SAFE public projects git trees. - safe/jmp/linux-2.6/blob - 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/ptrace.h>
  31 #include <linux/seq_file.h>
  32
  33 #include <asm/uaccess.h>
  34
  35 #include "util.h"
  36
  37 #define shm_flags       shm_perm.mode
  38
  39 static struct file_operations shm_file_operations;
  40 static struct vm_operations_struct shm_vm_ops;
  41
  42 static struct ipc_ids shm_ids;
  43
  44 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
  45 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
  46 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
  47 #define shm_buildid(id, seq) \
  48         ipc_buildid(&shm_ids, id, seq)
  49
  50 static int newseg (key_t key, int shmflg, size_t size);
  51 static void shm_open (struct vm_area_struct *shmd);
  52 static void shm_close (struct vm_area_struct *shmd);
  53 #ifdef CONFIG_PROC_FS
  54 static int sysvipc_shm_proc_show(struct seq_file *s, void *it);
  55 #endif
  56
  57 size_t  shm_ctlmax = SHMMAX;
  58 size_t  shm_ctlall = SHMALL;
  59 int     shm_ctlmni = SHMMNI;
  60
  61 static int shm_tot; /* total number of shared memory pages */
  62
  63 void __init shm_init (void)
  64 {
  65         ipc_init_ids(&shm_ids, 1);
  66         ipc_init_proc_interface("sysvipc/shm",
  67                                 "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n",
  68                                 &shm_ids,
  69                                 sysvipc_shm_proc_show);
  70 }
  71
  72 static inline int shm_checkid(struct shmid_kernel *s, int id)
  73 {
  74         if (ipc_checkid(&shm_ids,&s->shm_perm,id))
  75                 return -EIDRM;
  76         return 0;
  77 }
  78
  79 static inline struct shmid_kernel *shm_rmid(int id)
  80 {
  81         return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
  82 }
  83
  84 static inline int shm_addid(struct shmid_kernel *shp)
  85 {
  86         return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
  87 }
  88
  89
  90
  91 static inline void shm_inc (int id) {
  92         struct shmid_kernel *shp;
  93
  94         if(!(shp = shm_lock(id)))
  95                 BUG();
  96         shp->shm_atim = get_seconds();
  97         shp->shm_lprid = current->tgid;
  98         shp->shm_nattch++;
  99         shm_unlock(shp);
 100 }
 101
 102 /* This is called by fork, once for every shm attach. */
 103 static void shm_open (struct vm_area_struct *shmd)
 104 {
 105         shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
 106 }
 107
 108 /*
 109  * shm_destroy - free the struct shmid_kernel
 110  *
 111  * @shp: struct to free
 112  *
 113  * It has to be called with shp and shm_ids.sem locked,
 114  * but returns with shp unlocked and freed.
 115  */
 116 static void shm_destroy (struct shmid_kernel *shp)
 117 {
 118         shm_tot -= (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 119         shm_rmid (shp->id);
 120         shm_unlock(shp);
 121         if (!is_file_hugepages(shp->shm_file))
 122                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 123         else
 124                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
 125                                                 shp->mlock_user);
 126         fput (shp->shm_file);
 127         security_shm_free(shp);
 128         ipc_rcu_putref(shp);
 129 }
 130
 131 /*
 132  * remove the attach descriptor shmd.
 133  * free memory for segment if it is marked destroyed.
 134  * The descriptor has already been removed from the current->mm->mmap list
 135  * and will later be kfree()d.
 136  */
 137 static void shm_close (struct vm_area_struct *shmd)
 138 {
 139         struct file * file = shmd->vm_file;
 140         int id = file->f_dentry->d_inode->i_ino;
 141         struct shmid_kernel *shp;
 142
 143         down (&shm_ids.sem);
 144         /* remove from the list of attaches of the shm segment */
 145         if(!(shp = shm_lock(id)))
 146                 BUG();
 147         shp->shm_lprid = current->tgid;
 148         shp->shm_dtim = get_seconds();
 149         shp->shm_nattch--;
 150         if(shp->shm_nattch == 0 &&
 151            shp->shm_flags & SHM_DEST)
 152                 shm_destroy (shp);
 153         else
 154                 shm_unlock(shp);
 155         up (&shm_ids.sem);
 156 }
 157
 158 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 159 {
 160         file_accessed(file);
 161         vma->vm_ops = &shm_vm_ops;
 162         shm_inc(file->f_dentry->d_inode->i_ino);
 163         return 0;
 164 }
 165
 166 static struct file_operations shm_file_operations = {
 167         .mmap   = shm_mmap
 168 };
 169
 170 static struct vm_operations_struct shm_vm_ops = {
 171         .open   = shm_open,     /* callback for a new vm-area open */
 172         .close  = shm_close,    /* callback for when the vm-area is released */
 173         .nopage = shmem_nopage,
 174 #if defined(CONFIG_NUMA) && defined(CONFIG_SHMEM)
 175         .set_policy = shmem_set_policy,
 176         .get_policy = shmem_get_policy,
 177 #endif
 178 };
 179
 180 static int newseg (key_t key, int shmflg, size_t size)
 181 {
 182         int error;
 183         struct shmid_kernel *shp;
 184         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 185         struct file * file;
 186         char name[13];
 187         int id;
 188
 189         if (size < SHMMIN || size > shm_ctlmax)
 190                 return -EINVAL;
 191
 192         if (shm_tot + numpages >= shm_ctlall)
 193                 return -ENOSPC;
 194
 195         shp = ipc_rcu_alloc(sizeof(*shp));
 196         if (!shp)
 197                 return -ENOMEM;
 198
 199         shp->shm_perm.key = key;
 200         shp->shm_flags = (shmflg & S_IRWXUGO);
 201         shp->mlock_user = NULL;
 202
 203         shp->shm_perm.security = NULL;
 204         error = security_shm_alloc(shp);
 205         if (error) {
 206                 ipc_rcu_putref(shp);
 207                 return error;
 208         }
 209
 210         if (shmflg & SHM_HUGETLB) {
 211                 /* hugetlb_zero_setup takes care of mlock user accounting */
 212                 file = hugetlb_zero_setup(size);
 213                 shp->mlock_user = current->user;
 214         } else {
 215                 int acctflag = VM_ACCOUNT;
 216                 /*
 217                  * Do not allow no accounting for OVERCOMMIT_NEVER, even
 218                  * if it's asked for.
 219                  */
 220                 if  ((shmflg & SHM_NORESERVE) &&
 221                                 sysctl_overcommit_memory != OVERCOMMIT_NEVER)
 222                         acctflag = 0;
 223                 sprintf (name, "SYSV%08x", key);
 224                 file = shmem_file_setup(name, size, acctflag);
 225         }
 226         error = PTR_ERR(file);
 227         if (IS_ERR(file))
 228                 goto no_file;
 229
 230         error = -ENOSPC;
 231         id = shm_addid(shp);
 232         if(id == -1)
 233                 goto no_id;
 234
 235         shp->shm_cprid = current->tgid;
 236         shp->shm_lprid = 0;
 237         shp->shm_atim = shp->shm_dtim = 0;
 238         shp->shm_ctim = get_seconds();
 239         shp->shm_segsz = size;
 240         shp->shm_nattch = 0;
 241         shp->id = shm_buildid(id,shp->shm_perm.seq);
 242         shp->shm_file = file;
 243         file->f_dentry->d_inode->i_ino = shp->id;
 244
 245         /* Hugetlb ops would have already been assigned. */
 246         if (!(shmflg & SHM_HUGETLB))
 247                 file->f_op = &shm_file_operations;
 248
 249         shm_tot += numpages;
 250         shm_unlock(shp);
 251         return shp->id;
 252
 253 no_id:
 254         fput(file);
 255 no_file:
 256         security_shm_free(shp);
 257         ipc_rcu_putref(shp);
 258         return error;
 259 }
 260
 261 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 262 {
 263         struct shmid_kernel *shp;
 264         int err, id = 0;
 265
 266         down(&shm_ids.sem);
 267         if (key == IPC_PRIVATE) {
 268                 err = newseg(key, shmflg, size);
 269         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 270                 if (!(shmflg & IPC_CREAT))
 271                         err = -ENOENT;
 272                 else
 273                         err = newseg(key, shmflg, size);
 274         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 275                 err = -EEXIST;
 276         } else {
 277                 shp = shm_lock(id);
 278                 if(shp==NULL)
 279                         BUG();
 280                 if (shp->shm_segsz < size)
 281                         err = -EINVAL;
 282                 else if (ipcperms(&shp->shm_perm, shmflg))
 283                         err = -EACCES;
 284                 else {
 285                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 286                         err = security_shm_associate(shp, shmflg);
 287                         if (!err)
 288                                 err = shmid;
 289                 }
 290                 shm_unlock(shp);
 291         }
 292         up(&shm_ids.sem);
 293
 294         return err;
 295 }
 296
 297 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 298 {
 299         switch(version) {
 300         case IPC_64:
 301                 return copy_to_user(buf, in, sizeof(*in));
 302         case IPC_OLD:
 303             {
 304                 struct shmid_ds out;
 305
 306                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 307                 out.shm_segsz   = in->shm_segsz;
 308                 out.shm_atime   = in->shm_atime;
 309                 out.shm_dtime   = in->shm_dtime;
 310                 out.shm_ctime   = in->shm_ctime;
 311                 out.shm_cpid    = in->shm_cpid;
 312                 out.shm_lpid    = in->shm_lpid;
 313                 out.shm_nattch  = in->shm_nattch;
 314
 315                 return copy_to_user(buf, &out, sizeof(out));
 316             }
 317         default:
 318                 return -EINVAL;
 319         }
 320 }
 321
 322 struct shm_setbuf {
 323         uid_t   uid;
 324         gid_t   gid;
 325         mode_t  mode;
 326 };
 327
 328 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 329 {
 330         switch(version) {
 331         case IPC_64:
 332             {
 333                 struct shmid64_ds tbuf;
 334
 335                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 336                         return -EFAULT;
 337
 338                 out->uid        = tbuf.shm_perm.uid;
 339                 out->gid        = tbuf.shm_perm.gid;
 340                 out->mode       = tbuf.shm_flags;
 341
 342                 return 0;
 343             }
 344         case IPC_OLD:
 345             {
 346                 struct shmid_ds tbuf_old;
 347
 348                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 349                         return -EFAULT;
 350
 351                 out->uid        = tbuf_old.shm_perm.uid;
 352                 out->gid        = tbuf_old.shm_perm.gid;
 353                 out->mode       = tbuf_old.shm_flags;
 354
 355                 return 0;
 356             }
 357         default:
 358                 return -EINVAL;
 359         }
 360 }
 361
 362 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 363 {
 364         switch(version) {
 365         case IPC_64:
 366                 return copy_to_user(buf, in, sizeof(*in));
 367         case IPC_OLD:
 368             {
 369                 struct shminfo out;
 370
 371                 if(in->shmmax > INT_MAX)
 372                         out.shmmax = INT_MAX;
 373                 else
 374                         out.shmmax = (int)in->shmmax;
 375
 376                 out.shmmin      = in->shmmin;
 377                 out.shmmni      = in->shmmni;
 378                 out.shmseg      = in->shmseg;
 379                 out.shmall      = in->shmall;
 380
 381                 return copy_to_user(buf, &out, sizeof(out));
 382             }
 383         default:
 384                 return -EINVAL;
 385         }
 386 }
 387
 388 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 389 {
 390         int i;
 391
 392         *rss = 0;
 393         *swp = 0;
 394
 395         for (i = 0; i <= shm_ids.max_id; i++) {
 396                 struct shmid_kernel *shp;
 397                 struct inode *inode;
 398
 399                 shp = shm_get(i);
 400                 if(!shp)
 401                         continue;
 402
 403                 inode = shp->shm_file->f_dentry->d_inode;
 404
 405                 if (is_file_hugepages(shp->shm_file)) {
 406                         struct address_space *mapping = inode->i_mapping;
 407                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 408                 } else {
 409                         struct shmem_inode_info *info = SHMEM_I(inode);
 410                         spin_lock(&info->lock);
 411                         *rss += inode->i_mapping->nrpages;
 412                         *swp += info->swapped;
 413                         spin_unlock(&info->lock);
 414                 }
 415         }
 416 }
 417
 418 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 419 {
 420         struct shm_setbuf setbuf;
 421         struct shmid_kernel *shp;
 422         int err, version;
 423
 424         if (cmd < 0 || shmid < 0) {
 425                 err = -EINVAL;
 426                 goto out;
 427         }
 428
 429         version = ipc_parse_version(&cmd);
 430
 431         switch (cmd) { /* replace with proc interface ? */
 432         case IPC_INFO:
 433         {
 434                 struct shminfo64 shminfo;
 435
 436                 err = security_shm_shmctl(NULL, cmd);
 437                 if (err)
 438                         return err;
 439
 440                 memset(&shminfo,0,sizeof(shminfo));
 441                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 442                 shminfo.shmmax = shm_ctlmax;
 443                 shminfo.shmall = shm_ctlall;
 444
 445                 shminfo.shmmin = SHMMIN;
 446                 if(copy_shminfo_to_user (buf, &shminfo, version))
 447                         return -EFAULT;
 448                 /* reading a integer is always atomic */
 449                 err= shm_ids.max_id;
 450                 if(err<0)
 451                         err = 0;
 452                 goto out;
 453         }
 454         case SHM_INFO:
 455         {
 456                 struct shm_info shm_info;
 457
 458                 err = security_shm_shmctl(NULL, cmd);
 459                 if (err)
 460                         return err;
 461
 462                 memset(&shm_info,0,sizeof(shm_info));
 463                 down(&shm_ids.sem);
 464                 shm_info.used_ids = shm_ids.in_use;
 465                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 466                 shm_info.shm_tot = shm_tot;
 467                 shm_info.swap_attempts = 0;
 468                 shm_info.swap_successes = 0;
 469                 err = shm_ids.max_id;
 470                 up(&shm_ids.sem);
 471                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 472                         err = -EFAULT;
 473                         goto out;
 474                 }
 475
 476                 err = err < 0 ? 0 : err;
 477                 goto out;
 478         }
 479         case SHM_STAT:
 480         case IPC_STAT:
 481         {
 482                 struct shmid64_ds tbuf;
 483                 int result;
 484                 memset(&tbuf, 0, sizeof(tbuf));
 485                 shp = shm_lock(shmid);
 486                 if(shp==NULL) {
 487                         err = -EINVAL;
 488                         goto out;
 489                 } else if(cmd==SHM_STAT) {
 490                         err = -EINVAL;
 491                         if (shmid > shm_ids.max_id)
 492                                 goto out_unlock;
 493                         result = shm_buildid(shmid, shp->shm_perm.seq);
 494                 } else {
 495                         err = shm_checkid(shp,shmid);
 496                         if(err)
 497                                 goto out_unlock;
 498                         result = 0;
 499                 }
 500                 err=-EACCES;
 501                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 502                         goto out_unlock;
 503                 err = security_shm_shmctl(shp, cmd);
 504                 if (err)
 505                         goto out_unlock;
 506                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 507                 tbuf.shm_segsz  = shp->shm_segsz;
 508                 tbuf.shm_atime  = shp->shm_atim;
 509                 tbuf.shm_dtime  = shp->shm_dtim;
 510                 tbuf.shm_ctime  = shp->shm_ctim;
 511                 tbuf.shm_cpid   = shp->shm_cprid;
 512                 tbuf.shm_lpid   = shp->shm_lprid;
 513                 if (!is_file_hugepages(shp->shm_file))
 514                         tbuf.shm_nattch = shp->shm_nattch;
 515                 else
 516                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 517                 shm_unlock(shp);
 518                 if(copy_shmid_to_user (buf, &tbuf, version))
 519                         err = -EFAULT;
 520                 else
 521                         err = result;
 522                 goto out;
 523         }
 524         case SHM_LOCK:
 525         case SHM_UNLOCK:
 526         {
 527                 shp = shm_lock(shmid);
 528                 if(shp==NULL) {
 529                         err = -EINVAL;
 530                         goto out;
 531                 }
 532                 err = shm_checkid(shp,shmid);
 533                 if(err)
 534                         goto out_unlock;
 535
 536                 if (!capable(CAP_IPC_LOCK)) {
 537                         err = -EPERM;
 538                         if (current->euid != shp->shm_perm.uid &&
 539                             current->euid != shp->shm_perm.cuid)
 540                                 goto out_unlock;
 541                         if (cmd == SHM_LOCK &&
 542                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 543                                 goto out_unlock;
 544                 }
 545
 546                 err = security_shm_shmctl(shp, cmd);
 547                 if (err)
 548                         goto out_unlock;
 549
 550                 if(cmd==SHM_LOCK) {
 551                         struct user_struct * user = current->user;
 552                         if (!is_file_hugepages(shp->shm_file)) {
 553                                 err = shmem_lock(shp->shm_file, 1, user);
 554                                 if (!err) {
 555                                         shp->shm_flags |= SHM_LOCKED;
 556                                         shp->mlock_user = user;
 557                                 }
 558                         }
 559                 } else if (!is_file_hugepages(shp->shm_file)) {
 560                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 561                         shp->shm_flags &= ~SHM_LOCKED;
 562                         shp->mlock_user = NULL;
 563                 }
 564                 shm_unlock(shp);
 565                 goto out;
 566         }
 567         case IPC_RMID:
 568         {
 569                 /*
 570                  *      We cannot simply remove the file. The SVID states
 571                  *      that the block remains until the last person
 572                  *      detaches from it, then is deleted. A shmat() on
 573                  *      an RMID segment is legal in older Linux and if
 574                  *      we change it apps break...
 575                  *
 576                  *      Instead we set a destroyed flag, and then blow
 577                  *      the name away when the usage hits zero.
 578                  */
 579                 down(&shm_ids.sem);
 580                 shp = shm_lock(shmid);
 581                 err = -EINVAL;
 582                 if (shp == NULL)
 583                         goto out_up;
 584                 err = shm_checkid(shp, shmid);
 585                 if(err)
 586                         goto out_unlock_up;
 587
 588                 if (current->euid != shp->shm_perm.uid &&
 589                     current->euid != shp->shm_perm.cuid &&
 590                     !capable(CAP_SYS_ADMIN)) {
 591                         err=-EPERM;
 592                         goto out_unlock_up;
 593                 }
 594
 595                 err = security_shm_shmctl(shp, cmd);
 596                 if (err)
 597                         goto out_unlock_up;
 598
 599                 if (shp->shm_nattch){
 600                         shp->shm_flags |= SHM_DEST;
 601                         /* Do not find it any more */
 602                         shp->shm_perm.key = IPC_PRIVATE;
 603                         shm_unlock(shp);
 604                 } else
 605                         shm_destroy (shp);
 606                 up(&shm_ids.sem);
 607                 goto out;
 608         }
 609
 610         case IPC_SET:
 611         {
 612                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 613                         err = -EFAULT;
 614                         goto out;
 615                 }
 616                 if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode)))
 617                         return err;
 618                 down(&shm_ids.sem);
 619                 shp = shm_lock(shmid);
 620                 err=-EINVAL;
 621                 if(shp==NULL)
 622                         goto out_up;
 623                 err = shm_checkid(shp,shmid);
 624                 if(err)
 625                         goto out_unlock_up;
 626                 err=-EPERM;
 627                 if (current->euid != shp->shm_perm.uid &&
 628                     current->euid != shp->shm_perm.cuid &&
 629                     !capable(CAP_SYS_ADMIN)) {
 630                         goto out_unlock_up;
 631                 }
 632
 633                 err = security_shm_shmctl(shp, cmd);
 634                 if (err)
 635                         goto out_unlock_up;
 636
 637                 shp->shm_perm.uid = setbuf.uid;
 638                 shp->shm_perm.gid = setbuf.gid;
 639                 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
 640                         | (setbuf.mode & S_IRWXUGO);
 641                 shp->shm_ctim = get_seconds();
 642                 break;
 643         }
 644
 645         default:
 646                 err = -EINVAL;
 647                 goto out;
 648         }
 649
 650         err = 0;
 651 out_unlock_up:
 652         shm_unlock(shp);
 653 out_up:
 654         up(&shm_ids.sem);
 655         goto out;
 656 out_unlock:
 657         shm_unlock(shp);
 658 out:
 659         return err;
 660 }
 661
 662 /*
 663  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 664  *
 665  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 666  * "raddr" thing points to kernel space, and there has to be a wrapper around
 667  * this.
 668  */
 669 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 670 {
 671         struct shmid_kernel *shp;
 672         unsigned long addr;
 673         unsigned long size;
 674         struct file * file;
 675         int    err;
 676         unsigned long flags;
 677         unsigned long prot;
 678         unsigned long o_flags;
 679         int acc_mode;
 680         void *user_addr;
 681
 682         if (shmid < 0) {
 683                 err = -EINVAL;
 684                 goto out;
 685         } else if ((addr = (ulong)shmaddr)) {
 686                 if (addr & (SHMLBA-1)) {
 687                         if (shmflg & SHM_RND)
 688                                 addr &= ~(SHMLBA-1);       /* round down */
 689                         else
 690 #ifndef __ARCH_FORCE_SHMLBA
 691                                 if (addr & ~PAGE_MASK)
 692 #endif
 693                                         return -EINVAL;
 694                 }
 695                 flags = MAP_SHARED | MAP_FIXED;
 696         } else {
 697                 if ((shmflg & SHM_REMAP))
 698                         return -EINVAL;
 699
 700                 flags = MAP_SHARED;
 701         }
 702
 703         if (shmflg & SHM_RDONLY) {
 704                 prot = PROT_READ;
 705                 o_flags = O_RDONLY;
 706                 acc_mode = S_IRUGO;
 707         } else {
 708                 prot = PROT_READ | PROT_WRITE;
 709                 o_flags = O_RDWR;
 710                 acc_mode = S_IRUGO | S_IWUGO;
 711         }
 712         if (shmflg & SHM_EXEC) {
 713                 prot |= PROT_EXEC;
 714                 acc_mode |= S_IXUGO;
 715         }
 716
 717         /*
 718          * We cannot rely on the fs check since SYSV IPC does have an
 719          * additional creator id...
 720          */
 721         shp = shm_lock(shmid);
 722         if(shp == NULL) {
 723                 err = -EINVAL;
 724                 goto out;
 725         }
 726         err = shm_checkid(shp,shmid);
 727         if (err) {
 728                 shm_unlock(shp);
 729                 goto out;
 730         }
 731         if (ipcperms(&shp->shm_perm, acc_mode)) {
 732                 shm_unlock(shp);
 733                 err = -EACCES;
 734                 goto out;
 735         }
 736
 737         err = security_shm_shmat(shp, shmaddr, shmflg);
 738         if (err) {
 739                 shm_unlock(shp);
 740                 return err;
 741         }
 742
 743         file = shp->shm_file;
 744         size = i_size_read(file->f_dentry->d_inode);
 745         shp->shm_nattch++;
 746         shm_unlock(shp);
 747
 748         down_write(&current->mm->mmap_sem);
 749         if (addr && !(shmflg & SHM_REMAP)) {
 750                 user_addr = ERR_PTR(-EINVAL);
 751                 if (find_vma_intersection(current->mm, addr, addr + size))
 752                         goto invalid;
 753                 /*
 754                  * If shm segment goes below stack, make sure there is some
 755                  * space left for the stack to grow (at least 4 pages).
 756                  */
 757                 if (addr < current->mm->start_stack &&
 758                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 759                         goto invalid;
 760         }
 761
 762         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 763
 764 invalid:
 765         up_write(&current->mm->mmap_sem);
 766
 767         down (&shm_ids.sem);
 768         if(!(shp = shm_lock(shmid)))
 769                 BUG();
 770         shp->shm_nattch--;
 771         if(shp->shm_nattch == 0 &&
 772            shp->shm_flags & SHM_DEST)
 773                 shm_destroy (shp);
 774         else
 775                 shm_unlock(shp);
 776         up (&shm_ids.sem);
 777
 778         *raddr = (unsigned long) user_addr;
 779         err = 0;
 780         if (IS_ERR(user_addr))
 781                 err = PTR_ERR(user_addr);
 782 out:
 783         return err;
 784 }
 785
 786 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 787 {
 788         unsigned long ret;
 789         long err;
 790
 791         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 792         if (err)
 793                 return err;
 794         force_successful_syscall_return();
 795         return (long)ret;
 796 }
 797
 798 /*
 799  * detach and kill segment if marked destroyed.
 800  * The work is done in shm_close.
 801  */
 802 asmlinkage long sys_shmdt(char __user *shmaddr)
 803 {
 804         struct mm_struct *mm = current->mm;
 805         struct vm_area_struct *vma, *next;
 806         unsigned long addr = (unsigned long)shmaddr;
 807         loff_t size = 0;
 808         int retval = -EINVAL;
 809
 810         down_write(&mm->mmap_sem);
 811
 812         /*
 813          * This function tries to be smart and unmap shm segments that
 814          * were modified by partial mlock or munmap calls:
 815          * - It first determines the size of the shm segment that should be
 816          *   unmapped: It searches for a vma that is backed by shm and that
 817          *   started at address shmaddr. It records it's size and then unmaps
 818          *   it.
 819          * - Then it unmaps all shm vmas that started at shmaddr and that
 820          *   are within the initially determined size.
 821          * Errors from do_munmap are ignored: the function only fails if
 822          * it's called with invalid parameters or if it's called to unmap
 823          * a part of a vma. Both calls in this function are for full vmas,
 824          * the parameters are directly copied from the vma itself and always
 825          * valid - therefore do_munmap cannot fail. (famous last words?)
 826          */
 827         /*
 828          * If it had been mremap()'d, the starting address would not
 829          * match the usual checks anyway. So assume all vma's are
 830          * above the starting address given.
 831          */
 832         vma = find_vma(mm, addr);
 833
 834         while (vma) {
 835                 next = vma->vm_next;
 836
 837                 /*
 838                  * Check if the starting address would match, i.e. it's
 839                  * a fragment created by mprotect() and/or munmap(), or it
 840                  * otherwise it starts at this address with no hassles.
 841                  */
 842                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 843                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 844
 845
 846                         size = vma->vm_file->f_dentry->d_inode->i_size;
 847                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 848                         /*
 849                          * We discovered the size of the shm segment, so
 850                          * break out of here and fall through to the next
 851                          * loop that uses the size information to stop
 852                          * searching for matching vma's.
 853                          */
 854                         retval = 0;
 855                         vma = next;
 856                         break;
 857                 }
 858                 vma = next;
 859         }
 860
 861         /*
 862          * We need look no further than the maximum address a fragment
 863          * could possibly have landed at. Also cast things to loff_t to
 864          * prevent overflows and make comparisions vs. equal-width types.
 865          */
 866         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 867                 next = vma->vm_next;
 868
 869                 /* finding a matching vma now does not alter retval */
 870                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 871                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 872
 873                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 874                 vma = next;
 875         }
 876
 877         up_write(&mm->mmap_sem);
 878         return retval;
 879 }
 880
 881 #ifdef CONFIG_PROC_FS
 882 static int sysvipc_shm_proc_show(struct seq_file *s, void *it)
 883 {
 884         struct shmid_kernel *shp = it;
 885         char *format;
 886
 887 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 888 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 889
 890         if (sizeof(size_t) <= sizeof(int))
 891                 format = SMALL_STRING;
 892         else
 893                 format = BIG_STRING;
 894         return seq_printf(s, format,
 895                           shp->shm_perm.key,
 896                           shp->id,
 897                           shp->shm_flags,
 898                           shp->shm_segsz,
 899                           shp->shm_cprid,
 900                           shp->shm_lprid,
 901                           is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 902                           shp->shm_perm.uid,
 903                           shp->shm_perm.gid,
 904                           shp->shm_perm.cuid,
 905                           shp->shm_perm.cgid,
 906                           shp->shm_atim,
 907                           shp->shm_dtim,
 908                           shp->shm_ctim);
 909 }
 910 #endif