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