linux/ipc/shm.c
<<
>>
Prefs
   1/*
   2 * linux/ipc/shm.c
   3 * Copyright (C) 1992, 1993 Krishna Balasubramanian
   4 *       Many improvements/fixes by Bruno Haible.
   5 * Replaced `struct shm_desc' by `struct vm_area_struct', July 1994.
   6 * Fixed the shm swap deallocation (shm_unuse()), August 1998 Andrea Arcangeli.
   7 *
   8 * /proc/sysvipc/shm support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
   9 * BIGMEM support, Andrea Arcangeli <andrea@suse.de>
  10 * SMP thread shm, Jean-Luc Boyard <jean-luc.boyard@siemens.fr>
  11 * HIGHMEM support, Ingo Molnar <mingo@redhat.com>
  12 * Make shmmax, shmall, shmmni sysctl'able, Christoph Rohland <cr@sap.com>
  13 * Shared /dev/zero support, Kanoj Sarcar <kanoj@sgi.com>
  14 * Move the mm functionality over to mm/shmem.c, Christoph Rohland <cr@sap.com>
  15 *
  16 * support for audit of ipc object properties and permission changes
  17 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
  18 *
  19 * namespaces support
  20 * OpenVZ, SWsoft Inc.
  21 * Pavel Emelianov <xemul@openvz.org>
  22 */
  23
  24#include <linux/slab.h>
  25#include <linux/mm.h>
  26#include <linux/hugetlb.h>
  27#include <linux/shm.h>
  28#include <linux/init.h>
  29#include <linux/file.h>
  30#include <linux/mman.h>
  31#include <linux/shmem_fs.h>
  32#include <linux/security.h>
  33#include <linux/syscalls.h>
  34#include <linux/audit.h>
  35#include <linux/capability.h>
  36#include <linux/ptrace.h>
  37#include <linux/seq_file.h>
  38#include <linux/rwsem.h>
  39#include <linux/nsproxy.h>
  40#include <linux/mount.h>
  41#include <linux/ipc_namespace.h>
  42
  43#include <asm/uaccess.h>
  44
  45#include "util.h"
  46
  47struct shm_file_data {
  48        int id;
  49        struct ipc_namespace *ns;
  50        struct file *file;
  51        const struct vm_operations_struct *vm_ops;
  52};
  53
  54#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  55
  56static const struct file_operations shm_file_operations;
  57static const struct vm_operations_struct shm_vm_ops;
  58
  59#define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  60
  61#define shm_unlock(shp)                 \
  62        ipc_unlock(&(shp)->shm_perm)
  63
  64static int newseg(struct ipc_namespace *, struct ipc_params *);
  65static void shm_open(struct vm_area_struct *vma);
  66static void shm_close(struct vm_area_struct *vma);
  67static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  68#ifdef CONFIG_PROC_FS
  69static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  70#endif
  71
  72void shm_init_ns(struct ipc_namespace *ns)
  73{
  74        ns->shm_ctlmax = SHMMAX;
  75        ns->shm_ctlall = SHMALL;
  76        ns->shm_ctlmni = SHMMNI;
  77        ns->shm_rmid_forced = 0;
  78        ns->shm_tot = 0;
  79        ipc_init_ids(&shm_ids(ns));
  80}
  81
  82/*
  83 * Called with shm_ids.rw_mutex (writer) and the shp structure locked.
  84 * Only shm_ids.rw_mutex remains locked on exit.
  85 */
  86static void do_shm_rmid(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
  87{
  88        struct shmid_kernel *shp;
  89        shp = container_of(ipcp, struct shmid_kernel, shm_perm);
  90
  91        if (shp->shm_nattch){
  92                shp->shm_perm.mode |= SHM_DEST;
  93                /* Do not find it any more */
  94                shp->shm_perm.key = IPC_PRIVATE;
  95                shm_unlock(shp);
  96        } else
  97                shm_destroy(ns, shp);
  98}
  99
 100#ifdef CONFIG_IPC_NS
 101void shm_exit_ns(struct ipc_namespace *ns)
 102{
 103        free_ipcs(ns, &shm_ids(ns), do_shm_rmid);
 104        idr_destroy(&ns->ids[IPC_SHM_IDS].ipcs_idr);
 105}
 106#endif
 107
 108static int __init ipc_ns_init(void)
 109{
 110        shm_init_ns(&init_ipc_ns);
 111        return 0;
 112}
 113
 114pure_initcall(ipc_ns_init);
 115
 116void __init shm_init (void)
 117{
 118        ipc_init_proc_interface("sysvipc/shm",
 119#if BITS_PER_LONG <= 32
 120                                "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime        rss       swap\n",
 121#else
 122                                "       key      shmid perms                  size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime                   rss                  swap\n",
 123#endif
 124                                IPC_SHM_IDS, sysvipc_shm_proc_show);
 125}
 126
 127/*
 128 * shm_lock_(check_) routines are called in the paths where the rw_mutex
 129 * is not necessarily held.
 130 */
 131static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 132{
 133        struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 134
 135        if (IS_ERR(ipcp))
 136                return (struct shmid_kernel *)ipcp;
 137
 138        return container_of(ipcp, struct shmid_kernel, shm_perm);
 139}
 140
 141static inline void shm_lock_by_ptr(struct shmid_kernel *ipcp)
 142{
 143        rcu_read_lock();
 144        spin_lock(&ipcp->shm_perm.lock);
 145}
 146
 147static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 148                                                int id)
 149{
 150        struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 151
 152        if (IS_ERR(ipcp))
 153                return (struct shmid_kernel *)ipcp;
 154
 155        return container_of(ipcp, struct shmid_kernel, shm_perm);
 156}
 157
 158static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 159{
 160        ipc_rmid(&shm_ids(ns), &s->shm_perm);
 161}
 162
 163
 164/* This is called by fork, once for every shm attach. */
 165static void shm_open(struct vm_area_struct *vma)
 166{
 167        struct file *file = vma->vm_file;
 168        struct shm_file_data *sfd = shm_file_data(file);
 169        struct shmid_kernel *shp;
 170
 171        shp = shm_lock(sfd->ns, sfd->id);
 172        BUG_ON(IS_ERR(shp));
 173        shp->shm_atim = get_seconds();
 174        shp->shm_lprid = task_tgid_vnr(current);
 175        shp->shm_nattch++;
 176        shm_unlock(shp);
 177}
 178
 179/*
 180 * shm_destroy - free the struct shmid_kernel
 181 *
 182 * @ns: namespace
 183 * @shp: struct to free
 184 *
 185 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 186 * but returns with shp unlocked and freed.
 187 */
 188static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 189{
 190        ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 191        shm_rmid(ns, shp);
 192        shm_unlock(shp);
 193        if (!is_file_hugepages(shp->shm_file))
 194                shmem_lock(shp->shm_file, 0, shp->mlock_user);
 195        else if (shp->mlock_user)
 196                user_shm_unlock(file_inode(shp->shm_file)->i_size,
 197                                                shp->mlock_user);
 198        fput (shp->shm_file);
 199        security_shm_free(shp);
 200        ipc_rcu_putref(shp);
 201}
 202
 203/*
 204 * shm_may_destroy - identifies whether shm segment should be destroyed now
 205 *
 206 * Returns true if and only if there are no active users of the segment and
 207 * one of the following is true:
 208 *
 209 * 1) shmctl(id, IPC_RMID, NULL) was called for this shp
 210 *
 211 * 2) sysctl kernel.shm_rmid_forced is set to 1.
 212 */
 213static bool shm_may_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 214{
 215        return (shp->shm_nattch == 0) &&
 216               (ns->shm_rmid_forced ||
 217                (shp->shm_perm.mode & SHM_DEST));
 218}
 219
 220/*
 221 * remove the attach descriptor vma.
 222 * free memory for segment if it is marked destroyed.
 223 * The descriptor has already been removed from the current->mm->mmap list
 224 * and will later be kfree()d.
 225 */
 226static void shm_close(struct vm_area_struct *vma)
 227{
 228        struct file * file = vma->vm_file;
 229        struct shm_file_data *sfd = shm_file_data(file);
 230        struct shmid_kernel *shp;
 231        struct ipc_namespace *ns = sfd->ns;
 232
 233        down_write(&shm_ids(ns).rw_mutex);
 234        /* remove from the list of attaches of the shm segment */
 235        shp = shm_lock(ns, sfd->id);
 236        BUG_ON(IS_ERR(shp));
 237        shp->shm_lprid = task_tgid_vnr(current);
 238        shp->shm_dtim = get_seconds();
 239        shp->shm_nattch--;
 240        if (shm_may_destroy(ns, shp))
 241                shm_destroy(ns, shp);
 242        else
 243                shm_unlock(shp);
 244        up_write(&shm_ids(ns).rw_mutex);
 245}
 246
 247/* Called with ns->shm_ids(ns).rw_mutex locked */
 248static int shm_try_destroy_current(int id, void *p, void *data)
 249{
 250        struct ipc_namespace *ns = data;
 251        struct kern_ipc_perm *ipcp = p;
 252        struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 253
 254        if (shp->shm_creator != current)
 255                return 0;
 256
 257        /*
 258         * Mark it as orphaned to destroy the segment when
 259         * kernel.shm_rmid_forced is changed.
 260         * It is noop if the following shm_may_destroy() returns true.
 261         */
 262        shp->shm_creator = NULL;
 263
 264        /*
 265         * Don't even try to destroy it.  If shm_rmid_forced=0 and IPC_RMID
 266         * is not set, it shouldn't be deleted here.
 267         */
 268        if (!ns->shm_rmid_forced)
 269                return 0;
 270
 271        if (shm_may_destroy(ns, shp)) {
 272                shm_lock_by_ptr(shp);
 273                shm_destroy(ns, shp);
 274        }
 275        return 0;
 276}
 277
 278/* Called with ns->shm_ids(ns).rw_mutex locked */
 279static int shm_try_destroy_orphaned(int id, void *p, void *data)
 280{
 281        struct ipc_namespace *ns = data;
 282        struct kern_ipc_perm *ipcp = p;
 283        struct shmid_kernel *shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 284
 285        /*
 286         * We want to destroy segments without users and with already
 287         * exit'ed originating process.
 288         *
 289         * As shp->* are changed under rw_mutex, it's safe to skip shp locking.
 290         */
 291        if (shp->shm_creator != NULL)
 292                return 0;
 293
 294        if (shm_may_destroy(ns, shp)) {
 295                shm_lock_by_ptr(shp);
 296                shm_destroy(ns, shp);
 297        }
 298        return 0;
 299}
 300
 301void shm_destroy_orphaned(struct ipc_namespace *ns)
 302{
 303        down_write(&shm_ids(ns).rw_mutex);
 304        if (shm_ids(ns).in_use)
 305                idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_orphaned, ns);
 306        up_write(&shm_ids(ns).rw_mutex);
 307}
 308
 309
 310void exit_shm(struct task_struct *task)
 311{
 312        struct ipc_namespace *ns = task->nsproxy->ipc_ns;
 313
 314        if (shm_ids(ns).in_use == 0)
 315                return;
 316
 317        /* Destroy all already created segments, but not mapped yet */
 318        down_write(&shm_ids(ns).rw_mutex);
 319        if (shm_ids(ns).in_use)
 320                idr_for_each(&shm_ids(ns).ipcs_idr, &shm_try_destroy_current, ns);
 321        up_write(&shm_ids(ns).rw_mutex);
 322}
 323
 324static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 325{
 326        struct file *file = vma->vm_file;
 327        struct shm_file_data *sfd = shm_file_data(file);
 328
 329        return sfd->vm_ops->fault(vma, vmf);
 330}
 331
 332#ifdef CONFIG_NUMA
 333static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 334{
 335        struct file *file = vma->vm_file;
 336        struct shm_file_data *sfd = shm_file_data(file);
 337        int err = 0;
 338        if (sfd->vm_ops->set_policy)
 339                err = sfd->vm_ops->set_policy(vma, new);
 340        return err;
 341}
 342
 343static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 344                                        unsigned long addr)
 345{
 346        struct file *file = vma->vm_file;
 347        struct shm_file_data *sfd = shm_file_data(file);
 348        struct mempolicy *pol = NULL;
 349
 350        if (sfd->vm_ops->get_policy)
 351                pol = sfd->vm_ops->get_policy(vma, addr);
 352        else if (vma->vm_policy)
 353                pol = vma->vm_policy;
 354
 355        return pol;
 356}
 357#endif
 358
 359static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 360{
 361        struct shm_file_data *sfd = shm_file_data(file);
 362        int ret;
 363
 364        ret = sfd->file->f_op->mmap(sfd->file, vma);
 365        if (ret != 0)
 366                return ret;
 367        sfd->vm_ops = vma->vm_ops;
 368#ifdef CONFIG_MMU
 369        BUG_ON(!sfd->vm_ops->fault);
 370#endif
 371        vma->vm_ops = &shm_vm_ops;
 372        shm_open(vma);
 373
 374        return ret;
 375}
 376
 377static int shm_release(struct inode *ino, struct file *file)
 378{
 379        struct shm_file_data *sfd = shm_file_data(file);
 380
 381        put_ipc_ns(sfd->ns);
 382        shm_file_data(file) = NULL;
 383        kfree(sfd);
 384        return 0;
 385}
 386
 387static int shm_fsync(struct file *file, loff_t start, loff_t end, int datasync)
 388{
 389        struct shm_file_data *sfd = shm_file_data(file);
 390
 391        if (!sfd->file->f_op->fsync)
 392                return -EINVAL;
 393        return sfd->file->f_op->fsync(sfd->file, start, end, datasync);
 394}
 395
 396static long shm_fallocate(struct file *file, int mode, loff_t offset,
 397                          loff_t len)
 398{
 399        struct shm_file_data *sfd = shm_file_data(file);
 400
 401        if (!sfd->file->f_op->fallocate)
 402                return -EOPNOTSUPP;
 403        return sfd->file->f_op->fallocate(file, mode, offset, len);
 404}
 405
 406static unsigned long shm_get_unmapped_area(struct file *file,
 407        unsigned long addr, unsigned long len, unsigned long pgoff,
 408        unsigned long flags)
 409{
 410        struct shm_file_data *sfd = shm_file_data(file);
 411        return sfd->file->f_op->get_unmapped_area(sfd->file, addr, len,
 412                                                pgoff, flags);
 413}
 414
 415static const struct file_operations shm_file_operations = {
 416        .mmap           = shm_mmap,
 417        .fsync          = shm_fsync,
 418        .release        = shm_release,
 419#ifndef CONFIG_MMU
 420        .get_unmapped_area      = shm_get_unmapped_area,
 421#endif
 422        .llseek         = noop_llseek,
 423        .fallocate      = shm_fallocate,
 424};
 425
 426static const struct file_operations shm_file_operations_huge = {
 427        .mmap           = shm_mmap,
 428        .fsync          = shm_fsync,
 429        .release        = shm_release,
 430        .get_unmapped_area      = shm_get_unmapped_area,
 431        .llseek         = noop_llseek,
 432        .fallocate      = shm_fallocate,
 433};
 434
 435int is_file_shm_hugepages(struct file *file)
 436{
 437        return file->f_op == &shm_file_operations_huge;
 438}
 439
 440static const struct vm_operations_struct shm_vm_ops = {
 441        .open   = shm_open,     /* callback for a new vm-area open */
 442        .close  = shm_close,    /* callback for when the vm-area is released */
 443        .fault  = shm_fault,
 444#if defined(CONFIG_NUMA)
 445        .set_policy = shm_set_policy,
 446        .get_policy = shm_get_policy,
 447#endif
 448};
 449
 450/**
 451 * newseg - Create a new shared memory segment
 452 * @ns: namespace
 453 * @params: ptr to the structure that contains key, size and shmflg
 454 *
 455 * Called with shm_ids.rw_mutex held as a writer.
 456 */
 457
 458static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 459{
 460        key_t key = params->key;
 461        int shmflg = params->flg;
 462        size_t size = params->u.size;
 463        int error;
 464        struct shmid_kernel *shp;
 465        size_t numpages = (size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 466        struct file * file;
 467        char name[13];
 468        int id;
 469        vm_flags_t acctflag = 0;
 470
 471        if (size < SHMMIN || size > ns->shm_ctlmax)
 472                return -EINVAL;
 473
 474        if (ns->shm_tot + numpages > ns->shm_ctlall)
 475                return -ENOSPC;
 476
 477        shp = ipc_rcu_alloc(sizeof(*shp));
 478        if (!shp)
 479                return -ENOMEM;
 480
 481        shp->shm_perm.key = key;
 482        shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 483        shp->mlock_user = NULL;
 484
 485        shp->shm_perm.security = NULL;
 486        error = security_shm_alloc(shp);
 487        if (error) {
 488                ipc_rcu_putref(shp);
 489                return error;
 490        }
 491
 492        sprintf (name, "SYSV%08x", key);
 493        if (shmflg & SHM_HUGETLB) {
 494                struct hstate *hs = hstate_sizelog((shmflg >> SHM_HUGE_SHIFT)
 495                                                & SHM_HUGE_MASK);
 496                size_t hugesize;
 497
 498                if (!hs) {
 499                        error = -EINVAL;
 500                        goto no_file;
 501                }
 502                hugesize = ALIGN(size, huge_page_size(hs));
 503
 504                /* hugetlb_file_setup applies strict accounting */
 505                if (shmflg & SHM_NORESERVE)
 506                        acctflag = VM_NORESERVE;
 507                file = hugetlb_file_setup(name, hugesize, acctflag,
 508                                  &shp->mlock_user, HUGETLB_SHMFS_INODE,
 509                                (shmflg >> SHM_HUGE_SHIFT) & SHM_HUGE_MASK);
 510        } else {
 511                /*
 512                 * Do not allow no accounting for OVERCOMMIT_NEVER, even
 513                 * if it's asked for.
 514                 */
 515                if  ((shmflg & SHM_NORESERVE) &&
 516                                sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 517                        acctflag = VM_NORESERVE;
 518                file = shmem_file_setup(name, size, acctflag);
 519        }
 520        error = PTR_ERR(file);
 521        if (IS_ERR(file))
 522                goto no_file;
 523
 524        id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 525        if (id < 0) {
 526                error = id;
 527                goto no_id;
 528        }
 529
 530        shp->shm_cprid = task_tgid_vnr(current);
 531        shp->shm_lprid = 0;
 532        shp->shm_atim = shp->shm_dtim = 0;
 533        shp->shm_ctim = get_seconds();
 534        shp->shm_segsz = size;
 535        shp->shm_nattch = 0;
 536        shp->shm_file = file;
 537        shp->shm_creator = current;
 538        /*
 539         * shmid gets reported as "inode#" in /proc/pid/maps.
 540         * proc-ps tools use this. Changing this will break them.
 541         */
 542        file_inode(file)->i_ino = shp->shm_perm.id;
 543
 544        ns->shm_tot += numpages;
 545        error = shp->shm_perm.id;
 546        shm_unlock(shp);
 547        return error;
 548
 549no_id:
 550        if (is_file_hugepages(file) && shp->mlock_user)
 551                user_shm_unlock(size, shp->mlock_user);
 552        fput(file);
 553no_file:
 554        security_shm_free(shp);
 555        ipc_rcu_putref(shp);
 556        return error;
 557}
 558
 559/*
 560 * Called with shm_ids.rw_mutex and ipcp locked.
 561 */
 562static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 563{
 564        struct shmid_kernel *shp;
 565
 566        shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 567        return security_shm_associate(shp, shmflg);
 568}
 569
 570/*
 571 * Called with shm_ids.rw_mutex and ipcp locked.
 572 */
 573static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 574                                struct ipc_params *params)
 575{
 576        struct shmid_kernel *shp;
 577
 578        shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 579        if (shp->shm_segsz < params->u.size)
 580                return -EINVAL;
 581
 582        return 0;
 583}
 584
 585SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 586{
 587        struct ipc_namespace *ns;
 588        struct ipc_ops shm_ops;
 589        struct ipc_params shm_params;
 590
 591        ns = current->nsproxy->ipc_ns;
 592
 593        shm_ops.getnew = newseg;
 594        shm_ops.associate = shm_security;
 595        shm_ops.more_checks = shm_more_checks;
 596
 597        shm_params.key = key;
 598        shm_params.flg = shmflg;
 599        shm_params.u.size = size;
 600
 601        return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 602}
 603
 604static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 605{
 606        switch(version) {
 607        case IPC_64:
 608                return copy_to_user(buf, in, sizeof(*in));
 609        case IPC_OLD:
 610            {
 611                struct shmid_ds out;
 612
 613                memset(&out, 0, sizeof(out));
 614                ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 615                out.shm_segsz   = in->shm_segsz;
 616                out.shm_atime   = in->shm_atime;
 617                out.shm_dtime   = in->shm_dtime;
 618                out.shm_ctime   = in->shm_ctime;
 619                out.shm_cpid    = in->shm_cpid;
 620                out.shm_lpid    = in->shm_lpid;
 621                out.shm_nattch  = in->shm_nattch;
 622
 623                return copy_to_user(buf, &out, sizeof(out));
 624            }
 625        default:
 626                return -EINVAL;
 627        }
 628}
 629
 630static inline unsigned long
 631copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 632{
 633        switch(version) {
 634        case IPC_64:
 635                if (copy_from_user(out, buf, sizeof(*out)))
 636                        return -EFAULT;
 637                return 0;
 638        case IPC_OLD:
 639            {
 640                struct shmid_ds tbuf_old;
 641
 642                if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 643                        return -EFAULT;
 644
 645                out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 646                out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 647                out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 648
 649                return 0;
 650            }
 651        default:
 652                return -EINVAL;
 653        }
 654}
 655
 656static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 657{
 658        switch(version) {
 659        case IPC_64:
 660                return copy_to_user(buf, in, sizeof(*in));
 661        case IPC_OLD:
 662            {
 663                struct shminfo out;
 664
 665                if(in->shmmax > INT_MAX)
 666                        out.shmmax = INT_MAX;
 667                else
 668                        out.shmmax = (int)in->shmmax;
 669
 670                out.shmmin      = in->shmmin;
 671                out.shmmni      = in->shmmni;
 672                out.shmseg      = in->shmseg;
 673                out.shmall      = in->shmall; 
 674
 675                return copy_to_user(buf, &out, sizeof(out));
 676            }
 677        default:
 678                return -EINVAL;
 679        }
 680}
 681
 682/*
 683 * Calculate and add used RSS and swap pages of a shm.
 684 * Called with shm_ids.rw_mutex held as a reader
 685 */
 686static void shm_add_rss_swap(struct shmid_kernel *shp,
 687        unsigned long *rss_add, unsigned long *swp_add)
 688{
 689        struct inode *inode;
 690
 691        inode = file_inode(shp->shm_file);
 692
 693        if (is_file_hugepages(shp->shm_file)) {
 694                struct address_space *mapping = inode->i_mapping;
 695                struct hstate *h = hstate_file(shp->shm_file);
 696                *rss_add += pages_per_huge_page(h) * mapping->nrpages;
 697        } else {
 698#ifdef CONFIG_SHMEM
 699                struct shmem_inode_info *info = SHMEM_I(inode);
 700                spin_lock(&info->lock);
 701                *rss_add += inode->i_mapping->nrpages;
 702                *swp_add += info->swapped;
 703                spin_unlock(&info->lock);
 704#else
 705                *rss_add += inode->i_mapping->nrpages;
 706#endif
 707        }
 708}
 709
 710/*
 711 * Called with shm_ids.rw_mutex held as a reader
 712 */
 713static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 714                unsigned long *swp)
 715{
 716        int next_id;
 717        int total, in_use;
 718
 719        *rss = 0;
 720        *swp = 0;
 721
 722        in_use = shm_ids(ns).in_use;
 723
 724        for (total = 0, next_id = 0; total < in_use; next_id++) {
 725                struct kern_ipc_perm *ipc;
 726                struct shmid_kernel *shp;
 727
 728                ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 729                if (ipc == NULL)
 730                        continue;
 731                shp = container_of(ipc, struct shmid_kernel, shm_perm);
 732
 733                shm_add_rss_swap(shp, rss, swp);
 734
 735                total++;
 736        }
 737}
 738
 739/*
 740 * This function handles some shmctl commands which require the rw_mutex
 741 * to be held in write mode.
 742 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 743 */
 744static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 745                       struct shmid_ds __user *buf, int version)
 746{
 747        struct kern_ipc_perm *ipcp;
 748        struct shmid64_ds shmid64;
 749        struct shmid_kernel *shp;
 750        int err;
 751
 752        if (cmd == IPC_SET) {
 753                if (copy_shmid_from_user(&shmid64, buf, version))
 754                        return -EFAULT;
 755        }
 756
 757        ipcp = ipcctl_pre_down(ns, &shm_ids(ns), shmid, cmd,
 758                               &shmid64.shm_perm, 0);
 759        if (IS_ERR(ipcp))
 760                return PTR_ERR(ipcp);
 761
 762        shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 763
 764        err = security_shm_shmctl(shp, cmd);
 765        if (err)
 766                goto out_unlock;
 767        switch (cmd) {
 768        case IPC_RMID:
 769                do_shm_rmid(ns, ipcp);
 770                goto out_up;
 771        case IPC_SET:
 772                err = ipc_update_perm(&shmid64.shm_perm, ipcp);
 773                if (err)
 774                        goto out_unlock;
 775                shp->shm_ctim = get_seconds();
 776                break;
 777        default:
 778                err = -EINVAL;
 779        }
 780out_unlock:
 781        shm_unlock(shp);
 782out_up:
 783        up_write(&shm_ids(ns).rw_mutex);
 784        return err;
 785}
 786
 787SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 788{
 789        struct shmid_kernel *shp;
 790        int err, version;
 791        struct ipc_namespace *ns;
 792
 793        if (cmd < 0 || shmid < 0) {
 794                err = -EINVAL;
 795                goto out;
 796        }
 797
 798        version = ipc_parse_version(&cmd);
 799        ns = current->nsproxy->ipc_ns;
 800
 801        switch (cmd) { /* replace with proc interface ? */
 802        case IPC_INFO:
 803        {
 804                struct shminfo64 shminfo;
 805
 806                err = security_shm_shmctl(NULL, cmd);
 807                if (err)
 808                        return err;
 809
 810                memset(&shminfo, 0, sizeof(shminfo));
 811                shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 812                shminfo.shmmax = ns->shm_ctlmax;
 813                shminfo.shmall = ns->shm_ctlall;
 814
 815                shminfo.shmmin = SHMMIN;
 816                if(copy_shminfo_to_user (buf, &shminfo, version))
 817                        return -EFAULT;
 818
 819                down_read(&shm_ids(ns).rw_mutex);
 820                err = ipc_get_maxid(&shm_ids(ns));
 821                up_read(&shm_ids(ns).rw_mutex);
 822
 823                if(err<0)
 824                        err = 0;
 825                goto out;
 826        }
 827        case SHM_INFO:
 828        {
 829                struct shm_info shm_info;
 830
 831                err = security_shm_shmctl(NULL, cmd);
 832                if (err)
 833                        return err;
 834
 835                memset(&shm_info, 0, sizeof(shm_info));
 836                down_read(&shm_ids(ns).rw_mutex);
 837                shm_info.used_ids = shm_ids(ns).in_use;
 838                shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 839                shm_info.shm_tot = ns->shm_tot;
 840                shm_info.swap_attempts = 0;
 841                shm_info.swap_successes = 0;
 842                err = ipc_get_maxid(&shm_ids(ns));
 843                up_read(&shm_ids(ns).rw_mutex);
 844                if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 845                        err = -EFAULT;
 846                        goto out;
 847                }
 848
 849                err = err < 0 ? 0 : err;
 850                goto out;
 851        }
 852        case SHM_STAT:
 853        case IPC_STAT:
 854        {
 855                struct shmid64_ds tbuf;
 856                int result;
 857
 858                if (cmd == SHM_STAT) {
 859                        shp = shm_lock(ns, shmid);
 860                        if (IS_ERR(shp)) {
 861                                err = PTR_ERR(shp);
 862                                goto out;
 863                        }
 864                        result = shp->shm_perm.id;
 865                } else {
 866                        shp = shm_lock_check(ns, shmid);
 867                        if (IS_ERR(shp)) {
 868                                err = PTR_ERR(shp);
 869                                goto out;
 870                        }
 871                        result = 0;
 872                }
 873                err = -EACCES;
 874                if (ipcperms(ns, &shp->shm_perm, S_IRUGO))
 875                        goto out_unlock;
 876                err = security_shm_shmctl(shp, cmd);
 877                if (err)
 878                        goto out_unlock;
 879                memset(&tbuf, 0, sizeof(tbuf));
 880                kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 881                tbuf.shm_segsz  = shp->shm_segsz;
 882                tbuf.shm_atime  = shp->shm_atim;
 883                tbuf.shm_dtime  = shp->shm_dtim;
 884                tbuf.shm_ctime  = shp->shm_ctim;
 885                tbuf.shm_cpid   = shp->shm_cprid;
 886                tbuf.shm_lpid   = shp->shm_lprid;
 887                tbuf.shm_nattch = shp->shm_nattch;
 888                shm_unlock(shp);
 889                if(copy_shmid_to_user (buf, &tbuf, version))
 890                        err = -EFAULT;
 891                else
 892                        err = result;
 893                goto out;
 894        }
 895        case SHM_LOCK:
 896        case SHM_UNLOCK:
 897        {
 898                struct file *shm_file;
 899
 900                shp = shm_lock_check(ns, shmid);
 901                if (IS_ERR(shp)) {
 902                        err = PTR_ERR(shp);
 903                        goto out;
 904                }
 905
 906                audit_ipc_obj(&(shp->shm_perm));
 907
 908                if (!ns_capable(ns->user_ns, CAP_IPC_LOCK)) {
 909                        kuid_t euid = current_euid();
 910                        err = -EPERM;
 911                        if (!uid_eq(euid, shp->shm_perm.uid) &&
 912                            !uid_eq(euid, shp->shm_perm.cuid))
 913                                goto out_unlock;
 914                        if (cmd == SHM_LOCK && !rlimit(RLIMIT_MEMLOCK))
 915                                goto out_unlock;
 916                }
 917
 918                err = security_shm_shmctl(shp, cmd);
 919                if (err)
 920                        goto out_unlock;
 921
 922                shm_file = shp->shm_file;
 923                if (is_file_hugepages(shm_file))
 924                        goto out_unlock;
 925
 926                if (cmd == SHM_LOCK) {
 927                        struct user_struct *user = current_user();
 928                        err = shmem_lock(shm_file, 1, user);
 929                        if (!err && !(shp->shm_perm.mode & SHM_LOCKED)) {
 930                                shp->shm_perm.mode |= SHM_LOCKED;
 931                                shp->mlock_user = user;
 932                        }
 933                        goto out_unlock;
 934                }
 935
 936                /* SHM_UNLOCK */
 937                if (!(shp->shm_perm.mode & SHM_LOCKED))
 938                        goto out_unlock;
 939                shmem_lock(shm_file, 0, shp->mlock_user);
 940                shp->shm_perm.mode &= ~SHM_LOCKED;
 941                shp->mlock_user = NULL;
 942                get_file(shm_file);
 943                shm_unlock(shp);
 944                shmem_unlock_mapping(shm_file->f_mapping);
 945                fput(shm_file);
 946                goto out;
 947        }
 948        case IPC_RMID:
 949        case IPC_SET:
 950                err = shmctl_down(ns, shmid, cmd, buf, version);
 951                return err;
 952        default:
 953                return -EINVAL;
 954        }
 955
 956out_unlock:
 957        shm_unlock(shp);
 958out:
 959        return err;
 960}
 961
 962/*
 963 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 964 *
 965 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 966 * "raddr" thing points to kernel space, and there has to be a wrapper around
 967 * this.
 968 */
 969long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr,
 970              unsigned long shmlba)
 971{
 972        struct shmid_kernel *shp;
 973        unsigned long addr;
 974        unsigned long size;
 975        struct file * file;
 976        int    err;
 977        unsigned long flags;
 978        unsigned long prot;
 979        int acc_mode;
 980        struct ipc_namespace *ns;
 981        struct shm_file_data *sfd;
 982        struct path path;
 983        fmode_t f_mode;
 984        unsigned long populate = 0;
 985
 986        err = -EINVAL;
 987        if (shmid < 0)
 988                goto out;
 989        else if ((addr = (ulong)shmaddr)) {
 990                if (addr & (shmlba - 1)) {
 991                        if (shmflg & SHM_RND)
 992                                addr &= ~(shmlba - 1);     /* round down */
 993                        else
 994#ifndef __ARCH_FORCE_SHMLBA
 995                                if (addr & ~PAGE_MASK)
 996#endif
 997                                        goto out;
 998                }
 999                flags = MAP_SHARED | MAP_FIXED;
1000        } else {
1001                if ((shmflg & SHM_REMAP))
1002                        goto out;
1003
1004                flags = MAP_SHARED;
1005        }
1006
1007        if (shmflg & SHM_RDONLY) {
1008                prot = PROT_READ;
1009                acc_mode = S_IRUGO;
1010                f_mode = FMODE_READ;
1011        } else {
1012                prot = PROT_READ | PROT_WRITE;
1013                acc_mode = S_IRUGO | S_IWUGO;
1014                f_mode = FMODE_READ | FMODE_WRITE;
1015        }
1016        if (shmflg & SHM_EXEC) {
1017                prot |= PROT_EXEC;
1018                acc_mode |= S_IXUGO;
1019        }
1020
1021        /*
1022         * We cannot rely on the fs check since SYSV IPC does have an
1023         * additional creator id...
1024         */
1025        ns = current->nsproxy->ipc_ns;
1026        shp = shm_lock_check(ns, shmid);
1027        if (IS_ERR(shp)) {
1028                err = PTR_ERR(shp);
1029                goto out;
1030        }
1031
1032        err = -EACCES;
1033        if (ipcperms(ns, &shp->shm_perm, acc_mode))
1034                goto out_unlock;
1035
1036        err = security_shm_shmat(shp, shmaddr, shmflg);
1037        if (err)
1038                goto out_unlock;
1039
1040        path = shp->shm_file->f_path;
1041        path_get(&path);
1042        shp->shm_nattch++;
1043        size = i_size_read(path.dentry->d_inode);
1044        shm_unlock(shp);
1045
1046        err = -ENOMEM;
1047        sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
1048        if (!sfd)
1049                goto out_put_dentry;
1050
1051        file = alloc_file(&path, f_mode,
1052                          is_file_hugepages(shp->shm_file) ?
1053                                &shm_file_operations_huge :
1054                                &shm_file_operations);
1055        err = PTR_ERR(file);
1056        if (IS_ERR(file))
1057                goto out_free;
1058
1059        file->private_data = sfd;
1060        file->f_mapping = shp->shm_file->f_mapping;
1061        sfd->id = shp->shm_perm.id;
1062        sfd->ns = get_ipc_ns(ns);
1063        sfd->file = shp->shm_file;
1064        sfd->vm_ops = NULL;
1065
1066        err = security_mmap_file(file, prot, flags);
1067        if (err)
1068                goto out_fput;
1069
1070        down_write(&current->mm->mmap_sem);
1071        if (addr && !(shmflg & SHM_REMAP)) {
1072                err = -EINVAL;
1073                if (find_vma_intersection(current->mm, addr, addr + size))
1074                        goto invalid;
1075                /*
1076                 * If shm segment goes below stack, make sure there is some
1077                 * space left for the stack to grow (at least 4 pages).
1078                 */
1079                if (addr < current->mm->start_stack &&
1080                    addr > current->mm->start_stack - size - PAGE_SIZE * 5)
1081                        goto invalid;
1082        }
1083                
1084        addr = do_mmap_pgoff(file, addr, size, prot, flags, 0, &populate);
1085        *raddr = addr;
1086        err = 0;
1087        if (IS_ERR_VALUE(addr))
1088                err = (long)addr;
1089invalid:
1090        up_write(&current->mm->mmap_sem);
1091        if (populate)
1092                mm_populate(addr, populate);
1093
1094out_fput:
1095        fput(file);
1096
1097out_nattch:
1098        down_write(&shm_ids(ns).rw_mutex);
1099        shp = shm_lock(ns, shmid);
1100        BUG_ON(IS_ERR(shp));
1101        shp->shm_nattch--;
1102        if (shm_may_destroy(ns, shp))
1103                shm_destroy(ns, shp);
1104        else
1105                shm_unlock(shp);
1106        up_write(&shm_ids(ns).rw_mutex);
1107
1108out:
1109        return err;
1110
1111out_unlock:
1112        shm_unlock(shp);
1113        goto out;
1114
1115out_free:
1116        kfree(sfd);
1117out_put_dentry:
1118        path_put(&path);
1119        goto out_nattch;
1120}
1121
1122SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
1123{
1124        unsigned long ret;
1125        long err;
1126
1127        err = do_shmat(shmid, shmaddr, shmflg, &ret, SHMLBA);
1128        if (err)
1129                return err;
1130        force_successful_syscall_return();
1131        return (long)ret;
1132}
1133
1134/*
1135 * detach and kill segment if marked destroyed.
1136 * The work is done in shm_close.
1137 */
1138SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
1139{
1140        struct mm_struct *mm = current->mm;
1141        struct vm_area_struct *vma;
1142        unsigned long addr = (unsigned long)shmaddr;
1143        int retval = -EINVAL;
1144#ifdef CONFIG_MMU
1145        loff_t size = 0;
1146        struct vm_area_struct *next;
1147#endif
1148
1149        if (addr & ~PAGE_MASK)
1150                return retval;
1151
1152        down_write(&mm->mmap_sem);
1153
1154        /*
1155         * This function tries to be smart and unmap shm segments that
1156         * were modified by partial mlock or munmap calls:
1157         * - It first determines the size of the shm segment that should be
1158         *   unmapped: It searches for a vma that is backed by shm and that
1159         *   started at address shmaddr. It records it's size and then unmaps
1160         *   it.
1161         * - Then it unmaps all shm vmas that started at shmaddr and that
1162         *   are within the initially determined size.
1163         * Errors from do_munmap are ignored: the function only fails if
1164         * it's called with invalid parameters or if it's called to unmap
1165         * a part of a vma. Both calls in this function are for full vmas,
1166         * the parameters are directly copied from the vma itself and always
1167         * valid - therefore do_munmap cannot fail. (famous last words?)
1168         */
1169        /*
1170         * If it had been mremap()'d, the starting address would not
1171         * match the usual checks anyway. So assume all vma's are
1172         * above the starting address given.
1173         */
1174        vma = find_vma(mm, addr);
1175
1176#ifdef CONFIG_MMU
1177        while (vma) {
1178                next = vma->vm_next;
1179
1180                /*
1181                 * Check if the starting address would match, i.e. it's
1182                 * a fragment created by mprotect() and/or munmap(), or it
1183                 * otherwise it starts at this address with no hassles.
1184                 */
1185                if ((vma->vm_ops == &shm_vm_ops) &&
1186                        (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1187
1188
1189                        size = file_inode(vma->vm_file)->i_size;
1190                        do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1191                        /*
1192                         * We discovered the size of the shm segment, so
1193                         * break out of here and fall through to the next
1194                         * loop that uses the size information to stop
1195                         * searching for matching vma's.
1196                         */
1197                        retval = 0;
1198                        vma = next;
1199                        break;
1200                }
1201                vma = next;
1202        }
1203
1204        /*
1205         * We need look no further than the maximum address a fragment
1206         * could possibly have landed at. Also cast things to loff_t to
1207         * prevent overflows and make comparisons vs. equal-width types.
1208         */
1209        size = PAGE_ALIGN(size);
1210        while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1211                next = vma->vm_next;
1212
1213                /* finding a matching vma now does not alter retval */
1214                if ((vma->vm_ops == &shm_vm_ops) &&
1215                        (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1216
1217                        do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1218                vma = next;
1219        }
1220
1221#else /* CONFIG_MMU */
1222        /* under NOMMU conditions, the exact address to be destroyed must be
1223         * given */
1224        retval = -EINVAL;
1225        if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1226                do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1227                retval = 0;
1228        }
1229
1230#endif
1231
1232        up_write(&mm->mmap_sem);
1233        return retval;
1234}
1235
1236#ifdef CONFIG_PROC_FS
1237static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1238{
1239        struct user_namespace *user_ns = seq_user_ns(s);
1240        struct shmid_kernel *shp = it;
1241        unsigned long rss = 0, swp = 0;
1242
1243        shm_add_rss_swap(shp, &rss, &swp);
1244
1245#if BITS_PER_LONG <= 32
1246#define SIZE_SPEC "%10lu"
1247#else
1248#define SIZE_SPEC "%21lu"
1249#endif
1250
1251        return seq_printf(s,
1252                          "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1253                          "%5lu %5u %5u %5u %5u %10lu %10lu %10lu "
1254                          SIZE_SPEC " " SIZE_SPEC "\n",
1255                          shp->shm_perm.key,
1256                          shp->shm_perm.id,
1257                          shp->shm_perm.mode,
1258                          shp->shm_segsz,
1259                          shp->shm_cprid,
1260                          shp->shm_lprid,
1261                          shp->shm_nattch,
1262                          from_kuid_munged(user_ns, shp->shm_perm.uid),
1263                          from_kgid_munged(user_ns, shp->shm_perm.gid),
1264                          from_kuid_munged(user_ns, shp->shm_perm.cuid),
1265                          from_kgid_munged(user_ns, shp->shm_perm.cgid),
1266                          shp->shm_atim,
1267                          shp->shm_dtim,
1268                          shp->shm_ctim,
1269                          rss * PAGE_SIZE,
1270                          swp * PAGE_SIZE);
1271}
1272#endif
1273
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.