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