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