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  */
  17
  18 #include <linux/config.h>
  19 #include <linux/slab.h>
  20 #include <linux/mm.h>
  21 #include <linux/hugetlb.h>
  22 #include <linux/shm.h>
  23 #include <linux/init.h>
  24 #include <linux/file.h>
  25 #include <linux/mman.h>
  26 #include <linux/shmem_fs.h>
  27 #include <linux/security.h>
  28 #include <linux/syscalls.h>
  29 #include <linux/audit.h>
  30 #include <linux/capability.h>
  31 #include <linux/ptrace.h>
  32 #include <linux/seq_file.h>
  33 #include <linux/vs_context.h>
  34 #include <linux/vs_limit.h>
  35
  36 #include <asm/uaccess.h>
  37
  38 #include "util.h"
  39
  40 static struct file_operations shm_file_operations;
  41 static struct vm_operations_struct shm_vm_ops;
  42
  43 static struct ipc_ids shm_ids;
  44
  45 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
  46 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
  47 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
  48 #define shm_buildid(id, seq) \
  49         ipc_buildid(&shm_ids, id, seq)
  50
  51 static int newseg (key_t key, int shmflg, size_t size);
  52 static void shm_open (struct vm_area_struct *shmd);
  53 static void shm_close (struct vm_area_struct *shmd);
  54 #ifdef CONFIG_PROC_FS
  55 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  56 #endif
  57
  58 size_t  shm_ctlmax = SHMMAX;
  59 size_t  shm_ctlall = SHMALL;
  60 int     shm_ctlmni = SHMMNI;
  61
  62 static int shm_tot; /* total number of shared memory pages */
  63
  64 void __init shm_init (void)
  65 {
  66         ipc_init_ids(&shm_ids, 1);
  67         ipc_init_proc_interface("sysvipc/shm",
  68                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
  69                                 &shm_ids,
  70                                 sysvipc_shm_proc_show);
  71 }
  72
  73 static inline int shm_checkid(struct shmid_kernel *s, int id)
  74 {
  75         if (ipc_checkid(&shm_ids,&s->shm_perm,id))
  76                 return -EIDRM;
  77         return 0;
  78 }
  79
  80 static inline struct shmid_kernel *shm_rmid(int id)
  81 {
  82         return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
  83 }
  84
  85 static inline int shm_addid(struct shmid_kernel *shp)
  86 {
  87         return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
  88 }
  89
  90
  91
  92 static inline void shm_inc (int id) {
  93         struct shmid_kernel *shp;
  94
  95         if(!(shp = shm_lock(id)))
  96                 BUG();
  97         shp->shm_atim = get_seconds();
  98         shp->shm_lprid = current->tgid;
  99         shp->shm_nattch++;
 100         shm_unlock(shp);
 101 }
 102
 103 /* This is called by fork, once for every shm attach. */
 104 static void shm_open (struct vm_area_struct *shmd)
 105 {
 106         shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
 107 }
 108
 109 /*
 110  * shm_destroy - free the struct shmid_kernel
 111  *
 112  * @shp: struct to free
 113  *
 114  * It has to be called with shp and shm_ids.sem locked,
 115  * but returns with shp unlocked and freed.
 116  */
 117 static void shm_destroy (struct shmid_kernel *shp)
 118 {
 119         struct vx_info *vxi = lookup_vx_info(shp->shm_perm.xid);
 120         int numpages = (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 121
 122         vx_ipcshm_sub(vxi, shp, numpages);
 123         shm_tot -= numpages;
 124
 125         shm_rmid (shp->id);
 126         shm_unlock(shp);
 127         if (!is_file_hugepages(shp->shm_file))
 128                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 129         else
 130                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
 131                                                 shp->mlock_user);
 132         fput (shp->shm_file);
 133         security_shm_free(shp);
 134         put_vx_info(vxi);
 135         ipc_rcu_putref(shp);
 136 }
 137
 138 /*
 139  * remove the attach descriptor shmd.
 140  * free memory for segment if it is marked destroyed.
 141  * The descriptor has already been removed from the current->mm->mmap list
 142  * and will later be kfree()d.
 143  */
 144 static void shm_close (struct vm_area_struct *shmd)
 145 {
 146         struct file * file = shmd->vm_file;
 147         int id = file->f_dentry->d_inode->i_ino;
 148         struct shmid_kernel *shp;
 149
 150         down (&shm_ids.sem);
 151         /* remove from the list of attaches of the shm segment */
 152         if(!(shp = shm_lock(id)))
 153                 BUG();
 154         shp->shm_lprid = current->tgid;
 155         shp->shm_dtim = get_seconds();
 156         shp->shm_nattch--;
 157         if(shp->shm_nattch == 0 &&
 158            shp->shm_perm.mode & SHM_DEST)
 159                 shm_destroy (shp);
 160         else
 161                 shm_unlock(shp);
 162         up (&shm_ids.sem);
 163 }
 164
 165 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 166 {
 167         int ret;
 168
 169         ret = shmem_mmap(file, vma);
 170         if (ret == 0) {
 171                 vma->vm_ops = &shm_vm_ops;
 172                 if (!(vma->vm_flags & VM_WRITE))
 173                         vma->vm_flags &= ~VM_MAYWRITE;
 174                 shm_inc(file->f_dentry->d_inode->i_ino);
 175         }
 176
 177         return ret;
 178 }
 179
 180 static struct file_operations shm_file_operations = {
 181         .mmap   = shm_mmap,
 182 #ifndef CONFIG_MMU
 183         .get_unmapped_area = shmem_get_unmapped_area,
 184 #endif
 185 };
 186
 187 static struct vm_operations_struct shm_vm_ops = {
 188         .open   = shm_open,     /* callback for a new vm-area open */
 189         .close  = shm_close,    /* callback for when the vm-area is released */
 190         .nopage = shmem_nopage,
 191 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
 192         .set_policy = shmem_set_policy,
 193         .get_policy = shmem_get_policy,
 194 #endif
 195 };
 196
 197 static int newseg (key_t key, int shmflg, size_t size)
 198 {
 199         int error;
 200         struct shmid_kernel *shp;
 201         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 202         struct file * file;
 203         char name[13];
 204         int id;
 205
 206         if (size < SHMMIN || size > shm_ctlmax)
 207                 return -EINVAL;
 208
 209         if (shm_tot + numpages >= shm_ctlall)
 210                 return -ENOSPC;
 211         if (!vx_ipcshm_avail(current->vx_info, numpages))
 212                 return -ENOSPC;
 213
 214         shp = ipc_rcu_alloc(sizeof(*shp));
 215         if (!shp)
 216                 return -ENOMEM;
 217
 218         shp->shm_perm.key = key;
 219         shp->shm_perm.xid = vx_current_xid();
 220         shp->shm_perm.mode = (shmflg & S_IRWXUGO);
 221         shp->mlock_user = NULL;
 222
 223         shp->shm_perm.security = NULL;
 224         error = security_shm_alloc(shp);
 225         if (error) {
 226                 ipc_rcu_putref(shp);
 227                 return error;
 228         }
 229
 230         if (shmflg & SHM_HUGETLB) {
 231                 /* hugetlb_zero_setup takes care of mlock user accounting */
 232                 file = hugetlb_zero_setup(size);
 233                 shp->mlock_user = current->user;
 234         } else {
 235                 int acctflag = VM_ACCOUNT;
 236                 /*
 237                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 238                  * if it's asked for.
 239                  */
 240                 if  ((shmflg & SHM_NORESERVE) &&
 241                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 242                         acctflag = 0;
 243                 sprintf (name, "SYSV%08x", key);
 244                 file = shmem_file_setup(name, size, acctflag);
 245         }
 246         error = PTR_ERR(file);
 247         if (IS_ERR(file))
 248                 goto no_file;
 249
 250         error = -ENOSPC;
 251         id = shm_addid(shp);
 252         if(id == -1)
 253                 goto no_id;
 254
 255         shp->shm_cprid = current->tgid;
 256         shp->shm_lprid = 0;
 257         shp->shm_atim = shp->shm_dtim = 0;
 258         shp->shm_ctim = get_seconds();
 259         shp->shm_segsz = size;
 260         shp->shm_nattch = 0;
 261         shp->id = shm_buildid(id,shp->shm_perm.seq);
 262         shp->shm_file = file;
 263         file->f_dentry->d_inode->i_ino = shp->id;
 264
 265         /* Hugetlb ops would have already been assigned. */
 266         if (!(shmflg & SHM_HUGETLB))
 267                 file->f_op = &shm_file_operations;
 268
 269         shm_tot += numpages;
 270         vx_ipcshm_add(current->vx_info, key, numpages);
 271         shm_unlock(shp);
 272         return shp->id;
 273
 274 no_id:
 275         fput(file);
 276 no_file:
 277         security_shm_free(shp);
 278         ipc_rcu_putref(shp);
 279         return error;
 280 }
 281
 282 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 283 {
 284         struct shmid_kernel *shp;
 285         int err, id = 0;
 286
 287         down(&shm_ids.sem);
 288         if (key == IPC_PRIVATE) {
 289                 err = newseg(key, shmflg, size);
 290         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 291                 if (!(shmflg & IPC_CREAT))
 292                         err = -ENOENT;
 293                 else
 294                         err = newseg(key, shmflg, size);
 295         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 296                 err = -EEXIST;
 297         } else {
 298                 shp = shm_lock(id);
 299                 if(shp==NULL)
 300                         BUG();
 301                 if (shp->shm_segsz < size)
 302                         err = -EINVAL;
 303                 else if (ipcperms(&shp->shm_perm, shmflg))
 304                         err = -EACCES;
 305                 else {
 306                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 307                         err = security_shm_associate(shp, shmflg);
 308                         if (!err)
 309                                 err = shmid;
 310                 }
 311                 shm_unlock(shp);
 312         }
 313         up(&shm_ids.sem);
 314
 315         return err;
 316 }
 317
 318 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 319 {
 320         switch(version) {
 321         case IPC_64:
 322                 return copy_to_user(buf, in, sizeof(*in));
 323         case IPC_OLD:
 324             {
 325                 struct shmid_ds out;
 326
 327                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 328                 out.shm_segsz   = in->shm_segsz;
 329                 out.shm_atime   = in->shm_atime;
 330                 out.shm_dtime   = in->shm_dtime;
 331                 out.shm_ctime   = in->shm_ctime;
 332                 out.shm_cpid    = in->shm_cpid;
 333                 out.shm_lpid    = in->shm_lpid;
 334                 out.shm_nattch  = in->shm_nattch;
 335
 336                 return copy_to_user(buf, &out, sizeof(out));
 337             }
 338         default:
 339                 return -EINVAL;
 340         }
 341 }
 342
 343 struct shm_setbuf {
 344         uid_t   uid;
 345         gid_t   gid;
 346         mode_t  mode;
 347 };
 348
 349 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 350 {
 351         switch(version) {
 352         case IPC_64:
 353             {
 354                 struct shmid64_ds tbuf;
 355
 356                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 357                         return -EFAULT;
 358
 359                 out->uid        = tbuf.shm_perm.uid;
 360                 out->gid        = tbuf.shm_perm.gid;
 361                 out->mode       = tbuf.shm_perm.mode;
 362
 363                 return 0;
 364             }
 365         case IPC_OLD:
 366             {
 367                 struct shmid_ds tbuf_old;
 368
 369                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 370                         return -EFAULT;
 371
 372                 out->uid        = tbuf_old.shm_perm.uid;
 373                 out->gid        = tbuf_old.shm_perm.gid;
 374                 out->mode       = tbuf_old.shm_perm.mode;
 375
 376                 return 0;
 377             }
 378         default:
 379                 return -EINVAL;
 380         }
 381 }
 382
 383 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 384 {
 385         switch(version) {
 386         case IPC_64:
 387                 return copy_to_user(buf, in, sizeof(*in));
 388         case IPC_OLD:
 389             {
 390                 struct shminfo out;
 391
 392                 if(in->shmmax > INT_MAX)
 393                         out.shmmax = INT_MAX;
 394                 else
 395                         out.shmmax = (int)in->shmmax;
 396
 397                 out.shmmin      = in->shmmin;
 398                 out.shmmni      = in->shmmni;
 399                 out.shmseg      = in->shmseg;
 400                 out.shmall      = in->shmall;
 401
 402                 return copy_to_user(buf, &out, sizeof(out));
 403             }
 404         default:
 405                 return -EINVAL;
 406         }
 407 }
 408
 409 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 410 {
 411         int i;
 412
 413         *rss = 0;
 414         *swp = 0;
 415
 416         for (i = 0; i <= shm_ids.max_id; i++) {
 417                 struct shmid_kernel *shp;
 418                 struct inode *inode;
 419
 420                 shp = shm_get(i);
 421                 if(!shp)
 422                         continue;
 423
 424                 inode = shp->shm_file->f_dentry->d_inode;
 425
 426                 if (is_file_hugepages(shp->shm_file)) {
 427                         struct address_space *mapping = inode->i_mapping;
 428                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 429                 } else {
 430                         struct shmem_inode_info *info = SHMEM_I(inode);
 431                         spin_lock(&info->lock);
 432                         *rss += inode->i_mapping->nrpages;
 433                         *swp += info->swapped;
 434                         spin_unlock(&info->lock);
 435                 }
 436         }
 437 }
 438
 439 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 440 {
 441         struct shm_setbuf setbuf;
 442         struct shmid_kernel *shp;
 443         int err, version;
 444
 445         if (cmd < 0 || shmid < 0) {
 446                 err = -EINVAL;
 447                 goto out;
 448         }
 449
 450         version = ipc_parse_version(&cmd);
 451
 452         switch (cmd) { /* replace with proc interface ? */
 453         case IPC_INFO:
 454         {
 455                 struct shminfo64 shminfo;
 456
 457                 err = security_shm_shmctl(NULL, cmd);
 458                 if (err)
 459                         return err;
 460
 461                 memset(&shminfo,0,sizeof(shminfo));
 462                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 463                 shminfo.shmmax = shm_ctlmax;
 464                 shminfo.shmall = shm_ctlall;
 465
 466                 shminfo.shmmin = SHMMIN;
 467                 if(copy_shminfo_to_user (buf, &shminfo, version))
 468                         return -EFAULT;
 469                 /* reading a integer is always atomic */
 470                 err= shm_ids.max_id;
 471                 if(err<0)
 472                         err = 0;
 473                 goto out;
 474         }
 475         case SHM_INFO:
 476         {
 477                 struct shm_info shm_info;
 478
 479                 err = security_shm_shmctl(NULL, cmd);
 480                 if (err)
 481                         return err;
 482
 483                 memset(&shm_info,0,sizeof(shm_info));
 484                 down(&shm_ids.sem);
 485                 shm_info.used_ids = shm_ids.in_use;
 486                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 487                 shm_info.shm_tot = shm_tot;
 488                 shm_info.swap_attempts = 0;
 489                 shm_info.swap_successes = 0;
 490                 err = shm_ids.max_id;
 491                 up(&shm_ids.sem);
 492                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 493                         err = -EFAULT;
 494                         goto out;
 495                 }
 496
 497                 err = err < 0 ? 0 : err;
 498                 goto out;
 499         }
 500         case SHM_STAT:
 501         case IPC_STAT:
 502         {
 503                 struct shmid64_ds tbuf;
 504                 int result;
 505                 memset(&tbuf, 0, sizeof(tbuf));
 506                 shp = shm_lock(shmid);
 507                 if(shp==NULL) {
 508                         err = -EINVAL;
 509                         goto out;
 510                 } else if(cmd==SHM_STAT) {
 511                         err = -EINVAL;
 512                         if (shmid > shm_ids.max_id)
 513                                 goto out_unlock;
 514                         result = shm_buildid(shmid, shp->shm_perm.seq);
 515                 } else {
 516                         err = shm_checkid(shp,shmid);
 517                         if(err)
 518                                 goto out_unlock;
 519                         result = 0;
 520                 }
 521                 err=-EACCES;
 522                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 523                         goto out_unlock;
 524                 err = security_shm_shmctl(shp, cmd);
 525                 if (err)
 526                         goto out_unlock;
 527                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 528                 tbuf.shm_segsz  = shp->shm_segsz;
 529                 tbuf.shm_atime  = shp->shm_atim;
 530                 tbuf.shm_dtime  = shp->shm_dtim;
 531                 tbuf.shm_ctime  = shp->shm_ctim;
 532                 tbuf.shm_cpid   = shp->shm_cprid;
 533                 tbuf.shm_lpid   = shp->shm_lprid;
 534                 if (!is_file_hugepages(shp->shm_file))
 535                         tbuf.shm_nattch = shp->shm_nattch;
 536                 else
 537                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 538                 shm_unlock(shp);
 539                 if(copy_shmid_to_user (buf, &tbuf, version))
 540                         err = -EFAULT;
 541                 else
 542                         err = result;
 543                 goto out;
 544         }
 545         case SHM_LOCK:
 546         case SHM_UNLOCK:
 547         {
 548                 shp = shm_lock(shmid);
 549                 if(shp==NULL) {
 550                         err = -EINVAL;
 551                         goto out;
 552                 }
 553                 err = shm_checkid(shp,shmid);
 554                 if(err)
 555                         goto out_unlock;
 556
 557                 if (!capable(CAP_IPC_LOCK)) {
 558                         err = -EPERM;
 559                         if (current->euid != shp->shm_perm.uid &&
 560                             current->euid != shp->shm_perm.cuid)
 561                                 goto out_unlock;
 562                         if (cmd == SHM_LOCK &&
 563                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 564                                 goto out_unlock;
 565                 }
 566
 567                 err = security_shm_shmctl(shp, cmd);
 568                 if (err)
 569                         goto out_unlock;
 570
 571                 if(cmd==SHM_LOCK) {
 572                         struct user_struct * user = current->user;
 573                         if (!is_file_hugepages(shp->shm_file)) {
 574                                 err = shmem_lock(shp->shm_file, 1, user);
 575                                 if (!err) {
 576                                         shp->shm_perm.mode |= SHM_LOCKED;
 577                                         shp->mlock_user = user;
 578                                 }
 579                         }
 580                 } else if (!is_file_hugepages(shp->shm_file)) {
 581                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 582                         shp->shm_perm.mode &= ~SHM_LOCKED;
 583                         shp->mlock_user = NULL;
 584                 }
 585                 shm_unlock(shp);
 586                 goto out;
 587         }
 588         case IPC_RMID:
 589         {
 590                 /*
 591                  *      We cannot simply remove the file. The SVID states
 592                  *      that the block remains until the last person
 593                  *      detaches from it, then is deleted. A shmat() on
 594                  *      an RMID segment is legal in older Linux and if
 595                  *      we change it apps break...
 596                  *
 597                  *      Instead we set a destroyed flag, and then blow
 598                  *      the name away when the usage hits zero.
 599                  */
 600                 down(&shm_ids.sem);
 601                 shp = shm_lock(shmid);
 602                 err = -EINVAL;
 603                 if (shp == NULL)
 604                         goto out_up;
 605                 err = shm_checkid(shp, shmid);
 606                 if(err)
 607                         goto out_unlock_up;
 608
 609                 if (current->euid != shp->shm_perm.uid &&
 610                     current->euid != shp->shm_perm.cuid &&
 611                     !capable(CAP_SYS_ADMIN)) {
 612                         err=-EPERM;
 613                         goto out_unlock_up;
 614                 }
 615
 616                 err = security_shm_shmctl(shp, cmd);
 617                 if (err)
 618                         goto out_unlock_up;
 619
 620                 if (shp->shm_nattch){
 621                         shp->shm_perm.mode |= SHM_DEST;
 622                         /* Do not find it any more */
 623                         shp->shm_perm.key = IPC_PRIVATE;
 624                         shm_unlock(shp);
 625                 } else
 626                         shm_destroy (shp);
 627                 up(&shm_ids.sem);
 628                 goto out;
 629         }
 630
 631         case IPC_SET:
 632         {
 633                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 634                         err = -EFAULT;
 635                         goto out;
 636                 }
 637                 if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode)))
 638                         return err;
 639                 down(&shm_ids.sem);
 640                 shp = shm_lock(shmid);
 641                 err=-EINVAL;
 642                 if(shp==NULL)
 643                         goto out_up;
 644                 err = shm_checkid(shp,shmid);
 645                 if(err)
 646                         goto out_unlock_up;
 647                 err=-EPERM;
 648                 if (current->euid != shp->shm_perm.uid &&
 649                     current->euid != shp->shm_perm.cuid &&
 650                     !capable(CAP_SYS_ADMIN)) {
 651                         goto out_unlock_up;
 652                 }
 653
 654                 err = security_shm_shmctl(shp, cmd);
 655                 if (err)
 656                         goto out_unlock_up;
 657
 658                 shp->shm_perm.uid = setbuf.uid;
 659                 shp->shm_perm.gid = setbuf.gid;
 660                 shp->shm_perm.mode = (shp->shm_perm.mode & ~S_IRWXUGO)
 661                         | (setbuf.mode & S_IRWXUGO);
 662                 shp->shm_ctim = get_seconds();
 663                 break;
 664         }
 665
 666         default:
 667                 err = -EINVAL;
 668                 goto out;
 669         }
 670
 671         err = 0;
 672 out_unlock_up:
 673         shm_unlock(shp);
 674 out_up:
 675         up(&shm_ids.sem);
 676         goto out;
 677 out_unlock:
 678         shm_unlock(shp);
 679 out:
 680         return err;
 681 }
 682
 683 /*
 684  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 685  *
 686  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 687  * "raddr" thing points to kernel space, and there has to be a wrapper around
 688  * this.
 689  */
 690 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 691 {
 692         struct shmid_kernel *shp;
 693         unsigned long addr;
 694         unsigned long size;
 695         struct file * file;
 696         int    err;
 697         unsigned long flags;
 698         unsigned long prot;
 699         unsigned long o_flags;
 700         int acc_mode;
 701         void *user_addr;
 702
 703         if (shmid < 0) {
 704                 err = -EINVAL;
 705                 goto out;
 706         } else if ((addr = (ulong)shmaddr)) {
 707                 if (addr & (SHMLBA-1)) {
 708                         if (shmflg & SHM_RND)
 709                                 addr &= ~(SHMLBA-1);       /* round down */
 710                         else
 711 #ifndef __ARCH_FORCE_SHMLBA
 712                                 if (addr & ~PAGE_MASK)
 713 #endif
 714                                         return -EINVAL;
 715                 }
 716                 flags = MAP_SHARED | MAP_FIXED;
 717         } else {
 718                 if ((shmflg & SHM_REMAP))
 719                         return -EINVAL;
 720
 721                 flags = MAP_SHARED;
 722         }
 723
 724         if (shmflg & SHM_RDONLY) {
 725                 prot = PROT_READ;
 726                 o_flags = O_RDONLY;
 727                 acc_mode = S_IRUGO;
 728         } else {
 729                 prot = PROT_READ | PROT_WRITE;
 730                 o_flags = O_RDWR;
 731                 acc_mode = S_IRUGO | S_IWUGO;
 732         }
 733         if (shmflg & SHM_EXEC) {
 734                 prot |= PROT_EXEC;
 735                 acc_mode |= S_IXUGO;
 736         }
 737
 738         /*
 739          * We cannot rely on the fs check since SYSV IPC does have an
 740          * additional creator id...
 741          */
 742         shp = shm_lock(shmid);
 743         if(shp == NULL) {
 744                 err = -EINVAL;
 745                 goto out;
 746         }
 747         err = shm_checkid(shp,shmid);
 748         if (err) {
 749                 shm_unlock(shp);
 750                 goto out;
 751         }
 752         if (ipcperms(&shp->shm_perm, acc_mode)) {
 753                 shm_unlock(shp);
 754                 err = -EACCES;
 755                 goto out;
 756         }
 757
 758         err = security_shm_shmat(shp, shmaddr, shmflg);
 759         if (err) {
 760                 shm_unlock(shp);
 761                 return err;
 762         }
 763
 764         file = shp->shm_file;
 765         size = i_size_read(file->f_dentry->d_inode);
 766         shp->shm_nattch++;
 767         shm_unlock(shp);
 768
 769         down_write(&current->mm->mmap_sem);
 770         if (addr && !(shmflg & SHM_REMAP)) {
 771                 user_addr = ERR_PTR(-EINVAL);
 772                 if (find_vma_intersection(current->mm, addr, addr + size))
 773                         goto invalid;
 774                 /*
 775                  * If shm segment goes below stack, make sure there is some
 776                  * space left for the stack to grow (at least 4 pages).
 777                  */
 778                 if (addr < current->mm->start_stack &&
 779                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 780                         goto invalid;
 781         }
 782
 783         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 784
 785 invalid:
 786         up_write(&current->mm->mmap_sem);
 787
 788         down (&shm_ids.sem);
 789         if(!(shp = shm_lock(shmid)))
 790                 BUG();
 791         shp->shm_nattch--;
 792         if(shp->shm_nattch == 0 &&
 793            shp->shm_perm.mode & SHM_DEST)
 794                 shm_destroy (shp);
 795         else
 796                 shm_unlock(shp);
 797         up (&shm_ids.sem);
 798
 799         *raddr = (unsigned long) user_addr;
 800         err = 0;
 801         if (IS_ERR(user_addr))
 802                 err = PTR_ERR(user_addr);
 803 out:
 804         return err;
 805 }
 806
 807 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 808 {
 809         unsigned long ret;
 810         long err;
 811
 812         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 813         if (err)
 814                 return err;
 815         force_successful_syscall_return();
 816         return (long)ret;
 817 }
 818
 819 /*
 820  * detach and kill segment if marked destroyed.
 821  * The work is done in shm_close.
 822  */
 823 asmlinkage long sys_shmdt(char __user *shmaddr)
 824 {
 825         struct mm_struct *mm = current->mm;
 826         struct vm_area_struct *vma, *next;
 827         unsigned long addr = (unsigned long)shmaddr;
 828         loff_t size = 0;
 829         int retval = -EINVAL;
 830
 831         down_write(&mm->mmap_sem);
 832
 833         /*
 834          * This function tries to be smart and unmap shm segments that
 835          * were modified by partial mlock or munmap calls:
 836          * - It first determines the size of the shm segment that should be
 837          *   unmapped: It searches for a vma that is backed by shm and that
 838          *   started at address shmaddr. It records it's size and then unmaps
 839          *   it.
 840          * - Then it unmaps all shm vmas that started at shmaddr and that
 841          *   are within the initially determined size.
 842          * Errors from do_munmap are ignored: the function only fails if
 843          * it's called with invalid parameters or if it's called to unmap
 844          * a part of a vma. Both calls in this function are for full vmas,
 845          * the parameters are directly copied from the vma itself and always
 846          * valid - therefore do_munmap cannot fail. (famous last words?)
 847          */
 848         /*
 849          * If it had been mremap()'d, the starting address would not
 850          * match the usual checks anyway. So assume all vma's are
 851          * above the starting address given.
 852          */
 853         vma = find_vma(mm, addr);
 854
 855         while (vma) {
 856                 next = vma->vm_next;
 857
 858                 /*
 859                  * Check if the starting address would match, i.e. it's
 860                  * a fragment created by mprotect() and/or munmap(), or it
 861                  * otherwise it starts at this address with no hassles.
 862                  */
 863                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 864                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 865
 866
 867                         size = vma->vm_file->f_dentry->d_inode->i_size;
 868                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 869                         /*
 870                          * We discovered the size of the shm segment, so
 871                          * break out of here and fall through to the next
 872                          * loop that uses the size information to stop
 873                          * searching for matching vma's.
 874                          */
 875                         retval = 0;
 876                         vma = next;
 877                         break;
 878                 }
 879                 vma = next;
 880         }
 881
 882         /*
 883          * We need look no further than the maximum address a fragment
 884          * could possibly have landed at. Also cast things to loff_t to
 885          * prevent overflows and make comparisions vs. equal-width types.
 886          */
 887         size = PAGE_ALIGN(size);
 888         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 889                 next = vma->vm_next;
 890
 891                 /* finding a matching vma now does not alter retval */
 892                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 893                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 894
 895                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 896                 vma = next;
 897         }
 898
 899         up_write(&mm->mmap_sem);
 900         return retval;
 901 }
 902
 903 #ifdef CONFIG_PROC_FS
 904 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
 905 {
 906         struct shmid_kernel *shp = it;
 907         char *format;
 908
 909 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 910 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 911
 912         if (!vx_check(shp->shm_perm.xid, VX_IDENT))
 913                 return 0;
 914
 915         if (sizeof(size_t) <= sizeof(int))
 916                 format = SMALL_STRING;
 917         else
 918                 format = BIG_STRING;
 919         return seq_printf(s, format,
 920                           shp->shm_perm.key,
 921                           shp->id,
 922                           shp->shm_perm.mode,
 923                           shp->shm_segsz,
 924                           shp->shm_cprid,
 925                           shp->shm_lprid,
 926                           is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 927                           shp->shm_perm.uid,
 928                           shp->shm_perm.gid,
 929                           shp->shm_perm.cuid,
 930                           shp->shm_perm.cgid,
 931                           shp->shm_atim,
 932                           shp->shm_dtim,
 933                           shp->shm_ctim);
 934 }
 935 #endif