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