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