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
  20 /*
  21  * UID task count cache, to get fast user lookup in "alloc_uid"
  22  * when changing user ID's (ie setuid() and friends).
  23  */
  24
  25 #define UIDHASH_MASK            (UIDHASH_SZ - 1)
  26 #define __uidhashfn(uid)        (((uid >> UIDHASH_BITS) + uid) & UIDHASH_MASK)
  27 #define uidhashentry(ns, uid)   ((ns)->uidhash_table + __uidhashfn((uid)))
  28
  29 static struct kmem_cache *uid_cachep;
  30
  31 /*
  32  * The uidhash_lock is mostly taken from process context, but it is
  33  * occasionally also taken from softirq/tasklet context, when
  34  * task-structs get RCU-freed. Hence all locking must be softirq-safe.
  35  * But free_uid() is also called with local interrupts disabled, and running
  36  * local_bh_enable() with local interrupts disabled is an error - we'll run
  37  * softirq callbacks, and they can unconditionally enable interrupts, and
  38  * the caller of free_uid() didn't expect that..
  39  */
  40 static DEFINE_SPINLOCK(uidhash_lock);
  41
  42 struct user_struct root_user = {
  43         .__count        = ATOMIC_INIT(1),
  44         .processes      = ATOMIC_INIT(1),
  45         .files          = ATOMIC_INIT(0),
  46         .sigpending     = ATOMIC_INIT(0),
  47         .mq_bytes       = 0,
  48         .locked_shm     = 0,
  49 #ifdef CONFIG_KEYS
  50         .uid_keyring    = &root_user_keyring,
  51         .session_keyring = &root_session_keyring,
  52 #endif
  53 #ifdef CONFIG_FAIR_USER_SCHED
  54         .tg             = &init_task_group,
  55 #endif
  56 };
  57
  58 /*
  59  * These routines must be called with the uidhash spinlock held!
  60  */
  61 static inline void uid_hash_insert(struct user_struct *up,
  62                                                 struct hlist_head *hashent)
  63 {
  64         hlist_add_head(&up->uidhash_node, hashent);
  65 }
  66
  67 static inline void uid_hash_remove(struct user_struct *up)
  68 {
  69         hlist_del_init(&up->uidhash_node);
  70 }
  71
  72 static inline struct user_struct *uid_hash_find(uid_t uid,
  73                                                 struct hlist_head *hashent)
  74 {
  75         struct user_struct *user;
  76         struct hlist_node *h;
  77
  78         hlist_for_each_entry(user, h, hashent, uidhash_node) {
  79                 if (user->uid == uid) {
  80                         atomic_inc(&user->__count);
  81                         return user;
  82                 }
  83         }
  84
  85         return NULL;
  86 }
  87
  88 #ifdef CONFIG_FAIR_USER_SCHED
  89
  90 static void sched_destroy_user(struct user_struct *up)
  91 {
  92         sched_destroy_group(up->tg);
  93 }
  94
  95 static int sched_create_user(struct user_struct *up)
  96 {
  97         int rc = 0;
  98
  99         up->tg = sched_create_group();
 100         if (IS_ERR(up->tg))
 101                 rc = -ENOMEM;
 102
 103         return rc;
 104 }
 105
 106 static void sched_switch_user(struct task_struct *p)
 107 {
 108         sched_move_task(p);
 109 }
 110
 111 #else   /* CONFIG_FAIR_USER_SCHED */
 112
 113 static void sched_destroy_user(struct user_struct *up) { }
 114 static int sched_create_user(struct user_struct *up) { return 0; }
 115 static void sched_switch_user(struct task_struct *p) { }
 116
 117 #endif  /* CONFIG_FAIR_USER_SCHED */
 118
 119 #if defined(CONFIG_FAIR_USER_SCHED) && defined(CONFIG_SYSFS)
 120
 121 static struct kobject uids_kobject; /* represents /sys/kernel/uids directory */
 122 static DEFINE_MUTEX(uids_mutex);
 123
 124 static inline void uids_mutex_lock(void)
 125 {
 126         mutex_lock(&uids_mutex);
 127 }
 128
 129 static inline void uids_mutex_unlock(void)
 130 {
 131         mutex_unlock(&uids_mutex);
 132 }
 133
 134 /* return cpu shares held by the user */
 135 ssize_t cpu_shares_show(struct kset *kset, char *buffer)
 136 {
 137         struct user_struct *up = container_of(kset, struct user_struct, kset);
 138
 139         return sprintf(buffer, "%lu\n", sched_group_shares(up->tg));
 140 }
 141
 142 /* modify cpu shares held by the user */
 143 ssize_t cpu_shares_store(struct kset *kset, const char *buffer, size_t size)
 144 {
 145         struct user_struct *up = container_of(kset, struct user_struct, kset);
 146         unsigned long shares;
 147         int rc;
 148
 149         sscanf(buffer, "%lu", &shares);
 150
 151         rc = sched_group_set_shares(up->tg, shares);
 152
 153         return (rc ? rc : size);
 154 }
 155
 156 static void user_attr_init(struct subsys_attribute *sa, char *name, int mode)
 157 {
 158         sa->attr.name = name;
 159         sa->attr.mode = mode;
 160         sa->show = cpu_shares_show;
 161         sa->store = cpu_shares_store;
 162 }
 163
 164 /* Create "/sys/kernel/uids/<uid>" directory and
 165  *  "/sys/kernel/uids/<uid>/cpu_share" file for this user.
 166  */
 167 static int user_kobject_create(struct user_struct *up)
 168 {
 169         struct kset *kset = &up->kset;
 170         struct kobject *kobj = &kset->kobj;
 171         int error;
 172
 173         memset(kset, 0, sizeof(struct kset));
 174         kobj->parent = &uids_kobject;   /* create under /sys/kernel/uids dir */
 175         kobject_set_name(kobj, "%d", up->uid);
 176         kset_init(kset);
 177         user_attr_init(&up->user_attr, "cpu_share", 0644);
 178
 179         error = kobject_add(kobj);
 180         if (error)
 181                 goto done;
 182
 183         error = sysfs_create_file(kobj, &up->user_attr.attr);
 184         if (error)
 185                 kobject_del(kobj);
 186
 187         kobject_uevent(kobj, KOBJ_ADD);
 188
 189 done:
 190         return error;
 191 }
 192
 193 /* create these in sysfs filesystem:
 194  *      "/sys/kernel/uids" directory
 195  *      "/sys/kernel/uids/0" directory (for root user)
 196  *      "/sys/kernel/uids/0/cpu_share" file (for root user)
 197  */
 198 int __init uids_kobject_init(void)
 199 {
 200         int error;
 201
 202         /* create under /sys/kernel dir */
 203         uids_kobject.parent = &kernel_subsys.kobj;
 204         uids_kobject.kset = &kernel_subsys;
 205         kobject_set_name(&uids_kobject, "uids");
 206         kobject_init(&uids_kobject);
 207
 208         error = kobject_add(&uids_kobject);
 209         if (!error)
 210                 error = user_kobject_create(&root_user);
 211
 212         return error;
 213 }
 214
 215 /* work function to remove sysfs directory for a user and free up
 216  * corresponding structures.
 217  */
 218 static void remove_user_sysfs_dir(struct work_struct *w)
 219 {
 220         struct user_struct *up = container_of(w, struct user_struct, work);
 221         struct kobject *kobj = &up->kset.kobj;
 222         unsigned long flags;
 223         int remove_user = 0;
 224
 225         /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
 226          * atomic.
 227          */
 228         uids_mutex_lock();
 229
 230         local_irq_save(flags);
 231
 232         if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
 233                 uid_hash_remove(up);
 234                 remove_user = 1;
 235                 spin_unlock_irqrestore(&uidhash_lock, flags);
 236         } else {
 237                 local_irq_restore(flags);
 238         }
 239
 240         if (!remove_user)
 241                 goto done;
 242
 243         sysfs_remove_file(kobj, &up->user_attr.attr);
 244         kobject_uevent(kobj, KOBJ_REMOVE);
 245         kobject_del(kobj);
 246
 247         sched_destroy_user(up);
 248         key_put(up->uid_keyring);
 249         key_put(up->session_keyring);
 250         kmem_cache_free(uid_cachep, up);
 251
 252 done:
 253         uids_mutex_unlock();
 254 }
 255
 256 /* IRQs are disabled and uidhash_lock is held upon function entry.
 257  * IRQ state (as stored in flags) is restored and uidhash_lock released
 258  * upon function exit.
 259  */
 260 static inline void free_user(struct user_struct *up, unsigned long flags)
 261 {
 262         /* restore back the count */
 263         atomic_inc(&up->__count);
 264         spin_unlock_irqrestore(&uidhash_lock, flags);
 265
 266         INIT_WORK(&up->work, remove_user_sysfs_dir);
 267         schedule_work(&up->work);
 268 }
 269
 270 #else   /* CONFIG_FAIR_USER_SCHED && CONFIG_SYSFS */
 271
 272 static inline int user_kobject_create(struct user_struct *up) { return 0; }
 273 static inline void uids_mutex_lock(void) { }
 274 static inline void uids_mutex_unlock(void) { }
 275
 276 /* IRQs are disabled and uidhash_lock is held upon function entry.
 277  * IRQ state (as stored in flags) is restored and uidhash_lock released
 278  * upon function exit.
 279  */
 280 static inline void free_user(struct user_struct *up, unsigned long flags)
 281 {
 282         uid_hash_remove(up);
 283         spin_unlock_irqrestore(&uidhash_lock, flags);
 284         sched_destroy_user(up);
 285         key_put(up->uid_keyring);
 286         key_put(up->session_keyring);
 287         kmem_cache_free(uid_cachep, up);
 288 }
 289
 290 #endif
 291
 292 /*
 293  * Locate the user_struct for the passed UID.  If found, take a ref on it.  The
 294  * caller must undo that ref with free_uid().
 295  *
 296  * If the user_struct could not be found, return NULL.
 297  */
 298 struct user_struct *find_user(uid_t uid)
 299 {
 300         struct user_struct *ret;
 301         unsigned long flags;
 302         struct user_namespace *ns = current->nsproxy->user_ns;
 303
 304         spin_lock_irqsave(&uidhash_lock, flags);
 305         ret = uid_hash_find(uid, uidhashentry(ns, uid));
 306         spin_unlock_irqrestore(&uidhash_lock, flags);
 307         return ret;
 308 }
 309
 310 void free_uid(struct user_struct *up)
 311 {
 312         unsigned long flags;
 313
 314         if (!up)
 315                 return;
 316
 317         local_irq_save(flags);
 318         if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
 319                 free_user(up, flags);
 320         else
 321                 local_irq_restore(flags);
 322 }
 323
 324 struct user_struct * alloc_uid(struct user_namespace *ns, uid_t uid)
 325 {
 326         struct hlist_head *hashent = uidhashentry(ns, uid);
 327         struct user_struct *up;
 328
 329         /* Make uid_hash_find() + user_kobject_create() + uid_hash_insert()
 330          * atomic.
 331          */
 332         uids_mutex_lock();
 333
 334         spin_lock_irq(&uidhash_lock);
 335         up = uid_hash_find(uid, hashent);
 336         spin_unlock_irq(&uidhash_lock);
 337
 338         if (!up) {
 339                 struct user_struct *new;
 340
 341                 new = kmem_cache_alloc(uid_cachep, GFP_KERNEL);
 342                 if (!new)
 343                         return NULL;
 344                 new->uid = uid;
 345                 atomic_set(&new->__count, 1);
 346                 atomic_set(&new->processes, 0);
 347                 atomic_set(&new->files, 0);
 348                 atomic_set(&new->sigpending, 0);
 349 #ifdef CONFIG_INOTIFY_USER
 350                 atomic_set(&new->inotify_watches, 0);
 351                 atomic_set(&new->inotify_devs, 0);
 352 #endif
 353
 354                 new->mq_bytes = 0;
 355                 new->locked_shm = 0;
 356
 357                 if (alloc_uid_keyring(new, current) < 0) {
 358                         kmem_cache_free(uid_cachep, new);
 359                         return NULL;
 360                 }
 361
 362                 if (sched_create_user(new) < 0) {
 363                         key_put(new->uid_keyring);
 364                         key_put(new->session_keyring);
 365                         kmem_cache_free(uid_cachep, new);
 366                         return NULL;
 367                 }
 368
 369                 if (user_kobject_create(new)) {
 370                         sched_destroy_user(new);
 371                         key_put(new->uid_keyring);
 372                         key_put(new->session_keyring);
 373                         kmem_cache_free(uid_cachep, new);
 374                         uids_mutex_unlock();
 375                         return NULL;
 376                 }
 377
 378                 /*
 379                  * Before adding this, check whether we raced
 380                  * on adding the same user already..
 381                  */
 382                 spin_lock_irq(&uidhash_lock);
 383                 up = uid_hash_find(uid, hashent);
 384                 if (up) {
 385                         /* This case is not possible when CONFIG_FAIR_USER_SCHED
 386                          * is defined, since we serialize alloc_uid() using
 387                          * uids_mutex. Hence no need to call
 388                          * sched_destroy_user() or remove_user_sysfs_dir().
 389                          */
 390                         key_put(new->uid_keyring);
 391                         key_put(new->session_keyring);
 392                         kmem_cache_free(uid_cachep, new);
 393                 } else {
 394                         uid_hash_insert(new, hashent);
 395                         up = new;
 396                 }
 397                 spin_unlock_irq(&uidhash_lock);
 398
 399         }
 400
 401         uids_mutex_unlock();
 402
 403         return up;
 404 }
 405
 406 void switch_uid(struct user_struct *new_user)
 407 {
 408         struct user_struct *old_user;
 409
 410         /* What if a process setreuid()'s and this brings the
 411          * new uid over his NPROC rlimit?  We can check this now
 412          * cheaply with the new uid cache, so if it matters
 413          * we should be checking for it.  -DaveM
 414          */
 415         old_user = current->user;
 416         atomic_inc(&new_user->processes);
 417         atomic_dec(&old_user->processes);
 418         switch_uid_keyring(new_user);
 419         current->user = new_user;
 420         sched_switch_user(current);
 421
 422         /*
 423          * We need to synchronize with __sigqueue_alloc()
 424          * doing a get_uid(p->user).. If that saw the old
 425          * user value, we need to wait until it has exited
 426          * its critical region before we can free the old
 427          * structure.
 428          */
 429         smp_mb();
 430         spin_unlock_wait(&current->sighand->siglock);
 431
 432         free_uid(old_user);
 433         suid_keys(current);
 434 }
 435
 436 void release_uids(struct user_namespace *ns)
 437 {
 438         int i;
 439         unsigned long flags;
 440         struct hlist_head *head;
 441         struct hlist_node *nd;
 442
 443         spin_lock_irqsave(&uidhash_lock, flags);
 444         /*
 445          * collapse the chains so that the user_struct-s will
 446          * be still alive, but not in hashes. subsequent free_uid()
 447          * will free them.
 448          */
 449         for (i = 0; i < UIDHASH_SZ; i++) {
 450                 head = ns->uidhash_table + i;
 451                 while (!hlist_empty(head)) {
 452                         nd = head->first;
 453                         hlist_del_init(nd);
 454                 }
 455         }
 456         spin_unlock_irqrestore(&uidhash_lock, flags);
 457
 458         free_uid(ns->root_user);
 459 }
 460
 461 static int __init uid_cache_init(void)
 462 {
 463         int n;
 464
 465         uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
 466                         0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
 467
 468         for(n = 0; n < UIDHASH_SZ; ++n)
 469                 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
 470
 471         /* Insert the root user immediately (init already runs as root) */
 472         spin_lock_irq(&uidhash_lock);
 473         uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0));
 474         spin_unlock_irq(&uidhash_lock);
 475
 476         return 0;
 477 }
 478
 479 module_init(uid_cache_init);