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