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