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 /*
 243  * Create /sys/kernel/uids/<uid>/cpu_share file for this user
 244  * We do not create this file for users in a user namespace (until
 245  * sysfs tagging is implemented).
 246  *
 247  * See Documentation/scheduler/sched-design-CFS.txt for ramifications.
 248  */
 249 static int uids_user_create(struct user_struct *up)
 250 {
 251         struct kobject *kobj = &up->kobj;
 252         int error;
 253
 254         memset(kobj, 0, sizeof(struct kobject));
 255         if (up->user_ns != &init_user_ns)
 256                 return 0;
 257         kobj->kset = uids_kset;
 258         error = kobject_init_and_add(kobj, &uids_ktype, NULL, "%d", up->uid);
 259         if (error) {
 260                 kobject_put(kobj);
 261                 goto done;
 262         }
 263
 264         kobject_uevent(kobj, KOBJ_ADD);
 265 done:
 266         return error;
 267 }
 268
 269 /* create these entries in sysfs:
 270  *      "/sys/kernel/uids" directory
 271  *      "/sys/kernel/uids/0" directory (for root user)
 272  *      "/sys/kernel/uids/0/cpu_share" file (for root user)
 273  */
 274 int __init uids_sysfs_init(void)
 275 {
 276         uids_kset = kset_create_and_add("uids", NULL, kernel_kobj);
 277         if (!uids_kset)
 278                 return -ENOMEM;
 279
 280         return uids_user_create(&root_user);
 281 }
 282
 283 /* work function to remove sysfs directory for a user and free up
 284  * corresponding structures.
 285  */
 286 static void remove_user_sysfs_dir(struct work_struct *w)
 287 {
 288         struct user_struct *up = container_of(w, struct user_struct, work);
 289         unsigned long flags;
 290         int remove_user = 0;
 291
 292         if (up->user_ns != &init_user_ns)
 293                 return;
 294         /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
 295          * atomic.
 296          */
 297         uids_mutex_lock();
 298
 299         local_irq_save(flags);
 300
 301         if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
 302                 uid_hash_remove(up);
 303                 remove_user = 1;
 304                 spin_unlock_irqrestore(&uidhash_lock, flags);
 305         } else {
 306                 local_irq_restore(flags);
 307         }
 308
 309         if (!remove_user)
 310                 goto done;
 311
 312         kobject_uevent(&up->kobj, KOBJ_REMOVE);
 313         kobject_del(&up->kobj);
 314         kobject_put(&up->kobj);
 315
 316         sched_destroy_user(up);
 317         key_put(up->uid_keyring);
 318         key_put(up->session_keyring);
 319         kmem_cache_free(uid_cachep, up);
 320
 321 done:
 322         uids_mutex_unlock();
 323 }
 324
 325 /* IRQs are disabled and uidhash_lock is held upon function entry.
 326  * IRQ state (as stored in flags) is restored and uidhash_lock released
 327  * upon function exit.
 328  */
 329 static void free_user(struct user_struct *up, unsigned long flags)
 330 {
 331         /* restore back the count */
 332         atomic_inc(&up->__count);
 333         spin_unlock_irqrestore(&uidhash_lock, flags);
 334
 335         put_user_ns(up->user_ns);
 336         INIT_WORK(&up->work, remove_user_sysfs_dir);
 337         schedule_work(&up->work);
 338 }
 339
 340 #else   /* CONFIG_USER_SCHED && CONFIG_SYSFS */
 341
 342 int uids_sysfs_init(void) { return 0; }
 343 static inline int uids_user_create(struct user_struct *up) { return 0; }
 344 static inline void uids_mutex_lock(void) { }
 345 static inline void uids_mutex_unlock(void) { }
 346
 347 /* IRQs are disabled and uidhash_lock is held upon function entry.
 348  * IRQ state (as stored in flags) is restored and uidhash_lock released
 349  * upon function exit.
 350  */
 351 static void free_user(struct user_struct *up, unsigned long flags)
 352 {
 353         uid_hash_remove(up);
 354         spin_unlock_irqrestore(&uidhash_lock, flags);
 355         sched_destroy_user(up);
 356         key_put(up->uid_keyring);
 357         key_put(up->session_keyring);
 358         put_user_ns(up->user_ns);
 359         kmem_cache_free(uid_cachep, up);
 360 }
 361
 362 #endif
 363
 364 /*
 365  * Locate the user_struct for the passed UID.  If found, take a ref on it.  The
 366  * caller must undo that ref with free_uid().
 367  *
 368  * If the user_struct could not be found, return NULL.
 369  */
 370 struct user_struct *find_user(uid_t uid)
 371 {
 372         struct user_struct *ret;
 373         unsigned long flags;
 374         struct user_namespace *ns = current_user_ns();
 375
 376         spin_lock_irqsave(&uidhash_lock, flags);
 377         ret = uid_hash_find(uid, uidhashentry(ns, uid));
 378         spin_unlock_irqrestore(&uidhash_lock, flags);
 379         return ret;
 380 }
 381
 382 void free_uid(struct user_struct *up)
 383 {
 384         unsigned long flags;
 385
 386         if (!up)
 387                 return;
 388
 389         local_irq_save(flags);
 390         if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
 391                 free_user(up, flags);
 392         else
 393                 local_irq_restore(flags);
 394 }
 395
 396 struct user_struct *alloc_uid(struct user_namespace *ns, uid_t uid)
 397 {
 398         struct hlist_head *hashent = uidhashentry(ns, uid);
 399         struct user_struct *up, *new;
 400
 401         /* Make uid_hash_find() + uids_user_create() + uid_hash_insert()
 402          * atomic.
 403          */
 404         uids_mutex_lock();
 405
 406         spin_lock_irq(&uidhash_lock);
 407         up = uid_hash_find(uid, hashent);
 408         spin_unlock_irq(&uidhash_lock);
 409
 410         if (!up) {
 411                 new = kmem_cache_zalloc(uid_cachep, GFP_KERNEL);
 412                 if (!new)
 413                         goto out_unlock;
 414
 415                 new->uid = uid;
 416                 atomic_set(&new->__count, 1);
 417
 418                 if (sched_create_user(new) < 0)
 419                         goto out_free_user;
 420
 421                 new->user_ns = get_user_ns(ns);
 422
 423                 if (uids_user_create(new))
 424                         goto out_destoy_sched;
 425
 426                 /*
 427                  * Before adding this, check whether we raced
 428                  * on adding the same user already..
 429                  */
 430                 spin_lock_irq(&uidhash_lock);
 431                 up = uid_hash_find(uid, hashent);
 432                 if (up) {
 433                         /* This case is not possible when CONFIG_USER_SCHED
 434                          * is defined, since we serialize alloc_uid() using
 435                          * uids_mutex. Hence no need to call
 436                          * sched_destroy_user() or remove_user_sysfs_dir().
 437                          */
 438                         key_put(new->uid_keyring);
 439                         key_put(new->session_keyring);
 440                         kmem_cache_free(uid_cachep, new);
 441                 } else {
 442                         uid_hash_insert(new, hashent);
 443                         up = new;
 444                 }
 445                 spin_unlock_irq(&uidhash_lock);
 446         }
 447
 448         uids_mutex_unlock();
 449
 450         return up;
 451
 452 out_destoy_sched:
 453         sched_destroy_user(new);
 454         put_user_ns(new->user_ns);
 455 out_free_user:
 456         kmem_cache_free(uid_cachep, new);
 457 out_unlock:
 458         uids_mutex_unlock();
 459         return NULL;
 460 }
 461
 462 static int __init uid_cache_init(void)
 463 {
 464         int n;
 465
 466         uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
 467                         0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
 468
 469         for(n = 0; n < UIDHASH_SZ; ++n)
 470                 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
 471
 472         /* Insert the root user immediately (init already runs as root) */
 473         spin_lock_irq(&uidhash_lock);
 474         uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0));
 475         spin_unlock_irq(&uidhash_lock);
 476
 477         return 0;
 478 }
 479
 480 module_init(uid_cache_init);