502d4d4864SRam Pai 1051b4629fe2SChuck Lever static int show_vfsstat(struct seq_file *m, void *v) 1052b4629fe2SChuck Lever { 1053b0765fb8SPavel Emelianov struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list); 1054c32c2f63SJan Blunck struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt }; 1055b4629fe2SChuck Lever int err = 0; 1056b4629fe2SChuck Lever 1057b4629fe2SChuck Lever /* device */ 1058b4629fe2SChuck Lever if (mnt->mnt_devname) { 1059b4629fe2SChuck Lever seq_puts(m, "device "); 1060b4629fe2SChuck Lever mangle(m, mnt->mnt_devname); 1061b4629fe2SChuck Lever } else 1062b4629fe2SChuck Lever seq_puts(m, "no device"); 1063b4629fe2SChuck Lever 1064b4629fe2SChuck Lever /* mount point */ 1065b4629fe2SChuck Lever seq_puts(m, " mounted on "); 1066c32c2f63SJan Blunck seq_path(m, &mnt_path, " \t\n\\"); 1067b4629fe2SChuck Lever seq_putc(m, ' '); 1068b4629fe2SChuck Lever 1069b4629fe2SChuck Lever /* file system type */ 1070b4629fe2SChuck Lever seq_puts(m, "with fstype "); 10712d4d4864SRam Pai show_type(m, mnt->mnt_sb); 1072b4629fe2SChuck Lever 1073b4629fe2SChuck Lever /* optional statistics */ 1074b4629fe2SChuck Lever if (mnt->mnt_sb->s_op->show_stats) { 1075b4629fe2SChuck Lever seq_putc(m, ' '); 1076b4629fe2SChuck Lever err = mnt->mnt_sb->s_op->show_stats(m, mnt); 1077b4629fe2SChuck Lever } 1078b4629fe2SChuck Lever 1079b4629fe2SChuck Lever seq_putc(m, '\n'); 1080b4629fe2SChuck Lever return err; 1081b4629fe2SChuck Lever } 1082b4629fe2SChuck Lever 1083a1a2c409SMiklos Szeredi const struct seq_operations mountstats_op = { 1084b4629fe2SChuck Lever .start = m_start, 1085b4629fe2SChuck Lever .next = m_next, 1086b4629fe2SChuck Lever .stop = m_stop, 1087b4629fe2SChuck Lever .show = show_vfsstat, 1088b4629fe2SChuck Lever }; 1089a1a2c409SMiklos Szeredi #endif /* CONFIG_PROC_FS */ 1090b4629fe2SChuck Lever 10911da177e4SLinus Torvalds /** 10921da177e4SLinus Torvalds * may_umount_tree - check if a mount tree is busy 10931da177e4SLinus Torvalds * @mnt: root of mount tree 10941da177e4SLinus Torvalds * 10951da177e4SLinus Torvalds * This is called to check if a tree of mounts has any 10961da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts that are 10971da177e4SLinus Torvalds * busy. 10981da177e4SLinus Torvalds */ 10991da177e4SLinus Torvalds int may_umount_tree(struct vfsmount *mnt) 11001da177e4SLinus Torvalds { 110136341f64SRam Pai int actual_refs = 0; 110236341f64SRam Pai int minimum_refs = 0; 110336341f64SRam Pai struct vfsmount *p; 11041da177e4SLinus Torvalds 1105b3e19d92SNick Piggin /* write lock needed for mnt_get_count */ 1106b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 110736341f64SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 1108b3e19d92SNick Piggin actual_refs += mnt_get_count(p); 11091da177e4SLinus Torvalds minimum_refs += 2; 11101da177e4SLinus Torvalds } 1111b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 11121da177e4SLinus Torvalds 11131da177e4SLinus Torvalds if (actual_refs > minimum_refs) 11141da177e4SLinus Torvalds return 0; 1115e3474a8eSIan Kent 1116e3474a8eSIan Kent return 1; 11171da177e4SLinus Torvalds } 11181da177e4SLinus Torvalds 11191da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount_tree); 11201da177e4SLinus Torvalds 11211da177e4SLinus Torvalds /** 11221da177e4SLinus Torvalds * may_umount - check if a mount point is busy 11231da177e4SLinus Torvalds * @mnt: root of mount 11241da177e4SLinus Torvalds * 11251da177e4SLinus Torvalds * This is called to check if a mount point has any 11261da177e4SLinus Torvalds * open files, pwds, chroots or sub mounts. If the 11271da177e4SLinus Torvalds * mount has sub mounts this will return busy 11281da177e4SLinus Torvalds * regardless of whether the sub mounts are busy. 11291da177e4SLinus Torvalds * 11301da177e4SLinus Torvalds * Doesn't take quota and stuff into account. IOW, in some cases it will 11311da177e4SLinus Torvalds * give false negatives. The main reason why it's here is that we need 11321da177e4SLinus Torvalds * a non-destructive way to look for easily umountable filesystems. 11331da177e4SLinus Torvalds */ 11341da177e4SLinus Torvalds int may_umount(struct vfsmount *mnt) 11351da177e4SLinus Torvalds { 1136e3474a8eSIan Kent int ret = 1; 11378ad08d8aSAl Viro down_read(&namespace_sem); 1138b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 1139a05964f3SRam Pai if (propagate_mount_busy(mnt, 2)) 1140e3474a8eSIan Kent ret = 0; 1141b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 11428ad08d8aSAl Viro up_read(&namespace_sem); 1143a05964f3SRam Pai return ret; 11441da177e4SLinus Torvalds } 11451da177e4SLinus Torvalds 11461da177e4SLinus Torvalds EXPORT_SYMBOL(may_umount); 11471da177e4SLinus Torvalds 1148b90fa9aeSRam Pai void release_mounts(struct list_head *head) 11491da177e4SLinus Torvalds { 115070fbcdf4SRam Pai struct vfsmount *mnt; 115170fbcdf4SRam Pai while (!list_empty(head)) { 1152b5e61818SPavel Emelianov mnt = list_first_entry(head, struct vfsmount, mnt_hash); 115370fbcdf4SRam Pai list_del_init(&mnt->mnt_hash); 115470fbcdf4SRam Pai if (mnt->mnt_parent != mnt) { 115570fbcdf4SRam Pai struct dentry *dentry; 115670fbcdf4SRam Pai struct vfsmount *m; 115799b7db7bSNick Piggin 115899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 115970fbcdf4SRam Pai dentry = mnt->mnt_mountpoint; 116070fbcdf4SRam Pai m = mnt->mnt_parent; 116170fbcdf4SRam Pai mnt->mnt_mountpoint = mnt->mnt_root; 116270fbcdf4SRam Pai mnt->mnt_parent = mnt; 11637c4b93d8SAl Viro m->mnt_ghosts--; 116499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 116570fbcdf4SRam Pai dput(dentry); 116670fbcdf4SRam Pai mntput(m); 11671da177e4SLinus Torvalds } 1168f03c6599SAl Viro mntput(mnt); 116970fbcdf4SRam Pai } 117070fbcdf4SRam Pai } 117170fbcdf4SRam Pai 117299b7db7bSNick Piggin /* 117399b7db7bSNick Piggin * vfsmount lock must be held for write 117499b7db7bSNick Piggin * namespace_sem must be held for write 117599b7db7bSNick Piggin */ 1176a05964f3SRam Pai void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill) 117770fbcdf4SRam Pai { 11787b8a53fdSAl Viro LIST_HEAD(tmp_list); 117970fbcdf4SRam Pai struct vfsmount *p; 118070fbcdf4SRam Pai 11811bfba4e8SAkinobu Mita for (p = mnt; p; p = next_mnt(p, mnt)) 11827b8a53fdSAl Viro list_move(&p->mnt_hash, &tmp_list); 118370fbcdf4SRam Pai 1184a05964f3SRam Pai if (propagate) 11857b8a53fdSAl Viro propagate_umount(&tmp_list); 1186a05964f3SRam Pai 11877b8a53fdSAl Viro list_for_each_entry(p, &tmp_list, mnt_hash) { 118870fbcdf4SRam Pai list_del_init(&p->mnt_expire); 118970fbcdf4SRam Pai list_del_init(&p->mnt_list); 11906b3286edSKirill Korotaev __touch_mnt_namespace(p->mnt_ns); 11916b3286edSKirill Korotaev p->mnt_ns = NULL; 1192f03c6599SAl Viro atomic_dec(&p->mnt_longterm); 119370fbcdf4SRam Pai list_del_init(&p->mnt_child); 11947c4b93d8SAl Viro if (p->mnt_parent != p) { 11957c4b93d8SAl Viro p->mnt_parent->mnt_ghosts++; 11965f57cbccSNick Piggin dentry_reset_mounted(p->mnt_parent, p->mnt_mountpoint); 11977c4b93d8SAl Viro } 1198a05964f3SRam Pai change_mnt_propagation(p, MS_PRIVATE); 11991da177e4SLinus Torvalds } 12007b8a53fdSAl Viro list_splice(&tmp_list, kill); 12011da177e4SLinus Torvalds } 12021da177e4SLinus Torvalds 1203c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts); 1204c35038beSAl Viro 12051da177e4SLinus Torvalds static int do_umount(struct vfsmount *mnt, int flags) 12061da177e4SLinus Torvalds { 12071da177e4SLinus Torvalds struct super_block *sb = mnt->mnt_sb; 12081da177e4SLinus Torvalds int retval; 120970fbcdf4SRam Pai LIST_HEAD(umount_list); 12101da177e4SLinus Torvalds 12111da177e4SLinus Torvalds retval = security_sb_umount(mnt, flags); 12121da177e4SLinus Torvalds if (retval) 12131da177e4SLinus Torvalds return retval; 12141da177e4SLinus Torvalds 12151da177e4SLinus Torvalds /* 12161da177e4SLinus Torvalds * Allow userspace to request a mountpoint be expired rather than 12171da177e4SLinus Torvalds * unmounting unconditionally. Unmount only happens if: 12181da177e4SLinus Torvalds * (1) the mark is already set (the mark is cleared by mntput()) 12191da177e4SLinus Torvalds * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount] 12201da177e4SLinus Torvalds */ 12211da177e4SLinus Torvalds if (flags & MNT_EXPIRE) { 12226ac08c39SJan Blunck if (mnt == current->fs->root.mnt || 12231da177e4SLinus Torvalds flags & (MNT_FORCE | MNT_DETACH)) 12241da177e4SLinus Torvalds return -EINVAL; 12251da177e4SLinus Torvalds 1226b3e19d92SNick Piggin /* 1227b3e19d92SNick Piggin * probably don't strictly need the lock here if we examined 1228b3e19d92SNick Piggin * all race cases, but it's a slowpath. 1229b3e19d92SNick Piggin */ 1230b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 1231b3e19d92SNick Piggin if (mnt_get_count(mnt) != 2) { 1232b3e19d92SNick Piggin br_write_lock(vfsmount_lock); 12331da177e4SLinus Torvalds return -EBUSY; 1234b3e19d92SNick Piggin } 1235b3e19d92SNick Piggin br_write_unlock(vfsmount_lock); 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1)) 12381da177e4SLinus Torvalds return -EAGAIN; 12391da177e4SLinus Torvalds } 12401da177e4SLinus Torvalds 12411da177e4SLinus Torvalds /* 12421da177e4SLinus Torvalds * If we may have to abort operations to get out of this 12431da177e4SLinus Torvalds * mount, and they will themselves hold resources we must 12441da177e4SLinus Torvalds * allow the fs to do things. In the Unix tradition of 12451da177e4SLinus Torvalds * 'Gee thats tricky lets do it in userspace' the umount_begin 12461da177e4SLinus Torvalds * might fail to complete on the first run through as other tasks 12471da177e4SLinus Torvalds * must return, and the like. Thats for the mount program to worry 12481da177e4SLinus Torvalds * about for the moment. 12491da177e4SLinus Torvalds */ 12501da177e4SLinus Torvalds 125142faad99SAl Viro if (flags & MNT_FORCE && sb->s_op->umount_begin) { 125242faad99SAl Viro sb->s_op->umount_begin(sb); 125342faad99SAl Viro } 12541da177e4SLinus Torvalds 12551da177e4SLinus Torvalds /* 12561da177e4SLinus Torvalds * No sense to grab the lock for this test, but test itself looks 12571da177e4SLinus Torvalds * somewhat bogus. Suggestions for better replacement? 12581da177e4SLinus Torvalds * Ho-hum... In principle, we might treat that as umount + switch 12591da177e4SLinus Torvalds * to rootfs. GC would eventually take care of the old vfsmount. 12601da177e4SLinus Torvalds * Actually it makes sense, especially if rootfs would contain a 12611da177e4SLinus Torvalds * /reboot - static binary that would close all descriptors and 12621da177e4SLinus Torvalds * call reboot(9). Then init(8) could umount root and exec /reboot. 12631da177e4SLinus Torvalds */ 12646ac08c39SJan Blunck if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) { 12651da177e4SLinus Torvalds /* 12661da177e4SLinus Torvalds * Special case for "unmounting" root ... 12671da177e4SLinus Torvalds * we just try to remount it readonly. 12681da177e4SLinus Torvalds */ 12691da177e4SLinus Torvalds down_write(&sb->s_umount); 12704aa98cf7SAl Viro if (!(sb->s_flags & MS_RDONLY)) 12711da177e4SLinus Torvalds retval = do_remount_sb(sb, MS_RDONLY, NULL, 0); 12721da177e4SLinus Torvalds up_write(&sb->s_umount); 12731da177e4SLinus Torvalds return retval; 12741da177e4SLinus Torvalds } 12751da177e4SLinus Torvalds 1276390c6843SRam Pai down_write(&namespace_sem); 127799b7db7bSNick Piggin br_write_lock(vfsmount_lock); 12785addc5ddSAl Viro event++; 12791da177e4SLinus Torvalds 1280c35038beSAl Viro if (!(flags & MNT_DETACH)) 1281c35038beSAl Viro shrink_submounts(mnt, &umount_list); 1282c35038beSAl Viro 12831da177e4SLinus Torvalds retval = -EBUSY; 1284a05964f3SRam Pai if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) { 12851da177e4SLinus Torvalds if (!list_empty(&mnt->mnt_list)) 1286a05964f3SRam Pai umount_tree(mnt, 1, &umount_list); 12871da177e4SLinus Torvalds retval = 0; 12881da177e4SLinus Torvalds } 128999b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1290390c6843SRam Pai up_write(&namespace_sem); 129170fbcdf4SRam Pai release_mounts(&umount_list); 12921da177e4SLinus Torvalds return retval; 12931da177e4SLinus Torvalds } 12941da177e4SLinus Torvalds 12951da177e4SLinus Torvalds /* 12961da177e4SLinus Torvalds * Now umount can handle mount points as well as block devices. 12971da177e4SLinus Torvalds * This is important for filesystems which use unnamed block devices. 12981da177e4SLinus Torvalds * 12991da177e4SLinus Torvalds * We now support a flag for forced unmount like the other 'big iron' 13001da177e4SLinus Torvalds * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD 13011da177e4SLinus Torvalds */ 13021da177e4SLinus Torvalds 1303bdc480e3SHeiko Carstens SYSCALL_DEFINE2(umount, char __user *, name, int, flags) 13041da177e4SLinus Torvalds { 13052d8f3038SAl Viro struct path path; 13061da177e4SLinus Torvalds int retval; 1307db1f05bbSMiklos Szeredi int lookup_flags = 0; 13081da177e4SLinus Torvalds 1309db1f05bbSMiklos Szeredi if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW)) 1310db1f05bbSMiklos Szeredi return -EINVAL; 1311db1f05bbSMiklos Szeredi 1312db1f05bbSMiklos Szeredi if (!(flags & UMOUNT_NOFOLLOW)) 1313db1f05bbSMiklos Szeredi lookup_flags |= LOOKUP_FOLLOW; 1314db1f05bbSMiklos Szeredi 1315db1f05bbSMiklos Szeredi retval = user_path_at(AT_FDCWD, name, lookup_flags, &path); 13161da177e4SLinus Torvalds if (retval) 13171da177e4SLinus Torvalds goto out; 13181da177e4SLinus Torvalds retval = -EINVAL; 13192d8f3038SAl Viro if (path.dentry != path.mnt->mnt_root) 13201da177e4SLinus Torvalds goto dput_and_out; 13212d8f3038SAl Viro if (!check_mnt(path.mnt)) 13221da177e4SLinus Torvalds goto dput_and_out; 13231da177e4SLinus Torvalds 13241da177e4SLinus Torvalds retval = -EPERM; 13251da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 13261da177e4SLinus Torvalds goto dput_and_out; 13271da177e4SLinus Torvalds 13282d8f3038SAl Viro retval = do_umount(path.mnt, flags); 13291da177e4SLinus Torvalds dput_and_out: 1330429731b1SJan Blunck /* we mustn't call path_put() as that would clear mnt_expiry_mark */ 13312d8f3038SAl Viro dput(path.dentry); 13322d8f3038SAl Viro mntput_no_expire(path.mnt); 13331da177e4SLinus Torvalds out: 13341da177e4SLinus Torvalds return retval; 13351da177e4SLinus Torvalds } 13361da177e4SLinus Torvalds 13371da177e4SLinus Torvalds #ifdef __ARCH_WANT_SYS_OLDUMOUNT 13381da177e4SLinus Torvalds 13391da177e4SLinus Torvalds /* 13401da177e4SLinus Torvalds * The 2.0 compatible umount. No flags. 13411da177e4SLinus Torvalds */ 1342bdc480e3SHeiko Carstens SYSCALL_DEFINE1(oldumount, char __user *, name) 13431da177e4SLinus Torvalds { 13441da177e4SLinus Torvalds return sys_umount(name, 0); 13451da177e4SLinus Torvalds } 13461da177e4SLinus Torvalds 13471da177e4SLinus Torvalds #endif 13481da177e4SLinus Torvalds 13492d92ab3cSAl Viro static int mount_is_safe(struct path *path) 13501da177e4SLinus Torvalds { 13511da177e4SLinus Torvalds if (capable(CAP_SYS_ADMIN)) 13521da177e4SLinus Torvalds return 0; 13531da177e4SLinus Torvalds return -EPERM; 13541da177e4SLinus Torvalds #ifdef notyet 13552d92ab3cSAl Viro if (S_ISLNK(path->dentry->d_inode->i_mode)) 13561da177e4SLinus Torvalds return -EPERM; 13572d92ab3cSAl Viro if (path->dentry->d_inode->i_mode & S_ISVTX) { 1358da9592edSDavid Howells if (current_uid() != path->dentry->d_inode->i_uid) 13591da177e4SLinus Torvalds return -EPERM; 13601da177e4SLinus Torvalds } 13612d92ab3cSAl Viro if (inode_permission(path->dentry->d_inode, MAY_WRITE)) 13621da177e4SLinus Torvalds return -EPERM; 13631da177e4SLinus Torvalds return 0; 13641da177e4SLinus Torvalds #endif 13651da177e4SLinus Torvalds } 13661da177e4SLinus Torvalds 1367b90fa9aeSRam Pai struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry, 136836341f64SRam Pai int flag) 13691da177e4SLinus Torvalds { 13701da177e4SLinus Torvalds struct vfsmount *res, *p, *q, *r, *s; 13711a390689SAl Viro struct path path; 13721da177e4SLinus Torvalds 13739676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt)) 13749676f0c6SRam Pai return NULL; 13759676f0c6SRam Pai 137636341f64SRam Pai res = q = clone_mnt(mnt, dentry, flag); 13771da177e4SLinus Torvalds if (!q) 13781da177e4SLinus Torvalds goto Enomem; 13791da177e4SLinus Torvalds q->mnt_mountpoint = mnt->mnt_mountpoint; 13801da177e4SLinus Torvalds 13811da177e4SLinus Torvalds p = mnt; 1382fdadd65fSDomen Puncer list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) { 13837ec02ef1SJan Blunck if (!is_subdir(r->mnt_mountpoint, dentry)) 13841da177e4SLinus Torvalds continue; 13851da177e4SLinus Torvalds 13861da177e4SLinus Torvalds for (s = r; s; s = next_mnt(s, r)) { 13879676f0c6SRam Pai if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) { 13889676f0c6SRam Pai s = skip_mnt_tree(s); 13899676f0c6SRam Pai continue; 13909676f0c6SRam Pai } 13911da177e4SLinus Torvalds while (p != s->mnt_parent) { 13921da177e4SLinus Torvalds p = p->mnt_parent; 13931da177e4SLinus Torvalds q = q->mnt_parent; 13941da177e4SLinus Torvalds } 13951da177e4SLinus Torvalds p = s; 13961a390689SAl Viro path.mnt = q; 13971a390689SAl Viro path.dentry = p->mnt_mountpoint; 139836341f64SRam Pai q = clone_mnt(p, p->mnt_root, flag); 13991da177e4SLinus Torvalds if (!q) 14001da177e4SLinus Torvalds goto Enomem; 140199b7db7bSNick Piggin br_write_lock(vfsmount_lock); 14021da177e4SLinus Torvalds list_add_tail(&q->mnt_list, &res->mnt_list); 14031a390689SAl Viro attach_mnt(q, &path); 140499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 14051da177e4SLinus Torvalds } 14061da177e4SLinus Torvalds } 14071da177e4SLinus Torvalds return res; 14081da177e4SLinus Torvalds Enomem: 14091da177e4SLinus Torvalds if (res) { 141070fbcdf4SRam Pai LIST_HEAD(umount_list); 141199b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1412a05964f3SRam Pai umount_tree(res, 0, &umount_list); 141399b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 141470fbcdf4SRam Pai release_mounts(&umount_list); 14151da177e4SLinus Torvalds } 14161da177e4SLinus Torvalds return NULL; 14171da177e4SLinus Torvalds } 14181da177e4SLinus Torvalds 1419589ff870SAl Viro struct vfsmount *collect_mounts(struct path *path) 14208aec0809SAl Viro { 14218aec0809SAl Viro struct vfsmount *tree; 14221a60a280SAl Viro down_write(&namespace_sem); 1423589ff870SAl Viro tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE); 14241a60a280SAl Viro up_write(&namespace_sem); 14258aec0809SAl Viro return tree; 14268aec0809SAl Viro } 14278aec0809SAl Viro 14288aec0809SAl Viro void drop_collected_mounts(struct vfsmount *mnt) 14298aec0809SAl Viro { 14308aec0809SAl Viro LIST_HEAD(umount_list); 14311a60a280SAl Viro down_write(&namespace_sem); 143299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 14338aec0809SAl Viro umount_tree(mnt, 0, &umount_list); 143499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 14351a60a280SAl Viro up_write(&namespace_sem); 14368aec0809SAl Viro release_mounts(&umount_list); 14378aec0809SAl Viro } 14388aec0809SAl Viro 14391f707137SAl Viro int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg, 14401f707137SAl Viro struct vfsmount *root) 14411f707137SAl Viro { 14421f707137SAl Viro struct vfsmount *mnt; 14431f707137SAl Viro int res = f(root, arg); 14441f707137SAl Viro if (res) 14451f707137SAl Viro return res; 14461f707137SAl Viro list_for_each_entry(mnt, &root->mnt_list, mnt_list) { 14471f707137SAl Viro res = f(mnt, arg); 14481f707137SAl Viro if (res) 14491f707137SAl Viro return res; 14501f707137SAl Viro } 14511f707137SAl Viro return 0; 14521f707137SAl Viro } 14531f707137SAl Viro 1454719f5d7fSMiklos Szeredi static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end) 1455719f5d7fSMiklos Szeredi { 1456719f5d7fSMiklos Szeredi struct vfsmount *p; 1457719f5d7fSMiklos Szeredi 1458719f5d7fSMiklos Szeredi for (p = mnt; p != end; p = next_mnt(p, mnt)) { 1459719f5d7fSMiklos Szeredi if (p->mnt_group_id && !IS_MNT_SHARED(p)) 1460719f5d7fSMiklos Szeredi mnt_release_group_id(p); 1461719f5d7fSMiklos Szeredi } 1462719f5d7fSMiklos Szeredi } 1463719f5d7fSMiklos Szeredi 1464719f5d7fSMiklos Szeredi static int invent_group_ids(struct vfsmount *mnt, bool recurse) 1465719f5d7fSMiklos Szeredi { 1466719f5d7fSMiklos Szeredi struct vfsmount *p; 1467719f5d7fSMiklos Szeredi 1468719f5d7fSMiklos Szeredi for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) { 1469719f5d7fSMiklos Szeredi if (!p->mnt_group_id && !IS_MNT_SHARED(p)) { 1470719f5d7fSMiklos Szeredi int err = mnt_alloc_group_id(p); 1471719f5d7fSMiklos Szeredi if (err) { 1472719f5d7fSMiklos Szeredi cleanup_group_ids(mnt, p); 1473719f5d7fSMiklos Szeredi return err; 1474719f5d7fSMiklos Szeredi } 1475719f5d7fSMiklos Szeredi } 1476719f5d7fSMiklos Szeredi } 1477719f5d7fSMiklos Szeredi 1478719f5d7fSMiklos Szeredi return 0; 1479719f5d7fSMiklos Szeredi } 1480719f5d7fSMiklos Szeredi 1481b90fa9aeSRam Pai /* 1482b90fa9aeSRam Pai * @source_mnt : mount tree to be attached 1483b90fa9aeSRam Pai * @nd : place the mount tree @source_mnt is attached 148421444403SRam Pai * @parent_nd : if non-null, detach the source_mnt from its parent and 148521444403SRam Pai * store the parent mount and mountpoint dentry. 148621444403SRam Pai * (done when source_mnt is moved) 1487b90fa9aeSRam Pai * 1488b90fa9aeSRam Pai * NOTE: in the table below explains the semantics when a source mount 1489b90fa9aeSRam Pai * of a given type is attached to a destination mount of a given type. 14909676f0c6SRam Pai * --------------------------------------------------------------------------- 1491b90fa9aeSRam Pai * | BIND MOUNT OPERATION | 14929676f0c6SRam Pai * |************************************************************************** 14939676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 14949676f0c6SRam Pai * | dest | | | | | 14959676f0c6SRam Pai * | | | | | | | 14969676f0c6SRam Pai * | v | | | | | 14979676f0c6SRam Pai * |************************************************************************** 14989676f0c6SRam Pai * | shared | shared (++) | shared (+) | shared(+++)| invalid | 14995afe0022SRam Pai * | | | | | | 15009676f0c6SRam Pai * |non-shared| shared (+) | private | slave (*) | invalid | 15019676f0c6SRam Pai * *************************************************************************** 1502b90fa9aeSRam Pai * A bind operation clones the source mount and mounts the clone on the 1503b90fa9aeSRam Pai * destination mount. 1504b90fa9aeSRam Pai * 1505b90fa9aeSRam Pai * (++) the cloned mount is propagated to all the mounts in the propagation 1506b90fa9aeSRam Pai * tree of the destination mount and the cloned mount is added to 1507b90fa9aeSRam Pai * the peer group of the source mount. 1508b90fa9aeSRam Pai * (+) the cloned mount is created under the destination mount and is marked 1509b90fa9aeSRam Pai * as shared. The cloned mount is added to the peer group of the source 1510b90fa9aeSRam Pai * mount. 15115afe0022SRam Pai * (+++) the mount is propagated to all the mounts in the propagation tree 15125afe0022SRam Pai * of the destination mount and the cloned mount is made slave 15135afe0022SRam Pai * of the same master as that of the source mount. The cloned mount 15145afe0022SRam Pai * is marked as 'shared and slave'. 15155afe0022SRam Pai * (*) the cloned mount is made a slave of the same master as that of the 15165afe0022SRam Pai * source mount. 15175afe0022SRam Pai * 15189676f0c6SRam Pai * --------------------------------------------------------------------------- 151921444403SRam Pai * | MOVE MOUNT OPERATION | 15209676f0c6SRam Pai * |************************************************************************** 15219676f0c6SRam Pai * | source-->| shared | private | slave | unbindable | 15229676f0c6SRam Pai * | dest | | | | | 15239676f0c6SRam Pai * | | | | | | | 15249676f0c6SRam Pai * | v | | | | | 15259676f0c6SRam Pai * |************************************************************************** 15269676f0c6SRam Pai * | shared | shared (+) | shared (+) | shared(+++) | invalid | 15275afe0022SRam Pai * | | | | | | 15289676f0c6SRam Pai * |non-shared| shared (+*) | private | slave (*) | unbindable | 15299676f0c6SRam Pai * *************************************************************************** 15305afe0022SRam Pai * 15315afe0022SRam Pai * (+) the mount is moved to the destination. And is then propagated to 15325afe0022SRam Pai * all the mounts in the propagation tree of the destination mount. 153321444403SRam Pai * (+*) the mount is moved to the destination. 15345afe0022SRam Pai * (+++) the mount is moved to the destination and is then propagated to 15355afe0022SRam Pai * all the mounts belonging to the destination mount's propagation tree. 15365afe0022SRam Pai * the mount is marked as 'shared and slave'. 15375afe0022SRam Pai * (*) the mount continues to be a slave at the new location. 1538b90fa9aeSRam Pai * 1539b90fa9aeSRam Pai * if the source mount is a tree, the operations explained above is 1540b90fa9aeSRam Pai * applied to each mount in the tree. 1541b90fa9aeSRam Pai * Must be called without spinlocks held, since this function can sleep 1542b90fa9aeSRam Pai * in allocations. 1543b90fa9aeSRam Pai */ 1544b90fa9aeSRam Pai static int attach_recursive_mnt(struct vfsmount *source_mnt, 15451a390689SAl Viro struct path *path, struct path *parent_path) 1546b90fa9aeSRam Pai { 1547b90fa9aeSRam Pai LIST_HEAD(tree_list); 15481a390689SAl Viro struct vfsmount *dest_mnt = path->mnt; 15491a390689SAl Viro struct dentry *dest_dentry = path->dentry; 1550b90fa9aeSRam Pai struct vfsmount *child, *p; 1551719f5d7fSMiklos Szeredi int err; 1552b90fa9aeSRam Pai 1553719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) { 1554719f5d7fSMiklos Szeredi err = invent_group_ids(source_mnt, true); 1555719f5d7fSMiklos Szeredi if (err) 1556719f5d7fSMiklos Szeredi goto out; 1557719f5d7fSMiklos Szeredi } 1558719f5d7fSMiklos Szeredi err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list); 1559719f5d7fSMiklos Szeredi if (err) 1560719f5d7fSMiklos Szeredi goto out_cleanup_ids; 1561b90fa9aeSRam Pai 156299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1563df1a1ad2SAl Viro 1564b90fa9aeSRam Pai if (IS_MNT_SHARED(dest_mnt)) { 1565b90fa9aeSRam Pai for (p = source_mnt; p; p = next_mnt(p, source_mnt)) 1566b90fa9aeSRam Pai set_mnt_shared(p); 1567b90fa9aeSRam Pai } 15681a390689SAl Viro if (parent_path) { 15691a390689SAl Viro detach_mnt(source_mnt, parent_path); 15701a390689SAl Viro attach_mnt(source_mnt, path); 1571e5d67f07SAl Viro touch_mnt_namespace(parent_path->mnt->mnt_ns); 157221444403SRam Pai } else { 1573b90fa9aeSRam Pai mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt); 1574b90fa9aeSRam Pai commit_tree(source_mnt); 157521444403SRam Pai } 1576b90fa9aeSRam Pai 1577b90fa9aeSRam Pai list_for_each_entry_safe(child, p, &tree_list, mnt_hash) { 1578b90fa9aeSRam Pai list_del_init(&child->mnt_hash); 1579b90fa9aeSRam Pai commit_tree(child); 1580b90fa9aeSRam Pai } 158199b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 158299b7db7bSNick Piggin 1583b90fa9aeSRam Pai return 0; 1584719f5d7fSMiklos Szeredi 1585719f5d7fSMiklos Szeredi out_cleanup_ids: 1586719f5d7fSMiklos Szeredi if (IS_MNT_SHARED(dest_mnt)) 1587719f5d7fSMiklos Szeredi cleanup_group_ids(source_mnt, NULL); 1588719f5d7fSMiklos Szeredi out: 1589719f5d7fSMiklos Szeredi return err; 1590b90fa9aeSRam Pai } 1591b90fa9aeSRam Pai 15928c3ee42eSAl Viro static int graft_tree(struct vfsmount *mnt, struct path *path) 15931da177e4SLinus Torvalds { 15941da177e4SLinus Torvalds int err; 15951da177e4SLinus Torvalds if (mnt->mnt_sb->s_flags & MS_NOUSER) 15961da177e4SLinus Torvalds return -EINVAL; 15971da177e4SLinus Torvalds 15988c3ee42eSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 15991da177e4SLinus Torvalds S_ISDIR(mnt->mnt_root->d_inode->i_mode)) 16001da177e4SLinus Torvalds return -ENOTDIR; 16011da177e4SLinus Torvalds 16021da177e4SLinus Torvalds err = -ENOENT; 16038c3ee42eSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 1604d83c49f3SAl Viro if (cant_mount(path->dentry)) 16051da177e4SLinus Torvalds goto out_unlock; 16061da177e4SLinus Torvalds 1607f3da392eSAlexey Dobriyan if (!d_unlinked(path->dentry)) 16088c3ee42eSAl Viro err = attach_recursive_mnt(mnt, path, NULL); 16091da177e4SLinus Torvalds out_unlock: 16108c3ee42eSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 16111da177e4SLinus Torvalds return err; 16121da177e4SLinus Torvalds } 16131da177e4SLinus Torvalds 16141da177e4SLinus Torvalds /* 16157a2e8a8fSValerie Aurora * Sanity check the flags to change_mnt_propagation. 16167a2e8a8fSValerie Aurora */ 16177a2e8a8fSValerie Aurora 16187a2e8a8fSValerie Aurora static int flags_to_propagation_type(int flags) 16197a2e8a8fSValerie Aurora { 16207a2e8a8fSValerie Aurora int type = flags & ~MS_REC; 16217a2e8a8fSValerie Aurora 16227a2e8a8fSValerie Aurora /* Fail if any non-propagation flags are set */ 16237a2e8a8fSValerie Aurora if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 16247a2e8a8fSValerie Aurora return 0; 16257a2e8a8fSValerie Aurora /* Only one propagation flag should be set */ 16267a2e8a8fSValerie Aurora if (!is_power_of_2(type)) 16277a2e8a8fSValerie Aurora return 0; 16287a2e8a8fSValerie Aurora return type; 16297a2e8a8fSValerie Aurora } 16307a2e8a8fSValerie Aurora 16317a2e8a8fSValerie Aurora /* 163207b20889SRam Pai * recursively change the type of the mountpoint. 163307b20889SRam Pai */ 16340a0d8a46SAl Viro static int do_change_type(struct path *path, int flag) 163507b20889SRam Pai { 16362d92ab3cSAl Viro struct vfsmount *m, *mnt = path->mnt; 163707b20889SRam Pai int recurse = flag & MS_REC; 16387a2e8a8fSValerie Aurora int type; 1639719f5d7fSMiklos Szeredi int err = 0; 164007b20889SRam Pai 1641ee6f9582SMiklos Szeredi if (!capable(CAP_SYS_ADMIN)) 1642ee6f9582SMiklos Szeredi return -EPERM; 1643ee6f9582SMiklos Szeredi 16442d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 164507b20889SRam Pai return -EINVAL; 164607b20889SRam Pai 16477a2e8a8fSValerie Aurora type = flags_to_propagation_type(flag); 16487a2e8a8fSValerie Aurora if (!type) 16497a2e8a8fSValerie Aurora return -EINVAL; 16507a2e8a8fSValerie Aurora 165107b20889SRam Pai down_write(&namespace_sem); 1652719f5d7fSMiklos Szeredi if (type == MS_SHARED) { 1653719f5d7fSMiklos Szeredi err = invent_group_ids(mnt, recurse); 1654719f5d7fSMiklos Szeredi if (err) 1655719f5d7fSMiklos Szeredi goto out_unlock; 1656719f5d7fSMiklos Szeredi } 1657719f5d7fSMiklos Szeredi 165899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 165907b20889SRam Pai for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL)) 166007b20889SRam Pai change_mnt_propagation(m, type); 166199b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1662719f5d7fSMiklos Szeredi 1663719f5d7fSMiklos Szeredi out_unlock: 166407b20889SRam Pai up_write(&namespace_sem); 1665719f5d7fSMiklos Szeredi return err; 166607b20889SRam Pai } 166707b20889SRam Pai 166807b20889SRam Pai /* 16691da177e4SLinus Torvalds * do loopback mount. 16701da177e4SLinus Torvalds */ 16710a0d8a46SAl Viro static int do_loopback(struct path *path, char *old_name, 16722dafe1c4SEric Sandeen int recurse) 16731da177e4SLinus Torvalds { 16742d92ab3cSAl Viro struct path old_path; 16751da177e4SLinus Torvalds struct vfsmount *mnt = NULL; 16762d92ab3cSAl Viro int err = mount_is_safe(path); 16771da177e4SLinus Torvalds if (err) 16781da177e4SLinus Torvalds return err; 16791da177e4SLinus Torvalds if (!old_name || !*old_name) 16801da177e4SLinus Torvalds return -EINVAL; 16812d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 16821da177e4SLinus Torvalds if (err) 16831da177e4SLinus Torvalds return err; 16841da177e4SLinus Torvalds 1685390c6843SRam Pai down_write(&namespace_sem); 16861da177e4SLinus Torvalds err = -EINVAL; 16872d92ab3cSAl Viro if (IS_MNT_UNBINDABLE(old_path.mnt)) 16889676f0c6SRam Pai goto out; 16899676f0c6SRam Pai 16902d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 1691ccd48bc7SAl Viro goto out; 1692ccd48bc7SAl Viro 16931da177e4SLinus Torvalds err = -ENOMEM; 16941da177e4SLinus Torvalds if (recurse) 16952d92ab3cSAl Viro mnt = copy_tree(old_path.mnt, old_path.dentry, 0); 16961da177e4SLinus Torvalds else 16972d92ab3cSAl Viro mnt = clone_mnt(old_path.mnt, old_path.dentry, 0); 16981da177e4SLinus Torvalds 1699ccd48bc7SAl Viro if (!mnt) 1700ccd48bc7SAl Viro goto out; 1701ccd48bc7SAl Viro 17022d92ab3cSAl Viro err = graft_tree(mnt, path); 17031da177e4SLinus Torvalds if (err) { 170470fbcdf4SRam Pai LIST_HEAD(umount_list); 170599b7db7bSNick Piggin 170699b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1707a05964f3SRam Pai umount_tree(mnt, 0, &umount_list); 170899b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 170970fbcdf4SRam Pai release_mounts(&umount_list); 17105b83d2c5SRam Pai } 17111da177e4SLinus Torvalds 1712ccd48bc7SAl Viro out: 1713390c6843SRam Pai up_write(&namespace_sem); 17142d92ab3cSAl Viro path_put(&old_path); 17151da177e4SLinus Torvalds return err; 17161da177e4SLinus Torvalds } 17171da177e4SLinus Torvalds 17182e4b7fcdSDave Hansen static int change_mount_flags(struct vfsmount *mnt, int ms_flags) 17192e4b7fcdSDave Hansen { 17202e4b7fcdSDave Hansen int error = 0; 17212e4b7fcdSDave Hansen int readonly_request = 0; 17222e4b7fcdSDave Hansen 17232e4b7fcdSDave Hansen if (ms_flags & MS_RDONLY) 17242e4b7fcdSDave Hansen readonly_request = 1; 17252e4b7fcdSDave Hansen if (readonly_request == __mnt_is_readonly(mnt)) 17262e4b7fcdSDave Hansen return 0; 17272e4b7fcdSDave Hansen 17282e4b7fcdSDave Hansen if (readonly_request) 17292e4b7fcdSDave Hansen error = mnt_make_readonly(mnt); 17302e4b7fcdSDave Hansen else 17312e4b7fcdSDave Hansen __mnt_unmake_readonly(mnt); 17322e4b7fcdSDave Hansen return error; 17332e4b7fcdSDave Hansen } 17342e4b7fcdSDave Hansen 17351da177e4SLinus Torvalds /* 17361da177e4SLinus Torvalds * change filesystem flags. dir should be a physical root of filesystem. 17371da177e4SLinus Torvalds * If you've mounted a non-root directory somewhere and want to do remount 17381da177e4SLinus Torvalds * on it - tough luck. 17391da177e4SLinus Torvalds */ 17400a0d8a46SAl Viro static int do_remount(struct path *path, int flags, int mnt_flags, 17411da177e4SLinus Torvalds void *data) 17421da177e4SLinus Torvalds { 17431da177e4SLinus Torvalds int err; 17442d92ab3cSAl Viro struct super_block *sb = path->mnt->mnt_sb; 17451da177e4SLinus Torvalds 17461da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 17471da177e4SLinus Torvalds return -EPERM; 17481da177e4SLinus Torvalds 17492d92ab3cSAl Viro if (!check_mnt(path->mnt)) 17501da177e4SLinus Torvalds return -EINVAL; 17511da177e4SLinus Torvalds 17522d92ab3cSAl Viro if (path->dentry != path->mnt->mnt_root) 17531da177e4SLinus Torvalds return -EINVAL; 17541da177e4SLinus Torvalds 17551da177e4SLinus Torvalds down_write(&sb->s_umount); 17562e4b7fcdSDave Hansen if (flags & MS_BIND) 17572d92ab3cSAl Viro err = change_mount_flags(path->mnt, flags); 17584aa98cf7SAl Viro else 17591da177e4SLinus Torvalds err = do_remount_sb(sb, flags, data, 0); 17607b43a79fSAl Viro if (!err) { 176199b7db7bSNick Piggin br_write_lock(vfsmount_lock); 1762495d6c9cSValerie Aurora mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK; 17632d92ab3cSAl Viro path->mnt->mnt_flags = mnt_flags; 176499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 17657b43a79fSAl Viro } 17661da177e4SLinus Torvalds up_write(&sb->s_umount); 17670e55a7ccSDan Williams if (!err) { 176899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 17690e55a7ccSDan Williams touch_mnt_namespace(path->mnt->mnt_ns); 177099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 17710e55a7ccSDan Williams } 17721da177e4SLinus Torvalds return err; 17731da177e4SLinus Torvalds } 17741da177e4SLinus Torvalds 17759676f0c6SRam Pai static inline int tree_contains_unbindable(struct vfsmount *mnt) 17769676f0c6SRam Pai { 17779676f0c6SRam Pai struct vfsmount *p; 17789676f0c6SRam Pai for (p = mnt; p; p = next_mnt(p, mnt)) { 17799676f0c6SRam Pai if (IS_MNT_UNBINDABLE(p)) 17809676f0c6SRam Pai return 1; 17819676f0c6SRam Pai } 17829676f0c6SRam Pai return 0; 17839676f0c6SRam Pai } 17849676f0c6SRam Pai 17850a0d8a46SAl Viro static int do_move_mount(struct path *path, char *old_name) 17861da177e4SLinus Torvalds { 17872d92ab3cSAl Viro struct path old_path, parent_path; 17881da177e4SLinus Torvalds struct vfsmount *p; 17891da177e4SLinus Torvalds int err = 0; 17901da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 17911da177e4SLinus Torvalds return -EPERM; 17921da177e4SLinus Torvalds if (!old_name || !*old_name) 17931da177e4SLinus Torvalds return -EINVAL; 17942d92ab3cSAl Viro err = kern_path(old_name, LOOKUP_FOLLOW, &old_path); 17951da177e4SLinus Torvalds if (err) 17961da177e4SLinus Torvalds return err; 17971da177e4SLinus Torvalds 1798390c6843SRam Pai down_write(&namespace_sem); 1799cc53ce53SDavid Howells err = follow_down(path, true); 1800cc53ce53SDavid Howells if (err < 0) 1801cc53ce53SDavid Howells goto out; 1802cc53ce53SDavid Howells 18031da177e4SLinus Torvalds err = -EINVAL; 18042d92ab3cSAl Viro if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt)) 18051da177e4SLinus Torvalds goto out; 18061da177e4SLinus Torvalds 18071da177e4SLinus Torvalds err = -ENOENT; 18082d92ab3cSAl Viro mutex_lock(&path->dentry->d_inode->i_mutex); 1809d83c49f3SAl Viro if (cant_mount(path->dentry)) 18101da177e4SLinus Torvalds goto out1; 18111da177e4SLinus Torvalds 1812f3da392eSAlexey Dobriyan if (d_unlinked(path->dentry)) 181321444403SRam Pai goto out1; 18141da177e4SLinus Torvalds 18151da177e4SLinus Torvalds err = -EINVAL; 18162d92ab3cSAl Viro if (old_path.dentry != old_path.mnt->mnt_root) 181721444403SRam Pai goto out1; 18181da177e4SLinus Torvalds 18192d92ab3cSAl Viro if (old_path.mnt == old_path.mnt->mnt_parent) 182021444403SRam Pai goto out1; 18211da177e4SLinus Torvalds 18222d92ab3cSAl Viro if (S_ISDIR(path->dentry->d_inode->i_mode) != 18232d92ab3cSAl Viro S_ISDIR(old_path.dentry->d_inode->i_mode)) 182421444403SRam Pai goto out1; 182521444403SRam Pai /* 182621444403SRam Pai * Don't move a mount residing in a shared parent. 182721444403SRam Pai */ 18282d92ab3cSAl Viro if (old_path.mnt->mnt_parent && 18292d92ab3cSAl Viro IS_MNT_SHARED(old_path.mnt->mnt_parent)) 183021444403SRam Pai goto out1; 18319676f0c6SRam Pai /* 18329676f0c6SRam Pai * Don't move a mount tree containing unbindable mounts to a destination 18339676f0c6SRam Pai * mount which is shared. 18349676f0c6SRam Pai */ 18352d92ab3cSAl Viro if (IS_MNT_SHARED(path->mnt) && 18362d92ab3cSAl Viro tree_contains_unbindable(old_path.mnt)) 18379676f0c6SRam Pai goto out1; 18381da177e4SLinus Torvalds err = -ELOOP; 18392d92ab3cSAl Viro for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent) 18402d92ab3cSAl Viro if (p == old_path.mnt) 184121444403SRam Pai goto out1; 18421da177e4SLinus Torvalds 18432d92ab3cSAl Viro err = attach_recursive_mnt(old_path.mnt, path, &parent_path); 18444ac91378SJan Blunck if (err) 184521444403SRam Pai goto out1; 18461da177e4SLinus Torvalds 18471da177e4SLinus Torvalds /* if the mount is moved, it should no longer be expire 18481da177e4SLinus Torvalds * automatically */ 18492d92ab3cSAl Viro list_del_init(&old_path.mnt->mnt_expire); 18501da177e4SLinus Torvalds out1: 18512d92ab3cSAl Viro mutex_unlock(&path->dentry->d_inode->i_mutex); 18521da177e4SLinus Torvalds out: 1853390c6843SRam Pai up_write(&namespace_sem); 18541da177e4SLinus Torvalds if (!err) 18551a390689SAl Viro path_put(&parent_path); 18562d92ab3cSAl Viro path_put(&old_path); 18571da177e4SLinus Torvalds return err; 18581da177e4SLinus Torvalds } 18591da177e4SLinus Torvalds 18601da177e4SLinus Torvalds /* 18611da177e4SLinus Torvalds * create a new mount for userspace and request it to be added into the 18621da177e4SLinus Torvalds * namespace's tree 18631da177e4SLinus Torvalds */ 18640a0d8a46SAl Viro static int do_new_mount(struct path *path, char *type, int flags, 18651da177e4SLinus Torvalds int mnt_flags, char *name, void *data) 18661da177e4SLinus Torvalds { 18671da177e4SLinus Torvalds struct vfsmount *mnt; 18681da177e4SLinus Torvalds 1869eca6f534SVegard Nossum if (!type) 18701da177e4SLinus Torvalds return -EINVAL; 18711da177e4SLinus Torvalds 18721da177e4SLinus Torvalds /* we need capabilities... */ 18731da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 18741da177e4SLinus Torvalds return -EPERM; 18751da177e4SLinus Torvalds 18761da177e4SLinus Torvalds mnt = do_kern_mount(type, flags, name, data); 18771da177e4SLinus Torvalds if (IS_ERR(mnt)) 18781da177e4SLinus Torvalds return PTR_ERR(mnt); 18791da177e4SLinus Torvalds 1880ea5b778aSDavid Howells return do_add_mount(mnt, path, mnt_flags); 18811da177e4SLinus Torvalds } 18821da177e4SLinus Torvalds 18831da177e4SLinus Torvalds /* 18841da177e4SLinus Torvalds * add a mount into a namespace's mount tree 1885ea5b778aSDavid Howells * - this unconditionally eats one of the caller's references to newmnt. 18861da177e4SLinus Torvalds */ 1887ea5b778aSDavid Howells int do_add_mount(struct vfsmount *newmnt, struct path *path, int mnt_flags) 18881da177e4SLinus Torvalds { 18891da177e4SLinus Torvalds int err; 18901da177e4SLinus Torvalds 18918089352aSAl Viro mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL); 189227d55f1fSAl Viro 1893390c6843SRam Pai down_write(&namespace_sem); 18941da177e4SLinus Torvalds /* Something was mounted here while we slept */ 1895cc53ce53SDavid Howells err = follow_down(path, true); 1896cc53ce53SDavid Howells if (err < 0) 1897cc53ce53SDavid Howells goto unlock; 1898cc53ce53SDavid Howells 18991da177e4SLinus Torvalds err = -EINVAL; 1900dd5cae6eSAl Viro if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt)) 19011da177e4SLinus Torvalds goto unlock; 19021da177e4SLinus Torvalds 19031da177e4SLinus Torvalds /* Refuse the same filesystem on the same mount point */ 19041da177e4SLinus Torvalds err = -EBUSY; 19058d66bf54SAl Viro if (path->mnt->mnt_sb == newmnt->mnt_sb && 19068d66bf54SAl Viro path->mnt->mnt_root == path->dentry) 19071da177e4SLinus Torvalds goto unlock; 19081da177e4SLinus Torvalds 19091da177e4SLinus Torvalds err = -EINVAL; 19101da177e4SLinus Torvalds if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode)) 19111da177e4SLinus Torvalds goto unlock; 19121da177e4SLinus Torvalds 19131da177e4SLinus Torvalds newmnt->mnt_flags = mnt_flags; 19148d66bf54SAl Viro if ((err = graft_tree(newmnt, path))) 19155b83d2c5SRam Pai goto unlock; 19161da177e4SLinus Torvalds 1917390c6843SRam Pai up_write(&namespace_sem); 19185b83d2c5SRam Pai return 0; 19191da177e4SLinus Torvalds 19201da177e4SLinus Torvalds unlock: 1921390c6843SRam Pai up_write(&namespace_sem); 1922f03c6599SAl Viro mntput(newmnt); 19231da177e4SLinus Torvalds return err; 19241da177e4SLinus Torvalds } 19251da177e4SLinus Torvalds 1926ea5b778aSDavid Howells /** 1927ea5b778aSDavid Howells * mnt_set_expiry - Put a mount on an expiration list 1928ea5b778aSDavid Howells * @mnt: The mount to list. 1929ea5b778aSDavid Howells * @expiry_list: The list to add the mount to. 1930ea5b778aSDavid Howells */ 1931ea5b778aSDavid Howells void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list) 1932ea5b778aSDavid Howells { 1933ea5b778aSDavid Howells down_write(&namespace_sem); 1934ea5b778aSDavid Howells br_write_lock(vfsmount_lock); 1935ea5b778aSDavid Howells 1936ea5b778aSDavid Howells list_add_tail(&mnt->mnt_expire, expiry_list); 1937ea5b778aSDavid Howells 1938ea5b778aSDavid Howells br_write_unlock(vfsmount_lock); 1939ea5b778aSDavid Howells up_write(&namespace_sem); 1940ea5b778aSDavid Howells } 1941ea5b778aSDavid Howells EXPORT_SYMBOL(mnt_set_expiry); 1942ea5b778aSDavid Howells 1943ea5b778aSDavid Howells /* 1944ea5b778aSDavid Howells * Remove a vfsmount from any expiration list it may be on 1945ea5b778aSDavid Howells */ 1946ea5b778aSDavid Howells void mnt_clear_expiry(struct vfsmount *mnt) 1947ea5b778aSDavid Howells { 1948ea5b778aSDavid Howells if (!list_empty(&mnt->mnt_expire)) { 1949ea5b778aSDavid Howells down_write(&namespace_sem); 1950ea5b778aSDavid Howells br_write_lock(vfsmount_lock); 1951ea5b778aSDavid Howells list_del_init(&mnt->mnt_expire); 1952ea5b778aSDavid Howells br_write_unlock(vfsmount_lock); 1953ea5b778aSDavid Howells up_write(&namespace_sem); 1954ea5b778aSDavid Howells } 1955ea5b778aSDavid Howells } 19561da177e4SLinus Torvalds 19575528f911STrond Myklebust /* 19581da177e4SLinus Torvalds * process a list of expirable mountpoints with the intent of discarding any 19591da177e4SLinus Torvalds * mountpoints that aren't in use and haven't been touched since last we came 19601da177e4SLinus Torvalds * here 19611da177e4SLinus Torvalds */ 19621da177e4SLinus Torvalds void mark_mounts_for_expiry(struct list_head *mounts) 19631da177e4SLinus Torvalds { 19641da177e4SLinus Torvalds struct vfsmount *mnt, *next; 19651da177e4SLinus Torvalds LIST_HEAD(graveyard); 1966bcc5c7d2SAl Viro LIST_HEAD(umounts); 19671da177e4SLinus Torvalds 19681da177e4SLinus Torvalds if (list_empty(mounts)) 19691da177e4SLinus Torvalds return; 19701da177e4SLinus Torvalds 1971bcc5c7d2SAl Viro down_write(&namespace_sem); 197299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 19731da177e4SLinus Torvalds 19741da177e4SLinus Torvalds /* extract from the expiration list every vfsmount that matches the 19751da177e4SLinus Torvalds * following criteria: 19761da177e4SLinus Torvalds * - only referenced by its parent vfsmount 19771da177e4SLinus Torvalds * - still marked for expiry (marked on the last call here; marks are 19781da177e4SLinus Torvalds * cleared by mntput()) 19791da177e4SLinus Torvalds */ 198055e700b9SMiklos Szeredi list_for_each_entry_safe(mnt, next, mounts, mnt_expire) { 19811da177e4SLinus Torvalds if (!xchg(&mnt->mnt_expiry_mark, 1) || 1982bcc5c7d2SAl Viro propagate_mount_busy(mnt, 1)) 19831da177e4SLinus Torvalds continue; 198455e700b9SMiklos Szeredi list_move(&mnt->mnt_expire, &graveyard); 19851da177e4SLinus Torvalds } 1986bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 1987bcc5c7d2SAl Viro mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire); 1988bcc5c7d2SAl Viro touch_mnt_namespace(mnt->mnt_ns); 1989bcc5c7d2SAl Viro umount_tree(mnt, 1, &umounts); 1990bcc5c7d2SAl Viro } 199199b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 1992bcc5c7d2SAl Viro up_write(&namespace_sem); 1993bcc5c7d2SAl Viro 1994bcc5c7d2SAl Viro release_mounts(&umounts); 19951da177e4SLinus Torvalds } 19961da177e4SLinus Torvalds 19971da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(mark_mounts_for_expiry); 19981da177e4SLinus Torvalds 19991da177e4SLinus Torvalds /* 20005528f911STrond Myklebust * Ripoff of 'select_parent()' 20015528f911STrond Myklebust * 20025528f911STrond Myklebust * search the list of submounts for a given mountpoint, and move any 20035528f911STrond Myklebust * shrinkable submounts to the 'graveyard' list. 20045528f911STrond Myklebust */ 20055528f911STrond Myklebust static int select_submounts(struct vfsmount *parent, struct list_head *graveyard) 20065528f911STrond Myklebust { 20075528f911STrond Myklebust struct vfsmount *this_parent = parent; 20085528f911STrond Myklebust struct list_head *next; 20095528f911STrond Myklebust int found = 0; 20105528f911STrond Myklebust 20115528f911STrond Myklebust repeat: 20125528f911STrond Myklebust next = this_parent->mnt_mounts.next; 20135528f911STrond Myklebust resume: 20145528f911STrond Myklebust while (next != &this_parent->mnt_mounts) { 20155528f911STrond Myklebust struct list_head *tmp = next; 20165528f911STrond Myklebust struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child); 20175528f911STrond Myklebust 20185528f911STrond Myklebust next = tmp->next; 20195528f911STrond Myklebust if (!(mnt->mnt_flags & MNT_SHRINKABLE)) 20205528f911STrond Myklebust continue; 20215528f911STrond Myklebust /* 20225528f911STrond Myklebust * Descend a level if the d_mounts list is non-empty. 20235528f911STrond Myklebust */ 20245528f911STrond Myklebust if (!list_empty(&mnt->mnt_mounts)) { 20255528f911STrond Myklebust this_parent = mnt; 20265528f911STrond Myklebust goto repeat; 20275528f911STrond Myklebust } 20285528f911STrond Myklebust 20295528f911STrond Myklebust if (!propagate_mount_busy(mnt, 1)) { 20305528f911STrond Myklebust list_move_tail(&mnt->mnt_expire, graveyard); 20315528f911STrond Myklebust found++; 20325528f911STrond Myklebust } 20335528f911STrond Myklebust } 20345528f911STrond Myklebust /* 20355528f911STrond Myklebust * All done at this level ... ascend and resume the search 20365528f911STrond Myklebust */ 20375528f911STrond Myklebust if (this_parent != parent) { 20385528f911STrond Myklebust next = this_parent->mnt_child.next; 20395528f911STrond Myklebust this_parent = this_parent->mnt_parent; 20405528f911STrond Myklebust goto resume; 20415528f911STrond Myklebust } 20425528f911STrond Myklebust return found; 20435528f911STrond Myklebust } 20445528f911STrond Myklebust 20455528f911STrond Myklebust /* 20465528f911STrond Myklebust * process a list of expirable mountpoints with the intent of discarding any 20475528f911STrond Myklebust * submounts of a specific parent mountpoint 204899b7db7bSNick Piggin * 204999b7db7bSNick Piggin * vfsmount_lock must be held for write 20505528f911STrond Myklebust */ 2051c35038beSAl Viro static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts) 20525528f911STrond Myklebust { 20535528f911STrond Myklebust LIST_HEAD(graveyard); 2054c35038beSAl Viro struct vfsmount *m; 20555528f911STrond Myklebust 20565528f911STrond Myklebust /* extract submounts of 'mountpoint' from the expiration list */ 2057c35038beSAl Viro while (select_submounts(mnt, &graveyard)) { 2058bcc5c7d2SAl Viro while (!list_empty(&graveyard)) { 2059c35038beSAl Viro m = list_first_entry(&graveyard, struct vfsmount, 2060bcc5c7d2SAl Viro mnt_expire); 2061afef80b3SEric W. Biederman touch_mnt_namespace(m->mnt_ns); 2062afef80b3SEric W. Biederman umount_tree(m, 1, umounts); 2063bcc5c7d2SAl Viro } 2064bcc5c7d2SAl Viro } 20655528f911STrond Myklebust } 20665528f911STrond Myklebust 20675528f911STrond Myklebust /* 20681da177e4SLinus Torvalds * Some copy_from_user() implementations do not return the exact number of 20691da177e4SLinus Torvalds * bytes remaining to copy on a fault. But copy_mount_options() requires that. 20701da177e4SLinus Torvalds * Note that this function differs from copy_from_user() in that it will oops 20711da177e4SLinus Torvalds * on bad values of `to', rather than returning a short copy. 20721da177e4SLinus Torvalds */ 2073b58fed8bSRam Pai static long exact_copy_from_user(void *to, const void __user * from, 2074b58fed8bSRam Pai unsigned long n) 20751da177e4SLinus Torvalds { 20761da177e4SLinus Torvalds char *t = to; 20771da177e4SLinus Torvalds const char __user *f = from; 20781da177e4SLinus Torvalds char c; 20791da177e4SLinus Torvalds 20801da177e4SLinus Torvalds if (!access_ok(VERIFY_READ, from, n)) 20811da177e4SLinus Torvalds return n; 20821da177e4SLinus Torvalds 20831da177e4SLinus Torvalds while (n) { 20841da177e4SLinus Torvalds if (__get_user(c, f)) { 20851da177e4SLinus Torvalds memset(t, 0, n); 20861da177e4SLinus Torvalds break; 20871da177e4SLinus Torvalds } 20881da177e4SLinus Torvalds *t++ = c; 20891da177e4SLinus Torvalds f++; 20901da177e4SLinus Torvalds n--; 20911da177e4SLinus Torvalds } 20921da177e4SLinus Torvalds return n; 20931da177e4SLinus Torvalds } 20941da177e4SLinus Torvalds 20951da177e4SLinus Torvalds int copy_mount_options(const void __user * data, unsigned long *where) 20961da177e4SLinus Torvalds { 20971da177e4SLinus Torvalds int i; 20981da177e4SLinus Torvalds unsigned long page; 20991da177e4SLinus Torvalds unsigned long size; 21001da177e4SLinus Torvalds 21011da177e4SLinus Torvalds *where = 0; 21021da177e4SLinus Torvalds if (!data) 21031da177e4SLinus Torvalds return 0; 21041da177e4SLinus Torvalds 21051da177e4SLinus Torvalds if (!(page = __get_free_page(GFP_KERNEL))) 21061da177e4SLinus Torvalds return -ENOMEM; 21071da177e4SLinus Torvalds 21081da177e4SLinus Torvalds /* We only care that *some* data at the address the user 21091da177e4SLinus Torvalds * gave us is valid. Just in case, we'll zero 21101da177e4SLinus Torvalds * the remainder of the page. 21111da177e4SLinus Torvalds */ 21121da177e4SLinus Torvalds /* copy_from_user cannot cross TASK_SIZE ! */ 21131da177e4SLinus Torvalds size = TASK_SIZE - (unsigned long)data; 21141da177e4SLinus Torvalds if (size > PAGE_SIZE) 21151da177e4SLinus Torvalds size = PAGE_SIZE; 21161da177e4SLinus Torvalds 21171da177e4SLinus Torvalds i = size - exact_copy_from_user((void *)page, data, size); 21181da177e4SLinus Torvalds if (!i) { 21191da177e4SLinus Torvalds free_page(page); 21201da177e4SLinus Torvalds return -EFAULT; 21211da177e4SLinus Torvalds } 21221da177e4SLinus Torvalds if (i != PAGE_SIZE) 21231da177e4SLinus Torvalds memset((char *)page + i, 0, PAGE_SIZE - i); 21241da177e4SLinus Torvalds *where = page; 21251da177e4SLinus Torvalds return 0; 21261da177e4SLinus Torvalds } 21271da177e4SLinus Torvalds 2128eca6f534SVegard Nossum int copy_mount_string(const void __user *data, char **where) 2129eca6f534SVegard Nossum { 2130eca6f534SVegard Nossum char *tmp; 2131eca6f534SVegard Nossum 2132eca6f534SVegard Nossum if (!data) { 2133eca6f534SVegard Nossum *where = NULL; 2134eca6f534SVegard Nossum return 0; 2135eca6f534SVegard Nossum } 2136eca6f534SVegard Nossum 2137eca6f534SVegard Nossum tmp = strndup_user(data, PAGE_SIZE); 2138eca6f534SVegard Nossum if (IS_ERR(tmp)) 2139eca6f534SVegard Nossum return PTR_ERR(tmp); 2140eca6f534SVegard Nossum 2141eca6f534SVegard Nossum *where = tmp; 2142eca6f534SVegard Nossum return 0; 2143eca6f534SVegard Nossum } 2144eca6f534SVegard Nossum 21451da177e4SLinus Torvalds /* 21461da177e4SLinus Torvalds * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to 21471da177e4SLinus Torvalds * be given to the mount() call (ie: read-only, no-dev, no-suid etc). 21481da177e4SLinus Torvalds * 21491da177e4SLinus Torvalds * data is a (void *) that can point to any structure up to 21501da177e4SLinus Torvalds * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent 21511da177e4SLinus Torvalds * information (or be NULL). 21521da177e4SLinus Torvalds * 21531da177e4SLinus Torvalds * Pre-0.97 versions of mount() didn't have a flags word. 21541da177e4SLinus Torvalds * When the flags word was introduced its top half was required 21551da177e4SLinus Torvalds * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9. 21561da177e4SLinus Torvalds * Therefore, if this magic number is present, it carries no information 21571da177e4SLinus Torvalds * and must be discarded. 21581da177e4SLinus Torvalds */ 21591da177e4SLinus Torvalds long do_mount(char *dev_name, char *dir_name, char *type_page, 21601da177e4SLinus Torvalds unsigned long flags, void *data_page) 21611da177e4SLinus Torvalds { 21622d92ab3cSAl Viro struct path path; 21631da177e4SLinus Torvalds int retval = 0; 21641da177e4SLinus Torvalds int mnt_flags = 0; 21651da177e4SLinus Torvalds 21661da177e4SLinus Torvalds /* Discard magic */ 21671da177e4SLinus Torvalds if ((flags & MS_MGC_MSK) == MS_MGC_VAL) 21681da177e4SLinus Torvalds flags &= ~MS_MGC_MSK; 21691da177e4SLinus Torvalds 21701da177e4SLinus Torvalds /* Basic sanity checks */ 21711da177e4SLinus Torvalds 21721da177e4SLinus Torvalds if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE)) 21731da177e4SLinus Torvalds return -EINVAL; 21741da177e4SLinus Torvalds 21751da177e4SLinus Torvalds if (data_page) 21761da177e4SLinus Torvalds ((char *)data_page)[PAGE_SIZE - 1] = 0; 21771da177e4SLinus Torvalds 2178a27ab9f2STetsuo Handa /* ... and get the mountpoint */ 2179a27ab9f2STetsuo Handa retval = kern_path(dir_name, LOOKUP_FOLLOW, &path); 2180a27ab9f2STetsuo Handa if (retval) 2181a27ab9f2STetsuo Handa return retval; 2182a27ab9f2STetsuo Handa 2183a27ab9f2STetsuo Handa retval = security_sb_mount(dev_name, &path, 2184a27ab9f2STetsuo Handa type_page, flags, data_page); 2185a27ab9f2STetsuo Handa if (retval) 2186a27ab9f2STetsuo Handa goto dput_out; 2187a27ab9f2STetsuo Handa 2188613cbe3dSAndi Kleen /* Default to relatime unless overriden */ 2189613cbe3dSAndi Kleen if (!(flags & MS_NOATIME)) 21900a1c01c9SMatthew Garrett mnt_flags |= MNT_RELATIME; 21910a1c01c9SMatthew Garrett 21921da177e4SLinus Torvalds /* Separate the per-mountpoint flags */ 21931da177e4SLinus Torvalds if (flags & MS_NOSUID) 21941da177e4SLinus Torvalds mnt_flags |= MNT_NOSUID; 21951da177e4SLinus Torvalds if (flags & MS_NODEV) 21961da177e4SLinus Torvalds mnt_flags |= MNT_NODEV; 21971da177e4SLinus Torvalds if (flags & MS_NOEXEC) 21981da177e4SLinus Torvalds mnt_flags |= MNT_NOEXEC; 2199fc33a7bbSChristoph Hellwig if (flags & MS_NOATIME) 2200fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NOATIME; 2201fc33a7bbSChristoph Hellwig if (flags & MS_NODIRATIME) 2202fc33a7bbSChristoph Hellwig mnt_flags |= MNT_NODIRATIME; 2203d0adde57SMatthew Garrett if (flags & MS_STRICTATIME) 2204d0adde57SMatthew Garrett mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME); 22052e4b7fcdSDave Hansen if (flags & MS_RDONLY) 22062e4b7fcdSDave Hansen mnt_flags |= MNT_READONLY; 2207fc33a7bbSChristoph Hellwig 22087a4dec53SAl Viro flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN | 2209d0adde57SMatthew Garrett MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT | 2210d0adde57SMatthew Garrett MS_STRICTATIME); 22111da177e4SLinus Torvalds 22121da177e4SLinus Torvalds if (flags & MS_REMOUNT) 22132d92ab3cSAl Viro retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags, 22141da177e4SLinus Torvalds data_page); 22151da177e4SLinus Torvalds else if (flags & MS_BIND) 22162d92ab3cSAl Viro retval = do_loopback(&path, dev_name, flags & MS_REC); 22179676f0c6SRam Pai else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE)) 22182d92ab3cSAl Viro retval = do_change_type(&path, flags); 22191da177e4SLinus Torvalds else if (flags & MS_MOVE) 22202d92ab3cSAl Viro retval = do_move_mount(&path, dev_name); 22211da177e4SLinus Torvalds else 22222d92ab3cSAl Viro retval = do_new_mount(&path, type_page, flags, mnt_flags, 22231da177e4SLinus Torvalds dev_name, data_page); 22241da177e4SLinus Torvalds dput_out: 22252d92ab3cSAl Viro path_put(&path); 22261da177e4SLinus Torvalds return retval; 22271da177e4SLinus Torvalds } 22281da177e4SLinus Torvalds 2229cf8d2c11STrond Myklebust static struct mnt_namespace *alloc_mnt_ns(void) 2230cf8d2c11STrond Myklebust { 2231cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2232cf8d2c11STrond Myklebust 2233cf8d2c11STrond Myklebust new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL); 2234cf8d2c11STrond Myklebust if (!new_ns) 2235cf8d2c11STrond Myklebust return ERR_PTR(-ENOMEM); 2236cf8d2c11STrond Myklebust atomic_set(&new_ns->count, 1); 2237cf8d2c11STrond Myklebust new_ns->root = NULL; 2238cf8d2c11STrond Myklebust INIT_LIST_HEAD(&new_ns->list); 2239cf8d2c11STrond Myklebust init_waitqueue_head(&new_ns->poll); 2240cf8d2c11STrond Myklebust new_ns->event = 0; 2241cf8d2c11STrond Myklebust return new_ns; 2242cf8d2c11STrond Myklebust } 2243cf8d2c11STrond Myklebust 2244f03c6599SAl Viro void mnt_make_longterm(struct vfsmount *mnt) 2245f03c6599SAl Viro { 2246f03c6599SAl Viro atomic_inc(&mnt->mnt_longterm); 2247f03c6599SAl Viro } 2248f03c6599SAl Viro 2249f03c6599SAl Viro void mnt_make_shortterm(struct vfsmount *mnt) 2250f03c6599SAl Viro { 2251f03c6599SAl Viro if (atomic_add_unless(&mnt->mnt_longterm, -1, 1)) 2252f03c6599SAl Viro return; 2253f03c6599SAl Viro br_write_lock(vfsmount_lock); 2254f03c6599SAl Viro atomic_dec(&mnt->mnt_longterm); 2255f03c6599SAl Viro br_write_unlock(vfsmount_lock); 2256f03c6599SAl Viro } 2257f03c6599SAl Viro 2258741a2951SJANAK DESAI /* 2259741a2951SJANAK DESAI * Allocate a new namespace structure and populate it with contents 2260741a2951SJANAK DESAI * copied from the namespace of the passed in task structure. 2261741a2951SJANAK DESAI */ 2262e3222c4eSBadari Pulavarty static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns, 22636b3286edSKirill Korotaev struct fs_struct *fs) 22641da177e4SLinus Torvalds { 22656b3286edSKirill Korotaev struct mnt_namespace *new_ns; 22667f2da1e7SAl Viro struct vfsmount *rootmnt = NULL, *pwdmnt = NULL; 22671da177e4SLinus Torvalds struct vfsmount *p, *q; 22681da177e4SLinus Torvalds 2269cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2270cf8d2c11STrond Myklebust if (IS_ERR(new_ns)) 2271cf8d2c11STrond Myklebust return new_ns; 22721da177e4SLinus Torvalds 2273390c6843SRam Pai down_write(&namespace_sem); 22741da177e4SLinus Torvalds /* First pass: copy the tree topology */ 22756b3286edSKirill Korotaev new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root, 22769676f0c6SRam Pai CL_COPY_ALL | CL_EXPIRE); 22771da177e4SLinus Torvalds if (!new_ns->root) { 2278390c6843SRam Pai up_write(&namespace_sem); 22791da177e4SLinus Torvalds kfree(new_ns); 22805cc4a034SJulia Lawall return ERR_PTR(-ENOMEM); 22811da177e4SLinus Torvalds } 228299b7db7bSNick Piggin br_write_lock(vfsmount_lock); 22831da177e4SLinus Torvalds list_add_tail(&new_ns->list, &new_ns->root->mnt_list); 228499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 22851da177e4SLinus Torvalds 22861da177e4SLinus Torvalds /* 22871da177e4SLinus Torvalds * Second pass: switch the tsk->fs->* elements and mark new vfsmounts 22881da177e4SLinus Torvalds * as belonging to new namespace. We have already acquired a private 22891da177e4SLinus Torvalds * fs_struct, so tsk->fs->lock is not needed. 22901da177e4SLinus Torvalds */ 22916b3286edSKirill Korotaev p = mnt_ns->root; 22921da177e4SLinus Torvalds q = new_ns->root; 22931da177e4SLinus Torvalds while (p) { 22946b3286edSKirill Korotaev q->mnt_ns = new_ns; 2295f03c6599SAl Viro atomic_inc(&q->mnt_longterm); 22961da177e4SLinus Torvalds if (fs) { 22976ac08c39SJan Blunck if (p == fs->root.mnt) { 2298f03c6599SAl Viro fs->root.mnt = mntget(q); 2299f03c6599SAl Viro atomic_inc(&q->mnt_longterm); 2300f03c6599SAl Viro mnt_make_shortterm(p); 23011da177e4SLinus Torvalds rootmnt = p; 23021da177e4SLinus Torvalds } 23036ac08c39SJan Blunck if (p == fs->pwd.mnt) { 2304f03c6599SAl Viro fs->pwd.mnt = mntget(q); 2305f03c6599SAl Viro atomic_inc(&q->mnt_longterm); 2306f03c6599SAl Viro mnt_make_shortterm(p); 23071da177e4SLinus Torvalds pwdmnt = p; 23081da177e4SLinus Torvalds } 23091da177e4SLinus Torvalds } 23106b3286edSKirill Korotaev p = next_mnt(p, mnt_ns->root); 23111da177e4SLinus Torvalds q = next_mnt(q, new_ns->root); 23121da177e4SLinus Torvalds } 2313390c6843SRam Pai up_write(&namespace_sem); 23141da177e4SLinus Torvalds 23151da177e4SLinus Torvalds if (rootmnt) 2316f03c6599SAl Viro mntput(rootmnt); 23171da177e4SLinus Torvalds if (pwdmnt) 2318f03c6599SAl Viro mntput(pwdmnt); 23191da177e4SLinus Torvalds 2320741a2951SJANAK DESAI return new_ns; 2321741a2951SJANAK DESAI } 2322741a2951SJANAK DESAI 2323213dd266SEric W. Biederman struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns, 2324e3222c4eSBadari Pulavarty struct fs_struct *new_fs) 2325741a2951SJANAK DESAI { 23266b3286edSKirill Korotaev struct mnt_namespace *new_ns; 2327741a2951SJANAK DESAI 2328e3222c4eSBadari Pulavarty BUG_ON(!ns); 23296b3286edSKirill Korotaev get_mnt_ns(ns); 2330741a2951SJANAK DESAI 2331741a2951SJANAK DESAI if (!(flags & CLONE_NEWNS)) 2332e3222c4eSBadari Pulavarty return ns; 2333741a2951SJANAK DESAI 2334e3222c4eSBadari Pulavarty new_ns = dup_mnt_ns(ns, new_fs); 2335741a2951SJANAK DESAI 23366b3286edSKirill Korotaev put_mnt_ns(ns); 2337e3222c4eSBadari Pulavarty return new_ns; 23381da177e4SLinus Torvalds } 23391da177e4SLinus Torvalds 2340cf8d2c11STrond Myklebust /** 2341cf8d2c11STrond Myklebust * create_mnt_ns - creates a private namespace and adds a root filesystem 2342cf8d2c11STrond Myklebust * @mnt: pointer to the new root filesystem mountpoint 2343cf8d2c11STrond Myklebust */ 2344a2770d86SLinus Torvalds struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt) 2345cf8d2c11STrond Myklebust { 2346cf8d2c11STrond Myklebust struct mnt_namespace *new_ns; 2347cf8d2c11STrond Myklebust 2348cf8d2c11STrond Myklebust new_ns = alloc_mnt_ns(); 2349cf8d2c11STrond Myklebust if (!IS_ERR(new_ns)) { 2350cf8d2c11STrond Myklebust mnt->mnt_ns = new_ns; 2351f03c6599SAl Viro atomic_inc(&mnt->mnt_longterm); 2352cf8d2c11STrond Myklebust new_ns->root = mnt; 2353cf8d2c11STrond Myklebust list_add(&new_ns->list, &new_ns->root->mnt_list); 2354cf8d2c11STrond Myklebust } 2355cf8d2c11STrond Myklebust return new_ns; 2356cf8d2c11STrond Myklebust } 2357a2770d86SLinus Torvalds EXPORT_SYMBOL(create_mnt_ns); 2358cf8d2c11STrond Myklebust 2359bdc480e3SHeiko Carstens SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name, 2360bdc480e3SHeiko Carstens char __user *, type, unsigned long, flags, void __user *, data) 23611da177e4SLinus Torvalds { 2362eca6f534SVegard Nossum int ret; 2363eca6f534SVegard Nossum char *kernel_type; 2364eca6f534SVegard Nossum char *kernel_dir; 2365eca6f534SVegard Nossum char *kernel_dev; 23661da177e4SLinus Torvalds unsigned long data_page; 23671da177e4SLinus Torvalds 2368eca6f534SVegard Nossum ret = copy_mount_string(type, &kernel_type); 2369eca6f534SVegard Nossum if (ret < 0) 2370eca6f534SVegard Nossum goto out_type; 23711da177e4SLinus Torvalds 2372eca6f534SVegard Nossum kernel_dir = getname(dir_name); 2373eca6f534SVegard Nossum if (IS_ERR(kernel_dir)) { 2374eca6f534SVegard Nossum ret = PTR_ERR(kernel_dir); 2375eca6f534SVegard Nossum goto out_dir; 2376eca6f534SVegard Nossum } 23771da177e4SLinus Torvalds 2378eca6f534SVegard Nossum ret = copy_mount_string(dev_name, &kernel_dev); 2379eca6f534SVegard Nossum if (ret < 0) 2380eca6f534SVegard Nossum goto out_dev; 23811da177e4SLinus Torvalds 2382eca6f534SVegard Nossum ret = copy_mount_options(data, &data_page); 2383eca6f534SVegard Nossum if (ret < 0) 2384eca6f534SVegard Nossum goto out_data; 23851da177e4SLinus Torvalds 2386eca6f534SVegard Nossum ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags, 2387eca6f534SVegard Nossum (void *) data_page); 2388eca6f534SVegard Nossum 23891da177e4SLinus Torvalds free_page(data_page); 2390eca6f534SVegard Nossum out_data: 2391eca6f534SVegard Nossum kfree(kernel_dev); 2392eca6f534SVegard Nossum out_dev: 2393eca6f534SVegard Nossum putname(kernel_dir); 2394eca6f534SVegard Nossum out_dir: 2395eca6f534SVegard Nossum kfree(kernel_type); 2396eca6f534SVegard Nossum out_type: 2397eca6f534SVegard Nossum return ret; 23981da177e4SLinus Torvalds } 23991da177e4SLinus Torvalds 24001da177e4SLinus Torvalds /* 24011da177e4SLinus Torvalds * pivot_root Semantics: 24021da177e4SLinus Torvalds * Moves the root file system of the current process to the directory put_old, 24031da177e4SLinus Torvalds * makes new_root as the new root file system of the current process, and sets 24041da177e4SLinus Torvalds * root/cwd of all processes which had them on the current root to new_root. 24051da177e4SLinus Torvalds * 24061da177e4SLinus Torvalds * Restrictions: 24071da177e4SLinus Torvalds * The new_root and put_old must be directories, and must not be on the 24081da177e4SLinus Torvalds * same file system as the current process root. The put_old must be 24091da177e4SLinus Torvalds * underneath new_root, i.e. adding a non-zero number of /.. to the string 24101da177e4SLinus Torvalds * pointed to by put_old must yield the same directory as new_root. No other 24111da177e4SLinus Torvalds * file system may be mounted on put_old. After all, new_root is a mountpoint. 24121da177e4SLinus Torvalds * 24134a0d11faSNeil Brown * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem. 24144a0d11faSNeil Brown * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives 24154a0d11faSNeil Brown * in this situation. 24164a0d11faSNeil Brown * 24171da177e4SLinus Torvalds * Notes: 24181da177e4SLinus Torvalds * - we don't move root/cwd if they are not at the root (reason: if something 24191da177e4SLinus Torvalds * cared enough to change them, it's probably wrong to force them elsewhere) 24201da177e4SLinus Torvalds * - it's okay to pick a root that isn't the root of a file system, e.g. 24211da177e4SLinus Torvalds * /nfs/my_root where /nfs is the mount point. It must be a mountpoint, 24221da177e4SLinus Torvalds * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root 24231da177e4SLinus Torvalds * first. 24241da177e4SLinus Torvalds */ 24253480b257SHeiko Carstens SYSCALL_DEFINE2(pivot_root, const char __user *, new_root, 24263480b257SHeiko Carstens const char __user *, put_old) 24271da177e4SLinus Torvalds { 24281da177e4SLinus Torvalds struct vfsmount *tmp; 24292d8f3038SAl Viro struct path new, old, parent_path, root_parent, root; 24301da177e4SLinus Torvalds int error; 24311da177e4SLinus Torvalds 24321da177e4SLinus Torvalds if (!capable(CAP_SYS_ADMIN)) 24331da177e4SLinus Torvalds return -EPERM; 24341da177e4SLinus Torvalds 24352d8f3038SAl Viro error = user_path_dir(new_root, &new); 24361da177e4SLinus Torvalds if (error) 24371da177e4SLinus Torvalds goto out0; 24381da177e4SLinus Torvalds error = -EINVAL; 24392d8f3038SAl Viro if (!check_mnt(new.mnt)) 24401da177e4SLinus Torvalds goto out1; 24411da177e4SLinus Torvalds 24422d8f3038SAl Viro error = user_path_dir(put_old, &old); 24431da177e4SLinus Torvalds if (error) 24441da177e4SLinus Torvalds goto out1; 24451da177e4SLinus Torvalds 24462d8f3038SAl Viro error = security_sb_pivotroot(&old, &new); 24471da177e4SLinus Torvalds if (error) { 24482d8f3038SAl Viro path_put(&old); 24491da177e4SLinus Torvalds goto out1; 24501da177e4SLinus Torvalds } 24511da177e4SLinus Torvalds 2452f7ad3c6bSMiklos Szeredi get_fs_root(current->fs, &root); 2453390c6843SRam Pai down_write(&namespace_sem); 24542d8f3038SAl Viro mutex_lock(&old.dentry->d_inode->i_mutex); 24551da177e4SLinus Torvalds error = -EINVAL; 24562d8f3038SAl Viro if (IS_MNT_SHARED(old.mnt) || 24572d8f3038SAl Viro IS_MNT_SHARED(new.mnt->mnt_parent) || 24588c3ee42eSAl Viro IS_MNT_SHARED(root.mnt->mnt_parent)) 245921444403SRam Pai goto out2; 24608c3ee42eSAl Viro if (!check_mnt(root.mnt)) 24611da177e4SLinus Torvalds goto out2; 24621da177e4SLinus Torvalds error = -ENOENT; 2463d83c49f3SAl Viro if (cant_mount(old.dentry)) 24641da177e4SLinus Torvalds goto out2; 2465f3da392eSAlexey Dobriyan if (d_unlinked(new.dentry)) 24661da177e4SLinus Torvalds goto out2; 2467f3da392eSAlexey Dobriyan if (d_unlinked(old.dentry)) 24681da177e4SLinus Torvalds goto out2; 24691da177e4SLinus Torvalds error = -EBUSY; 24702d8f3038SAl Viro if (new.mnt == root.mnt || 24712d8f3038SAl Viro old.mnt == root.mnt) 24721da177e4SLinus Torvalds goto out2; /* loop, on the same file system */ 24731da177e4SLinus Torvalds error = -EINVAL; 24748c3ee42eSAl Viro if (root.mnt->mnt_root != root.dentry) 24751da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 24768c3ee42eSAl Viro if (root.mnt->mnt_parent == root.mnt) 24770bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 24782d8f3038SAl Viro if (new.mnt->mnt_root != new.dentry) 24791da177e4SLinus Torvalds goto out2; /* not a mountpoint */ 24802d8f3038SAl Viro if (new.mnt->mnt_parent == new.mnt) 24810bb6fcc1SMiklos Szeredi goto out2; /* not attached */ 24824ac91378SJan Blunck /* make sure we can reach put_old from new_root */ 24832d8f3038SAl Viro tmp = old.mnt; 248499b7db7bSNick Piggin br_write_lock(vfsmount_lock); 24852d8f3038SAl Viro if (tmp != new.mnt) { 24861da177e4SLinus Torvalds for (;;) { 24871da177e4SLinus Torvalds if (tmp->mnt_parent == tmp) 24881da177e4SLinus Torvalds goto out3; /* already mounted on put_old */ 24892d8f3038SAl Viro if (tmp->mnt_parent == new.mnt) 24901da177e4SLinus Torvalds break; 24911da177e4SLinus Torvalds tmp = tmp->mnt_parent; 24921da177e4SLinus Torvalds } 24932d8f3038SAl Viro if (!is_subdir(tmp->mnt_mountpoint, new.dentry)) 24941da177e4SLinus Torvalds goto out3; 24952d8f3038SAl Viro } else if (!is_subdir(old.dentry, new.dentry)) 24961da177e4SLinus Torvalds goto out3; 24972d8f3038SAl Viro detach_mnt(new.mnt, &parent_path); 24988c3ee42eSAl Viro detach_mnt(root.mnt, &root_parent); 24994ac91378SJan Blunck /* mount old root on put_old */ 25002d8f3038SAl Viro attach_mnt(root.mnt, &old); 25014ac91378SJan Blunck /* mount new_root on / */ 25022d8f3038SAl Viro attach_mnt(new.mnt, &root_parent); 25036b3286edSKirill Korotaev touch_mnt_namespace(current->nsproxy->mnt_ns); 250499b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 25052d8f3038SAl Viro chroot_fs_refs(&root, &new); 2506b3e19d92SNick Piggin 25071da177e4SLinus Torvalds error = 0; 25081a390689SAl Viro path_put(&root_parent); 25091a390689SAl Viro path_put(&parent_path); 25101da177e4SLinus Torvalds out2: 25112d8f3038SAl Viro mutex_unlock(&old.dentry->d_inode->i_mutex); 2512390c6843SRam Pai up_write(&namespace_sem); 25138c3ee42eSAl Viro path_put(&root); 25142d8f3038SAl Viro path_put(&old); 25151da177e4SLinus Torvalds out1: 25162d8f3038SAl Viro path_put(&new); 25171da177e4SLinus Torvalds out0: 25181da177e4SLinus Torvalds return error; 25191da177e4SLinus Torvalds out3: 252099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 25211da177e4SLinus Torvalds goto out2; 25221da177e4SLinus Torvalds } 25231da177e4SLinus Torvalds 25241da177e4SLinus Torvalds static void __init init_mount_tree(void) 25251da177e4SLinus Torvalds { 25261da177e4SLinus Torvalds struct vfsmount *mnt; 25276b3286edSKirill Korotaev struct mnt_namespace *ns; 2528ac748a09SJan Blunck struct path root; 25291da177e4SLinus Torvalds 25301da177e4SLinus Torvalds mnt = do_kern_mount("rootfs", 0, "rootfs", NULL); 25311da177e4SLinus Torvalds if (IS_ERR(mnt)) 25321da177e4SLinus Torvalds panic("Can't create rootfs"); 2533b3e19d92SNick Piggin 25343b22edc5STrond Myklebust ns = create_mnt_ns(mnt); 25353b22edc5STrond Myklebust if (IS_ERR(ns)) 25361da177e4SLinus Torvalds panic("Can't allocate initial namespace"); 25371da177e4SLinus Torvalds 25386b3286edSKirill Korotaev init_task.nsproxy->mnt_ns = ns; 25396b3286edSKirill Korotaev get_mnt_ns(ns); 25401da177e4SLinus Torvalds 2541ac748a09SJan Blunck root.mnt = ns->root; 2542ac748a09SJan Blunck root.dentry = ns->root->mnt_root; 2543ac748a09SJan Blunck 2544ac748a09SJan Blunck set_fs_pwd(current->fs, &root); 2545ac748a09SJan Blunck set_fs_root(current->fs, &root); 25461da177e4SLinus Torvalds } 25471da177e4SLinus Torvalds 254874bf17cfSDenis Cheng void __init mnt_init(void) 25491da177e4SLinus Torvalds { 255013f14b4dSEric Dumazet unsigned u; 255115a67dd8SRandy Dunlap int err; 25521da177e4SLinus Torvalds 2553390c6843SRam Pai init_rwsem(&namespace_sem); 2554390c6843SRam Pai 25551da177e4SLinus Torvalds mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount), 255620c2df83SPaul Mundt 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL); 25571da177e4SLinus Torvalds 2558b58fed8bSRam Pai mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC); 25591da177e4SLinus Torvalds 25601da177e4SLinus Torvalds if (!mount_hashtable) 25611da177e4SLinus Torvalds panic("Failed to allocate mount hash table\n"); 25621da177e4SLinus Torvalds 256313f14b4dSEric Dumazet printk("Mount-cache hash table entries: %lu\n", HASH_SIZE); 25641da177e4SLinus Torvalds 256513f14b4dSEric Dumazet for (u = 0; u < HASH_SIZE; u++) 256613f14b4dSEric Dumazet INIT_LIST_HEAD(&mount_hashtable[u]); 25671da177e4SLinus Torvalds 256899b7db7bSNick Piggin br_lock_init(vfsmount_lock); 256999b7db7bSNick Piggin 257015a67dd8SRandy Dunlap err = sysfs_init(); 257115a67dd8SRandy Dunlap if (err) 257215a67dd8SRandy Dunlap printk(KERN_WARNING "%s: sysfs_init error: %d\n", 25738e24eea7SHarvey Harrison __func__, err); 257400d26666SGreg Kroah-Hartman fs_kobj = kobject_create_and_add("fs", NULL); 257500d26666SGreg Kroah-Hartman if (!fs_kobj) 25768e24eea7SHarvey Harrison printk(KERN_WARNING "%s: kobj create error\n", __func__); 25771da177e4SLinus Torvalds init_rootfs(); 25781da177e4SLinus Torvalds init_mount_tree(); 25791da177e4SLinus Torvalds } 25801da177e4SLinus Torvalds 2581616511d0STrond Myklebust void put_mnt_ns(struct mnt_namespace *ns) 25821da177e4SLinus Torvalds { 258370fbcdf4SRam Pai LIST_HEAD(umount_list); 2584616511d0STrond Myklebust 2585d498b25aSAl Viro if (!atomic_dec_and_test(&ns->count)) 2586616511d0STrond Myklebust return; 2587390c6843SRam Pai down_write(&namespace_sem); 258899b7db7bSNick Piggin br_write_lock(vfsmount_lock); 2589d498b25aSAl Viro umount_tree(ns->root, 0, &umount_list); 259099b7db7bSNick Piggin br_write_unlock(vfsmount_lock); 2591390c6843SRam Pai up_write(&namespace_sem); 259270fbcdf4SRam Pai release_mounts(&umount_list); 25936b3286edSKirill Korotaev kfree(ns); 25941da177e4SLinus Torvalds } 2595cf8d2c11STrond Myklebust EXPORT_SYMBOL(put_mnt_ns); 2596