1 // SPDX-License-Identifier: GPL-2.0-only 2 3 #include <linux/stat.h> 4 #include <linux/sysctl.h> 5 #include <linux/slab.h> 6 #include <linux/cred.h> 7 #include <linux/hash.h> 8 #include <linux/kmemleak.h> 9 #include <linux/user_namespace.h> 10 11 #define UCOUNTS_HASHTABLE_BITS 10 12 static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)]; 13 static DEFINE_SPINLOCK(ucounts_lock); 14 15 #define ucounts_hashfn(ns, uid) \ 16 hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \ 17 UCOUNTS_HASHTABLE_BITS) 18 #define ucounts_hashentry(ns, uid) \ 19 (ucounts_hashtable + ucounts_hashfn(ns, uid)) 20 21 22 #ifdef CONFIG_SYSCTL 23 static struct ctl_table_set * 24 set_lookup(struct ctl_table_root *root) 25 { 26 return ¤t_user_ns()->set; 27 } 28 29 static int set_is_seen(struct ctl_table_set *set) 30 { 31 return ¤t_user_ns()->set == set; 32 } 33 34 static int set_permissions(struct ctl_table_header *head, 35 struct ctl_table *table) 36 { 37 struct user_namespace *user_ns = 38 container_of(head->set, struct user_namespace, set); 39 int mode; 40 41 /* Allow users with CAP_SYS_RESOURCE unrestrained access */ 42 if (ns_capable(user_ns, CAP_SYS_RESOURCE)) 43 mode = (table->mode & S_IRWXU) >> 6; 44 else 45 /* Allow all others at most read-only access */ 46 mode = table->mode & S_IROTH; 47 return (mode << 6) | (mode << 3) | mode; 48 } 49 50 static struct ctl_table_root set_root = { 51 .lookup = set_lookup, 52 .permissions = set_permissions, 53 }; 54 55 #define UCOUNT_ENTRY(name) \ 56 { \ 57 .procname = name, \ 58 .maxlen = sizeof(int), \ 59 .mode = 0644, \ 60 .proc_handler = proc_dointvec_minmax, \ 61 .extra1 = SYSCTL_ZERO, \ 62 .extra2 = SYSCTL_INT_MAX, \ 63 } 64 static struct ctl_table user_table[] = { 65 UCOUNT_ENTRY("max_user_namespaces"), 66 UCOUNT_ENTRY("max_pid_namespaces"), 67 UCOUNT_ENTRY("max_uts_namespaces"), 68 UCOUNT_ENTRY("max_ipc_namespaces"), 69 UCOUNT_ENTRY("max_net_namespaces"), 70 UCOUNT_ENTRY("max_mnt_namespaces"), 71 UCOUNT_ENTRY("max_cgroup_namespaces"), 72 UCOUNT_ENTRY("max_time_namespaces"), 73 #ifdef CONFIG_INOTIFY_USER 74 UCOUNT_ENTRY("max_inotify_instances"), 75 UCOUNT_ENTRY("max_inotify_watches"), 76 #endif 77 #ifdef CONFIG_FANOTIFY 78 UCOUNT_ENTRY("max_fanotify_groups"), 79 UCOUNT_ENTRY("max_fanotify_marks"), 80 #endif 81 { } 82 }; 83 #endif /* CONFIG_SYSCTL */ 84 85 bool setup_userns_sysctls(struct user_namespace *ns) 86 { 87 #ifdef CONFIG_SYSCTL 88 struct ctl_table *tbl; 89 90 BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1); 91 setup_sysctl_set(&ns->set, &set_root, set_is_seen); 92 tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL); 93 if (tbl) { 94 int i; 95 for (i = 0; i < UCOUNT_COUNTS; i++) { 96 tbl[i].data = &ns->ucount_max[i]; 97 } 98 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl); 99 } 100 if (!ns->sysctls) { 101 kfree(tbl); 102 retire_sysctl_set(&ns->set); 103 return false; 104 } 105 #endif 106 return true; 107 } 108 109 void retire_userns_sysctls(struct user_namespace *ns) 110 { 111 #ifdef CONFIG_SYSCTL 112 struct ctl_table *tbl; 113 114 tbl = ns->sysctls->ctl_table_arg; 115 unregister_sysctl_table(ns->sysctls); 116 retire_sysctl_set(&ns->set); 117 kfree(tbl); 118 #endif 119 } 120 121 static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent) 122 { 123 struct ucounts *ucounts; 124 125 hlist_for_each_entry(ucounts, hashent, node) { 126 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns)) 127 return ucounts; 128 } 129 return NULL; 130 } 131 132 static struct ucounts *get_ucounts(struct user_namespace *ns, kuid_t uid) 133 { 134 struct hlist_head *hashent = ucounts_hashentry(ns, uid); 135 struct ucounts *ucounts, *new; 136 137 spin_lock_irq(&ucounts_lock); 138 ucounts = find_ucounts(ns, uid, hashent); 139 if (!ucounts) { 140 spin_unlock_irq(&ucounts_lock); 141 142 new = kzalloc(sizeof(*new), GFP_KERNEL); 143 if (!new) 144 return NULL; 145 146 new->ns = ns; 147 new->uid = uid; 148 new->count = 0; 149 150 spin_lock_irq(&ucounts_lock); 151 ucounts = find_ucounts(ns, uid, hashent); 152 if (ucounts) { 153 kfree(new); 154 } else { 155 hlist_add_head(&new->node, hashent); 156 ucounts = new; 157 } 158 } 159 if (ucounts->count == INT_MAX) 160 ucounts = NULL; 161 else 162 ucounts->count += 1; 163 spin_unlock_irq(&ucounts_lock); 164 return ucounts; 165 } 166 167 static void put_ucounts(struct ucounts *ucounts) 168 { 169 unsigned long flags; 170 171 spin_lock_irqsave(&ucounts_lock, flags); 172 ucounts->count -= 1; 173 if (!ucounts->count) 174 hlist_del_init(&ucounts->node); 175 else 176 ucounts = NULL; 177 spin_unlock_irqrestore(&ucounts_lock, flags); 178 179 kfree(ucounts); 180 } 181 182 static inline bool atomic_inc_below(atomic_t *v, int u) 183 { 184 int c, old; 185 c = atomic_read(v); 186 for (;;) { 187 if (unlikely(c >= u)) 188 return false; 189 old = atomic_cmpxchg(v, c, c+1); 190 if (likely(old == c)) 191 return true; 192 c = old; 193 } 194 } 195 196 struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid, 197 enum ucount_type type) 198 { 199 struct ucounts *ucounts, *iter, *bad; 200 struct user_namespace *tns; 201 ucounts = get_ucounts(ns, uid); 202 for (iter = ucounts; iter; iter = tns->ucounts) { 203 int max; 204 tns = iter->ns; 205 max = READ_ONCE(tns->ucount_max[type]); 206 if (!atomic_inc_below(&iter->ucount[type], max)) 207 goto fail; 208 } 209 return ucounts; 210 fail: 211 bad = iter; 212 for (iter = ucounts; iter != bad; iter = iter->ns->ucounts) 213 atomic_dec(&iter->ucount[type]); 214 215 put_ucounts(ucounts); 216 return NULL; 217 } 218 219 void dec_ucount(struct ucounts *ucounts, enum ucount_type type) 220 { 221 struct ucounts *iter; 222 for (iter = ucounts; iter; iter = iter->ns->ucounts) { 223 int dec = atomic_dec_if_positive(&iter->ucount[type]); 224 WARN_ON_ONCE(dec < 0); 225 } 226 put_ucounts(ucounts); 227 } 228 229 static __init int user_namespace_sysctl_init(void) 230 { 231 #ifdef CONFIG_SYSCTL 232 static struct ctl_table_header *user_header; 233 static struct ctl_table empty[1]; 234 /* 235 * It is necessary to register the user directory in the 236 * default set so that registrations in the child sets work 237 * properly. 238 */ 239 user_header = register_sysctl("user", empty); 240 kmemleak_ignore(user_header); 241 BUG_ON(!user_header); 242 BUG_ON(!setup_userns_sysctls(&init_user_ns)); 243 #endif 244 return 0; 245 } 246 subsys_initcall(user_namespace_sysctl_init); 247