kernel/user.c

   1 /*
   2  * The "user cache".
   3  *
   4  * (C) Copyright 1991-2000 Linus Torvalds
   5  *
   6  * We have a per-user structure to keep track of how many
   7  * processes, files etc the user has claimed, in order to be
   8  * able to have per-user limits for system resources.
   9  */
  10
  11 #include <linux/init.h>
  12 #include <linux/sched.h>
  13 #include <linux/slab.h>
  14 #include <linux/bitops.h>
  15 #include <linux/key.h>
  16 #include <linux/interrupt.h>
  17 #include <linux/module.h>
  18 #include <linux/user_namespace.h>
  19 #include "cred-internals.h"
  20
  21 struct user_namespace init_user_ns = {
  22         .kref = {
  23                 .refcount       = ATOMIC_INIT(1),
  24         },
  25         .creator = &root_user,
  26 };
  27 EXPORT_SYMBOL_GPL(init_user_ns);
  28
  29 /*
  30  * UID task count cache, to get fast user lookup in "alloc_uid"
  31  * when changing user ID's (ie setuid() and friends).
  32  */
  33
  34 #define UIDHASH_MASK            (UIDHASH_SZ - 1)
  35 #define __uidhashfn(uid)        (((uid >> UIDHASH_BITS) + uid) & UIDHASH_MASK)
  36 #define uidhashentry(ns, uid)   ((ns)->uidhash_table + __uidhashfn((uid)))
  37
  38 static struct kmem_cache *uid_cachep;
  39
  40 /*
  41  * The uidhash_lock is mostly taken from process context, but it is
  42  * occasionally also taken from softirq/tasklet context, when
  43  * task-structs get RCU-freed. Hence all locking must be softirq-safe.
  44  * But free_uid() is also called with local interrupts disabled, and running
  45  * local_bh_enable() with local interrupts disabled is an error - we'll run
  46  * softirq callbacks, and they can unconditionally enable interrupts, and
  47  * the caller of free_uid() didn't expect that..
  48  */
  49 static DEFINE_SPINLOCK(uidhash_lock);
  50
  51 /* root_user.__count is 2, 1 for init task cred, 1 for init_user_ns->creator */
  52 struct user_struct root_user = {
  53         .__count        = ATOMIC_INIT(2),
  54         .processes      = ATOMIC_INIT(1),
  55         .files          = ATOMIC_INIT(0),
  56         .sigpending     = ATOMIC_INIT(0),
  57         .locked_shm     = 0,
  58         .user_ns        = &init_user_ns,
  59 #ifdef CONFIG_USER_SCHED
  60         .tg             = &init_task_group,
  61 #endif
  62 };
  63
  64 /*
  65  * These routines must be called with the uidhash spinlock held!
  66  */
  67 static void uid_hash_insert(struct user_struct *up, struct hlist_head *hashent)
  68 {
  69         hlist_add_head(&up->uidhash_node, hashent);
  70 }
  71
  72 static void uid_hash_remove(struct user_struct *up)
  73 {
  74         hlist_del_init(&up->uidhash_node);
  75 }
  76
  77 static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent)
  78 {
  79         struct user_struct *user;
  80         struct hlist_node *h;
  81
  82         hlist_for_each_entry(user, h, hashent, uidhash_node) {
  83                 if (user->uid == uid) {
  84                         atomic_inc(&user->__count);
  85                         return user;
  86                 }
  87         }
  88
  89         return NULL;
  90 }
  91
  92 #ifdef CONFIG_USER_SCHED
  93
  94 static void sched_destroy_user(struct user_struct *up)
  95 {
  96         sched_destroy_group(up->tg);
  97 }
  98
  99 static int sched_create_user(struct user_struct *up)
 100 {
 101         int rc = 0;
 102
 103         up->tg = sched_create_group(&root_task_group);
 104         if (IS_ERR(up->tg))
 105                 rc = -ENOMEM;
 106
 107         return rc;
 108 }
 109
 110 #else   /* CONFIG_USER_SCHED */
 111
 112 static void sched_destroy_user(struct user_struct *up) { }
 113 static int sched_create_user(struct user_struct *up) { return 0; }
 114
 115 #endif  /* CONFIG_USER_SCHED */
 116
 117 #if defined(CONFIG_USER_SCHED) && defined(CONFIG_SYSFS)
 118
 119 static struct kset *uids_kset; /* represents the /sys/kernel/uids/ directory */
 120 static DEFINE_MUTEX(uids_mutex);
 121
 122 static inline void uids_mutex_lock(void)
 123 {
 124         mutex_lock(&uids_mutex);
 125 }
 126
 127 static inline void uids_mutex_unlock(void)
 128 {
 129         mutex_unlock(&uids_mutex);
 130 }
 131
 132 /* uid directory attributes */
 133 #ifdef CONFIG_FAIR_GROUP_SCHED
 134 static ssize_t cpu_shares_show(struct kobject *kobj,
 135                                struct kobj_attribute *attr,
 136                                char *buf)
 137 {
 138         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 139
 140         return sprintf(buf, "%lu\n", sched_group_shares(up->tg));
 141 }
 142
 143 static ssize_t cpu_shares_store(struct kobject *kobj,
 144                                 struct kobj_attribute *attr,
 145                                 const char *buf, size_t size)
 146 {
 147         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 148         unsigned long shares;
 149         int rc;
 150
 151         sscanf(buf, "%lu", &shares);
 152
 153         rc = sched_group_set_shares(up->tg, shares);
 154
 155         return (rc ? rc : size);
 156 }
 157
 158 static struct kobj_attribute cpu_share_attr =
 159         __ATTR(cpu_share, 0644, cpu_shares_show, cpu_shares_store);
 160 #endif
 161
 162 #ifdef CONFIG_RT_GROUP_SCHED
 163 static ssize_t cpu_rt_runtime_show(struct kobject *kobj,
 164                                    struct kobj_attribute *attr,
 165                                    char *buf)
 166 {
 167         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 168
 169         return sprintf(buf, "%ld\n", sched_group_rt_runtime(up->tg));
 170 }
 171
 172 static ssize_t cpu_rt_runtime_store(struct kobject *kobj,
 173                                     struct kobj_attribute *attr,
 174                                     const char *buf, size_t size)
 175 {
 176         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 177         unsigned long rt_runtime;
 178         int rc;
 179
 180         sscanf(buf, "%ld", &rt_runtime);
 181
 182         rc = sched_group_set_rt_runtime(up->tg, rt_runtime);
 183
 184         return (rc ? rc : size);
 185 }
 186
 187 static struct kobj_attribute cpu_rt_runtime_attr =
 188         __ATTR(cpu_rt_runtime, 0644, cpu_rt_runtime_show, cpu_rt_runtime_store);
 189
 190 static ssize_t cpu_rt_period_show(struct kobject *kobj,
 191                                    struct kobj_attribute *attr,
 192                                    char *buf)
 193 {
 194         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 195
 196         return sprintf(buf, "%lu\n", sched_group_rt_period(up->tg));
 197 }
 198
 199 static ssize_t cpu_rt_period_store(struct kobject *kobj,
 200                                     struct kobj_attribute *attr,
 201                                     const char *buf, size_t size)
 202 {
 203         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 204         unsigned long rt_period;
 205         int rc;
 206
 207         sscanf(buf, "%lu", &rt_period);
 208
 209         rc = sched_group_set_rt_period(up->tg, rt_period);
 210
 211         return (rc ? rc : size);
 212 }
 213
 214 static struct kobj_attribute cpu_rt_period_attr =
 215         __ATTR(cpu_rt_period, 0644, cpu_rt_period_show, cpu_rt_period_store);
 216 #endif
 217
 218 /* default attributes per uid directory */
 219 static struct attribute *uids_attributes[] = {
 220 #ifdef CONFIG_FAIR_GROUP_SCHED
 221         &cpu_share_attr.attr,
 222 #endif
 223 #ifdef CONFIG_RT_GROUP_SCHED
 224         &cpu_rt_runtime_attr.attr,
 225         &cpu_rt_period_attr.attr,
 226 #endif
 227         NULL
 228 };
 229
 230 /* the lifetime of user_struct is not managed by the core (now) */
 231 static void uids_release(struct kobject *kobj)
 232 {
 233         return;
 234 }
 235
 236 static struct kobj_type uids_ktype = {
 237         .sysfs_ops = &kobj_sysfs_ops,
 238         .default_attrs = uids_attributes,
 239         .release = uids_release,
 240 };
 241
 242 /* create /sys/kernel/uids/<uid>/cpu_share file for this user */
 243 static int uids_user_create(struct user_struct *up)
 244 {
 245         struct kobject *kobj = &up->kobj;
 246         int error;
 247
 248         memset(kobj, 0, sizeof(struct kobject));
 249         kobj->kset = uids_kset;
 250         error = kobject_init_and_add(kobj, &uids_ktype, NULL, "%d", up->uid);
 251         if (error) {
 252                 kobject_put(kobj);
 253                 goto done;
 254         }
 255
 256         kobject_uevent(kobj, KOBJ_ADD);
 257 done:
 258         return error;
 259 }
 260
 261 /* create these entries in sysfs:
 262  *      "/sys/kernel/uids" directory
 263  *      "/sys/kernel/uids/0" directory (for root user)
 264  *      "/sys/kernel/uids/0/cpu_share" file (for root user)
 265  */
 266 int __init uids_sysfs_init(void)
 267 {
 268         uids_kset = kset_create_and_add("uids", NULL, kernel_kobj);
 269         if (!uids_kset)
 270                 return -ENOMEM;
 271
 272         return uids_user_create(&root_user);
 273 }
 274
 275 /* work function to remove sysfs directory for a user and free up
 276  * corresponding structures.
 277  */
 278 static void remove_user_sysfs_dir(struct work_struct *w)
 279 {
 280         struct user_struct *up = container_of(w, struct user_struct, work);
 281         unsigned long flags;
 282         int remove_user = 0;
 283
 284         /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
 285          * atomic.
 286          */
 287         uids_mutex_lock();
 288
 289         local_irq_save(flags);
 290
 291         if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
 292                 uid_hash_remove(up);
 293                 remove_user = 1;
 294                 spin_unlock_irqrestore(&uidhash_lock, flags);
 295         } else {
 296                 local_irq_restore(flags);
 297         }
 298
 299         if (!remove_user)
 300                 goto done;
 301
 302         kobject_uevent(&up->kobj, KOBJ_REMOVE);
 303         kobject_del(&up->kobj);
 304         kobject_put(&up->kobj);
 305
 306         sched_destroy_user(up);
 307         key_put(up->uid_keyring);
 308         key_put(up->session_keyring);
 309         kmem_cache_free(uid_cachep, up);
 310
 311 done:
 312         uids_mutex_unlock();
 313 }
 314
 315 /* IRQs are disabled and uidhash_lock is held upon function entry.
 316  * IRQ state (as stored in flags) is restored and uidhash_lock released
 317  * upon function exit.
 318  */
 319 static void free_user(struct user_struct *up, unsigned long flags)
 320 {
 321         /* restore back the count */
 322         atomic_inc(&up->__count);
 323         spin_unlock_irqrestore(&uidhash_lock, flags);
 324
 325         put_user_ns(up->user_ns);
 326         INIT_WORK(&up->work, remove_user_sysfs_dir);
 327         schedule_work(&up->work);
 328 }
 329
 330 #else   /* CONFIG_USER_SCHED && CONFIG_SYSFS */
 331
 332 int uids_sysfs_init(void) { return 0; }
 333 static inline int uids_user_create(struct user_struct *up) { return 0; }
 334 static inline void uids_mutex_lock(void) { }
 335 static inline void uids_mutex_unlock(void) { }
 336
 337 /* IRQs are disabled and uidhash_lock is held upon function entry.
 338  * IRQ state (as stored in flags) is restored and uidhash_lock released
 339  * upon function exit.
 340  */
 341 static void free_user(struct user_struct *up, unsigned long flags)
 342 {
 343         uid_hash_remove(up);
 344         spin_unlock_irqrestore(&uidhash_lock, flags);
 345         sched_destroy_user(up);
 346         key_put(up->uid_keyring);
 347         key_put(up->session_keyring);
 348         put_user_ns(up->user_ns);
 349         kmem_cache_free(uid_cachep, up);
 350 }
 351
 352 #endif
 353
 354 /*
 355  * Locate the user_struct for the passed UID.  If found, take a ref on it.  The
 356  * caller must undo that ref with free_uid().
 357  *
 358  * If the user_struct could not be found, return NULL.
 359  */
 360 struct user_struct *find_user(uid_t uid)
 361 {
 362         struct user_struct *ret;
 363         unsigned long flags;
 364         struct user_namespace *ns = current_user_ns();
 365
 366         spin_lock_irqsave(&uidhash_lock, flags);
 367         ret = uid_hash_find(uid, uidhashentry(ns, uid));
 368         spin_unlock_irqrestore(&uidhash_lock, flags);
 369         return ret;
 370 }
 371
 372 void free_uid(struct user_struct *up)
 373 {
 374         unsigned long flags;
 375
 376         if (!up)
 377                 return;
 378
 379         local_irq_save(flags);
 380         if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
 381                 free_user(up, flags);
 382         else
 383                 local_irq_restore(flags);
 384 }
 385
 386 struct user_struct *alloc_uid(struct user_namespace *ns, uid_t uid)
 387 {
 388         struct hlist_head *hashent = uidhashentry(ns, uid);
 389         struct user_struct *up, *new;
 390
 391         /* Make uid_hash_find() + uids_user_create() + uid_hash_insert()
 392          * atomic.
 393          */
 394         uids_mutex_lock();
 395
 396         spin_lock_irq(&uidhash_lock);
 397         up = uid_hash_find(uid, hashent);
 398         spin_unlock_irq(&uidhash_lock);
 399
 400         if (!up) {
 401                 new = kmem_cache_zalloc(uid_cachep, GFP_KERNEL);
 402                 if (!new)
 403                         goto out_unlock;
 404
 405                 new->uid = uid;
 406                 atomic_set(&new->__count, 1);
 407
 408                 if (sched_create_user(new) < 0)
 409                         goto out_free_user;
 410
 411                 new->user_ns = get_user_ns(ns);
 412
 413                 if (uids_user_create(new))
 414                         goto out_destoy_sched;
 415
 416                 /*
 417                  * Before adding this, check whether we raced
 418                  * on adding the same user already..
 419                  */
 420                 spin_lock_irq(&uidhash_lock);
 421                 up = uid_hash_find(uid, hashent);
 422                 if (up) {
 423                         /* This case is not possible when CONFIG_USER_SCHED
 424                          * is defined, since we serialize alloc_uid() using
 425                          * uids_mutex. Hence no need to call
 426                          * sched_destroy_user() or remove_user_sysfs_dir().
 427                          */
 428                         key_put(new->uid_keyring);
 429                         key_put(new->session_keyring);
 430                         kmem_cache_free(uid_cachep, new);
 431                 } else {
 432                         uid_hash_insert(new, hashent);
 433                         up = new;
 434                 }
 435                 spin_unlock_irq(&uidhash_lock);
 436         }
 437
 438         uids_mutex_unlock();
 439
 440         return up;
 441
 442 out_destoy_sched:
 443         sched_destroy_user(new);
 444         put_user_ns(new->user_ns);
 445 out_free_user:
 446         kmem_cache_free(uid_cachep, new);
 447 out_unlock:
 448         uids_mutex_unlock();
 449         return NULL;
 450 }
 451
 452 static int __init uid_cache_init(void)
 453 {
 454         int n;
 455
 456         uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
 457                         0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
 458
 459         for(n = 0; n < UIDHASH_SZ; ++n)
 460                 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
 461
 462         /* Insert the root user immediately (init already runs as root) */
 463         spin_lock_irq(&uidhash_lock);
 464         uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0));
 465         spin_unlock_irq(&uidhash_lock);
 466
 467         return 0;
 468 }
 469
 470 module_init(uid_cache_init);