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/syscalls.h>
  30 #include <linux/audit.h>
  31 #include <linux/ptrace.h>
  32 #include <linux/vs_limit.h>
  33
  34 #include <asm/uaccess.h>
  35
  36 #include "util.h"
  37
  38 #define shm_flags       shm_perm.mode
  39
  40 static struct file_operations shm_file_operations;
  41 static struct vm_operations_struct shm_vm_ops;
  42
  43 static struct ipc_ids shm_ids;
  44
  45 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
  46 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
  47 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
  48 #define shm_buildid(id, seq) \
  49         ipc_buildid(&shm_ids, id, seq)
  50
  51 static int newseg (key_t key, int shmflg, size_t size);
  52 static void shm_open (struct vm_area_struct *shmd);
  53 static void shm_close (struct vm_area_struct *shmd);
  54 #ifdef CONFIG_PROC_FS
  55 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
  56 #endif
  57
  58 size_t  shm_ctlmax = SHMMAX;
  59 size_t  shm_ctlall = SHMALL;
  60 int     shm_ctlmni = SHMMNI;
  61
  62 static int shm_tot; /* total number of shared memory pages */
  63
  64 void __init shm_init (void)
  65 {
  66         ipc_init_ids(&shm_ids, 1);
  67 #ifdef CONFIG_PROC_FS
  68         create_proc_read_entry("sysvipc/shm", 0, NULL, sysvipc_shm_read_proc, NULL);
  69 #endif
  70 }
  71
  72 static inline int shm_checkid(struct shmid_kernel *s, int id)
  73 {
  74         if (ipc_checkid(&shm_ids,&s->shm_perm,id))
  75                 return -EIDRM;
  76         return 0;
  77 }
  78
  79 static inline struct shmid_kernel *shm_rmid(int id)
  80 {
  81         return (struct shmid_kernel *)ipc_rmid(&shm_ids,id);
  82 }
  83
  84 static inline int shm_addid(struct shmid_kernel *shp)
  85 {
  86         return ipc_addid(&shm_ids, &shp->shm_perm, shm_ctlmni);
  87 }
  88
  89
  90
  91 static inline void shm_inc (int id) {
  92         struct shmid_kernel *shp;
  93
  94         if(!(shp = shm_lock(id)))
  95                 BUG();
  96         shp->shm_atim = get_seconds();
  97         shp->shm_lprid = current->tgid;
  98         shp->shm_nattch++;
  99         shm_unlock(shp);
 100 }
 101
 102 /* This is called by fork, once for every shm attach. */
 103 static void shm_open (struct vm_area_struct *shmd)
 104 {
 105         shm_inc (shmd->vm_file->f_dentry->d_inode->i_ino);
 106 }
 107
 108 /*
 109  * shm_destroy - free the struct shmid_kernel
 110  *
 111  * @shp: struct to free
 112  *
 113  * It has to be called with shp and shm_ids.sem locked,
 114  * but returns with shp unlocked and freed.
 115  */
 116 static void shm_destroy (struct shmid_kernel *shp)
 117 {
 118         struct vx_info *vxi = locate_vx_info(shp->shm_perm.xid);
 119         int numpages = (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 120
 121         vx_ipcshm_sub(vxi, shp, numpages);
 122         shm_tot -= numpages;
 123
 124         shm_rmid (shp->id);
 125         shm_unlock(shp);
 126         if (!is_file_hugepages(shp->shm_file))
 127                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 128         else
 129                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
 130                                                 shp->mlock_user);
 131         fput (shp->shm_file);
 132         security_shm_free(shp);
 133         put_vx_info(vxi);
 134         ipc_rcu_putref(shp);
 135 }
 136
 137 /*
 138  * remove the attach descriptor shmd.
 139  * free memory for segment if it is marked destroyed.
 140  * The descriptor has already been removed from the current->mm->mmap list
 141  * and will later be kfree()d.
 142  */
 143 static void shm_close (struct vm_area_struct *shmd)
 144 {
 145         struct file * file = shmd->vm_file;
 146         int id = file->f_dentry->d_inode->i_ino;
 147         struct shmid_kernel *shp;
 148
 149         down (&shm_ids.sem);
 150         /* remove from the list of attaches of the shm segment */
 151         if(!(shp = shm_lock(id)))
 152                 BUG();
 153         shp->shm_lprid = current->tgid;
 154         shp->shm_dtim = get_seconds();
 155         shp->shm_nattch--;
 156         if(shp->shm_nattch == 0 &&
 157            shp->shm_flags & SHM_DEST)
 158                 shm_destroy (shp);
 159         else
 160                 shm_unlock(shp);
 161         up (&shm_ids.sem);
 162 }
 163
 164 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 165 {
 166         file_accessed(file);
 167         vma->vm_ops = &shm_vm_ops;
 168         shm_inc(file->f_dentry->d_inode->i_ino);
 169         return 0;
 170 }
 171
 172 static struct file_operations shm_file_operations = {
 173         .mmap   = shm_mmap
 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 #ifdef CONFIG_NUMA
 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         if (!vx_ipcshm_avail(current->vx_info, numpages))
 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.xid = vx_current_xid();
 209         shp->shm_flags = (shmflg & S_IRWXUGO);
 210         shp->mlock_user = NULL;
 211
 212         shp->shm_perm.security = NULL;
 213         error = security_shm_alloc(shp);
 214         if (error) {
 215                 ipc_rcu_putref(shp);
 216                 return error;
 217         }
 218
 219         if (shmflg & SHM_HUGETLB) {
 220                 /* hugetlb_zero_setup takes care of mlock user accounting */
 221                 file = hugetlb_zero_setup(size);
 222                 shp->mlock_user = current->user;
 223         } else {
 224                 sprintf (name, "SYSV%08x", key);
 225                 file = shmem_file_setup(name, size, VM_ACCOUNT);
 226         }
 227         error = PTR_ERR(file);
 228         if (IS_ERR(file))
 229                 goto no_file;
 230
 231         error = -ENOSPC;
 232         id = shm_addid(shp);
 233         if(id == -1)
 234                 goto no_id;
 235
 236         shp->shm_cprid = current->tgid;
 237         shp->shm_lprid = 0;
 238         shp->shm_atim = shp->shm_dtim = 0;
 239         shp->shm_ctim = get_seconds();
 240         shp->shm_segsz = size;
 241         shp->shm_nattch = 0;
 242         shp->id = shm_buildid(id,shp->shm_perm.seq);
 243         shp->shm_file = file;
 244         file->f_dentry->d_inode->i_ino = shp->id;
 245         if (shmflg & SHM_HUGETLB)
 246                 set_file_hugepages(file);
 247         else
 248                 file->f_op = &shm_file_operations;
 249         shm_tot += numpages;
 250         vx_ipcshm_add(current->vx_info, key, numpages);
 251         shm_unlock(shp);
 252         return shp->id;
 253
 254 no_id:
 255         fput(file);
 256 no_file:
 257         security_shm_free(shp);
 258         ipc_rcu_putref(shp);
 259         return error;
 260 }
 261
 262 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 263 {
 264         struct shmid_kernel *shp;
 265         int err, id = 0;
 266
 267         down(&shm_ids.sem);
 268         if (key == IPC_PRIVATE) {
 269                 err = newseg(key, shmflg, size);
 270         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 271                 if (!(shmflg & IPC_CREAT))
 272                         err = -ENOENT;
 273                 else
 274                         err = newseg(key, shmflg, size);
 275         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 276                 err = -EEXIST;
 277         } else {
 278                 shp = shm_lock(id);
 279                 if(shp==NULL)
 280                         BUG();
 281                 if (shp->shm_segsz < size)
 282                         err = -EINVAL;
 283                 else if (ipcperms(&shp->shm_perm, shmflg))
 284                         err = -EACCES;
 285                 else {
 286                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 287                         err = security_shm_associate(shp, shmflg);
 288                         if (!err)
 289                                 err = shmid;
 290                 }
 291                 shm_unlock(shp);
 292         }
 293         up(&shm_ids.sem);
 294
 295         return err;
 296 }
 297
 298 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 299 {
 300         switch(version) {
 301         case IPC_64:
 302                 return copy_to_user(buf, in, sizeof(*in));
 303         case IPC_OLD:
 304             {
 305                 struct shmid_ds out;
 306
 307                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 308                 out.shm_segsz   = in->shm_segsz;
 309                 out.shm_atime   = in->shm_atime;
 310                 out.shm_dtime   = in->shm_dtime;
 311                 out.shm_ctime   = in->shm_ctime;
 312                 out.shm_cpid    = in->shm_cpid;
 313                 out.shm_lpid    = in->shm_lpid;
 314                 out.shm_nattch  = in->shm_nattch;
 315
 316                 return copy_to_user(buf, &out, sizeof(out));
 317             }
 318         default:
 319                 return -EINVAL;
 320         }
 321 }
 322
 323 struct shm_setbuf {
 324         uid_t   uid;
 325         gid_t   gid;
 326         mode_t  mode;
 327 };
 328
 329 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 330 {
 331         switch(version) {
 332         case IPC_64:
 333             {
 334                 struct shmid64_ds tbuf;
 335
 336                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 337                         return -EFAULT;
 338
 339                 out->uid        = tbuf.shm_perm.uid;
 340                 out->gid        = tbuf.shm_perm.gid;
 341                 out->mode       = tbuf.shm_flags;
 342
 343                 return 0;
 344             }
 345         case IPC_OLD:
 346             {
 347                 struct shmid_ds tbuf_old;
 348
 349                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 350                         return -EFAULT;
 351
 352                 out->uid        = tbuf_old.shm_perm.uid;
 353                 out->gid        = tbuf_old.shm_perm.gid;
 354                 out->mode       = tbuf_old.shm_flags;
 355
 356                 return 0;
 357             }
 358         default:
 359                 return -EINVAL;
 360         }
 361 }
 362
 363 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 364 {
 365         switch(version) {
 366         case IPC_64:
 367                 return copy_to_user(buf, in, sizeof(*in));
 368         case IPC_OLD:
 369             {
 370                 struct shminfo out;
 371
 372                 if(in->shmmax > INT_MAX)
 373                         out.shmmax = INT_MAX;
 374                 else
 375                         out.shmmax = (int)in->shmmax;
 376
 377                 out.shmmin      = in->shmmin;
 378                 out.shmmni      = in->shmmni;
 379                 out.shmseg      = in->shmseg;
 380                 out.shmall      = in->shmall;
 381
 382                 return copy_to_user(buf, &out, sizeof(out));
 383             }
 384         default:
 385                 return -EINVAL;
 386         }
 387 }
 388
 389 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 390 {
 391         int i;
 392
 393         *rss = 0;
 394         *swp = 0;
 395
 396         for (i = 0; i <= shm_ids.max_id; i++) {
 397                 struct shmid_kernel *shp;
 398                 struct inode *inode;
 399
 400                 shp = shm_get(i);
 401                 if(!shp)
 402                         continue;
 403
 404                 inode = shp->shm_file->f_dentry->d_inode;
 405
 406                 if (is_file_hugepages(shp->shm_file)) {
 407                         struct address_space *mapping = inode->i_mapping;
 408                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 409                 } else {
 410                         struct shmem_inode_info *info = SHMEM_I(inode);
 411                         spin_lock(&info->lock);
 412                         *rss += inode->i_mapping->nrpages;
 413                         *swp += info->swapped;
 414                         spin_unlock(&info->lock);
 415                 }
 416         }
 417 }
 418
 419 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 420 {
 421         struct shm_setbuf setbuf;
 422         struct shmid_kernel *shp;
 423         int err, version;
 424
 425         if (cmd < 0 || shmid < 0) {
 426                 err = -EINVAL;
 427                 goto out;
 428         }
 429
 430         version = ipc_parse_version(&cmd);
 431
 432         switch (cmd) { /* replace with proc interface ? */
 433         case IPC_INFO:
 434         {
 435                 struct shminfo64 shminfo;
 436
 437                 err = security_shm_shmctl(NULL, cmd);
 438                 if (err)
 439                         return err;
 440
 441                 memset(&shminfo,0,sizeof(shminfo));
 442                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 443                 shminfo.shmmax = shm_ctlmax;
 444                 shminfo.shmall = shm_ctlall;
 445
 446                 shminfo.shmmin = SHMMIN;
 447                 if(copy_shminfo_to_user (buf, &shminfo, version))
 448                         return -EFAULT;
 449                 /* reading a integer is always atomic */
 450                 err= shm_ids.max_id;
 451                 if(err<0)
 452                         err = 0;
 453                 goto out;
 454         }
 455         case SHM_INFO:
 456         {
 457                 struct shm_info shm_info;
 458
 459                 err = security_shm_shmctl(NULL, cmd);
 460                 if (err)
 461                         return err;
 462
 463                 memset(&shm_info,0,sizeof(shm_info));
 464                 down(&shm_ids.sem);
 465                 shm_info.used_ids = shm_ids.in_use;
 466                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 467                 shm_info.shm_tot = shm_tot;
 468                 shm_info.swap_attempts = 0;
 469                 shm_info.swap_successes = 0;
 470                 err = shm_ids.max_id;
 471                 up(&shm_ids.sem);
 472                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 473                         err = -EFAULT;
 474                         goto out;
 475                 }
 476
 477                 err = err < 0 ? 0 : err;
 478                 goto out;
 479         }
 480         case SHM_STAT:
 481         case IPC_STAT:
 482         {
 483                 struct shmid64_ds tbuf;
 484                 int result;
 485                 memset(&tbuf, 0, sizeof(tbuf));
 486                 shp = shm_lock(shmid);
 487                 if(shp==NULL) {
 488                         err = -EINVAL;
 489                         goto out;
 490                 } else if(cmd==SHM_STAT) {
 491                         err = -EINVAL;
 492                         if (shmid > shm_ids.max_id)
 493                                 goto out_unlock;
 494                         result = shm_buildid(shmid, shp->shm_perm.seq);
 495                 } else {
 496                         err = shm_checkid(shp,shmid);
 497                         if(err)
 498                                 goto out_unlock;
 499                         result = 0;
 500                 }
 501                 err=-EACCES;
 502                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 503                         goto out_unlock;
 504                 err = security_shm_shmctl(shp, cmd);
 505                 if (err)
 506                         goto out_unlock;
 507                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 508                 tbuf.shm_segsz  = shp->shm_segsz;
 509                 tbuf.shm_atime  = shp->shm_atim;
 510                 tbuf.shm_dtime  = shp->shm_dtim;
 511                 tbuf.shm_ctime  = shp->shm_ctim;
 512                 tbuf.shm_cpid   = shp->shm_cprid;
 513                 tbuf.shm_lpid   = shp->shm_lprid;
 514                 if (!is_file_hugepages(shp->shm_file))
 515                         tbuf.shm_nattch = shp->shm_nattch;
 516                 else
 517                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 518                 shm_unlock(shp);
 519                 if(copy_shmid_to_user (buf, &tbuf, version))
 520                         err = -EFAULT;
 521                 else
 522                         err = result;
 523                 goto out;
 524         }
 525         case SHM_LOCK:
 526         case SHM_UNLOCK:
 527         {
 528                 shp = shm_lock(shmid);
 529                 if(shp==NULL) {
 530                         err = -EINVAL;
 531                         goto out;
 532                 }
 533                 err = shm_checkid(shp,shmid);
 534                 if(err)
 535                         goto out_unlock;
 536
 537                 if (!capable(CAP_IPC_LOCK)) {
 538                         err = -EPERM;
 539                         if (current->euid != shp->shm_perm.uid &&
 540                             current->euid != shp->shm_perm.cuid)
 541                                 goto out_unlock;
 542                         if (cmd == SHM_LOCK &&
 543                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 544                                 goto out_unlock;
 545                 }
 546
 547                 err = security_shm_shmctl(shp, cmd);
 548                 if (err)
 549                         goto out_unlock;
 550
 551                 if(cmd==SHM_LOCK) {
 552                         struct user_struct * user = current->user;
 553                         if (!is_file_hugepages(shp->shm_file)) {
 554                                 err = shmem_lock(shp->shm_file, 1, user);
 555                                 if (!err) {
 556                                         shp->shm_flags |= SHM_LOCKED;
 557                                         shp->mlock_user = user;
 558                                 }
 559                         }
 560                 } else if (!is_file_hugepages(shp->shm_file)) {
 561                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 562                         shp->shm_flags &= ~SHM_LOCKED;
 563                         shp->mlock_user = NULL;
 564                 }
 565                 shm_unlock(shp);
 566                 goto out;
 567         }
 568         case IPC_RMID:
 569         {
 570                 /*
 571                  *      We cannot simply remove the file. The SVID states
 572                  *      that the block remains until the last person
 573                  *      detaches from it, then is deleted. A shmat() on
 574                  *      an RMID segment is legal in older Linux and if
 575                  *      we change it apps break...
 576                  *
 577                  *      Instead we set a destroyed flag, and then blow
 578                  *      the name away when the usage hits zero.
 579                  */
 580                 down(&shm_ids.sem);
 581                 shp = shm_lock(shmid);
 582                 err = -EINVAL;
 583                 if (shp == NULL)
 584                         goto out_up;
 585                 err = shm_checkid(shp, shmid);
 586                 if(err)
 587                         goto out_unlock_up;
 588
 589                 if (current->euid != shp->shm_perm.uid &&
 590                     current->euid != shp->shm_perm.cuid &&
 591                     !capable(CAP_SYS_ADMIN)) {
 592                         err=-EPERM;
 593                         goto out_unlock_up;
 594                 }
 595
 596                 err = security_shm_shmctl(shp, cmd);
 597                 if (err)
 598                         goto out_unlock_up;
 599
 600                 if (shp->shm_nattch){
 601                         shp->shm_flags |= SHM_DEST;
 602                         /* Do not find it any more */
 603                         shp->shm_perm.key = IPC_PRIVATE;
 604                         shm_unlock(shp);
 605                 } else
 606                         shm_destroy (shp);
 607                 up(&shm_ids.sem);
 608                 goto out;
 609         }
 610
 611         case IPC_SET:
 612         {
 613                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 614                         err = -EFAULT;
 615                         goto out;
 616                 }
 617                 if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode)))
 618                         return err;
 619                 down(&shm_ids.sem);
 620                 shp = shm_lock(shmid);
 621                 err=-EINVAL;
 622                 if(shp==NULL)
 623                         goto out_up;
 624                 err = shm_checkid(shp,shmid);
 625                 if(err)
 626                         goto out_unlock_up;
 627                 err=-EPERM;
 628                 if (current->euid != shp->shm_perm.uid &&
 629                     current->euid != shp->shm_perm.cuid &&
 630                     !capable(CAP_SYS_ADMIN)) {
 631                         goto out_unlock_up;
 632                 }
 633
 634                 err = security_shm_shmctl(shp, cmd);
 635                 if (err)
 636                         goto out_unlock_up;
 637
 638                 shp->shm_perm.uid = setbuf.uid;
 639                 shp->shm_perm.gid = setbuf.gid;
 640                 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
 641                         | (setbuf.mode & S_IRWXUGO);
 642                 shp->shm_ctim = get_seconds();
 643                 break;
 644         }
 645
 646         default:
 647                 err = -EINVAL;
 648                 goto out;
 649         }
 650
 651         err = 0;
 652 out_unlock_up:
 653         shm_unlock(shp);
 654 out_up:
 655         up(&shm_ids.sem);
 656         goto out;
 657 out_unlock:
 658         shm_unlock(shp);
 659 out:
 660         return err;
 661 }
 662
 663 /*
 664  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 665  *
 666  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 667  * "raddr" thing points to kernel space, and there has to be a wrapper around
 668  * this.
 669  */
 670 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 671 {
 672         struct shmid_kernel *shp;
 673         unsigned long addr;
 674         unsigned long size;
 675         struct file * file;
 676         int    err;
 677         unsigned long flags;
 678         unsigned long prot;
 679         unsigned long o_flags;
 680         int acc_mode;
 681         void *user_addr;
 682
 683         if (shmid < 0) {
 684                 err = -EINVAL;
 685                 goto out;
 686         } else if ((addr = (ulong)shmaddr)) {
 687                 if (addr & (SHMLBA-1)) {
 688                         if (shmflg & SHM_RND)
 689                                 addr &= ~(SHMLBA-1);       /* round down */
 690                         else
 691 #ifndef __ARCH_FORCE_SHMLBA
 692                                 if (addr & ~PAGE_MASK)
 693 #endif
 694                                         return -EINVAL;
 695                 }
 696                 flags = MAP_SHARED | MAP_FIXED;
 697         } else {
 698                 if ((shmflg & SHM_REMAP))
 699                         return -EINVAL;
 700
 701                 flags = MAP_SHARED;
 702         }
 703
 704         if (shmflg & SHM_RDONLY) {
 705                 prot = PROT_READ;
 706                 o_flags = O_RDONLY;
 707                 acc_mode = S_IRUGO;
 708         } else {
 709                 prot = PROT_READ | PROT_WRITE;
 710                 o_flags = O_RDWR;
 711                 acc_mode = S_IRUGO | S_IWUGO;
 712         }
 713         if (shmflg & SHM_EXEC) {
 714                 prot |= PROT_EXEC;
 715                 acc_mode |= S_IXUGO;
 716         }
 717
 718         /*
 719          * We cannot rely on the fs check since SYSV IPC does have an
 720          * additional creator id...
 721          */
 722         shp = shm_lock(shmid);
 723         if(shp == NULL) {
 724                 err = -EINVAL;
 725                 goto out;
 726         }
 727         err = shm_checkid(shp,shmid);
 728         if (err) {
 729                 shm_unlock(shp);
 730                 goto out;
 731         }
 732         if (ipcperms(&shp->shm_perm, acc_mode)) {
 733                 shm_unlock(shp);
 734                 err = -EACCES;
 735                 goto out;
 736         }
 737
 738         err = security_shm_shmat(shp, shmaddr, shmflg);
 739         if (err) {
 740                 shm_unlock(shp);
 741                 return err;
 742         }
 743
 744         file = shp->shm_file;
 745         size = i_size_read(file->f_dentry->d_inode);
 746         shp->shm_nattch++;
 747         shm_unlock(shp);
 748
 749         down_write(&current->mm->mmap_sem);
 750         if (addr && !(shmflg & SHM_REMAP)) {
 751                 user_addr = ERR_PTR(-EINVAL);
 752                 if (find_vma_intersection(current->mm, addr, addr + size))
 753                         goto invalid;
 754                 /*
 755                  * If shm segment goes below stack, make sure there is some
 756                  * space left for the stack to grow (at least 4 pages).
 757                  */
 758                 if (addr < current->mm->start_stack &&
 759                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 760                         goto invalid;
 761         }
 762
 763         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 764
 765 invalid:
 766         up_write(&current->mm->mmap_sem);
 767
 768         down (&shm_ids.sem);
 769         if(!(shp = shm_lock(shmid)))
 770                 BUG();
 771         shp->shm_nattch--;
 772         if(shp->shm_nattch == 0 &&
 773            shp->shm_flags & SHM_DEST)
 774                 shm_destroy (shp);
 775         else
 776                 shm_unlock(shp);
 777         up (&shm_ids.sem);
 778
 779         *raddr = (unsigned long) user_addr;
 780         err = 0;
 781         if (IS_ERR(user_addr))
 782                 err = PTR_ERR(user_addr);
 783 out:
 784         return err;
 785 }
 786
 787 asmlinkage long sys_shmat(int shmid, char __user *shmaddr, int shmflg)
 788 {
 789         unsigned long ret;
 790         long err;
 791
 792         err = do_shmat(shmid, shmaddr, shmflg, &ret);
 793         if (err)
 794                 return err;
 795         force_successful_syscall_return();
 796         return (long)ret;
 797 }
 798
 799 /*
 800  * detach and kill segment if marked destroyed.
 801  * The work is done in shm_close.
 802  */
 803 asmlinkage long sys_shmdt(char __user *shmaddr)
 804 {
 805         struct mm_struct *mm = current->mm;
 806         struct vm_area_struct *vma, *next;
 807         unsigned long addr = (unsigned long)shmaddr;
 808         loff_t size = 0;
 809         int retval = -EINVAL;
 810
 811         down_write(&mm->mmap_sem);
 812
 813         /*
 814          * This function tries to be smart and unmap shm segments that
 815          * were modified by partial mlock or munmap calls:
 816          * - It first determines the size of the shm segment that should be
 817          *   unmapped: It searches for a vma that is backed by shm and that
 818          *   started at address shmaddr. It records it's size and then unmaps
 819          *   it.
 820          * - Then it unmaps all shm vmas that started at shmaddr and that
 821          *   are within the initially determined size.
 822          * Errors from do_munmap are ignored: the function only fails if
 823          * it's called with invalid parameters or if it's called to unmap
 824          * a part of a vma. Both calls in this function are for full vmas,
 825          * the parameters are directly copied from the vma itself and always
 826          * valid - therefore do_munmap cannot fail. (famous last words?)
 827          */
 828         /*
 829          * If it had been mremap()'d, the starting address would not
 830          * match the usual checks anyway. So assume all vma's are
 831          * above the starting address given.
 832          */
 833         vma = find_vma(mm, addr);
 834
 835         while (vma) {
 836                 next = vma->vm_next;
 837
 838                 /*
 839                  * Check if the starting address would match, i.e. it's
 840                  * a fragment created by mprotect() and/or munmap(), or it
 841                  * otherwise it starts at this address with no hassles.
 842                  */
 843                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 844                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 845
 846
 847                         size = vma->vm_file->f_dentry->d_inode->i_size;
 848                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 849                         /*
 850                          * We discovered the size of the shm segment, so
 851                          * break out of here and fall through to the next
 852                          * loop that uses the size information to stop
 853                          * searching for matching vma's.
 854                          */
 855                         retval = 0;
 856                         vma = next;
 857                         break;
 858                 }
 859                 vma = next;
 860         }
 861
 862         /*
 863          * We need look no further than the maximum address a fragment
 864          * could possibly have landed at. Also cast things to loff_t to
 865          * prevent overflows and make comparisions vs. equal-width types.
 866          */
 867         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 868                 next = vma->vm_next;
 869
 870                 /* finding a matching vma now does not alter retval */
 871                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 872                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 873
 874                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 875                 vma = next;
 876         }
 877
 878         up_write(&mm->mmap_sem);
 879         return retval;
 880 }
 881
 882 #ifdef CONFIG_PROC_FS
 883 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
 884 {
 885         off_t pos = 0;
 886         off_t begin = 0;
 887         int i, len = 0;
 888
 889         down(&shm_ids.sem);
 890         len += sprintf(buffer, "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n");
 891
 892         for(i = 0; i <= shm_ids.max_id; i++) {
 893                 struct shmid_kernel* shp;
 894
 895                 shp = shm_lock(i);
 896                 if (shp) {
 897 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 898 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 899                         char *format;
 900
 901                         if (!vx_check(shp->shm_perm.xid, VX_IDENT)) {
 902                                 shm_unlock(shp);
 903                                 continue;
 904                         }
 905                         if (sizeof(size_t) <= sizeof(int))
 906                                 format = SMALL_STRING;
 907                         else
 908                                 format = BIG_STRING;
 909                         len += sprintf(buffer + len, format,
 910                                 shp->shm_perm.key,
 911                                 shm_buildid(i, shp->shm_perm.seq),
 912                                 shp->shm_flags,
 913                                 shp->shm_segsz,
 914                                 shp->shm_cprid,
 915                                 shp->shm_lprid,
 916                                 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 917                                 shp->shm_perm.uid,
 918                                 shp->shm_perm.gid,
 919                                 shp->shm_perm.cuid,
 920                                 shp->shm_perm.cgid,
 921                                 shp->shm_atim,
 922                                 shp->shm_dtim,
 923                                 shp->shm_ctim);
 924                         shm_unlock(shp);
 925
 926                         pos += len;
 927                         if(pos < offset) {
 928                                 len = 0;
 929                                 begin = pos;
 930                         }
 931                         if(pos > offset + length)
 932                                 goto done;
 933                 }
 934         }
 935         *eof = 1;
 936 done:
 937         up(&shm_ids.sem);
 938         *start = buffer + (offset - begin);
 939         len -= (offset - begin);
 940         if(len > length)
 941                 len = length;
 942         if(len < 0)
 943                 len = 0;
 944         return len;
 945 }
 946 #endif