ipc/shm.c

   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
  20 #include <linux/config.h>
  21 #include <linux/slab.h>
  22 #include <linux/mm.h>
  23 #include <linux/hugetlb.h>
  24 #include <linux/shm.h>
  25 #include <linux/init.h>
  26 #include <linux/file.h>
  27 #include <linux/mman.h>
  28 #include <linux/shmem_fs.h>
  29 #include <linux/security.h>
  30 #include <linux/syscalls.h>
  31 #include <linux/audit.h>
  32 #include <linux/capability.h>
  33 #include <linux/ptrace.h>
  34 #include <linux/seq_file.h>
  35 #include <linux/mutex.h>
  36 #include <linux/vs_context.h>
  37 #include <linux/vs_limit.h>
  38
  39 #include <asm/uaccess.h>
  40
  41 #include "util.h"
  42
  43 static struct file_operations shm_file_operations;
  44 static struct vm_operations_struct shm_vm_ops;
  45
  46 static struct ipc_ids shm_ids;
  47
  48 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
  49 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
  50 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
  51 #define shm_buildid(id, seq) \
  52         ipc_buildid(&shm_ids, id, seq)
  53
  54 static int newseg (key_t key, int shmflg, size_t size);
  55 static void shm_open (struct vm_area_struct *shmd);
  56 static void shm_close (struct vm_area_struct *shmd);
  57 #ifdef CONFIG_PROC_FS
  58 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  59 #endif
  60
  61 size_t  shm_ctlmax = SHMMAX;
  62 size_t  shm_ctlall = SHMALL;
  63 int     shm_ctlmni = SHMMNI;
  64
  65 static int shm_tot; /* total number of shared memory pages */
  66
  67 void __init shm_init (void)
  68 {
  69         ipc_init_ids(&shm_ids, 1);
  70         ipc_init_proc_interface("sysvipc/shm",
  71                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
  72                                 &shm_ids,
  73                                 sysvipc_shm_proc_show);
  74 }
  75
  76 static inline int shm_checkid(struct shmid_kernel *s, int id)
  77 {
  78         if (ipc_checkid(&shm_ids,&s->shm_perm,id))
  79                 return -EIDRM;
  80         return 0;
  81 }
  82
  83 static inline struct shmid_kernel *shm_rmid(int id)
  84 {
  85         return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
  86 }
  87
  88 static inline int shm_addid(struct shmid_kernel *shp)
  89 {
  90         return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
  91 }
  92
  93
  94
  95 static inline void shm_inc (int id) {
  96         struct shmid_kernel *shp;
  97
  98         shp = shm_lock(id);
  99         BUG_ON(!shp);
 100         shp->shm_atim = get_seconds();
 101         shp->shm_lprid = current->tgid;
 102         shp->shm_nattch++;
 103         shm_unlock(shp);
 104 }
 105
 106 /* This is called by fork, once for every shm attach. */
 107 static void shm_open (struct vm_area_struct *shmd)
 108 {
 109         shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
 110 }
 111
 112 /*
 113  * shm_destroy - free the struct shmid_kernel
 114  *
 115  * @shp: struct to free
 116  *
 117  * It has to be called with shp and shm_ids.mutex locked,
 118  * but returns with shp unlocked and freed.
 119  */
 120 static void shm_destroy (struct shmid_kernel *shp)
 121 {
 122         struct vx_info *vxi = lookup_vx_info(shp->shm_perm.xid);
 123         int numpages = (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 124
 125         vx_ipcshm_sub(vxi, shp, numpages);
 126         shm_tot -= numpages;
 127
 128         shm_rmid (shp->id);
 129         shm_unlock(shp);
 130         if (!is_file_hugepages(shp->shm_file))
 131                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 132         else
 133                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
 134                                                 shp->mlock_user);
 135         fput (shp->shm_file);
 136         security_shm_free(shp);
 137         put_vx_info(vxi);
 138         ipc_rcu_putref(shp);
 139 }
 140
 141 /*
 142  * remove the attach descriptor shmd.
 143  * free memory for segment if it is marked destroyed.
 144  * The descriptor has already been removed from the current->mm->mmap list
 145  * and will later be kfree()d.
 146  */
 147 static void shm_close (struct vm_area_struct *shmd)
 148 {
 149         struct file * file = shmd->vm_file;
 150         int id = file->f_dentry->d_inode->i_ino;
 151         struct shmid_kernel *shp;
 152
 153         mutex_lock(&shm_ids.mutex);
 154         /* remove from the list of attaches of the shm segment */
 155         shp = shm_lock(id);
 156         BUG_ON(!shp);
 157         shp->shm_lprid = current->tgid;
 158         shp->shm_dtim = get_seconds();
 159         shp->shm_nattch--;
 160         if(shp->shm_nattch == 0 &&
 161            shp->shm_perm.mode & SHM_DEST)
 162                 shm_destroy (shp);
 163         else
 164                 shm_unlock(shp);
 165         mutex_unlock(&shm_ids.mutex);
 166 }
 167
 168 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 169 {
 170         int ret;
 171
 172         ret = shmem_mmap(file, vma);
 173         if (ret == 0) {
 174                 vma->vm_ops = &shm_vm_ops;
 175                 if (!(vma->vm_flags & VM_WRITE))
 176                         vma->vm_flags &= ~VM_MAYWRITE;
 177                 shm_inc(file->f_dentry->d_inode->i_ino);
 178         }
 179
 180         return ret;
 181 }
 182
 183 static struct file_operations shm_file_operations = {
 184         .mmap   = shm_mmap,
 185 #ifndef CONFIG_MMU
 186         .get_unmapped_area = shmem_get_unmapped_area,
 187 #endif
 188 };
 189
 190 static struct vm_operations_struct shm_vm_ops = {
 191         .open   = shm_open,     /* callback for a new vm-area open */
 192         .close  = shm_close,    /* callback for when the vm-area is released */
 193         .nopage = shmem_nopage,
 194 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
 195         .set_policy = shmem_set_policy,
 196         .get_policy = shmem_get_policy,
 197 #endif
 198 };
 199
 200 static int newseg (key_t key, int shmflg, size_t size)
 201 {
 202         int error;
 203         struct shmid_kernel *shp;
 204         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 205         struct file * file;
 206         char name[13];
 207         int id;
 208
 209         if (size < SHMMIN || size > shm_ctlmax)
 210                 return -EINVAL;
 211
 212         if (shm_tot + numpages >= shm_ctlall)
 213                 return -ENOSPC;
 214         if (!vx_ipcshm_avail(current->vx_info, numpages))
 215                 return -ENOSPC;
 216
 217         shp = ipc_rcu_alloc(sizeof(*shp));
 218         if (!shp)
 219                 return -ENOMEM;
 220
 221         shp->shm_perm.key = key;
 222         shp->shm_perm.xid = vx_current_xid();
 223         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 224         shp->mlock_user = NULL;
 225
 226         shp->shm_perm.security = NULL;
 227         error = security_shm_alloc(shp);
 228         if (error) {
 229                 ipc_rcu_putref(shp);
 230                 return error;
 231         }
 232
 233         if (shmflg & SHM_HUGETLB) {
 234                 /* hugetlb_zero_setup takes care of mlock user accounting */
 235                 file = hugetlb_zero_setup(size);
 236                 shp->mlock_user = current->user;
 237         } else {
 238                 int acctflag = VM_ACCOUNT;
 239                 /*
 240                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 241                  * if it's asked for.
 242                  */
 243                 if  ((shmflg & SHM_NORESERVE) &&
 244                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 245                         acctflag = 0;
 246                 sprintf (name, "SYSV%08x", key);
 247                 file = shmem_file_setup(name, size, acctflag);
 248         }
 249         error = PTR_ERR(file);
 250         if (IS_ERR(file))
 251                 goto no_file;
 252
 253         error = -ENOSPC;
 254         id = shm_addid(shp);
 255         if(id == -1)
 256                 goto no_id;
 257
 258         shp->shm_cprid = current->tgid;
 259         shp->shm_lprid = 0;
 260         shp->shm_atim = shp->shm_dtim = 0;
 261         shp->shm_ctim = get_seconds();
 262         shp->shm_segsz = size;
 263         shp->shm_nattch = 0;
 264         shp->id = shm_buildid(id,shp->shm_perm.seq);
 265         shp->shm_file = file;
 266         file->f_dentry->d_inode->i_ino = shp->id;
 267
 268         /* Hugetlb ops would have already been assigned. */
 269         if (!(shmflg & SHM_HUGETLB))
 270                 file->f_op = &shm_file_operations;
 271
 272         shm_tot += numpages;
 273         vx_ipcshm_add(current->vx_info, key, numpages);
 274         shm_unlock(shp);
 275         return shp->id;
 276
 277 no_id:
 278         fput(file);
 279 no_file:
 280         security_shm_free(shp);
 281         ipc_rcu_putref(shp);
 282         return error;
 283 }
 284
 285 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 286 {
 287         struct shmid_kernel *shp;
 288         int err, id = 0;
 289
 290         mutex_lock(&shm_ids.mutex);
 291         if (key == IPC_PRIVATE) {
 292                 err = newseg(key, shmflg, size);
 293         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 294                 if (!(shmflg & IPC_CREAT))
 295                         err = -ENOENT;
 296                 else
 297                         err = newseg(key, shmflg, size);
 298         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 299                 err = -EEXIST;
 300         } else {
 301                 shp = shm_lock(id);
 302                 BUG_ON(shp==NULL);
 303                 if (shp->shm_segsz < size)
 304                         err = -EINVAL;
 305                 else if (ipcperms(&shp->shm_perm, shmflg))
 306                         err = -EACCES;
 307                 else {
 308                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 309                         err = security_shm_associate(shp, shmflg);
 310                         if (!err)
 311                                 err = shmid;
 312                 }
 313                 shm_unlock(shp);
 314         }
 315         mutex_unlock(&shm_ids.mutex);
 316
 317         return err;
 318 }
 319
 320 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 321 {
 322         switch(version) {
 323         case IPC_64:
 324                 return copy_to_user(buf, in, sizeof(*in));
 325         case IPC_OLD:
 326             {
 327                 struct shmid_ds out;
 328
 329                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 330                 out.shm_segsz   = in->shm_segsz;
 331                 out.shm_atime   = in->shm_atime;
 332                 out.shm_dtime   = in->shm_dtime;
 333                 out.shm_ctime   = in->shm_ctime;
 334                 out.shm_cpid    = in->shm_cpid;
 335                 out.shm_lpid    = in->shm_lpid;
 336                 out.shm_nattch  = in->shm_nattch;
 337
 338                 return copy_to_user(buf, &out, sizeof(out));
 339             }
 340         default:
 341                 return -EINVAL;
 342         }
 343 }
 344
 345 struct shm_setbuf {
 346         uid_t   uid;
 347         gid_t   gid;
 348         mode_t  mode;
 349 };
 350
 351 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 352 {
 353         switch(version) {
 354         case IPC_64:
 355             {
 356                 struct shmid64_ds tbuf;
 357
 358                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 359                         return -EFAULT;
 360
 361                 out->uid        = tbuf.shm_perm.uid;
 362                 out->gid        = tbuf.shm_perm.gid;
 363                 out->mode       = tbuf.shm_perm.mode;
 364
 365                 return 0;
 366             }
 367         case IPC_OLD:
 368             {
 369                 struct shmid_ds tbuf_old;
 370
 371                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 372                         return -EFAULT;
 373
 374                 out->uid        = tbuf_old.shm_perm.uid;
 375                 out->gid        = tbuf_old.shm_perm.gid;
 376                 out->mode       = tbuf_old.shm_perm.mode;
 377
 378                 return 0;
 379             }
 380         default:
 381                 return -EINVAL;
 382         }
 383 }
 384
 385 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 386 {
 387         switch(version) {
 388         case IPC_64:
 389                 return copy_to_user(buf, in, sizeof(*in));
 390         case IPC_OLD:
 391             {
 392                 struct shminfo out;
 393
 394                 if(in->shmmax > INT_MAX)
 395                         out.shmmax = INT_MAX;
 396                 else
 397                         out.shmmax = (int)in->shmmax;
 398
 399                 out.shmmin      = in->shmmin;
 400                 out.shmmni      = in->shmmni;
 401                 out.shmseg      = in->shmseg;
 402                 out.shmall      = in->shmall;
 403
 404                 return copy_to_user(buf, &out, sizeof(out));
 405             }
 406         default:
 407                 return -EINVAL;
 408         }
 409 }
 410
 411 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 412 {
 413         int i;
 414
 415         *rss = 0;
 416         *swp = 0;
 417
 418         for (i = 0; i <= shm_ids.max_id; i++) {
 419                 struct shmid_kernel *shp;
 420                 struct inode *inode;
 421
 422                 shp = shm_get(i);
 423                 if(!shp)
 424                         continue;
 425
 426                 inode = shp->shm_file->f_dentry->d_inode;
 427
 428                 if (is_file_hugepages(shp->shm_file)) {
 429                         struct address_space *mapping = inode->i_mapping;
 430                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 431                 } else {
 432                         struct shmem_inode_info *info = SHMEM_I(inode);
 433                         spin_lock(&info->lock);
 434                         *rss += inode->i_mapping->nrpages;
 435                         *swp += info->swapped;
 436                         spin_unlock(&info->lock);
 437                 }
 438         }
 439 }
 440
 441 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 442 {
 443         struct shm_setbuf setbuf;
 444         struct shmid_kernel *shp;
 445         int err, version;
 446
 447         if (cmd < 0 || shmid < 0) {
 448                 err = -EINVAL;
 449                 goto out;
 450         }
 451
 452         version = ipc_parse_version(&cmd);
 453
 454         switch (cmd) { /* replace with proc interface ? */
 455         case IPC_INFO:
 456         {
 457                 struct shminfo64 shminfo;
 458
 459                 err = security_shm_shmctl(NULL, cmd);
 460                 if (err)
 461                         return err;
 462
 463                 memset(&shminfo,0,sizeof(shminfo));
 464                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 465                 shminfo.shmmax = shm_ctlmax;
 466                 shminfo.shmall = shm_ctlall;
 467
 468                 shminfo.shmmin = SHMMIN;
 469                 if(copy_shminfo_to_user (buf, &shminfo, version))
 470                         return -EFAULT;
 471                 /* reading a integer is always atomic */
 472                 err= shm_ids.max_id;
 473                 if(err<0)
 474                         err = 0;
 475                 goto out;
 476         }
 477         case SHM_INFO:
 478         {
 479                 struct shm_info shm_info;
 480
 481                 err = security_shm_shmctl(NULL, cmd);
 482                 if (err)
 483                         return err;
 484
 485                 memset(&shm_info,0,sizeof(shm_info));
 486                 mutex_lock(&shm_ids.mutex);
 487                 shm_info.used_ids = shm_ids.in_use;
 488                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 489                 shm_info.shm_tot = shm_tot;
 490                 shm_info.swap_attempts = 0;
 491                 shm_info.swap_successes = 0;
 492                 err = shm_ids.max_id;
 493                 mutex_unlock(&shm_ids.mutex);
 494                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 495                         err = -EFAULT;
 496                         goto out;
 497                 }
 498
 499                 err = err < 0 ? 0 : err;
 500                 goto out;
 501         }
 502         case SHM_STAT:
 503         case IPC_STAT:
 504         {
 505                 struct shmid64_ds tbuf;
 506                 int result;
 507                 memset(&tbuf, 0, sizeof(tbuf));
 508                 shp = shm_lock(shmid);
 509                 if(shp==NULL) {
 510                         err = -EINVAL;
 511                         goto out;
 512                 } else if(cmd==SHM_STAT) {
 513                         err = -EINVAL;
 514                         if (shmid > shm_ids.max_id)
 515                                 goto out_unlock;
 516                         result = shm_buildid(shmid, shp->shm_perm.seq);
 517                 } else {
 518                         err = shm_checkid(shp,shmid);
 519                         if(err)
 520                                 goto out_unlock;
 521                         result = 0;
 522                 }
 523                 err=-EACCES;
 524                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 525                         goto out_unlock;
 526                 err = security_shm_shmctl(shp, cmd);
 527                 if (err)
 528                         goto out_unlock;
 529                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 530                 tbuf.shm_segsz  = shp->shm_segsz;
 531                 tbuf.shm_atime  = shp->shm_atim;
 532                 tbuf.shm_dtime  = shp->shm_dtim;
 533                 tbuf.shm_ctime  = shp->shm_ctim;
 534                 tbuf.shm_cpid   = shp->shm_cprid;
 535                 tbuf.shm_lpid   = shp->shm_lprid;
 536                 if (!is_file_hugepages(shp->shm_file))
 537                         tbuf.shm_nattch = shp->shm_nattch;
 538                 else
 539                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 540                 shm_unlock(shp);
 541                 if(copy_shmid_to_user (buf, &tbuf, version))
 542                         err = -EFAULT;
 543                 else
 544                         err = result;
 545                 goto out;
 546         }
 547         case SHM_LOCK:
 548         case SHM_UNLOCK:
 549         {
 550                 shp = shm_lock(shmid);
 551                 if(shp==NULL) {
 552                         err = -EINVAL;
 553                         goto out;
 554                 }
 555                 err = shm_checkid(shp,shmid);
 556                 if(err)
 557                         goto out_unlock;
 558
 559                 err = audit_ipc_obj(&(shp->shm_perm));
 560                 if (err)
 561                         goto out_unlock;
 562
 563                 if (!capable(CAP_IPC_LOCK)) {
 564                         err = -EPERM;
 565                         if (current->euid != shp->shm_perm.uid &&
 566                             current->euid != shp->shm_perm.cuid)
 567                                 goto out_unlock;
 568                         if (cmd == SHM_LOCK &&
 569                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 570                                 goto out_unlock;
 571                 }
 572
 573                 err = security_shm_shmctl(shp, cmd);
 574                 if (err)
 575                         goto out_unlock;
 576
 577                 if(cmd==SHM_LOCK) {
 578                         struct user_struct * user = current->user;
 579                         if (!is_file_hugepages(shp->shm_file)) {
 580                                 err = shmem_lock(shp->shm_file, 1, user);
 581                                 if (!err) {
 582                                         shp->shm_perm.mode |= SHM_LOCKED;
 583                                         shp->mlock_user = user;
 584                                 }
 585                         }
 586                 } else if (!is_file_hugepages(shp->shm_file)) {
 587                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 588                         shp->shm_perm.mode &= ~SHM_LOCKED;
 589                         shp->mlock_user = NULL;
 590                 }
 591                 shm_unlock(shp);
 592                 goto out;
 593         }
 594         case IPC_RMID:
 595         {
 596                 /*
 597                  *      We cannot simply remove the file. The SVID states
 598                  *      that the block remains until the last person
 599                  *      detaches from it, then is deleted. A shmat() on
 600                  *      an RMID segment is legal in older Linux and if
 601                  *      we change it apps break...
 602                  *
 603                  *      Instead we set a destroyed flag, and then blow
 604                  *      the name away when the usage hits zero.
 605                  */
 606                 mutex_lock(&shm_ids.mutex);
 607                 shp = shm_lock(shmid);
 608                 err = -EINVAL;
 609                 if (shp == NULL)
 610                         goto out_up;
 611                 err = shm_checkid(shp, shmid);
 612                 if(err)
 613                         goto out_unlock_up;
 614
 615                 err = audit_ipc_obj(&(shp->shm_perm));
 616                 if (err)
 617                         goto out_unlock_up;
 618
 619                 if (current->euid != shp->shm_perm.uid &&
 620                     current->euid != shp->shm_perm.cuid &&
 621                     !capable(CAP_SYS_ADMIN)) {
 622                         err=-EPERM;
 623                         goto out_unlock_up;
 624                 }
 625
 626                 err = security_shm_shmctl(shp, cmd);
 627                 if (err)
 628                         goto out_unlock_up;
 629
 630                 if (shp->shm_nattch){
 631                         shp->shm_perm.mode |= SHM_DEST;
 632                         /* Do not find it any more */
 633                         shp->shm_perm.key = IPC_PRIVATE;
 634                         shm_unlock(shp);
 635                 } else
 636                         shm_destroy (shp);
 637                 mutex_unlock(&shm_ids.mutex);
 638                 goto out;
 639         }
 640
 641         case IPC_SET:
 642         {
 643                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 644                         err = -EFAULT;
 645                         goto out;
 646                 }
 647                 mutex_lock(&shm_ids.mutex);
 648                 shp = shm_lock(shmid);
 649                 err=-EINVAL;
 650                 if(shp==NULL)
 651                         goto out_up;
 652                 err = shm_checkid(shp,shmid);
 653                 if(err)
 654                         goto out_unlock_up;
 655                 err = audit_ipc_obj(&(shp->shm_perm));
 656                 if (err)
 657                         goto out_unlock_up;
 658                 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode, &(shp->shm_perm));
 659                 if (err)
 660                         goto out_unlock_up;
 661                 err=-EPERM;
 662                 if (current->euid != shp->shm_perm.uid &&
 663                     current->euid != shp->shm_perm.cuid &&
 664                     !capable(CAP_SYS_ADMIN)) {
 665                         goto out_unlock_up;
 666                 }
 667
 668                 err = security_shm_shmctl(shp, cmd);
 669                 if (err)
 670                         goto out_unlock_up;
 671
 672                 shp->shm_perm.uid = setbuf.uid;
 673                 shp->shm_perm.gid = setbuf.gid;
 674                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 675                         | (setbuf.mode & S_IRWXUGO);
 676                 shp->shm_ctim = get_seconds();
 677                 break;
 678         }
 679
 680         default:
 681                 err = -EINVAL;
 682                 goto out;
 683         }
 684
 685         err = 0;
 686 out_unlock_up:
 687         shm_unlock(shp);
 688 out_up:
 689         mutex_unlock(&shm_ids.mutex);
 690         goto out;
 691 out_unlock:
 692         shm_unlock(shp);
 693 out:
 694         return err;
 695 }
 696
 697 /*
 698  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 699  *
 700  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 701  * "raddr" thing points to kernel space, and there has to be a wrapper around
 702  * this.
 703  */
 704 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 705 {
 706         struct shmid_kernel *shp;
 707         unsigned long addr;
 708         unsigned long size;
 709         struct file * file;
 710         int    err;
 711         unsigned long flags;
 712         unsigned long prot;
 713         unsigned long o_flags;
 714         int acc_mode;
 715         void *user_addr;
 716
 717         if (shmid < 0) {
 718                 err = -EINVAL;
 719                 goto out;
 720         } else if ((addr = (ulong)shmaddr)) {
 721                 if (addr & (SHMLBA-1)) {
 722                         if (shmflg & SHM_RND)
 723                                 addr &= ~(SHMLBA-1);       /* round down */
 724                         else
 725 #ifndef __ARCH_FORCE_SHMLBA
 726                                 if (addr & ~PAGE_MASK)
 727 #endif
 728                                         return -EINVAL;
 729                 }
 730                 flags = MAP_SHARED | MAP_FIXED;
 731         } else {
 732                 if ((shmflg & SHM_REMAP))
 733                         return -EINVAL;
 734
 735                 flags = MAP_SHARED;
 736         }
 737
 738         if (shmflg & SHM_RDONLY) {
 739                 prot = PROT_READ;
 740                 o_flags = O_RDONLY;
 741                 acc_mode = S_IRUGO;
 742         } else {
 743                 prot = PROT_READ | PROT_WRITE;
 744                 o_flags = O_RDWR;
 745                 acc_mode = S_IRUGO | S_IWUGO;
 746         }
 747         if (shmflg & SHM_EXEC) {
 748                 prot |= PROT_EXEC;
 749                 acc_mode |= S_IXUGO;
 750         }
 751
 752         /*
 753          * We cannot rely on the fs check since SYSV IPC does have an
 754          * additional creator id...
 755          */
 756         shp = shm_lock(shmid);
 757         if(shp == NULL) {
 758                 err = -EINVAL;
 759                 goto out;
 760         }
 761         err = shm_checkid(shp,shmid);
 762         if (err) {
 763                 shm_unlock(shp);
 764                 goto out;
 765         }
 766         if (ipcperms(&shp->shm_perm, acc_mode)) {
 767                 shm_unlock(shp);
 768                 err = -EACCES;
 769                 goto out;
 770         }
 771
 772         err = security_shm_shmat(shp, shmaddr, shmflg);
 773         if (err) {
 774                 shm_unlock(shp);
 775                 return err;
 776         }
 777
 778         file = shp->shm_file;
 779         size = i_size_read(file->f_dentry->d_inode);
 780         shp->shm_nattch++;
 781         shm_unlock(shp);
 782
 783         down_write(&current->mm->mmap_sem);
 784         if (addr && !(shmflg & SHM_REMAP)) {
 785                 user_addr = ERR_PTR(-EINVAL);
 786                 if (find_vma_intersection(current->mm, addr, addr + size))
 787                         goto invalid;
 788                 /*
 789                  * If shm segment goes below stack, make sure there is some
 790                  * space left for the stack to grow (at least 4 pages).
 791                  */
 792                 if (addr < current->mm->start_stack &&
 793                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 794                         goto invalid;
 795         }
 796
 797         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 798
 799 invalid:
 800         up_write(&current->mm->mmap_sem);
 801
 802         mutex_lock(&shm_ids.mutex);
 803         shp = shm_lock(shmid);
 804         BUG_ON(!shp);
 805         shp->shm_nattch--;
 806         if(shp->shm_nattch == 0 &&
 807            shp->shm_perm.mode & SHM_DEST)
 808                 shm_destroy (shp);
 809         else
 810                 shm_unlock(shp);
 811         mutex_unlock(&shm_ids.mutex);
 812
 813         *raddr = (unsigned long) user_addr;
 814         err = 0;
 815         if (IS_ERR(user_addr))
 816                 err = PTR_ERR(user_addr);
 817 out:
 818         return err;
 819 }
 820
 821 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 822 {
 823         unsigned long ret;
 824         long err;
 825
 826         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 827         if (err)
 828                 return err;
 829         force_successful_syscall_return();
 830         return (long)ret;
 831 }
 832
 833 /*
 834  * detach and kill segment if marked destroyed.
 835  * The work is done in shm_close.
 836  */
 837 asmlinkage long sys_shmdt(char __user *shmaddr)
 838 {
 839         struct mm_struct *mm = current->mm;
 840         struct vm_area_struct *vma, *next;
 841         unsigned long addr = (unsigned long)shmaddr;
 842         loff_t size = 0;
 843         int retval = -EINVAL;
 844
 845         if (addr & ~PAGE_MASK)
 846                 return retval;
 847
 848         down_write(&mm->mmap_sem);
 849
 850         /*
 851          * This function tries to be smart and unmap shm segments that
 852          * were modified by partial mlock or munmap calls:
 853          * - It first determines the size of the shm segment that should be
 854          *   unmapped: It searches for a vma that is backed by shm and that
 855          *   started at address shmaddr. It records it's size and then unmaps
 856          *   it.
 857          * - Then it unmaps all shm vmas that started at shmaddr and that
 858          *   are within the initially determined size.
 859          * Errors from do_munmap are ignored: the function only fails if
 860          * it's called with invalid parameters or if it's called to unmap
 861          * a part of a vma. Both calls in this function are for full vmas,
 862          * the parameters are directly copied from the vma itself and always
 863          * valid - therefore do_munmap cannot fail. (famous last words?)
 864          */
 865         /*
 866          * If it had been mremap()'d, the starting address would not
 867          * match the usual checks anyway. So assume all vma's are
 868          * above the starting address given.
 869          */
 870         vma = find_vma(mm, addr);
 871
 872         while (vma) {
 873                 next = vma->vm_next;
 874
 875                 /*
 876                  * Check if the starting address would match, i.e. it's
 877                  * a fragment created by mprotect() and/or munmap(), or it
 878                  * otherwise it starts at this address with no hassles.
 879                  */
 880                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 881                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 882
 883
 884                         size = vma->vm_file->f_dentry->d_inode->i_size;
 885                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 886                         /*
 887                          * We discovered the size of the shm segment, so
 888                          * break out of here and fall through to the next
 889                          * loop that uses the size information to stop
 890                          * searching for matching vma's.
 891                          */
 892                         retval = 0;
 893                         vma = next;
 894                         break;
 895                 }
 896                 vma = next;
 897         }
 898
 899         /*
 900          * We need look no further than the maximum address a fragment
 901          * could possibly have landed at. Also cast things to loff_t to
 902          * prevent overflows and make comparisions vs. equal-width types.
 903          */
 904         size = PAGE_ALIGN(size);
 905         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 906                 next = vma->vm_next;
 907
 908                 /* finding a matching vma now does not alter retval */
 909                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 910                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 911
 912                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 913                 vma = next;
 914         }
 915
 916         up_write(&mm->mmap_sem);
 917         return retval;
 918 }
 919
 920 #ifdef CONFIG_PROC_FS
 921 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
 922 {
 923         struct shmid_kernel *shp = it;
 924         char *format;
 925
 926 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 927 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 928
 929         if (!vx_check(shp->shm_perm.xid, VX_IDENT))
 930                 return 0;
 931
 932         if (sizeof(size_t) <= sizeof(int))
 933                 format = SMALL_STRING;
 934         else
 935                 format = BIG_STRING;
 936         return seq_printf(s, format,
 937                           shp->shm_perm.key,
 938                           shp->id,
 939                           shp->shm_perm.mode,
 940                           shp->shm_segsz,
 941                           shp->shm_cprid,
 942                           shp->shm_lprid,
 943                           is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 944                           shp->shm_perm.uid,
 945                           shp->shm_perm.gid,
 946                           shp->shm_perm.cuid,
 947                           shp->shm_perm.cgid,
 948                           shp->shm_atim,
 949                           shp->shm_dtim,
 950                           shp->shm_ctim);
 951 }
 952 #endif