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