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#include <linux/ima.h>
  43
  44#include <asm/uaccess.h>
  45
  46#include "util.h"
  47
  48struct shm_file_data {
  49        int id;
  50        struct ipc_namespace *ns;
  51        struct file *file;
  52        const struct vm_operations_struct *vm_ops;
  53};
  54
  55#define shm_file_data(file) (*((struct shm_file_data **)&(file)->private_data))
  56
  57static const struct file_operations shm_file_operations;
  58static struct vm_operations_struct shm_vm_ops;
  59
  60#define shm_ids(ns)     ((ns)->ids[IPC_SHM_IDS])
  61
  62#define shm_unlock(shp)                 \
  63        ipc_unlock(&(shp)->shm_perm)
  64
  65static int newseg(struct ipc_namespace *, struct ipc_params *);
  66static void shm_open(struct vm_area_struct *vma);
  67static void shm_close(struct vm_area_struct *vma);
  68static void shm_destroy (struct ipc_namespace *ns, struct shmid_kernel *shp);
  69#ifdef CONFIG_PROC_FS
  70static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  71#endif
  72
  73void shm_init_ns(struct ipc_namespace *ns)
  74{
  75        ns->shm_ctlmax = SHMMAX;
  76        ns->shm_ctlall = SHMALL;
  77        ns->shm_ctlmni = SHMMNI;
  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}
 105#endif
 106
 107void __init shm_init (void)
 108{
 109        shm_init_ns(&init_ipc_ns);
 110        ipc_init_proc_interface("sysvipc/shm",
 111                                "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
 112                                IPC_SHM_IDS, sysvipc_shm_proc_show);
 113}
 114
 115/*
 116 * shm_lock_(check_) routines are called in the paths where the rw_mutex
 117 * is not necessarily held.
 118 */
 119static inline struct shmid_kernel *shm_lock(struct ipc_namespace *ns, int id)
 120{
 121        struct kern_ipc_perm *ipcp = ipc_lock(&shm_ids(ns), id);
 122
 123        if (IS_ERR(ipcp))
 124                return (struct shmid_kernel *)ipcp;
 125
 126        return container_of(ipcp, struct shmid_kernel, shm_perm);
 127}
 128
 129static inline struct shmid_kernel *shm_lock_check(struct ipc_namespace *ns,
 130                                                int id)
 131{
 132        struct kern_ipc_perm *ipcp = ipc_lock_check(&shm_ids(ns), id);
 133
 134        if (IS_ERR(ipcp))
 135                return (struct shmid_kernel *)ipcp;
 136
 137        return container_of(ipcp, struct shmid_kernel, shm_perm);
 138}
 139
 140static inline void shm_rmid(struct ipc_namespace *ns, struct shmid_kernel *s)
 141{
 142        ipc_rmid(&shm_ids(ns), &s->shm_perm);
 143}
 144
 145
 146/* This is called by fork, once for every shm attach. */
 147static void shm_open(struct vm_area_struct *vma)
 148{
 149        struct file *file = vma->vm_file;
 150        struct shm_file_data *sfd = shm_file_data(file);
 151        struct shmid_kernel *shp;
 152
 153        shp = shm_lock(sfd->ns, sfd->id);
 154        BUG_ON(IS_ERR(shp));
 155        shp->shm_atim = get_seconds();
 156        shp->shm_lprid = task_tgid_vnr(current);
 157        shp->shm_nattch++;
 158        shm_unlock(shp);
 159}
 160
 161/*
 162 * shm_destroy - free the struct shmid_kernel
 163 *
 164 * @ns: namespace
 165 * @shp: struct to free
 166 *
 167 * It has to be called with shp and shm_ids.rw_mutex (writer) locked,
 168 * but returns with shp unlocked and freed.
 169 */
 170static void shm_destroy(struct ipc_namespace *ns, struct shmid_kernel *shp)
 171{
 172        ns->shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 173        shm_rmid(ns, shp);
 174        shm_unlock(shp);
 175        if (!is_file_hugepages(shp->shm_file))
 176                shmem_lock(shp->shm_file, 0, shp->mlock_user);
 177        else if (shp->mlock_user)
 178                user_shm_unlock(shp->shm_file->f_path.dentry->d_inode->i_size,
 179                                                shp->mlock_user);
 180        fput (shp->shm_file);
 181        security_shm_free(shp);
 182        ipc_rcu_putref(shp);
 183}
 184
 185/*
 186 * remove the attach descriptor vma.
 187 * free memory for segment if it is marked destroyed.
 188 * The descriptor has already been removed from the current->mm->mmap list
 189 * and will later be kfree()d.
 190 */
 191static void shm_close(struct vm_area_struct *vma)
 192{
 193        struct file * file = vma->vm_file;
 194        struct shm_file_data *sfd = shm_file_data(file);
 195        struct shmid_kernel *shp;
 196        struct ipc_namespace *ns = sfd->ns;
 197
 198        down_write(&shm_ids(ns).rw_mutex);
 199        /* remove from the list of attaches of the shm segment */
 200        shp = shm_lock(ns, sfd->id);
 201        BUG_ON(IS_ERR(shp));
 202        shp->shm_lprid = task_tgid_vnr(current);
 203        shp->shm_dtim = get_seconds();
 204        shp->shm_nattch--;
 205        if(shp->shm_nattch == 0 &&
 206           shp->shm_perm.mode & SHM_DEST)
 207                shm_destroy(ns, shp);
 208        else
 209                shm_unlock(shp);
 210        up_write(&shm_ids(ns).rw_mutex);
 211}
 212
 213static int shm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 214{
 215        struct file *file = vma->vm_file;
 216        struct shm_file_data *sfd = shm_file_data(file);
 217
 218        return sfd->vm_ops->fault(vma, vmf);
 219}
 220
 221#ifdef CONFIG_NUMA
 222static int shm_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
 223{
 224        struct file *file = vma->vm_file;
 225        struct shm_file_data *sfd = shm_file_data(file);
 226        int err = 0;
 227        if (sfd->vm_ops->set_policy)
 228                err = sfd->vm_ops->set_policy(vma, new);
 229        return err;
 230}
 231
 232static struct mempolicy *shm_get_policy(struct vm_area_struct *vma,
 233                                        unsigned long addr)
 234{
 235        struct file *file = vma->vm_file;
 236        struct shm_file_data *sfd = shm_file_data(file);
 237        struct mempolicy *pol = NULL;
 238
 239        if (sfd->vm_ops->get_policy)
 240                pol = sfd->vm_ops->get_policy(vma, addr);
 241        else if (vma->vm_policy)
 242                pol = vma->vm_policy;
 243
 244        return pol;
 245}
 246#endif
 247
 248static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 249{
 250        struct shm_file_data *sfd = shm_file_data(file);
 251        int ret;
 252
 253        ret = sfd->file->f_op->mmap(sfd->file, vma);
 254        if (ret != 0)
 255                return ret;
 256        sfd->vm_ops = vma->vm_ops;
 257#ifdef CONFIG_MMU
 258        BUG_ON(!sfd->vm_ops->fault);
 259#endif
 260        vma->vm_ops = &shm_vm_ops;
 261        shm_open(vma);
 262
 263        return ret;
 264}
 265
 266static int shm_release(struct inode *ino, struct file *file)
 267{
 268        struct shm_file_data *sfd = shm_file_data(file);
 269
 270        put_ipc_ns(sfd->ns);
 271        shm_file_data(file) = NULL;
 272        kfree(sfd);
 273        return 0;
 274}
 275
 276static int shm_fsync(struct file *file, struct dentry *dentry, int datasync)
 277{
 278        int (*fsync) (struct file *, struct dentry *, int datasync);
 279        struct shm_file_data *sfd = shm_file_data(file);
 280        int ret = -EINVAL;
 281
 282        fsync = sfd->file->f_op->fsync;
 283        if (fsync)
 284                ret = fsync(sfd->file, sfd->file->f_path.dentry, datasync);
 285        return ret;
 286}
 287
 288static unsigned long shm_get_unmapped_area(struct file *file,
 289        unsigned long addr, unsigned long len, unsigned long pgoff,
 290        unsigned long flags)
 291{
 292        struct shm_file_data *sfd = shm_file_data(file);
 293        return get_unmapped_area(sfd->file, addr, len, pgoff, flags);
 294}
 295
 296int is_file_shm_hugepages(struct file *file)
 297{
 298        int ret = 0;
 299
 300        if (file->f_op == &shm_file_operations) {
 301                struct shm_file_data *sfd;
 302                sfd = shm_file_data(file);
 303                ret = is_file_hugepages(sfd->file);
 304        }
 305        return ret;
 306}
 307
 308static const struct file_operations shm_file_operations = {
 309        .mmap           = shm_mmap,
 310        .fsync          = shm_fsync,
 311        .release        = shm_release,
 312        .get_unmapped_area      = shm_get_unmapped_area,
 313};
 314
 315static struct vm_operations_struct shm_vm_ops = {
 316        .open   = shm_open,     /* callback for a new vm-area open */
 317        .close  = shm_close,    /* callback for when the vm-area is released */
 318        .fault  = shm_fault,
 319#if defined(CONFIG_NUMA)
 320        .set_policy = shm_set_policy,
 321        .get_policy = shm_get_policy,
 322#endif
 323};
 324
 325/**
 326 * newseg - Create a new shared memory segment
 327 * @ns: namespace
 328 * @params: ptr to the structure that contains key, size and shmflg
 329 *
 330 * Called with shm_ids.rw_mutex held as a writer.
 331 */
 332
 333static int newseg(struct ipc_namespace *ns, struct ipc_params *params)
 334{
 335        key_t key = params->key;
 336        int shmflg = params->flg;
 337        size_t size = params->u.size;
 338        int error;
 339        struct shmid_kernel *shp;
 340        int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 341        struct file * file;
 342        char name[13];
 343        int id;
 344        int acctflag = 0;
 345
 346        if (size < SHMMIN || size > ns->shm_ctlmax)
 347                return -EINVAL;
 348
 349        if (ns->shm_tot + numpages > ns->shm_ctlall)
 350                return -ENOSPC;
 351
 352        shp = ipc_rcu_alloc(sizeof(*shp));
 353        if (!shp)
 354                return -ENOMEM;
 355
 356        shp->shm_perm.key = key;
 357        shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 358        shp->mlock_user = NULL;
 359
 360        shp->shm_perm.security = NULL;
 361        error = security_shm_alloc(shp);
 362        if (error) {
 363                ipc_rcu_putref(shp);
 364                return error;
 365        }
 366
 367        sprintf (name, "SYSV%08x", key);
 368        if (shmflg & SHM_HUGETLB) {
 369                /* hugetlb_file_setup applies strict accounting */
 370                if (shmflg & SHM_NORESERVE)
 371                        acctflag = VM_NORESERVE;
 372                file = hugetlb_file_setup(name, size, acctflag,
 373                                                        &shp->mlock_user);
 374        } else {
 375                /*
 376                 * Do not allow no accounting for OVERCOMMIT_NEVER, even
 377                 * if it's asked for.
 378                 */
 379                if  ((shmflg & SHM_NORESERVE) &&
 380                                sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 381                        acctflag = VM_NORESERVE;
 382                file = shmem_file_setup(name, size, acctflag);
 383        }
 384        error = PTR_ERR(file);
 385        if (IS_ERR(file))
 386                goto no_file;
 387        ima_shm_check(file);
 388
 389        id = ipc_addid(&shm_ids(ns), &shp->shm_perm, ns->shm_ctlmni);
 390        if (id < 0) {
 391                error = id;
 392                goto no_id;
 393        }
 394
 395        shp->shm_cprid = task_tgid_vnr(current);
 396        shp->shm_lprid = 0;
 397        shp->shm_atim = shp->shm_dtim = 0;
 398        shp->shm_ctim = get_seconds();
 399        shp->shm_segsz = size;
 400        shp->shm_nattch = 0;
 401        shp->shm_file = file;
 402        /*
 403         * shmid gets reported as "inode#" in /proc/pid/maps.
 404         * proc-ps tools use this. Changing this will break them.
 405         */
 406        file->f_dentry->d_inode->i_ino = shp->shm_perm.id;
 407
 408        ns->shm_tot += numpages;
 409        error = shp->shm_perm.id;
 410        shm_unlock(shp);
 411        return error;
 412
 413no_id:
 414        if (shp->mlock_user)    /* shmflg & SHM_HUGETLB case */
 415                user_shm_unlock(size, shp->mlock_user);
 416        fput(file);
 417no_file:
 418        security_shm_free(shp);
 419        ipc_rcu_putref(shp);
 420        return error;
 421}
 422
 423/*
 424 * Called with shm_ids.rw_mutex and ipcp locked.
 425 */
 426static inline int shm_security(struct kern_ipc_perm *ipcp, int shmflg)
 427{
 428        struct shmid_kernel *shp;
 429
 430        shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 431        return security_shm_associate(shp, shmflg);
 432}
 433
 434/*
 435 * Called with shm_ids.rw_mutex and ipcp locked.
 436 */
 437static inline int shm_more_checks(struct kern_ipc_perm *ipcp,
 438                                struct ipc_params *params)
 439{
 440        struct shmid_kernel *shp;
 441
 442        shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 443        if (shp->shm_segsz < params->u.size)
 444                return -EINVAL;
 445
 446        return 0;
 447}
 448
 449SYSCALL_DEFINE3(shmget, key_t, key, size_t, size, int, shmflg)
 450{
 451        struct ipc_namespace *ns;
 452        struct ipc_ops shm_ops;
 453        struct ipc_params shm_params;
 454
 455        ns = current->nsproxy->ipc_ns;
 456
 457        shm_ops.getnew = newseg;
 458        shm_ops.associate = shm_security;
 459        shm_ops.more_checks = shm_more_checks;
 460
 461        shm_params.key = key;
 462        shm_params.flg = shmflg;
 463        shm_params.u.size = size;
 464
 465        return ipcget(ns, &shm_ids(ns), &shm_ops, &shm_params);
 466}
 467
 468static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 469{
 470        switch(version) {
 471        case IPC_64:
 472                return copy_to_user(buf, in, sizeof(*in));
 473        case IPC_OLD:
 474            {
 475                struct shmid_ds out;
 476
 477                ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 478                out.shm_segsz   = in->shm_segsz;
 479                out.shm_atime   = in->shm_atime;
 480                out.shm_dtime   = in->shm_dtime;
 481                out.shm_ctime   = in->shm_ctime;
 482                out.shm_cpid    = in->shm_cpid;
 483                out.shm_lpid    = in->shm_lpid;
 484                out.shm_nattch  = in->shm_nattch;
 485
 486                return copy_to_user(buf, &out, sizeof(out));
 487            }
 488        default:
 489                return -EINVAL;
 490        }
 491}
 492
 493static inline unsigned long
 494copy_shmid_from_user(struct shmid64_ds *out, void __user *buf, int version)
 495{
 496        switch(version) {
 497        case IPC_64:
 498                if (copy_from_user(out, buf, sizeof(*out)))
 499                        return -EFAULT;
 500                return 0;
 501        case IPC_OLD:
 502            {
 503                struct shmid_ds tbuf_old;
 504
 505                if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 506                        return -EFAULT;
 507
 508                out->shm_perm.uid       = tbuf_old.shm_perm.uid;
 509                out->shm_perm.gid       = tbuf_old.shm_perm.gid;
 510                out->shm_perm.mode      = tbuf_old.shm_perm.mode;
 511
 512                return 0;
 513            }
 514        default:
 515                return -EINVAL;
 516        }
 517}
 518
 519static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 520{
 521        switch(version) {
 522        case IPC_64:
 523                return copy_to_user(buf, in, sizeof(*in));
 524        case IPC_OLD:
 525            {
 526                struct shminfo out;
 527
 528                if(in->shmmax > INT_MAX)
 529                        out.shmmax = INT_MAX;
 530                else
 531                        out.shmmax = (int)in->shmmax;
 532
 533                out.shmmin      = in->shmmin;
 534                out.shmmni      = in->shmmni;
 535                out.shmseg      = in->shmseg;
 536                out.shmall      = in->shmall; 
 537
 538                return copy_to_user(buf, &out, sizeof(out));
 539            }
 540        default:
 541                return -EINVAL;
 542        }
 543}
 544
 545/*
 546 * Called with shm_ids.rw_mutex held as a reader
 547 */
 548static void shm_get_stat(struct ipc_namespace *ns, unsigned long *rss,
 549                unsigned long *swp)
 550{
 551        int next_id;
 552        int total, in_use;
 553
 554        *rss = 0;
 555        *swp = 0;
 556
 557        in_use = shm_ids(ns).in_use;
 558
 559        for (total = 0, next_id = 0; total < in_use; next_id++) {
 560                struct kern_ipc_perm *ipc;
 561                struct shmid_kernel *shp;
 562                struct inode *inode;
 563
 564                ipc = idr_find(&shm_ids(ns).ipcs_idr, next_id);
 565                if (ipc == NULL)
 566                        continue;
 567                shp = container_of(ipc, struct shmid_kernel, shm_perm);
 568
 569                inode = shp->shm_file->f_path.dentry->d_inode;
 570
 571                if (is_file_hugepages(shp->shm_file)) {
 572                        struct address_space *mapping = inode->i_mapping;
 573                        struct hstate *h = hstate_file(shp->shm_file);
 574                        *rss += pages_per_huge_page(h) * mapping->nrpages;
 575                } else {
 576#ifdef CONFIG_SHMEM
 577                        struct shmem_inode_info *info = SHMEM_I(inode);
 578                        spin_lock(&info->lock);
 579                        *rss += inode->i_mapping->nrpages;
 580                        *swp += info->swapped;
 581                        spin_unlock(&info->lock);
 582#else
 583                        *rss += inode->i_mapping->nrpages;
 584#endif
 585                }
 586
 587                total++;
 588        }
 589}
 590
 591/*
 592 * This function handles some shmctl commands which require the rw_mutex
 593 * to be held in write mode.
 594 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
 595 */
 596static int shmctl_down(struct ipc_namespace *ns, int shmid, int cmd,
 597                       struct shmid_ds __user *buf, int version)
 598{
 599        struct kern_ipc_perm *ipcp;
 600        struct shmid64_ds shmid64;
 601        struct shmid_kernel *shp;
 602        int err;
 603
 604        if (cmd == IPC_SET) {
 605                if (copy_shmid_from_user(&shmid64, buf, version))
 606                        return -EFAULT;
 607        }
 608
 609        ipcp = ipcctl_pre_down(&shm_ids(ns), shmid, cmd, &shmid64.shm_perm, 0);
 610        if (IS_ERR(ipcp))
 611                return PTR_ERR(ipcp);
 612
 613        shp = container_of(ipcp, struct shmid_kernel, shm_perm);
 614
 615        err = security_shm_shmctl(shp, cmd);
 616        if (err)
 617                goto out_unlock;
 618        switch (cmd) {
 619        case IPC_RMID:
 620                do_shm_rmid(ns, ipcp);
 621                goto out_up;
 622        case IPC_SET:
 623                ipc_update_perm(&shmid64.shm_perm, ipcp);
 624                shp->shm_ctim = get_seconds();
 625                break;
 626        default:
 627                err = -EINVAL;
 628        }
 629out_unlock:
 630        shm_unlock(shp);
 631out_up:
 632        up_write(&shm_ids(ns).rw_mutex);
 633        return err;
 634}
 635
 636SYSCALL_DEFINE3(shmctl, int, shmid, int, cmd, struct shmid_ds __user *, buf)
 637{
 638        struct shmid_kernel *shp;
 639        int err, version;
 640        struct ipc_namespace *ns;
 641
 642        if (cmd < 0 || shmid < 0) {
 643                err = -EINVAL;
 644                goto out;
 645        }
 646
 647        version = ipc_parse_version(&cmd);
 648        ns = current->nsproxy->ipc_ns;
 649
 650        switch (cmd) { /* replace with proc interface ? */
 651        case IPC_INFO:
 652        {
 653                struct shminfo64 shminfo;
 654
 655                err = security_shm_shmctl(NULL, cmd);
 656                if (err)
 657                        return err;
 658
 659                memset(&shminfo, 0, sizeof(shminfo));
 660                shminfo.shmmni = shminfo.shmseg = ns->shm_ctlmni;
 661                shminfo.shmmax = ns->shm_ctlmax;
 662                shminfo.shmall = ns->shm_ctlall;
 663
 664                shminfo.shmmin = SHMMIN;
 665                if(copy_shminfo_to_user (buf, &shminfo, version))
 666                        return -EFAULT;
 667
 668                down_read(&shm_ids(ns).rw_mutex);
 669                err = ipc_get_maxid(&shm_ids(ns));
 670                up_read(&shm_ids(ns).rw_mutex);
 671
 672                if(err<0)
 673                        err = 0;
 674                goto out;
 675        }
 676        case SHM_INFO:
 677        {
 678                struct shm_info shm_info;
 679
 680                err = security_shm_shmctl(NULL, cmd);
 681                if (err)
 682                        return err;
 683
 684                memset(&shm_info, 0, sizeof(shm_info));
 685                down_read(&shm_ids(ns).rw_mutex);
 686                shm_info.used_ids = shm_ids(ns).in_use;
 687                shm_get_stat (ns, &shm_info.shm_rss, &shm_info.shm_swp);
 688                shm_info.shm_tot = ns->shm_tot;
 689                shm_info.swap_attempts = 0;
 690                shm_info.swap_successes = 0;
 691                err = ipc_get_maxid(&shm_ids(ns));
 692                up_read(&shm_ids(ns).rw_mutex);
 693                if (copy_to_user(buf, &shm_info, sizeof(shm_info))) {
 694                        err = -EFAULT;
 695                        goto out;
 696                }
 697
 698                err = err < 0 ? 0 : err;
 699                goto out;
 700        }
 701        case SHM_STAT:
 702        case IPC_STAT:
 703        {
 704                struct shmid64_ds tbuf;
 705                int result;
 706
 707                if (cmd == SHM_STAT) {
 708                        shp = shm_lock(ns, shmid);
 709                        if (IS_ERR(shp)) {
 710                                err = PTR_ERR(shp);
 711                                goto out;
 712                        }
 713                        result = shp->shm_perm.id;
 714                } else {
 715                        shp = shm_lock_check(ns, shmid);
 716                        if (IS_ERR(shp)) {
 717                                err = PTR_ERR(shp);
 718                                goto out;
 719                        }
 720                        result = 0;
 721                }
 722                err = -EACCES;
 723                if (ipcperms (&shp->shm_perm, S_IRUGO))
 724                        goto out_unlock;
 725                err = security_shm_shmctl(shp, cmd);
 726                if (err)
 727                        goto out_unlock;
 728                memset(&tbuf, 0, sizeof(tbuf));
 729                kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 730                tbuf.shm_segsz  = shp->shm_segsz;
 731                tbuf.shm_atime  = shp->shm_atim;
 732                tbuf.shm_dtime  = shp->shm_dtim;
 733                tbuf.shm_ctime  = shp->shm_ctim;
 734                tbuf.shm_cpid   = shp->shm_cprid;
 735                tbuf.shm_lpid   = shp->shm_lprid;
 736                tbuf.shm_nattch = shp->shm_nattch;
 737                shm_unlock(shp);
 738                if(copy_shmid_to_user (buf, &tbuf, version))
 739                        err = -EFAULT;
 740                else
 741                        err = result;
 742                goto out;
 743        }
 744        case SHM_LOCK:
 745        case SHM_UNLOCK:
 746        {
 747                struct file *uninitialized_var(shm_file);
 748
 749                lru_add_drain_all();  /* drain pagevecs to lru lists */
 750
 751                shp = shm_lock_check(ns, shmid);
 752                if (IS_ERR(shp)) {
 753                        err = PTR_ERR(shp);
 754                        goto out;
 755                }
 756
 757                audit_ipc_obj(&(shp->shm_perm));
 758
 759                if (!capable(CAP_IPC_LOCK)) {
 760                        uid_t euid = current_euid();
 761                        err = -EPERM;
 762                        if (euid != shp->shm_perm.uid &&
 763                            euid != shp->shm_perm.cuid)
 764                                goto out_unlock;
 765                        if (cmd == SHM_LOCK &&
 766                            !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 767                                goto out_unlock;
 768                }
 769
 770                err = security_shm_shmctl(shp, cmd);
 771                if (err)
 772                        goto out_unlock;
 773                
 774                if(cmd==SHM_LOCK) {
 775                        struct user_struct *user = current_user();
 776                        if (!is_file_hugepages(shp->shm_file)) {
 777                                err = shmem_lock(shp->shm_file, 1, user);
 778                                if (!err && !(shp->shm_perm.mode & SHM_LOCKED)){
 779                                        shp->shm_perm.mode |= SHM_LOCKED;
 780                                        shp->mlock_user = user;
 781                                }
 782                        }
 783                } else if (!is_file_hugepages(shp->shm_file)) {
 784                        shmem_lock(shp->shm_file, 0, shp->mlock_user);
 785                        shp->shm_perm.mode &= ~SHM_LOCKED;
 786                        shp->mlock_user = NULL;
 787                }
 788                shm_unlock(shp);
 789                goto out;
 790        }
 791        case IPC_RMID:
 792        case IPC_SET:
 793                err = shmctl_down(ns, shmid, cmd, buf, version);
 794                return err;
 795        default:
 796                return -EINVAL;
 797        }
 798
 799out_unlock:
 800        shm_unlock(shp);
 801out:
 802        return err;
 803}
 804
 805/*
 806 * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 807 *
 808 * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 809 * "raddr" thing points to kernel space, and there has to be a wrapper around
 810 * this.
 811 */
 812long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 813{
 814        struct shmid_kernel *shp;
 815        unsigned long addr;
 816        unsigned long size;
 817        struct file * file;
 818        int    err;
 819        unsigned long flags;
 820        unsigned long prot;
 821        int acc_mode;
 822        unsigned long user_addr;
 823        struct ipc_namespace *ns;
 824        struct shm_file_data *sfd;
 825        struct path path;
 826        fmode_t f_mode;
 827
 828        err = -EINVAL;
 829        if (shmid < 0)
 830                goto out;
 831        else if ((addr = (ulong)shmaddr)) {
 832                if (addr & (SHMLBA-1)) {
 833                        if (shmflg & SHM_RND)
 834                                addr &= ~(SHMLBA-1);       /* round down */
 835                        else
 836#ifndef __ARCH_FORCE_SHMLBA
 837                                if (addr & ~PAGE_MASK)
 838#endif
 839                                        goto out;
 840                }
 841                flags = MAP_SHARED | MAP_FIXED;
 842        } else {
 843                if ((shmflg & SHM_REMAP))
 844                        goto out;
 845
 846                flags = MAP_SHARED;
 847        }
 848
 849        if (shmflg & SHM_RDONLY) {
 850                prot = PROT_READ;
 851                acc_mode = S_IRUGO;
 852                f_mode = FMODE_READ;
 853        } else {
 854                prot = PROT_READ | PROT_WRITE;
 855                acc_mode = S_IRUGO | S_IWUGO;
 856                f_mode = FMODE_READ | FMODE_WRITE;
 857        }
 858        if (shmflg & SHM_EXEC) {
 859                prot |= PROT_EXEC;
 860                acc_mode |= S_IXUGO;
 861        }
 862
 863        /*
 864         * We cannot rely on the fs check since SYSV IPC does have an
 865         * additional creator id...
 866         */
 867        ns = current->nsproxy->ipc_ns;
 868        shp = shm_lock_check(ns, shmid);
 869        if (IS_ERR(shp)) {
 870                err = PTR_ERR(shp);
 871                goto out;
 872        }
 873
 874        err = -EACCES;
 875        if (ipcperms(&shp->shm_perm, acc_mode))
 876                goto out_unlock;
 877
 878        err = security_shm_shmat(shp, shmaddr, shmflg);
 879        if (err)
 880                goto out_unlock;
 881
 882        path.dentry = dget(shp->shm_file->f_path.dentry);
 883        path.mnt    = shp->shm_file->f_path.mnt;
 884        shp->shm_nattch++;
 885        size = i_size_read(path.dentry->d_inode);
 886        shm_unlock(shp);
 887
 888        err = -ENOMEM;
 889        sfd = kzalloc(sizeof(*sfd), GFP_KERNEL);
 890        if (!sfd)
 891                goto out_put_dentry;
 892
 893        file = alloc_file(path.mnt, path.dentry, f_mode, &shm_file_operations);
 894        if (!file)
 895                goto out_free;
 896        ima_shm_check(file);
 897
 898        file->private_data = sfd;
 899        file->f_mapping = shp->shm_file->f_mapping;
 900        sfd->id = shp->shm_perm.id;
 901        sfd->ns = get_ipc_ns(ns);
 902        sfd->file = shp->shm_file;
 903        sfd->vm_ops = NULL;
 904
 905        down_write(&current->mm->mmap_sem);
 906        if (addr && !(shmflg & SHM_REMAP)) {
 907                err = -EINVAL;
 908                if (find_vma_intersection(current->mm, addr, addr + size))
 909                        goto invalid;
 910                /*
 911                 * If shm segment goes below stack, make sure there is some
 912                 * space left for the stack to grow (at least 4 pages).
 913                 */
 914                if (addr < current->mm->start_stack &&
 915                    addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 916                        goto invalid;
 917        }
 918                
 919        user_addr = do_mmap (file, addr, size, prot, flags, 0);
 920        *raddr = user_addr;
 921        err = 0;
 922        if (IS_ERR_VALUE(user_addr))
 923                err = (long)user_addr;
 924invalid:
 925        up_write(&current->mm->mmap_sem);
 926
 927        fput(file);
 928
 929out_nattch:
 930        down_write(&shm_ids(ns).rw_mutex);
 931        shp = shm_lock(ns, shmid);
 932        BUG_ON(IS_ERR(shp));
 933        shp->shm_nattch--;
 934        if(shp->shm_nattch == 0 &&
 935           shp->shm_perm.mode & SHM_DEST)
 936                shm_destroy(ns, shp);
 937        else
 938                shm_unlock(shp);
 939        up_write(&shm_ids(ns).rw_mutex);
 940
 941out:
 942        return err;
 943
 944out_unlock:
 945        shm_unlock(shp);
 946        goto out;
 947
 948out_free:
 949        kfree(sfd);
 950out_put_dentry:
 951        dput(path.dentry);
 952        goto out_nattch;
 953}
 954
 955SYSCALL_DEFINE3(shmat, int, shmid, char __user *, shmaddr, int, shmflg)
 956{
 957        unsigned long ret;
 958        long err;
 959
 960        err = do_shmat(shmid, shmaddr, shmflg, &ret);
 961        if (err)
 962                return err;
 963        force_successful_syscall_return();
 964        return (long)ret;
 965}
 966
 967/*
 968 * detach and kill segment if marked destroyed.
 969 * The work is done in shm_close.
 970 */
 971SYSCALL_DEFINE1(shmdt, char __user *, shmaddr)
 972{
 973        struct mm_struct *mm = current->mm;
 974        struct vm_area_struct *vma;
 975        unsigned long addr = (unsigned long)shmaddr;
 976        int retval = -EINVAL;
 977#ifdef CONFIG_MMU
 978        loff_t size = 0;
 979        struct vm_area_struct *next;
 980#endif
 981
 982        if (addr & ~PAGE_MASK)
 983                return retval;
 984
 985        down_write(&mm->mmap_sem);
 986
 987        /*
 988         * This function tries to be smart and unmap shm segments that
 989         * were modified by partial mlock or munmap calls:
 990         * - It first determines the size of the shm segment that should be
 991         *   unmapped: It searches for a vma that is backed by shm and that
 992         *   started at address shmaddr. It records it's size and then unmaps
 993         *   it.
 994         * - Then it unmaps all shm vmas that started at shmaddr and that
 995         *   are within the initially determined size.
 996         * Errors from do_munmap are ignored: the function only fails if
 997         * it's called with invalid parameters or if it's called to unmap
 998         * a part of a vma. Both calls in this function are for full vmas,
 999         * the parameters are directly copied from the vma itself and always
1000         * valid - therefore do_munmap cannot fail. (famous last words?)
1001         */
1002        /*
1003         * If it had been mremap()'d, the starting address would not
1004         * match the usual checks anyway. So assume all vma's are
1005         * above the starting address given.
1006         */
1007        vma = find_vma(mm, addr);
1008
1009#ifdef CONFIG_MMU
1010        while (vma) {
1011                next = vma->vm_next;
1012
1013                /*
1014                 * Check if the starting address would match, i.e. it's
1015                 * a fragment created by mprotect() and/or munmap(), or it
1016                 * otherwise it starts at this address with no hassles.
1017                 */
1018                if ((vma->vm_ops == &shm_vm_ops) &&
1019                        (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
1020
1021
1022                        size = vma->vm_file->f_path.dentry->d_inode->i_size;
1023                        do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1024                        /*
1025                         * We discovered the size of the shm segment, so
1026                         * break out of here and fall through to the next
1027                         * loop that uses the size information to stop
1028                         * searching for matching vma's.
1029                         */
1030                        retval = 0;
1031                        vma = next;
1032                        break;
1033                }
1034                vma = next;
1035        }
1036
1037        /*
1038         * We need look no further than the maximum address a fragment
1039         * could possibly have landed at. Also cast things to loff_t to
1040         * prevent overflows and make comparisions vs. equal-width types.
1041         */
1042        size = PAGE_ALIGN(size);
1043        while (vma && (loff_t)(vma->vm_end - addr) <= size) {
1044                next = vma->vm_next;
1045
1046                /* finding a matching vma now does not alter retval */
1047                if ((vma->vm_ops == &shm_vm_ops) &&
1048                        (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
1049
1050                        do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1051                vma = next;
1052        }
1053
1054#else /* CONFIG_MMU */
1055        /* under NOMMU conditions, the exact address to be destroyed must be
1056         * given */
1057        retval = -EINVAL;
1058        if (vma->vm_start == addr && vma->vm_ops == &shm_vm_ops) {
1059                do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
1060                retval = 0;
1061        }
1062
1063#endif
1064
1065        up_write(&mm->mmap_sem);
1066        return retval;
1067}
1068
1069#ifdef CONFIG_PROC_FS
1070static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
1071{
1072        struct shmid_kernel *shp = it;
1073
1074#if BITS_PER_LONG <= 32
1075#define SIZE_SPEC "%10lu"
1076#else
1077#define SIZE_SPEC "%21lu"
1078#endif
1079
1080        return seq_printf(s,
1081                          "%10d %10d  %4o " SIZE_SPEC " %5u %5u  "
1082                          "%5lu %5u %5u %5u %5u %10lu %10lu %10lu\n",
1083                          shp->shm_perm.key,
1084                          shp->shm_perm.id,
1085                          shp->shm_perm.mode,
1086                          shp->shm_segsz,
1087                          shp->shm_cprid,
1088                          shp->shm_lprid,
1089                          shp->shm_nattch,
1090                          shp->shm_perm.uid,
1091                          shp->shm_perm.gid,
1092                          shp->shm_perm.cuid,
1093                          shp->shm_perm.cgid,
1094                          shp->shm_atim,
1095                          shp->shm_dtim,
1096                          shp->shm_ctim);
1097}
1098#endif
1099