linux/kernel/user_namespace.c
<<
>>
Prefs
   1/*
   2 *  This program is free software; you can redistribute it and/or
   3 *  modify it under the terms of the GNU General Public License as
   4 *  published by the Free Software Foundation, version 2 of the
   5 *  License.
   6 */
   7
   8#include <linux/export.h>
   9#include <linux/nsproxy.h>
  10#include <linux/slab.h>
  11#include <linux/user_namespace.h>
  12#include <linux/proc_ns.h>
  13#include <linux/highuid.h>
  14#include <linux/cred.h>
  15#include <linux/securebits.h>
  16#include <linux/keyctl.h>
  17#include <linux/key-type.h>
  18#include <keys/user-type.h>
  19#include <linux/seq_file.h>
  20#include <linux/fs.h>
  21#include <linux/uaccess.h>
  22#include <linux/ctype.h>
  23#include <linux/projid.h>
  24#include <linux/fs_struct.h>
  25
  26static struct kmem_cache *user_ns_cachep __read_mostly;
  27
  28static bool new_idmap_permitted(const struct file *file,
  29                                struct user_namespace *ns, int cap_setid,
  30                                struct uid_gid_map *map);
  31
  32static void set_cred_user_ns(struct cred *cred, struct user_namespace *user_ns)
  33{
  34        /* Start with the same capabilities as init but useless for doing
  35         * anything as the capabilities are bound to the new user namespace.
  36         */
  37        cred->securebits = SECUREBITS_DEFAULT;
  38        cred->cap_inheritable = CAP_EMPTY_SET;
  39        cred->cap_permitted = CAP_FULL_SET;
  40        cred->cap_effective = CAP_FULL_SET;
  41        cred->cap_bset = CAP_FULL_SET;
  42#ifdef CONFIG_KEYS
  43        key_put(cred->request_key_auth);
  44        cred->request_key_auth = NULL;
  45#endif
  46        /* tgcred will be cleared in our caller bc CLONE_THREAD won't be set */
  47        cred->user_ns = user_ns;
  48}
  49
  50/*
  51 * Create a new user namespace, deriving the creator from the user in the
  52 * passed credentials, and replacing that user with the new root user for the
  53 * new namespace.
  54 *
  55 * This is called by copy_creds(), which will finish setting the target task's
  56 * credentials.
  57 */
  58int create_user_ns(struct cred *new)
  59{
  60        struct user_namespace *ns, *parent_ns = new->user_ns;
  61        kuid_t owner = new->euid;
  62        kgid_t group = new->egid;
  63        int ret;
  64
  65        if (parent_ns->level > 32)
  66                return -EUSERS;
  67
  68        /*
  69         * Verify that we can not violate the policy of which files
  70         * may be accessed that is specified by the root directory,
  71         * by verifing that the root directory is at the root of the
  72         * mount namespace which allows all files to be accessed.
  73         */
  74        if (current_chrooted())
  75                return -EPERM;
  76
  77        /* The creator needs a mapping in the parent user namespace
  78         * or else we won't be able to reasonably tell userspace who
  79         * created a user_namespace.
  80         */
  81        if (!kuid_has_mapping(parent_ns, owner) ||
  82            !kgid_has_mapping(parent_ns, group))
  83                return -EPERM;
  84
  85        ns = kmem_cache_zalloc(user_ns_cachep, GFP_KERNEL);
  86        if (!ns)
  87                return -ENOMEM;
  88
  89        ret = proc_alloc_inum(&ns->proc_inum);
  90        if (ret) {
  91                kmem_cache_free(user_ns_cachep, ns);
  92                return ret;
  93        }
  94
  95        atomic_set(&ns->count, 1);
  96        /* Leave the new->user_ns reference with the new user namespace. */
  97        ns->parent = parent_ns;
  98        ns->level = parent_ns->level + 1;
  99        ns->owner = owner;
 100        ns->group = group;
 101
 102        set_cred_user_ns(new, ns);
 103
 104        update_mnt_policy(ns);
 105
 106        return 0;
 107}
 108
 109int unshare_userns(unsigned long unshare_flags, struct cred **new_cred)
 110{
 111        struct cred *cred;
 112        int err = -ENOMEM;
 113
 114        if (!(unshare_flags & CLONE_NEWUSER))
 115                return 0;
 116
 117        cred = prepare_creds();
 118        if (cred) {
 119                err = create_user_ns(cred);
 120                if (err)
 121                        put_cred(cred);
 122                else
 123                        *new_cred = cred;
 124        }
 125
 126        return err;
 127}
 128
 129void free_user_ns(struct user_namespace *ns)
 130{
 131        struct user_namespace *parent;
 132
 133        do {
 134                parent = ns->parent;
 135                proc_free_inum(ns->proc_inum);
 136                kmem_cache_free(user_ns_cachep, ns);
 137                ns = parent;
 138        } while (atomic_dec_and_test(&parent->count));
 139}
 140EXPORT_SYMBOL(free_user_ns);
 141
 142static u32 map_id_range_down(struct uid_gid_map *map, u32 id, u32 count)
 143{
 144        unsigned idx, extents;
 145        u32 first, last, id2;
 146
 147        id2 = id + count - 1;
 148
 149        /* Find the matching extent */
 150        extents = map->nr_extents;
 151        smp_read_barrier_depends();
 152        for (idx = 0; idx < extents; idx++) {
 153                first = map->extent[idx].first;
 154                last = first + map->extent[idx].count - 1;
 155                if (id >= first && id <= last &&
 156                    (id2 >= first && id2 <= last))
 157                        break;
 158        }
 159        /* Map the id or note failure */
 160        if (idx < extents)
 161                id = (id - first) + map->extent[idx].lower_first;
 162        else
 163                id = (u32) -1;
 164
 165        return id;
 166}
 167
 168static u32 map_id_down(struct uid_gid_map *map, u32 id)
 169{
 170        unsigned idx, extents;
 171        u32 first, last;
 172
 173        /* Find the matching extent */
 174        extents = map->nr_extents;
 175        smp_read_barrier_depends();
 176        for (idx = 0; idx < extents; idx++) {
 177                first = map->extent[idx].first;
 178                last = first + map->extent[idx].count - 1;
 179                if (id >= first && id <= last)
 180                        break;
 181        }
 182        /* Map the id or note failure */
 183        if (idx < extents)
 184                id = (id - first) + map->extent[idx].lower_first;
 185        else
 186                id = (u32) -1;
 187
 188        return id;
 189}
 190
 191static u32 map_id_up(struct uid_gid_map *map, u32 id)
 192{
 193        unsigned idx, extents;
 194        u32 first, last;
 195
 196        /* Find the matching extent */
 197        extents = map->nr_extents;
 198        smp_read_barrier_depends();
 199        for (idx = 0; idx < extents; idx++) {
 200                first = map->extent[idx].lower_first;
 201                last = first + map->extent[idx].count - 1;
 202                if (id >= first && id <= last)
 203                        break;
 204        }
 205        /* Map the id or note failure */
 206        if (idx < extents)
 207                id = (id - first) + map->extent[idx].first;
 208        else
 209                id = (u32) -1;
 210
 211        return id;
 212}
 213
 214/**
 215 *      make_kuid - Map a user-namespace uid pair into a kuid.
 216 *      @ns:  User namespace that the uid is in
 217 *      @uid: User identifier
 218 *
 219 *      Maps a user-namespace uid pair into a kernel internal kuid,
 220 *      and returns that kuid.
 221 *
 222 *      When there is no mapping defined for the user-namespace uid
 223 *      pair INVALID_UID is returned.  Callers are expected to test
 224 *      for and handle handle INVALID_UID being returned.  INVALID_UID
 225 *      may be tested for using uid_valid().
 226 */
 227kuid_t make_kuid(struct user_namespace *ns, uid_t uid)
 228{
 229        /* Map the uid to a global kernel uid */
 230        return KUIDT_INIT(map_id_down(&ns->uid_map, uid));
 231}
 232EXPORT_SYMBOL(make_kuid);
 233
 234/**
 235 *      from_kuid - Create a uid from a kuid user-namespace pair.
 236 *      @targ: The user namespace we want a uid in.
 237 *      @kuid: The kernel internal uid to start with.
 238 *
 239 *      Map @kuid into the user-namespace specified by @targ and
 240 *      return the resulting uid.
 241 *
 242 *      There is always a mapping into the initial user_namespace.
 243 *
 244 *      If @kuid has no mapping in @targ (uid_t)-1 is returned.
 245 */
 246uid_t from_kuid(struct user_namespace *targ, kuid_t kuid)
 247{
 248        /* Map the uid from a global kernel uid */
 249        return map_id_up(&targ->uid_map, __kuid_val(kuid));
 250}
 251EXPORT_SYMBOL(from_kuid);
 252
 253/**
 254 *      from_kuid_munged - Create a uid from a kuid user-namespace pair.
 255 *      @targ: The user namespace we want a uid in.
 256 *      @kuid: The kernel internal uid to start with.
 257 *
 258 *      Map @kuid into the user-namespace specified by @targ and
 259 *      return the resulting uid.
 260 *
 261 *      There is always a mapping into the initial user_namespace.
 262 *
 263 *      Unlike from_kuid from_kuid_munged never fails and always
 264 *      returns a valid uid.  This makes from_kuid_munged appropriate
 265 *      for use in syscalls like stat and getuid where failing the
 266 *      system call and failing to provide a valid uid are not an
 267 *      options.
 268 *
 269 *      If @kuid has no mapping in @targ overflowuid is returned.
 270 */
 271uid_t from_kuid_munged(struct user_namespace *targ, kuid_t kuid)
 272{
 273        uid_t uid;
 274        uid = from_kuid(targ, kuid);
 275
 276        if (uid == (uid_t) -1)
 277                uid = overflowuid;
 278        return uid;
 279}
 280EXPORT_SYMBOL(from_kuid_munged);
 281
 282/**
 283 *      make_kgid - Map a user-namespace gid pair into a kgid.
 284 *      @ns:  User namespace that the gid is in
 285 *      @uid: group identifier
 286 *
 287 *      Maps a user-namespace gid pair into a kernel internal kgid,
 288 *      and returns that kgid.
 289 *
 290 *      When there is no mapping defined for the user-namespace gid
 291 *      pair INVALID_GID is returned.  Callers are expected to test
 292 *      for and handle INVALID_GID being returned.  INVALID_GID may be
 293 *      tested for using gid_valid().
 294 */
 295kgid_t make_kgid(struct user_namespace *ns, gid_t gid)
 296{
 297        /* Map the gid to a global kernel gid */
 298        return KGIDT_INIT(map_id_down(&ns->gid_map, gid));
 299}
 300EXPORT_SYMBOL(make_kgid);
 301
 302/**
 303 *      from_kgid - Create a gid from a kgid user-namespace pair.
 304 *      @targ: The user namespace we want a gid in.
 305 *      @kgid: The kernel internal gid to start with.
 306 *
 307 *      Map @kgid into the user-namespace specified by @targ and
 308 *      return the resulting gid.
 309 *
 310 *      There is always a mapping into the initial user_namespace.
 311 *
 312 *      If @kgid has no mapping in @targ (gid_t)-1 is returned.
 313 */
 314gid_t from_kgid(struct user_namespace *targ, kgid_t kgid)
 315{
 316        /* Map the gid from a global kernel gid */
 317        return map_id_up(&targ->gid_map, __kgid_val(kgid));
 318}
 319EXPORT_SYMBOL(from_kgid);
 320
 321/**
 322 *      from_kgid_munged - Create a gid from a kgid user-namespace pair.
 323 *      @targ: The user namespace we want a gid in.
 324 *      @kgid: The kernel internal gid to start with.
 325 *
 326 *      Map @kgid into the user-namespace specified by @targ and
 327 *      return the resulting gid.
 328 *
 329 *      There is always a mapping into the initial user_namespace.
 330 *
 331 *      Unlike from_kgid from_kgid_munged never fails and always
 332 *      returns a valid gid.  This makes from_kgid_munged appropriate
 333 *      for use in syscalls like stat and getgid where failing the
 334 *      system call and failing to provide a valid gid are not options.
 335 *
 336 *      If @kgid has no mapping in @targ overflowgid is returned.
 337 */
 338gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
 339{
 340        gid_t gid;
 341        gid = from_kgid(targ, kgid);
 342
 343        if (gid == (gid_t) -1)
 344                gid = overflowgid;
 345        return gid;
 346}
 347EXPORT_SYMBOL(from_kgid_munged);
 348
 349/**
 350 *      make_kprojid - Map a user-namespace projid pair into a kprojid.
 351 *      @ns:  User namespace that the projid is in
 352 *      @projid: Project identifier
 353 *
 354 *      Maps a user-namespace uid pair into a kernel internal kuid,
 355 *      and returns that kuid.
 356 *
 357 *      When there is no mapping defined for the user-namespace projid
 358 *      pair INVALID_PROJID is returned.  Callers are expected to test
 359 *      for and handle handle INVALID_PROJID being returned.  INVALID_PROJID
 360 *      may be tested for using projid_valid().
 361 */
 362kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
 363{
 364        /* Map the uid to a global kernel uid */
 365        return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
 366}
 367EXPORT_SYMBOL(make_kprojid);
 368
 369/**
 370 *      from_kprojid - Create a projid from a kprojid user-namespace pair.
 371 *      @targ: The user namespace we want a projid in.
 372 *      @kprojid: The kernel internal project identifier to start with.
 373 *
 374 *      Map @kprojid into the user-namespace specified by @targ and
 375 *      return the resulting projid.
 376 *
 377 *      There is always a mapping into the initial user_namespace.
 378 *
 379 *      If @kprojid has no mapping in @targ (projid_t)-1 is returned.
 380 */
 381projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
 382{
 383        /* Map the uid from a global kernel uid */
 384        return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
 385}
 386EXPORT_SYMBOL(from_kprojid);
 387
 388/**
 389 *      from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
 390 *      @targ: The user namespace we want a projid in.
 391 *      @kprojid: The kernel internal projid to start with.
 392 *
 393 *      Map @kprojid into the user-namespace specified by @targ and
 394 *      return the resulting projid.
 395 *
 396 *      There is always a mapping into the initial user_namespace.
 397 *
 398 *      Unlike from_kprojid from_kprojid_munged never fails and always
 399 *      returns a valid projid.  This makes from_kprojid_munged
 400 *      appropriate for use in syscalls like stat and where
 401 *      failing the system call and failing to provide a valid projid are
 402 *      not an options.
 403 *
 404 *      If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
 405 */
 406projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
 407{
 408        projid_t projid;
 409        projid = from_kprojid(targ, kprojid);
 410
 411        if (projid == (projid_t) -1)
 412                projid = OVERFLOW_PROJID;
 413        return projid;
 414}
 415EXPORT_SYMBOL(from_kprojid_munged);
 416
 417
 418static int uid_m_show(struct seq_file *seq, void *v)
 419{
 420        struct user_namespace *ns = seq->private;
 421        struct uid_gid_extent *extent = v;
 422        struct user_namespace *lower_ns;
 423        uid_t lower;
 424
 425        lower_ns = seq_user_ns(seq);
 426        if ((lower_ns == ns) && lower_ns->parent)
 427                lower_ns = lower_ns->parent;
 428
 429        lower = from_kuid(lower_ns, KUIDT_INIT(extent->lower_first));
 430
 431        seq_printf(seq, "%10u %10u %10u\n",
 432                extent->first,
 433                lower,
 434                extent->count);
 435
 436        return 0;
 437}
 438
 439static int gid_m_show(struct seq_file *seq, void *v)
 440{
 441        struct user_namespace *ns = seq->private;
 442        struct uid_gid_extent *extent = v;
 443        struct user_namespace *lower_ns;
 444        gid_t lower;
 445
 446        lower_ns = seq_user_ns(seq);
 447        if ((lower_ns == ns) && lower_ns->parent)
 448                lower_ns = lower_ns->parent;
 449
 450        lower = from_kgid(lower_ns, KGIDT_INIT(extent->lower_first));
 451
 452        seq_printf(seq, "%10u %10u %10u\n",
 453                extent->first,
 454                lower,
 455                extent->count);
 456
 457        return 0;
 458}
 459
 460static int projid_m_show(struct seq_file *seq, void *v)
 461{
 462        struct user_namespace *ns = seq->private;
 463        struct uid_gid_extent *extent = v;
 464        struct user_namespace *lower_ns;
 465        projid_t lower;
 466
 467        lower_ns = seq_user_ns(seq);
 468        if ((lower_ns == ns) && lower_ns->parent)
 469                lower_ns = lower_ns->parent;
 470
 471        lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
 472
 473        seq_printf(seq, "%10u %10u %10u\n",
 474                extent->first,
 475                lower,
 476                extent->count);
 477
 478        return 0;
 479}
 480
 481static void *m_start(struct seq_file *seq, loff_t *ppos, struct uid_gid_map *map)
 482{
 483        struct uid_gid_extent *extent = NULL;
 484        loff_t pos = *ppos;
 485
 486        if (pos < map->nr_extents)
 487                extent = &map->extent[pos];
 488
 489        return extent;
 490}
 491
 492static void *uid_m_start(struct seq_file *seq, loff_t *ppos)
 493{
 494        struct user_namespace *ns = seq->private;
 495
 496        return m_start(seq, ppos, &ns->uid_map);
 497}
 498
 499static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
 500{
 501        struct user_namespace *ns = seq->private;
 502
 503        return m_start(seq, ppos, &ns->gid_map);
 504}
 505
 506static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
 507{
 508        struct user_namespace *ns = seq->private;
 509
 510        return m_start(seq, ppos, &ns->projid_map);
 511}
 512
 513static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
 514{
 515        (*pos)++;
 516        return seq->op->start(seq, pos);
 517}
 518
 519static void m_stop(struct seq_file *seq, void *v)
 520{
 521        return;
 522}
 523
 524struct seq_operations proc_uid_seq_operations = {
 525        .start = uid_m_start,
 526        .stop = m_stop,
 527        .next = m_next,
 528        .show = uid_m_show,
 529};
 530
 531struct seq_operations proc_gid_seq_operations = {
 532        .start = gid_m_start,
 533        .stop = m_stop,
 534        .next = m_next,
 535        .show = gid_m_show,
 536};
 537
 538struct seq_operations proc_projid_seq_operations = {
 539        .start = projid_m_start,
 540        .stop = m_stop,
 541        .next = m_next,
 542        .show = projid_m_show,
 543};
 544
 545static bool mappings_overlap(struct uid_gid_map *new_map, struct uid_gid_extent *extent)
 546{
 547        u32 upper_first, lower_first, upper_last, lower_last;
 548        unsigned idx;
 549
 550        upper_first = extent->first;
 551        lower_first = extent->lower_first;
 552        upper_last = upper_first + extent->count - 1;
 553        lower_last = lower_first + extent->count - 1;
 554
 555        for (idx = 0; idx < new_map->nr_extents; idx++) {
 556                u32 prev_upper_first, prev_lower_first;
 557                u32 prev_upper_last, prev_lower_last;
 558                struct uid_gid_extent *prev;
 559
 560                prev = &new_map->extent[idx];
 561
 562                prev_upper_first = prev->first;
 563                prev_lower_first = prev->lower_first;
 564                prev_upper_last = prev_upper_first + prev->count - 1;
 565                prev_lower_last = prev_lower_first + prev->count - 1;
 566
 567                /* Does the upper range intersect a previous extent? */
 568                if ((prev_upper_first <= upper_last) &&
 569                    (prev_upper_last >= upper_first))
 570                        return true;
 571
 572                /* Does the lower range intersect a previous extent? */
 573                if ((prev_lower_first <= lower_last) &&
 574                    (prev_lower_last >= lower_first))
 575                        return true;
 576        }
 577        return false;
 578}
 579
 580
 581static DEFINE_MUTEX(id_map_mutex);
 582
 583static ssize_t map_write(struct file *file, const char __user *buf,
 584                         size_t count, loff_t *ppos,
 585                         int cap_setid,
 586                         struct uid_gid_map *map,
 587                         struct uid_gid_map *parent_map)
 588{
 589        struct seq_file *seq = file->private_data;
 590        struct user_namespace *ns = seq->private;
 591        struct uid_gid_map new_map;
 592        unsigned idx;
 593        struct uid_gid_extent *extent = NULL;
 594        unsigned long page = 0;
 595        char *kbuf, *pos, *next_line;
 596        ssize_t ret = -EINVAL;
 597
 598        /*
 599         * The id_map_mutex serializes all writes to any given map.
 600         *
 601         * Any map is only ever written once.
 602         *
 603         * An id map fits within 1 cache line on most architectures.
 604         *
 605         * On read nothing needs to be done unless you are on an
 606         * architecture with a crazy cache coherency model like alpha.
 607         *
 608         * There is a one time data dependency between reading the
 609         * count of the extents and the values of the extents.  The
 610         * desired behavior is to see the values of the extents that
 611         * were written before the count of the extents.
 612         *
 613         * To achieve this smp_wmb() is used on guarantee the write
 614         * order and smp_read_barrier_depends() is guaranteed that we
 615         * don't have crazy architectures returning stale data.
 616         *
 617         */
 618        mutex_lock(&id_map_mutex);
 619
 620        ret = -EPERM;
 621        /* Only allow one successful write to the map */
 622        if (map->nr_extents != 0)
 623                goto out;
 624
 625        /*
 626         * Adjusting namespace settings requires capabilities on the target.
 627         */
 628        if (cap_valid(cap_setid) && !file_ns_capable(file, ns, CAP_SYS_ADMIN))
 629                goto out;
 630
 631        /* Get a buffer */
 632        ret = -ENOMEM;
 633        page = __get_free_page(GFP_TEMPORARY);
 634        kbuf = (char *) page;
 635        if (!page)
 636                goto out;
 637
 638        /* Only allow <= page size writes at the beginning of the file */
 639        ret = -EINVAL;
 640        if ((*ppos != 0) || (count >= PAGE_SIZE))
 641                goto out;
 642
 643        /* Slurp in the user data */
 644        ret = -EFAULT;
 645        if (copy_from_user(kbuf, buf, count))
 646                goto out;
 647        kbuf[count] = '\0';
 648
 649        /* Parse the user data */
 650        ret = -EINVAL;
 651        pos = kbuf;
 652        new_map.nr_extents = 0;
 653        for (;pos; pos = next_line) {
 654                extent = &new_map.extent[new_map.nr_extents];
 655
 656                /* Find the end of line and ensure I don't look past it */
 657                next_line = strchr(pos, '\n');
 658                if (next_line) {
 659                        *next_line = '\0';
 660                        next_line++;
 661                        if (*next_line == '\0')
 662                                next_line = NULL;
 663                }
 664
 665                pos = skip_spaces(pos);
 666                extent->first = simple_strtoul(pos, &pos, 10);
 667                if (!isspace(*pos))
 668                        goto out;
 669
 670                pos = skip_spaces(pos);
 671                extent->lower_first = simple_strtoul(pos, &pos, 10);
 672                if (!isspace(*pos))
 673                        goto out;
 674
 675                pos = skip_spaces(pos);
 676                extent->count = simple_strtoul(pos, &pos, 10);
 677                if (*pos && !isspace(*pos))
 678                        goto out;
 679
 680                /* Verify there is not trailing junk on the line */
 681                pos = skip_spaces(pos);
 682                if (*pos != '\0')
 683                        goto out;
 684
 685                /* Verify we have been given valid starting values */
 686                if ((extent->first == (u32) -1) ||
 687                    (extent->lower_first == (u32) -1 ))
 688                        goto out;
 689
 690                /* Verify count is not zero and does not cause the extent to wrap */
 691                if ((extent->first + extent->count) <= extent->first)
 692                        goto out;
 693                if ((extent->lower_first + extent->count) <= extent->lower_first)
 694                        goto out;
 695
 696                /* Do the ranges in extent overlap any previous extents? */
 697                if (mappings_overlap(&new_map, extent))
 698                        goto out;
 699
 700                new_map.nr_extents++;
 701
 702                /* Fail if the file contains too many extents */
 703                if ((new_map.nr_extents == UID_GID_MAP_MAX_EXTENTS) &&
 704                    (next_line != NULL))
 705                        goto out;
 706        }
 707        /* Be very certaint the new map actually exists */
 708        if (new_map.nr_extents == 0)
 709                goto out;
 710
 711        ret = -EPERM;
 712        /* Validate the user is allowed to use user id's mapped to. */
 713        if (!new_idmap_permitted(file, ns, cap_setid, &new_map))
 714                goto out;
 715
 716        /* Map the lower ids from the parent user namespace to the
 717         * kernel global id space.
 718         */
 719        for (idx = 0; idx < new_map.nr_extents; idx++) {
 720                u32 lower_first;
 721                extent = &new_map.extent[idx];
 722
 723                lower_first = map_id_range_down(parent_map,
 724                                                extent->lower_first,
 725                                                extent->count);
 726
 727                /* Fail if we can not map the specified extent to
 728                 * the kernel global id space.
 729                 */
 730                if (lower_first == (u32) -1)
 731                        goto out;
 732
 733                extent->lower_first = lower_first;
 734        }
 735
 736        /* Install the map */
 737        memcpy(map->extent, new_map.extent,
 738                new_map.nr_extents*sizeof(new_map.extent[0]));
 739        smp_wmb();
 740        map->nr_extents = new_map.nr_extents;
 741
 742        *ppos = count;
 743        ret = count;
 744out:
 745        mutex_unlock(&id_map_mutex);
 746        if (page)
 747                free_page(page);
 748        return ret;
 749}
 750
 751ssize_t proc_uid_map_write(struct file *file, const char __user *buf, size_t size, loff_t *ppos)
 752{
 753        struct seq_file *seq = file->private_data;
 754        struct user_namespace *ns = seq->private;
 755        struct user_namespace *seq_ns = seq_user_ns(seq);
 756
 757        if (!ns->parent)
 758                return -EPERM;
 759
 760        if ((seq_ns != ns) && (seq_ns != ns->parent))
 761                return -EPERM;
 762
 763        return map_write(file, buf, size, ppos, CAP_SETUID,
 764                         &ns->uid_map, &ns->parent->uid_map);
 765}
 766
 767ssize_t proc_gid_map_write(struct file *file, const char __user *buf, size_t size, loff_t *ppos)
 768{
 769        struct seq_file *seq = file->private_data;
 770        struct user_namespace *ns = seq->private;
 771        struct user_namespace *seq_ns = seq_user_ns(seq);
 772
 773        if (!ns->parent)
 774                return -EPERM;
 775
 776        if ((seq_ns != ns) && (seq_ns != ns->parent))
 777                return -EPERM;
 778
 779        return map_write(file, buf, size, ppos, CAP_SETGID,
 780                         &ns->gid_map, &ns->parent->gid_map);
 781}
 782
 783ssize_t proc_projid_map_write(struct file *file, const char __user *buf, size_t size, loff_t *ppos)
 784{
 785        struct seq_file *seq = file->private_data;
 786        struct user_namespace *ns = seq->private;
 787        struct user_namespace *seq_ns = seq_user_ns(seq);
 788
 789        if (!ns->parent)
 790                return -EPERM;
 791
 792        if ((seq_ns != ns) && (seq_ns != ns->parent))
 793                return -EPERM;
 794
 795        /* Anyone can set any valid project id no capability needed */
 796        return map_write(file, buf, size, ppos, -1,
 797                         &ns->projid_map, &ns->parent->projid_map);
 798}
 799
 800static bool new_idmap_permitted(const struct file *file, 
 801                                struct user_namespace *ns, int cap_setid,
 802                                struct uid_gid_map *new_map)
 803{
 804        /* Allow mapping to your own filesystem ids */
 805        if ((new_map->nr_extents == 1) && (new_map->extent[0].count == 1)) {
 806                u32 id = new_map->extent[0].lower_first;
 807                if (cap_setid == CAP_SETUID) {
 808                        kuid_t uid = make_kuid(ns->parent, id);
 809                        if (uid_eq(uid, file->f_cred->fsuid))
 810                                return true;
 811                }
 812                else if (cap_setid == CAP_SETGID) {
 813                        kgid_t gid = make_kgid(ns->parent, id);
 814                        if (gid_eq(gid, file->f_cred->fsgid))
 815                                return true;
 816                }
 817        }
 818
 819        /* Allow anyone to set a mapping that doesn't require privilege */
 820        if (!cap_valid(cap_setid))
 821                return true;
 822
 823        /* Allow the specified ids if we have the appropriate capability
 824         * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
 825         * And the opener of the id file also had the approprpiate capability.
 826         */
 827        if (ns_capable(ns->parent, cap_setid) &&
 828            file_ns_capable(file, ns->parent, cap_setid))
 829                return true;
 830
 831        return false;
 832}
 833
 834static void *userns_get(struct task_struct *task)
 835{
 836        struct user_namespace *user_ns;
 837
 838        rcu_read_lock();
 839        user_ns = get_user_ns(__task_cred(task)->user_ns);
 840        rcu_read_unlock();
 841
 842        return user_ns;
 843}
 844
 845static void userns_put(void *ns)
 846{
 847        put_user_ns(ns);
 848}
 849
 850static int userns_install(struct nsproxy *nsproxy, void *ns)
 851{
 852        struct user_namespace *user_ns = ns;
 853        struct cred *cred;
 854
 855        /* Don't allow gaining capabilities by reentering
 856         * the same user namespace.
 857         */
 858        if (user_ns == current_user_ns())
 859                return -EINVAL;
 860
 861        /* Threaded processes may not enter a different user namespace */
 862        if (atomic_read(&current->mm->mm_users) > 1)
 863                return -EINVAL;
 864
 865        if (current->fs->users != 1)
 866                return -EINVAL;
 867
 868        if (!ns_capable(user_ns, CAP_SYS_ADMIN))
 869                return -EPERM;
 870
 871        cred = prepare_creds();
 872        if (!cred)
 873                return -ENOMEM;
 874
 875        put_user_ns(cred->user_ns);
 876        set_cred_user_ns(cred, get_user_ns(user_ns));
 877
 878        return commit_creds(cred);
 879}
 880
 881static unsigned int userns_inum(void *ns)
 882{
 883        struct user_namespace *user_ns = ns;
 884        return user_ns->proc_inum;
 885}
 886
 887const struct proc_ns_operations userns_operations = {
 888        .name           = "user",
 889        .type           = CLONE_NEWUSER,
 890        .get            = userns_get,
 891        .put            = userns_put,
 892        .install        = userns_install,
 893        .inum           = userns_inum,
 894};
 895
 896static __init int user_namespaces_init(void)
 897{
 898        user_ns_cachep = KMEM_CACHE(user_namespace, SLAB_PANIC);
 899        return 0;
 900}
 901module_init(user_namespaces_init);
 902
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.