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