LCOV - code coverage report
Current view: top level - kernel - ucount.c (source / functions) Hit Total Coverage
Test: coverage.info Lines: 83 154 53.9 %
Date: 2023-04-06 08:38:28 Functions: 10 18 55.6 %

          Line data    Source code
       1             : // SPDX-License-Identifier: GPL-2.0-only
       2             : 
       3             : #include <linux/stat.h>
       4             : #include <linux/sysctl.h>
       5             : #include <linux/slab.h>
       6             : #include <linux/cred.h>
       7             : #include <linux/hash.h>
       8             : #include <linux/kmemleak.h>
       9             : #include <linux/user_namespace.h>
      10             : 
      11             : struct ucounts init_ucounts = {
      12             :         .ns    = &init_user_ns,
      13             :         .uid   = GLOBAL_ROOT_UID,
      14             :         .count = ATOMIC_INIT(1),
      15             : };
      16             : 
      17             : #define UCOUNTS_HASHTABLE_BITS 10
      18             : static struct hlist_head ucounts_hashtable[(1 << UCOUNTS_HASHTABLE_BITS)];
      19             : static DEFINE_SPINLOCK(ucounts_lock);
      20             : 
      21             : #define ucounts_hashfn(ns, uid)                                         \
      22             :         hash_long((unsigned long)__kuid_val(uid) + (unsigned long)(ns), \
      23             :                   UCOUNTS_HASHTABLE_BITS)
      24             : #define ucounts_hashentry(ns, uid)      \
      25             :         (ucounts_hashtable + ucounts_hashfn(ns, uid))
      26             : 
      27             : 
      28             : #ifdef CONFIG_SYSCTL
      29             : static struct ctl_table_set *
      30           0 : set_lookup(struct ctl_table_root *root)
      31             : {
      32           0 :         return &current_user_ns()->set;
      33             : }
      34             : 
      35           0 : static int set_is_seen(struct ctl_table_set *set)
      36             : {
      37           0 :         return &current_user_ns()->set == set;
      38             : }
      39             : 
      40           0 : static int set_permissions(struct ctl_table_header *head,
      41             :                                   struct ctl_table *table)
      42             : {
      43           0 :         struct user_namespace *user_ns =
      44           0 :                 container_of(head->set, struct user_namespace, set);
      45             :         int mode;
      46             : 
      47             :         /* Allow users with CAP_SYS_RESOURCE unrestrained access */
      48           0 :         if (ns_capable(user_ns, CAP_SYS_RESOURCE))
      49           0 :                 mode = (table->mode & S_IRWXU) >> 6;
      50             :         else
      51             :         /* Allow all others at most read-only access */
      52           0 :                 mode = table->mode & S_IROTH;
      53           0 :         return (mode << 6) | (mode << 3) | mode;
      54             : }
      55             : 
      56             : static struct ctl_table_root set_root = {
      57             :         .lookup = set_lookup,
      58             :         .permissions = set_permissions,
      59             : };
      60             : 
      61             : static long ue_zero = 0;
      62             : static long ue_int_max = INT_MAX;
      63             : 
      64             : #define UCOUNT_ENTRY(name)                                      \
      65             :         {                                                       \
      66             :                 .procname       = name,                         \
      67             :                 .maxlen         = sizeof(long),                 \
      68             :                 .mode           = 0644,                         \
      69             :                 .proc_handler   = proc_doulongvec_minmax,       \
      70             :                 .extra1         = &ue_zero,                 \
      71             :                 .extra2         = &ue_int_max,                      \
      72             :         }
      73             : static struct ctl_table user_table[] = {
      74             :         UCOUNT_ENTRY("max_user_namespaces"),
      75             :         UCOUNT_ENTRY("max_pid_namespaces"),
      76             :         UCOUNT_ENTRY("max_uts_namespaces"),
      77             :         UCOUNT_ENTRY("max_ipc_namespaces"),
      78             :         UCOUNT_ENTRY("max_net_namespaces"),
      79             :         UCOUNT_ENTRY("max_mnt_namespaces"),
      80             :         UCOUNT_ENTRY("max_cgroup_namespaces"),
      81             :         UCOUNT_ENTRY("max_time_namespaces"),
      82             : #ifdef CONFIG_INOTIFY_USER
      83             :         UCOUNT_ENTRY("max_inotify_instances"),
      84             :         UCOUNT_ENTRY("max_inotify_watches"),
      85             : #endif
      86             : #ifdef CONFIG_FANOTIFY
      87             :         UCOUNT_ENTRY("max_fanotify_groups"),
      88             :         UCOUNT_ENTRY("max_fanotify_marks"),
      89             : #endif
      90             :         { }
      91             : };
      92             : #endif /* CONFIG_SYSCTL */
      93             : 
      94           1 : bool setup_userns_sysctls(struct user_namespace *ns)
      95             : {
      96             : #ifdef CONFIG_SYSCTL
      97             :         struct ctl_table *tbl;
      98             : 
      99             :         BUILD_BUG_ON(ARRAY_SIZE(user_table) != UCOUNT_COUNTS + 1);
     100           1 :         setup_sysctl_set(&ns->set, &set_root, set_is_seen);
     101           1 :         tbl = kmemdup(user_table, sizeof(user_table), GFP_KERNEL);
     102           1 :         if (tbl) {
     103             :                 int i;
     104          10 :                 for (i = 0; i < UCOUNT_COUNTS; i++) {
     105          10 :                         tbl[i].data = &ns->ucount_max[i];
     106             :                 }
     107           1 :                 ns->sysctls = __register_sysctl_table(&ns->set, "user", tbl);
     108             :         }
     109           1 :         if (!ns->sysctls) {
     110           0 :                 kfree(tbl);
     111           0 :                 retire_sysctl_set(&ns->set);
     112           0 :                 return false;
     113             :         }
     114             : #endif
     115             :         return true;
     116             : }
     117             : 
     118           0 : void retire_userns_sysctls(struct user_namespace *ns)
     119             : {
     120             : #ifdef CONFIG_SYSCTL
     121             :         struct ctl_table *tbl;
     122             : 
     123           0 :         tbl = ns->sysctls->ctl_table_arg;
     124           0 :         unregister_sysctl_table(ns->sysctls);
     125           0 :         retire_sysctl_set(&ns->set);
     126           0 :         kfree(tbl);
     127             : #endif
     128           0 : }
     129             : 
     130             : static struct ucounts *find_ucounts(struct user_namespace *ns, kuid_t uid, struct hlist_head *hashent)
     131             : {
     132             :         struct ucounts *ucounts;
     133             : 
     134           2 :         hlist_for_each_entry(ucounts, hashent, node) {
     135           0 :                 if (uid_eq(ucounts->uid, uid) && (ucounts->ns == ns))
     136             :                         return ucounts;
     137             :         }
     138             :         return NULL;
     139             : }
     140             : 
     141           1 : static void hlist_add_ucounts(struct ucounts *ucounts)
     142             : {
     143           2 :         struct hlist_head *hashent = ucounts_hashentry(ucounts->ns, ucounts->uid);
     144           1 :         spin_lock_irq(&ucounts_lock);
     145           2 :         hlist_add_head(&ucounts->node, hashent);
     146           1 :         spin_unlock_irq(&ucounts_lock);
     147           1 : }
     148             : 
     149             : static inline bool get_ucounts_or_wrap(struct ucounts *ucounts)
     150             : {
     151             :         /* Returns true on a successful get, false if the count wraps. */
     152         696 :         return !atomic_add_negative(1, &ucounts->count);
     153             : }
     154             : 
     155         348 : struct ucounts *get_ucounts(struct ucounts *ucounts)
     156             : {
     157         348 :         if (!get_ucounts_or_wrap(ucounts)) {
     158           0 :                 put_ucounts(ucounts);
     159           0 :                 ucounts = NULL;
     160             :         }
     161         348 :         return ucounts;
     162             : }
     163             : 
     164           1 : struct ucounts *alloc_ucounts(struct user_namespace *ns, kuid_t uid)
     165             : {
     166           2 :         struct hlist_head *hashent = ucounts_hashentry(ns, uid);
     167             :         struct ucounts *ucounts, *new;
     168             :         bool wrapped;
     169             : 
     170           1 :         spin_lock_irq(&ucounts_lock);
     171           1 :         ucounts = find_ucounts(ns, uid, hashent);
     172           1 :         if (!ucounts) {
     173           1 :                 spin_unlock_irq(&ucounts_lock);
     174             : 
     175           1 :                 new = kzalloc(sizeof(*new), GFP_KERNEL);
     176           1 :                 if (!new)
     177             :                         return NULL;
     178             : 
     179           1 :                 new->ns = ns;
     180           1 :                 new->uid = uid;
     181           2 :                 atomic_set(&new->count, 1);
     182             : 
     183           1 :                 spin_lock_irq(&ucounts_lock);
     184           1 :                 ucounts = find_ucounts(ns, uid, hashent);
     185           1 :                 if (ucounts) {
     186           0 :                         kfree(new);
     187             :                 } else {
     188           2 :                         hlist_add_head(&new->node, hashent);
     189           1 :                         get_user_ns(new->ns);
     190           1 :                         spin_unlock_irq(&ucounts_lock);
     191           1 :                         return new;
     192             :                 }
     193             :         }
     194           0 :         wrapped = !get_ucounts_or_wrap(ucounts);
     195           0 :         spin_unlock_irq(&ucounts_lock);
     196           0 :         if (wrapped) {
     197           0 :                 put_ucounts(ucounts);
     198           0 :                 return NULL;
     199             :         }
     200             :         return ucounts;
     201             : }
     202             : 
     203         332 : void put_ucounts(struct ucounts *ucounts)
     204             : {
     205             :         unsigned long flags;
     206             : 
     207         332 :         if (atomic_dec_and_lock_irqsave(&ucounts->count, &ucounts_lock, flags)) {
     208           0 :                 hlist_del_init(&ucounts->node);
     209           0 :                 spin_unlock_irqrestore(&ucounts_lock, flags);
     210           0 :                 put_user_ns(ucounts->ns);
     211           0 :                 kfree(ucounts);
     212             :         }
     213         332 : }
     214             : 
     215             : static inline bool atomic_long_inc_below(atomic_long_t *v, int u)
     216             : {
     217             :         long c, old;
     218           1 :         c = atomic_long_read(v);
     219             :         for (;;) {
     220           1 :                 if (unlikely(c >= u))
     221             :                         return false;
     222           2 :                 old = atomic_long_cmpxchg(v, c, c+1);
     223           1 :                 if (likely(old == c))
     224             :                         return true;
     225             :                 c = old;
     226             :         }
     227             : }
     228             : 
     229           1 : struct ucounts *inc_ucount(struct user_namespace *ns, kuid_t uid,
     230             :                            enum ucount_type type)
     231             : {
     232             :         struct ucounts *ucounts, *iter, *bad;
     233             :         struct user_namespace *tns;
     234           1 :         ucounts = alloc_ucounts(ns, uid);
     235           2 :         for (iter = ucounts; iter; iter = tns->ucounts) {
     236             :                 long max;
     237           1 :                 tns = iter->ns;
     238           1 :                 max = READ_ONCE(tns->ucount_max[type]);
     239           2 :                 if (!atomic_long_inc_below(&iter->ucount[type], max))
     240             :                         goto fail;
     241             :         }
     242             :         return ucounts;
     243             : fail:
     244             :         bad = iter;
     245           0 :         for (iter = ucounts; iter != bad; iter = iter->ns->ucounts)
     246           0 :                 atomic_long_dec(&iter->ucount[type]);
     247             : 
     248           0 :         put_ucounts(ucounts);
     249           0 :         return NULL;
     250             : }
     251             : 
     252           0 : void dec_ucount(struct ucounts *ucounts, enum ucount_type type)
     253             : {
     254             :         struct ucounts *iter;
     255           0 :         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
     256           0 :                 long dec = atomic_long_dec_if_positive(&iter->ucount[type]);
     257           0 :                 WARN_ON_ONCE(dec < 0);
     258             :         }
     259           0 :         put_ucounts(ucounts);
     260           0 : }
     261             : 
     262         348 : long inc_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v)
     263             : {
     264             :         struct ucounts *iter;
     265         349 :         long max = LONG_MAX;
     266         349 :         long ret = 0;
     267             : 
     268         698 :         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
     269         698 :                 long new = atomic_long_add_return(v, &iter->rlimit[type]);
     270         348 :                 if (new < 0 || new > max)
     271             :                         ret = LONG_MAX;
     272         348 :                 else if (iter == ucounts)
     273         348 :                         ret = new;
     274         698 :                 max = get_userns_rlimit_max(iter->ns, type);
     275             :         }
     276         348 :         return ret;
     277             : }
     278             : 
     279         333 : bool dec_rlimit_ucounts(struct ucounts *ucounts, enum rlimit_type type, long v)
     280             : {
     281             :         struct ucounts *iter;
     282         333 :         long new = -1; /* Silence compiler warning */
     283         666 :         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
     284         666 :                 long dec = atomic_long_sub_return(v, &iter->rlimit[type]);
     285         333 :                 WARN_ON_ONCE(dec < 0);
     286         333 :                 if (iter == ucounts)
     287         333 :                         new = dec;
     288             :         }
     289         333 :         return (new == 0);
     290             : }
     291             : 
     292           0 : static void do_dec_rlimit_put_ucounts(struct ucounts *ucounts,
     293             :                                 struct ucounts *last, enum rlimit_type type)
     294             : {
     295             :         struct ucounts *iter, *next;
     296           0 :         for (iter = ucounts; iter != last; iter = next) {
     297           0 :                 long dec = atomic_long_sub_return(1, &iter->rlimit[type]);
     298           0 :                 WARN_ON_ONCE(dec < 0);
     299           0 :                 next = iter->ns->ucounts;
     300           0 :                 if (dec == 0)
     301           0 :                         put_ucounts(iter);
     302             :         }
     303           0 : }
     304             : 
     305           0 : void dec_rlimit_put_ucounts(struct ucounts *ucounts, enum rlimit_type type)
     306             : {
     307           0 :         do_dec_rlimit_put_ucounts(ucounts, NULL, type);
     308           0 : }
     309             : 
     310           0 : long inc_rlimit_get_ucounts(struct ucounts *ucounts, enum rlimit_type type)
     311             : {
     312             :         /* Caller must hold a reference to ucounts */
     313             :         struct ucounts *iter;
     314           0 :         long max = LONG_MAX;
     315           0 :         long dec, ret = 0;
     316             : 
     317           0 :         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
     318           0 :                 long new = atomic_long_add_return(1, &iter->rlimit[type]);
     319           0 :                 if (new < 0 || new > max)
     320             :                         goto unwind;
     321           0 :                 if (iter == ucounts)
     322           0 :                         ret = new;
     323           0 :                 max = get_userns_rlimit_max(iter->ns, type);
     324             :                 /*
     325             :                  * Grab an extra ucount reference for the caller when
     326             :                  * the rlimit count was previously 0.
     327             :                  */
     328           0 :                 if (new != 1)
     329           0 :                         continue;
     330           0 :                 if (!get_ucounts(iter))
     331             :                         goto dec_unwind;
     332             :         }
     333             :         return ret;
     334             : dec_unwind:
     335           0 :         dec = atomic_long_sub_return(1, &iter->rlimit[type]);
     336           0 :         WARN_ON_ONCE(dec < 0);
     337             : unwind:
     338           0 :         do_dec_rlimit_put_ucounts(ucounts, iter, type);
     339           0 :         return 0;
     340             : }
     341             : 
     342         348 : bool is_rlimit_overlimit(struct ucounts *ucounts, enum rlimit_type type, unsigned long rlimit)
     343             : {
     344             :         struct ucounts *iter;
     345         348 :         long max = rlimit;
     346         348 :         if (rlimit > LONG_MAX)
     347           0 :                 max = LONG_MAX;
     348         348 :         for (iter = ucounts; iter; iter = iter->ns->ucounts) {
     349         348 :                 long val = get_rlimit_value(iter, type);
     350         348 :                 if (val < 0 || val > max)
     351             :                         return true;
     352         696 :                 max = get_userns_rlimit_max(iter->ns, type);
     353             :         }
     354             :         return false;
     355             : }
     356             : 
     357           1 : static __init int user_namespace_sysctl_init(void)
     358             : {
     359             : #ifdef CONFIG_SYSCTL
     360             :         static struct ctl_table_header *user_header;
     361             :         static struct ctl_table empty[1];
     362             :         /*
     363             :          * It is necessary to register the user directory in the
     364             :          * default set so that registrations in the child sets work
     365             :          * properly.
     366             :          */
     367           1 :         user_header = register_sysctl("user", empty);
     368           1 :         kmemleak_ignore(user_header);
     369           1 :         BUG_ON(!user_header);
     370           1 :         BUG_ON(!setup_userns_sysctls(&init_user_ns));
     371             : #endif
     372           1 :         hlist_add_ucounts(&init_ucounts);
     373           1 :         inc_rlimit_ucounts(&init_ucounts, UCOUNT_RLIMIT_NPROC, 1);
     374           1 :         return 0;
     375             : }
     376             : subsys_initcall(user_namespace_sysctl_init);

Generated by: LCOV version 1.14