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 #include "cred-internals.h"
  20
  21 struct user_namespace init_user_ns = {
  22         .kref = {
  23                 .refcount       = ATOMIC_INIT(2),
  24         },
  25         .root_user = &root_user,
  26 };
  27 EXPORT_SYMBOL_GPL(init_user_ns);
  28
  29 /*
  30  * UID task count cache, to get fast user lookup in "alloc_uid"
  31  * when changing user ID's (ie setuid() and friends).
  32  */
  33
  34 #define UIDHASH_MASK            (UIDHASH_SZ - 1)
  35 #define __uidhashfn(uid)        (((uid >> UIDHASH_BITS) + uid) & UIDHASH_MASK)
  36 #define uidhashentry(ns, uid)   ((ns)->uidhash_table + __uidhashfn((uid)))
  37
  38 static struct kmem_cache *uid_cachep;
  39
  40 /*
  41  * The uidhash_lock is mostly taken from process context, but it is
  42  * occasionally also taken from softirq/tasklet context, when
  43  * task-structs get RCU-freed. Hence all locking must be softirq-safe.
  44  * But free_uid() is also called with local interrupts disabled, and running
  45  * local_bh_enable() with local interrupts disabled is an error - we'll run
  46  * softirq callbacks, and they can unconditionally enable interrupts, and
  47  * the caller of free_uid() didn't expect that..
  48  */
  49 static DEFINE_SPINLOCK(uidhash_lock);
  50
  51 struct user_struct root_user = {
  52         .__count        = ATOMIC_INIT(1),
  53         .processes      = ATOMIC_INIT(1),
  54         .files          = ATOMIC_INIT(0),
  55         .sigpending     = ATOMIC_INIT(0),
  56         .locked_shm     = 0,
  57 #ifdef CONFIG_USER_SCHED
  58         .tg             = &init_task_group,
  59 #endif
  60 };
  61
  62 /*
  63  * These routines must be called with the uidhash spinlock held!
  64  */
  65 static void uid_hash_insert(struct user_struct *up, struct hlist_head *hashent)
  66 {
  67         hlist_add_head(&up->uidhash_node, hashent);
  68 }
  69
  70 static void uid_hash_remove(struct user_struct *up)
  71 {
  72         hlist_del_init(&up->uidhash_node);
  73 }
  74
  75 static struct user_struct *uid_hash_find(uid_t uid, struct hlist_head *hashent)
  76 {
  77         struct user_struct *user;
  78         struct hlist_node *h;
  79
  80         hlist_for_each_entry(user, h, hashent, uidhash_node) {
  81                 if (user->uid == uid) {
  82                         atomic_inc(&user->__count);
  83                         return user;
  84                 }
  85         }
  86
  87         return NULL;
  88 }
  89
  90 #ifdef CONFIG_USER_SCHED
  91
  92 static void sched_destroy_user(struct user_struct *up)
  93 {
  94         sched_destroy_group(up->tg);
  95 }
  96
  97 static int sched_create_user(struct user_struct *up)
  98 {
  99         int rc = 0;
 100
 101         up->tg = sched_create_group(&root_task_group);
 102         if (IS_ERR(up->tg))
 103                 rc = -ENOMEM;
 104
 105         return rc;
 106 }
 107
 108 #else   /* CONFIG_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
 113 #endif  /* CONFIG_USER_SCHED */
 114
 115 #if defined(CONFIG_USER_SCHED) && defined(CONFIG_SYSFS)
 116
 117 static struct kset *uids_kset; /* represents the /sys/kernel/uids/ directory */
 118 static DEFINE_MUTEX(uids_mutex);
 119
 120 static inline void uids_mutex_lock(void)
 121 {
 122         mutex_lock(&uids_mutex);
 123 }
 124
 125 static inline void uids_mutex_unlock(void)
 126 {
 127         mutex_unlock(&uids_mutex);
 128 }
 129
 130 /* uid directory attributes */
 131 #ifdef CONFIG_FAIR_GROUP_SCHED
 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 #endif
 159
 160 #ifdef CONFIG_RT_GROUP_SCHED
 161 static ssize_t cpu_rt_runtime_show(struct kobject *kobj,
 162                                    struct kobj_attribute *attr,
 163                                    char *buf)
 164 {
 165         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 166
 167         return sprintf(buf, "%ld\n", sched_group_rt_runtime(up->tg));
 168 }
 169
 170 static ssize_t cpu_rt_runtime_store(struct kobject *kobj,
 171                                     struct kobj_attribute *attr,
 172                                     const char *buf, size_t size)
 173 {
 174         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 175         unsigned long rt_runtime;
 176         int rc;
 177
 178         sscanf(buf, "%ld", &rt_runtime);
 179
 180         rc = sched_group_set_rt_runtime(up->tg, rt_runtime);
 181
 182         return (rc ? rc : size);
 183 }
 184
 185 static struct kobj_attribute cpu_rt_runtime_attr =
 186         __ATTR(cpu_rt_runtime, 0644, cpu_rt_runtime_show, cpu_rt_runtime_store);
 187
 188 static ssize_t cpu_rt_period_show(struct kobject *kobj,
 189                                    struct kobj_attribute *attr,
 190                                    char *buf)
 191 {
 192         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 193
 194         return sprintf(buf, "%lu\n", sched_group_rt_period(up->tg));
 195 }
 196
 197 static ssize_t cpu_rt_period_store(struct kobject *kobj,
 198                                     struct kobj_attribute *attr,
 199                                     const char *buf, size_t size)
 200 {
 201         struct user_struct *up = container_of(kobj, struct user_struct, kobj);
 202         unsigned long rt_period;
 203         int rc;
 204
 205         sscanf(buf, "%lu", &rt_period);
 206
 207         rc = sched_group_set_rt_period(up->tg, rt_period);
 208
 209         return (rc ? rc : size);
 210 }
 211
 212 static struct kobj_attribute cpu_rt_period_attr =
 213         __ATTR(cpu_rt_period, 0644, cpu_rt_period_show, cpu_rt_period_store);
 214 #endif
 215
 216 /* default attributes per uid directory */
 217 static struct attribute *uids_attributes[] = {
 218 #ifdef CONFIG_FAIR_GROUP_SCHED
 219         &cpu_share_attr.attr,
 220 #endif
 221 #ifdef CONFIG_RT_GROUP_SCHED
 222         &cpu_rt_runtime_attr.attr,
 223         &cpu_rt_period_attr.attr,
 224 #endif
 225         NULL
 226 };
 227
 228 /* the lifetime of user_struct is not managed by the core (now) */
 229 static void uids_release(struct kobject *kobj)
 230 {
 231         return;
 232 }
 233
 234 static struct kobj_type uids_ktype = {
 235         .sysfs_ops = &kobj_sysfs_ops,
 236         .default_attrs = uids_attributes,
 237         .release = uids_release,
 238 };
 239
 240 /* create /sys/kernel/uids/<uid>/cpu_share file for this user */
 241 static int uids_user_create(struct user_struct *up)
 242 {
 243         struct kobject *kobj = &up->kobj;
 244         int error;
 245
 246         memset(kobj, 0, sizeof(struct kobject));
 247         kobj->kset = uids_kset;
 248         error = kobject_init_and_add(kobj, &uids_ktype, NULL, "%d", up->uid);
 249         if (error) {
 250                 kobject_put(kobj);
 251                 goto done;
 252         }
 253
 254         kobject_uevent(kobj, KOBJ_ADD);
 255 done:
 256         return error;
 257 }
 258
 259 /* create these entries in sysfs:
 260  *      "/sys/kernel/uids" directory
 261  *      "/sys/kernel/uids/0" directory (for root user)
 262  *      "/sys/kernel/uids/0/cpu_share" file (for root user)
 263  */
 264 int __init uids_sysfs_init(void)
 265 {
 266         uids_kset = kset_create_and_add("uids", NULL, kernel_kobj);
 267         if (!uids_kset)
 268                 return -ENOMEM;
 269
 270         return uids_user_create(&root_user);
 271 }
 272
 273 /* work function to remove sysfs directory for a user and free up
 274  * corresponding structures.
 275  */
 276 static void remove_user_sysfs_dir(struct work_struct *w)
 277 {
 278         struct user_struct *up = container_of(w, struct user_struct, work);
 279         unsigned long flags;
 280         int remove_user = 0;
 281
 282         /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
 283          * atomic.
 284          */
 285         uids_mutex_lock();
 286
 287         local_irq_save(flags);
 288
 289         if (atomic_dec_and_lock(&up->__count, &uidhash_lock)) {
 290                 uid_hash_remove(up);
 291                 remove_user = 1;
 292                 spin_unlock_irqrestore(&uidhash_lock, flags);
 293         } else {
 294                 local_irq_restore(flags);
 295         }
 296
 297         if (!remove_user)
 298                 goto done;
 299
 300         kobject_uevent(&up->kobj, KOBJ_REMOVE);
 301         kobject_del(&up->kobj);
 302         kobject_put(&up->kobj);
 303
 304         sched_destroy_user(up);
 305         key_put(up->uid_keyring);
 306         key_put(up->session_keyring);
 307         kmem_cache_free(uid_cachep, up);
 308
 309 done:
 310         uids_mutex_unlock();
 311 }
 312
 313 /* IRQs are disabled and uidhash_lock is held upon function entry.
 314  * IRQ state (as stored in flags) is restored and uidhash_lock released
 315  * upon function exit.
 316  */
 317 static inline void free_user(struct user_struct *up, unsigned long flags)
 318 {
 319         /* restore back the count */
 320         atomic_inc(&up->__count);
 321         spin_unlock_irqrestore(&uidhash_lock, flags);
 322
 323         INIT_WORK(&up->work, remove_user_sysfs_dir);
 324         schedule_work(&up->work);
 325 }
 326
 327 #else   /* CONFIG_USER_SCHED && CONFIG_SYSFS */
 328
 329 int uids_sysfs_init(void) { return 0; }
 330 static inline int uids_user_create(struct user_struct *up) { return 0; }
 331 static inline void uids_mutex_lock(void) { }
 332 static inline void uids_mutex_unlock(void) { }
 333
 334 /* IRQs are disabled and uidhash_lock is held upon function entry.
 335  * IRQ state (as stored in flags) is restored and uidhash_lock released
 336  * upon function exit.
 337  */
 338 static inline void free_user(struct user_struct *up, unsigned long flags)
 339 {
 340         uid_hash_remove(up);
 341         spin_unlock_irqrestore(&uidhash_lock, flags);
 342         sched_destroy_user(up);
 343         key_put(up->uid_keyring);
 344         key_put(up->session_keyring);
 345         kmem_cache_free(uid_cachep, up);
 346 }
 347
 348 #endif
 349
 350 /*
 351  * Locate the user_struct for the passed UID.  If found, take a ref on it.  The
 352  * caller must undo that ref with free_uid().
 353  *
 354  * If the user_struct could not be found, return NULL.
 355  */
 356 struct user_struct *find_user(uid_t uid)
 357 {
 358         struct user_struct *ret;
 359         unsigned long flags;
 360         struct user_namespace *ns = current->nsproxy->user_ns;
 361
 362         spin_lock_irqsave(&uidhash_lock, flags);
 363         ret = uid_hash_find(uid, uidhashentry(ns, uid));
 364         spin_unlock_irqrestore(&uidhash_lock, flags);
 365         return ret;
 366 }
 367
 368 void free_uid(struct user_struct *up)
 369 {
 370         unsigned long flags;
 371
 372         if (!up)
 373                 return;
 374
 375         local_irq_save(flags);
 376         if (atomic_dec_and_lock(&up->__count, &uidhash_lock))
 377                 free_user(up, flags);
 378         else
 379                 local_irq_restore(flags);
 380 }
 381
 382 struct user_struct *alloc_uid(struct user_namespace *ns, uid_t uid)
 383 {
 384         struct hlist_head *hashent = uidhashentry(ns, uid);
 385         struct user_struct *up, *new;
 386
 387         /* Make uid_hash_find() + uids_user_create() + uid_hash_insert()
 388          * atomic.
 389          */
 390         uids_mutex_lock();
 391
 392         spin_lock_irq(&uidhash_lock);
 393         up = uid_hash_find(uid, hashent);
 394         spin_unlock_irq(&uidhash_lock);
 395
 396         if (!up) {
 397                 new = kmem_cache_zalloc(uid_cachep, GFP_KERNEL);
 398                 if (!new)
 399                         goto out_unlock;
 400
 401                 new->uid = uid;
 402                 atomic_set(&new->__count, 1);
 403
 404                 if (sched_create_user(new) < 0)
 405                         goto out_free_user;
 406
 407                 if (uids_user_create(new))
 408                         goto out_destoy_sched;
 409
 410                 /*
 411                  * Before adding this, check whether we raced
 412                  * on adding the same user already..
 413                  */
 414                 spin_lock_irq(&uidhash_lock);
 415                 up = uid_hash_find(uid, hashent);
 416                 if (up) {
 417                         /* This case is not possible when CONFIG_USER_SCHED
 418                          * is defined, since we serialize alloc_uid() using
 419                          * uids_mutex. Hence no need to call
 420                          * sched_destroy_user() or remove_user_sysfs_dir().
 421                          */
 422                         key_put(new->uid_keyring);
 423                         key_put(new->session_keyring);
 424                         kmem_cache_free(uid_cachep, new);
 425                 } else {
 426                         uid_hash_insert(new, hashent);
 427                         up = new;
 428                 }
 429                 spin_unlock_irq(&uidhash_lock);
 430
 431         }
 432
 433         uids_mutex_unlock();
 434
 435         return up;
 436
 437 out_destoy_sched:
 438         sched_destroy_user(new);
 439 out_free_user:
 440         kmem_cache_free(uid_cachep, new);
 441 out_unlock:
 442         uids_mutex_unlock();
 443         return NULL;
 444 }
 445
 446 #ifdef CONFIG_USER_NS
 447 void release_uids(struct user_namespace *ns)
 448 {
 449         int i;
 450         unsigned long flags;
 451         struct hlist_head *head;
 452         struct hlist_node *nd;
 453
 454         spin_lock_irqsave(&uidhash_lock, flags);
 455         /*
 456          * collapse the chains so that the user_struct-s will
 457          * be still alive, but not in hashes. subsequent free_uid()
 458          * will free them.
 459          */
 460         for (i = 0; i < UIDHASH_SZ; i++) {
 461                 head = ns->uidhash_table + i;
 462                 while (!hlist_empty(head)) {
 463                         nd = head->first;
 464                         hlist_del_init(nd);
 465                 }
 466         }
 467         spin_unlock_irqrestore(&uidhash_lock, flags);
 468
 469         free_uid(ns->root_user);
 470 }
 471 #endif
 472
 473 static int __init uid_cache_init(void)
 474 {
 475         int n;
 476
 477         uid_cachep = kmem_cache_create("uid_cache", sizeof(struct user_struct),
 478                         0, SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
 479
 480         for(n = 0; n < UIDHASH_SZ; ++n)
 481                 INIT_HLIST_HEAD(init_user_ns.uidhash_table + n);
 482
 483         /* Insert the root user immediately (init already runs as root) */
 484         spin_lock_irq(&uidhash_lock);
 485         uid_hash_insert(&root_user, uidhashentry(&init_user_ns, 0));
 486         spin_unlock_irq(&uidhash_lock);
 487
 488         return 0;
 489 }
 490
 491 module_init(uid_cache_init);