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