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