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/vs_context.h>
  31 #include <linux/vs_limit.h>
  32 #include <asm/uaccess.h>
  33
  34 #include "util.h"
  35
  36 #define shm_flags       shm_perm.mode
  37
  38 static struct file_operations shm_file_operations;
  39 static struct vm_operations_struct shm_vm_ops;
  40
  41 static struct ipc_ids shm_ids;
  42
  43 #define shm_lock(id)    ((struct shmid_kernel*)ipc_lock(&shm_ids,id))
  44 #define shm_unlock(shp) ipc_unlock(&(shp)->shm_perm)
  45 #define shm_get(id)     ((struct shmid_kernel*)ipc_get(&shm_ids,id))
  46 #define shm_buildid(id, seq) \
  47         ipc_buildid(&shm_ids, id, seq)
  48
  49 static int newseg (key_t key, int shmflg, size_t size);
  50 static void shm_open (struct vm_area_struct *shmd);
  51 static void shm_close (struct vm_area_struct *shmd);
  52 #ifdef CONFIG_PROC_FS
  53 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
  54 #endif
  55
  56 size_t  shm_ctlmax = SHMMAX;
  57 size_t  shm_ctlall = SHMALL;
  58 int     shm_ctlmni = SHMMNI;
  59
  60 static int shm_tot; /* total number of shared memory pages */
  61
  62 void __init shm_init (void)
  63 {
  64         ipc_init_ids(&shm_ids, 1);
  65 #ifdef CONFIG_PROC_FS
  66         create_proc_read_entry("sysvipc/shm", 0, NULL, sysvipc_shm_read_proc, NULL);
  67 #endif
  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         struct vx_info *vxi = locate_vx_info(shp->shm_perm.xid);
 117         int numpages = (shp->shm_segsz + PAGE_SIZE - 1) >> PAGE_SHIFT;
 118
 119         vx_ipcshm_sub(vxi, shp, numpages);
 120         shm_tot -= numpages;
 121
 122         shm_rmid (shp->id);
 123         shm_unlock(shp);
 124         if (!is_file_hugepages(shp->shm_file))
 125                 shmem_lock(shp->shm_file, 0, shp->mlock_user);
 126         else
 127                 user_shm_unlock(shp->shm_file->f_dentry->d_inode->i_size,
 128                                                 shp->mlock_user);
 129         fput (shp->shm_file);
 130         security_shm_free(shp);
 131         put_vx_info(vxi);
 132         ipc_rcu_putref(shp);
 133 }
 134
 135 /*
 136  * remove the attach descriptor shmd.
 137  * free memory for segment if it is marked destroyed.
 138  * The descriptor has already been removed from the current->mm->mmap list
 139  * and will later be kfree()d.
 140  */
 141 static void shm_close (struct vm_area_struct *shmd)
 142 {
 143         struct file * file = shmd->vm_file;
 144         int id = file->f_dentry->d_inode->i_ino;
 145         struct shmid_kernel *shp;
 146
 147         down (&shm_ids.sem);
 148         /* remove from the list of attaches of the shm segment */
 149         if(!(shp = shm_lock(id)))
 150                 BUG();
 151         shp->shm_lprid = current->tgid;
 152         shp->shm_dtim = get_seconds();
 153         shp->shm_nattch--;
 154         if(shp->shm_nattch == 0 &&
 155            shp->shm_flags & SHM_DEST)
 156                 shm_destroy (shp);
 157         else
 158                 shm_unlock(shp);
 159         up (&shm_ids.sem);
 160 }
 161
 162 static int shm_mmap(struct file * file, struct vm_area_struct * vma)
 163 {
 164         file_accessed(file);
 165         vma->vm_ops = &shm_vm_ops;
 166         shm_inc(file->f_dentry->d_inode->i_ino);
 167         return 0;
 168 }
 169
 170 static struct file_operations shm_file_operations = {
 171         .mmap   = shm_mmap
 172 };
 173
 174 static struct vm_operations_struct shm_vm_ops = {
 175         .open   = shm_open,     /* callback for a new vm-area open */
 176         .close  = shm_close,    /* callback for when the vm-area is released */
 177         .nopage = shmem_nopage,
 178 #ifdef CONFIG_NUMA
 179         .set_policy = shmem_set_policy,
 180         .get_policy = shmem_get_policy,
 181 #endif
 182 };
 183
 184 static int newseg (key_t key, int shmflg, size_t size)
 185 {
 186         int error;
 187         struct shmid_kernel *shp;
 188         int numpages = (size + PAGE_SIZE -1) >> PAGE_SHIFT;
 189         struct file * file;
 190         char name[13];
 191         int id;
 192
 193         if (size < SHMMIN || size > shm_ctlmax)
 194                 return -EINVAL;
 195
 196         if (shm_tot + numpages >= shm_ctlall)
 197                 return -ENOSPC;
 198         if (!vx_ipcshm_avail(current->vx_info, numpages))
 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.xid = vx_current_xid();
 207         shp->shm_flags = (shmflg & S_IRWXUGO);
 208         shp->mlock_user = NULL;
 209
 210         shp->shm_perm.security = NULL;
 211         error = security_shm_alloc(shp);
 212         if (error) {
 213                 ipc_rcu_putref(shp);
 214                 return error;
 215         }
 216
 217         if (shmflg & SHM_HUGETLB) {
 218                 /* hugetlb_zero_setup takes care of mlock user accounting */
 219                 file = hugetlb_zero_setup(size);
 220                 shp->mlock_user = current->user;
 221         } else {
 222                 sprintf (name, "SYSV%08x", key);
 223                 file = shmem_file_setup(name, size, VM_ACCOUNT);
 224         }
 225         error = PTR_ERR(file);
 226         if (IS_ERR(file))
 227                 goto no_file;
 228
 229         error = -ENOSPC;
 230         id = shm_addid(shp);
 231         if(id == -1)
 232                 goto no_id;
 233
 234         shp->shm_cprid = current->tgid;
 235         shp->shm_lprid = 0;
 236         shp->shm_atim = shp->shm_dtim = 0;
 237         shp->shm_ctim = get_seconds();
 238         shp->shm_segsz = size;
 239         shp->shm_nattch = 0;
 240         shp->id = shm_buildid(id,shp->shm_perm.seq);
 241         shp->shm_file = file;
 242         file->f_dentry->d_inode->i_ino = shp->id;
 243         if (shmflg & SHM_HUGETLB)
 244                 set_file_hugepages(file);
 245         else
 246                 file->f_op = &shm_file_operations;
 247         shm_tot += numpages;
 248         vx_ipcshm_add(current->vx_info, key, numpages);
 249         shm_unlock(shp);
 250         return shp->id;
 251
 252 no_id:
 253         fput(file);
 254 no_file:
 255         security_shm_free(shp);
 256         ipc_rcu_putref(shp);
 257         return error;
 258 }
 259
 260 asmlinkage long sys_shmget (key_t key, size_t size, int shmflg)
 261 {
 262         struct shmid_kernel *shp;
 263         int err, id = 0;
 264
 265         down(&shm_ids.sem);
 266         if (key == IPC_PRIVATE) {
 267                 err = newseg(key, shmflg, size);
 268         } else if ((id = ipc_findkey(&shm_ids, key)) == -1) {
 269                 if (!(shmflg & IPC_CREAT))
 270                         err = -ENOENT;
 271                 else
 272                         err = newseg(key, shmflg, size);
 273         } else if ((shmflg & IPC_CREAT) && (shmflg & IPC_EXCL)) {
 274                 err = -EEXIST;
 275         } else {
 276                 shp = shm_lock(id);
 277                 if(shp==NULL)
 278                         BUG();
 279                 if (shp->shm_segsz < size)
 280                         err = -EINVAL;
 281                 else if (ipcperms(&shp->shm_perm, shmflg))
 282                         err = -EACCES;
 283                 else {
 284                         int shmid = shm_buildid(id, shp->shm_perm.seq);
 285                         err = security_shm_associate(shp, shmflg);
 286                         if (!err)
 287                                 err = shmid;
 288                 }
 289                 shm_unlock(shp);
 290         }
 291         up(&shm_ids.sem);
 292
 293         return err;
 294 }
 295
 296 static inline unsigned long copy_shmid_to_user(void __user *buf, struct shmid64_ds *in, int version)
 297 {
 298         switch(version) {
 299         case IPC_64:
 300                 return copy_to_user(buf, in, sizeof(*in));
 301         case IPC_OLD:
 302             {
 303                 struct shmid_ds out;
 304
 305                 ipc64_perm_to_ipc_perm(&in->shm_perm, &out.shm_perm);
 306                 out.shm_segsz   = in->shm_segsz;
 307                 out.shm_atime   = in->shm_atime;
 308                 out.shm_dtime   = in->shm_dtime;
 309                 out.shm_ctime   = in->shm_ctime;
 310                 out.shm_cpid    = in->shm_cpid;
 311                 out.shm_lpid    = in->shm_lpid;
 312                 out.shm_nattch  = in->shm_nattch;
 313
 314                 return copy_to_user(buf, &out, sizeof(out));
 315             }
 316         default:
 317                 return -EINVAL;
 318         }
 319 }
 320
 321 struct shm_setbuf {
 322         uid_t   uid;
 323         gid_t   gid;
 324         mode_t  mode;
 325 };
 326
 327 static inline unsigned long copy_shmid_from_user(struct shm_setbuf *out, void __user *buf, int version)
 328 {
 329         switch(version) {
 330         case IPC_64:
 331             {
 332                 struct shmid64_ds tbuf;
 333
 334                 if (copy_from_user(&tbuf, buf, sizeof(tbuf)))
 335                         return -EFAULT;
 336
 337                 out->uid        = tbuf.shm_perm.uid;
 338                 out->gid        = tbuf.shm_perm.gid;
 339                 out->mode       = tbuf.shm_flags;
 340
 341                 return 0;
 342             }
 343         case IPC_OLD:
 344             {
 345                 struct shmid_ds tbuf_old;
 346
 347                 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
 348                         return -EFAULT;
 349
 350                 out->uid        = tbuf_old.shm_perm.uid;
 351                 out->gid        = tbuf_old.shm_perm.gid;
 352                 out->mode       = tbuf_old.shm_flags;
 353
 354                 return 0;
 355             }
 356         default:
 357                 return -EINVAL;
 358         }
 359 }
 360
 361 static inline unsigned long copy_shminfo_to_user(void __user *buf, struct shminfo64 *in, int version)
 362 {
 363         switch(version) {
 364         case IPC_64:
 365                 return copy_to_user(buf, in, sizeof(*in));
 366         case IPC_OLD:
 367             {
 368                 struct shminfo out;
 369
 370                 if(in->shmmax > INT_MAX)
 371                         out.shmmax = INT_MAX;
 372                 else
 373                         out.shmmax = (int)in->shmmax;
 374
 375                 out.shmmin      = in->shmmin;
 376                 out.shmmni      = in->shmmni;
 377                 out.shmseg      = in->shmseg;
 378                 out.shmall      = in->shmall;
 379
 380                 return copy_to_user(buf, &out, sizeof(out));
 381             }
 382         default:
 383                 return -EINVAL;
 384         }
 385 }
 386
 387 static void shm_get_stat(unsigned long *rss, unsigned long *swp)
 388 {
 389         int i;
 390
 391         *rss = 0;
 392         *swp = 0;
 393
 394         for (i = 0; i <= shm_ids.max_id; i++) {
 395                 struct shmid_kernel *shp;
 396                 struct inode *inode;
 397
 398                 shp = shm_get(i);
 399                 if(!shp)
 400                         continue;
 401
 402                 inode = shp->shm_file->f_dentry->d_inode;
 403
 404                 if (is_file_hugepages(shp->shm_file)) {
 405                         struct address_space *mapping = inode->i_mapping;
 406                         *rss += (HPAGE_SIZE/PAGE_SIZE)*mapping->nrpages;
 407                 } else {
 408                         struct shmem_inode_info *info = SHMEM_I(inode);
 409                         spin_lock(&info->lock);
 410                         *rss += inode->i_mapping->nrpages;
 411                         *swp += info->swapped;
 412                         spin_unlock(&info->lock);
 413                 }
 414         }
 415 }
 416
 417 asmlinkage long sys_shmctl (int shmid, int cmd, struct shmid_ds __user *buf)
 418 {
 419         struct shm_setbuf setbuf;
 420         struct shmid_kernel *shp;
 421         int err, version;
 422
 423         if (cmd < 0 || shmid < 0) {
 424                 err = -EINVAL;
 425                 goto out;
 426         }
 427
 428         version = ipc_parse_version(&cmd);
 429
 430         switch (cmd) { /* replace with proc interface ? */
 431         case IPC_INFO:
 432         {
 433                 struct shminfo64 shminfo;
 434
 435                 err = security_shm_shmctl(NULL, cmd);
 436                 if (err)
 437                         return err;
 438
 439                 memset(&shminfo,0,sizeof(shminfo));
 440                 shminfo.shmmni = shminfo.shmseg = shm_ctlmni;
 441                 shminfo.shmmax = shm_ctlmax;
 442                 shminfo.shmall = shm_ctlall;
 443
 444                 shminfo.shmmin = SHMMIN;
 445                 if(copy_shminfo_to_user (buf, &shminfo, version))
 446                         return -EFAULT;
 447                 /* reading a integer is always atomic */
 448                 err= shm_ids.max_id;
 449                 if(err<0)
 450                         err = 0;
 451                 goto out;
 452         }
 453         case SHM_INFO:
 454         {
 455                 struct shm_info shm_info;
 456
 457                 err = security_shm_shmctl(NULL, cmd);
 458                 if (err)
 459                         return err;
 460
 461                 memset(&shm_info,0,sizeof(shm_info));
 462                 down(&shm_ids.sem);
 463                 shm_info.used_ids = shm_ids.in_use;
 464                 shm_get_stat (&shm_info.shm_rss, &shm_info.shm_swp);
 465                 shm_info.shm_tot = shm_tot;
 466                 shm_info.swap_attempts = 0;
 467                 shm_info.swap_successes = 0;
 468                 err = shm_ids.max_id;
 469                 up(&shm_ids.sem);
 470                 if(copy_to_user (buf, &shm_info, sizeof(shm_info))) {
 471                         err = -EFAULT;
 472                         goto out;
 473                 }
 474
 475                 err = err < 0 ? 0 : err;
 476                 goto out;
 477         }
 478         case SHM_STAT:
 479         case IPC_STAT:
 480         {
 481                 struct shmid64_ds tbuf;
 482                 int result;
 483                 memset(&tbuf, 0, sizeof(tbuf));
 484                 shp = shm_lock(shmid);
 485                 if(shp==NULL) {
 486                         err = -EINVAL;
 487                         goto out;
 488                 } else if(cmd==SHM_STAT) {
 489                         err = -EINVAL;
 490                         if (shmid > shm_ids.max_id)
 491                                 goto out_unlock;
 492                         result = shm_buildid(shmid, shp->shm_perm.seq);
 493                 } else {
 494                         err = shm_checkid(shp,shmid);
 495                         if(err)
 496                                 goto out_unlock;
 497                         result = 0;
 498                 }
 499                 err=-EACCES;
 500                 if (ipcperms (&shp->shm_perm, S_IRUGO))
 501                         goto out_unlock;
 502                 err = security_shm_shmctl(shp, cmd);
 503                 if (err)
 504                         goto out_unlock;
 505                 kernel_to_ipc64_perm(&shp->shm_perm, &tbuf.shm_perm);
 506                 tbuf.shm_segsz  = shp->shm_segsz;
 507                 tbuf.shm_atime  = shp->shm_atim;
 508                 tbuf.shm_dtime  = shp->shm_dtim;
 509                 tbuf.shm_ctime  = shp->shm_ctim;
 510                 tbuf.shm_cpid   = shp->shm_cprid;
 511                 tbuf.shm_lpid   = shp->shm_lprid;
 512                 if (!is_file_hugepages(shp->shm_file))
 513                         tbuf.shm_nattch = shp->shm_nattch;
 514                 else
 515                         tbuf.shm_nattch = file_count(shp->shm_file) - 1;
 516                 shm_unlock(shp);
 517                 if(copy_shmid_to_user (buf, &tbuf, version))
 518                         err = -EFAULT;
 519                 else
 520                         err = result;
 521                 goto out;
 522         }
 523         case SHM_LOCK:
 524         case SHM_UNLOCK:
 525         {
 526                 shp = shm_lock(shmid);
 527                 if(shp==NULL) {
 528                         err = -EINVAL;
 529                         goto out;
 530                 }
 531                 err = shm_checkid(shp,shmid);
 532                 if(err)
 533                         goto out_unlock;
 534
 535                 if (!capable(CAP_IPC_LOCK)) {
 536                         err = -EPERM;
 537                         if (current->euid != shp->shm_perm.uid &&
 538                             current->euid != shp->shm_perm.cuid)
 539                                 goto out_unlock;
 540                         if (cmd == SHM_LOCK &&
 541                             !current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur)
 542                                 goto out_unlock;
 543                 }
 544
 545                 err = security_shm_shmctl(shp, cmd);
 546                 if (err)
 547                         goto out_unlock;
 548
 549                 if(cmd==SHM_LOCK) {
 550                         struct user_struct * user = current->user;
 551                         if (!is_file_hugepages(shp->shm_file)) {
 552                                 err = shmem_lock(shp->shm_file, 1, user);
 553                                 if (!err) {
 554                                         shp->shm_flags |= SHM_LOCKED;
 555                                         shp->mlock_user = user;
 556                                 }
 557                         }
 558                 } else if (!is_file_hugepages(shp->shm_file)) {
 559                         shmem_lock(shp->shm_file, 0, shp->mlock_user);
 560                         shp->shm_flags &= ~SHM_LOCKED;
 561                         shp->mlock_user = NULL;
 562                 }
 563                 shm_unlock(shp);
 564                 goto out;
 565         }
 566         case IPC_RMID:
 567         {
 568                 /*
 569                  *      We cannot simply remove the file. The SVID states
 570                  *      that the block remains until the last person
 571                  *      detaches from it, then is deleted. A shmat() on
 572                  *      an RMID segment is legal in older Linux and if
 573                  *      we change it apps break...
 574                  *
 575                  *      Instead we set a destroyed flag, and then blow
 576                  *      the name away when the usage hits zero.
 577                  */
 578                 down(&shm_ids.sem);
 579                 shp = shm_lock(shmid);
 580                 err = -EINVAL;
 581                 if (shp == NULL)
 582                         goto out_up;
 583                 err = shm_checkid(shp, shmid);
 584                 if(err)
 585                         goto out_unlock_up;
 586
 587                 if (current->euid != shp->shm_perm.uid &&
 588                     current->euid != shp->shm_perm.cuid &&
 589                     !capable(CAP_SYS_ADMIN)) {
 590                         err=-EPERM;
 591                         goto out_unlock_up;
 592                 }
 593
 594                 err = security_shm_shmctl(shp, cmd);
 595                 if (err)
 596                         goto out_unlock_up;
 597
 598                 if (shp->shm_nattch){
 599                         shp->shm_flags |= SHM_DEST;
 600                         /* Do not find it any more */
 601                         shp->shm_perm.key = IPC_PRIVATE;
 602                         shm_unlock(shp);
 603                 } else
 604                         shm_destroy (shp);
 605                 up(&shm_ids.sem);
 606                 goto out;
 607         }
 608
 609         case IPC_SET:
 610         {
 611                 if (copy_shmid_from_user (&setbuf, buf, version)) {
 612                         err = -EFAULT;
 613                         goto out;
 614                 }
 615                 down(&shm_ids.sem);
 616                 shp = shm_lock(shmid);
 617                 err=-EINVAL;
 618                 if(shp==NULL)
 619                         goto out_up;
 620                 err = shm_checkid(shp,shmid);
 621                 if(err)
 622                         goto out_unlock_up;
 623                 err=-EPERM;
 624                 if (current->euid != shp->shm_perm.uid &&
 625                     current->euid != shp->shm_perm.cuid &&
 626                     !capable(CAP_SYS_ADMIN)) {
 627                         goto out_unlock_up;
 628                 }
 629
 630                 err = security_shm_shmctl(shp, cmd);
 631                 if (err)
 632                         goto out_unlock_up;
 633
 634                 shp->shm_perm.uid = setbuf.uid;
 635                 shp->shm_perm.gid = setbuf.gid;
 636                 shp->shm_flags = (shp->shm_flags & ~S_IRWXUGO)
 637                         | (setbuf.mode & S_IRWXUGO);
 638                 shp->shm_ctim = get_seconds();
 639                 break;
 640         }
 641
 642         default:
 643                 err = -EINVAL;
 644                 goto out;
 645         }
 646
 647         err = 0;
 648 out_unlock_up:
 649         shm_unlock(shp);
 650 out_up:
 651         up(&shm_ids.sem);
 652         goto out;
 653 out_unlock:
 654         shm_unlock(shp);
 655 out:
 656         return err;
 657 }
 658
 659 /*
 660  * Fix shmaddr, allocate descriptor, map shm, add attach descriptor to lists.
 661  *
 662  * NOTE! Despite the name, this is NOT a direct system call entrypoint. The
 663  * "raddr" thing points to kernel space, and there has to be a wrapper around
 664  * this.
 665  */
 666 long do_shmat(int shmid, char __user *shmaddr, int shmflg, ulong *raddr)
 667 {
 668         struct shmid_kernel *shp;
 669         unsigned long addr;
 670         unsigned long size;
 671         struct file * file;
 672         int    err;
 673         unsigned long flags;
 674         unsigned long prot;
 675         unsigned long o_flags;
 676         int acc_mode;
 677         void *user_addr;
 678
 679         if (shmid < 0) {
 680                 err = -EINVAL;
 681                 goto out;
 682         } else if ((addr = (ulong)shmaddr)) {
 683                 if (addr & (SHMLBA-1)) {
 684                         if (shmflg & SHM_RND)
 685                                 addr &= ~(SHMLBA-1);       /* round down */
 686                         else
 687 #ifndef __ARCH_FORCE_SHMLBA
 688                                 if (addr & ~PAGE_MASK)
 689 #endif
 690                                         return -EINVAL;
 691                 }
 692                 flags = MAP_SHARED | MAP_FIXED;
 693         } else {
 694                 if ((shmflg & SHM_REMAP))
 695                         return -EINVAL;
 696
 697                 flags = MAP_SHARED;
 698         }
 699
 700         if (shmflg & SHM_RDONLY) {
 701                 prot = PROT_READ;
 702                 o_flags = O_RDONLY;
 703                 acc_mode = S_IRUGO;
 704         } else {
 705                 prot = PROT_READ | PROT_WRITE;
 706                 o_flags = O_RDWR;
 707                 acc_mode = S_IRUGO | S_IWUGO;
 708         }
 709         if (shmflg & SHM_EXEC) {
 710                 prot |= PROT_EXEC;
 711                 acc_mode |= S_IXUGO;
 712         }
 713
 714         /*
 715          * We cannot rely on the fs check since SYSV IPC does have an
 716          * additional creator id...
 717          */
 718         shp = shm_lock(shmid);
 719         if(shp == NULL) {
 720                 err = -EINVAL;
 721                 goto out;
 722         }
 723         err = shm_checkid(shp,shmid);
 724         if (err) {
 725                 shm_unlock(shp);
 726                 goto out;
 727         }
 728         if (ipcperms(&shp->shm_perm, acc_mode)) {
 729                 shm_unlock(shp);
 730                 err = -EACCES;
 731                 goto out;
 732         }
 733
 734         err = security_shm_shmat(shp, shmaddr, shmflg);
 735         if (err) {
 736                 shm_unlock(shp);
 737                 return err;
 738         }
 739
 740         file = shp->shm_file;
 741         size = i_size_read(file->f_dentry->d_inode);
 742         shp->shm_nattch++;
 743         shm_unlock(shp);
 744
 745         down_write(&current->mm->mmap_sem);
 746         if (addr && !(shmflg & SHM_REMAP)) {
 747                 user_addr = ERR_PTR(-EINVAL);
 748                 if (find_vma_intersection(current->mm, addr, addr + size))
 749                         goto invalid;
 750                 /*
 751                  * If shm segment goes below stack, make sure there is some
 752                  * space left for the stack to grow (at least 4 pages).
 753                  */
 754                 if (addr < current->mm->start_stack &&
 755                     addr > current->mm->start_stack - size - PAGE_SIZE * 5)
 756                         goto invalid;
 757         }
 758
 759         user_addr = (void*) do_mmap (file, addr, size, prot, flags, 0);
 760
 761 invalid:
 762         up_write(&current->mm->mmap_sem);
 763
 764         down (&shm_ids.sem);
 765         if(!(shp = shm_lock(shmid)))
 766                 BUG();
 767         shp->shm_nattch--;
 768         if(shp->shm_nattch == 0 &&
 769            shp->shm_flags & SHM_DEST)
 770                 shm_destroy (shp);
 771         else
 772                 shm_unlock(shp);
 773         up (&shm_ids.sem);
 774
 775         *raddr = (unsigned long) user_addr;
 776         err = 0;
 777         if (IS_ERR(user_addr))
 778                 err = PTR_ERR(user_addr);
 779 out:
 780         return err;
 781 }
 782
 783 /*
 784  * detach and kill segment if marked destroyed.
 785  * The work is done in shm_close.
 786  */
 787 asmlinkage long sys_shmdt(char __user *shmaddr)
 788 {
 789         struct mm_struct *mm = current->mm;
 790         struct vm_area_struct *vma, *next;
 791         unsigned long addr = (unsigned long)shmaddr;
 792         loff_t size = 0;
 793         int retval = -EINVAL;
 794
 795         down_write(&mm->mmap_sem);
 796
 797         /*
 798          * This function tries to be smart and unmap shm segments that
 799          * were modified by partial mlock or munmap calls:
 800          * - It first determines the size of the shm segment that should be
 801          *   unmapped: It searches for a vma that is backed by shm and that
 802          *   started at address shmaddr. It records it's size and then unmaps
 803          *   it.
 804          * - Then it unmaps all shm vmas that started at shmaddr and that
 805          *   are within the initially determined size.
 806          * Errors from do_munmap are ignored: the function only fails if
 807          * it's called with invalid parameters or if it's called to unmap
 808          * a part of a vma. Both calls in this function are for full vmas,
 809          * the parameters are directly copied from the vma itself and always
 810          * valid - therefore do_munmap cannot fail. (famous last words?)
 811          */
 812         /*
 813          * If it had been mremap()'d, the starting address would not
 814          * match the usual checks anyway. So assume all vma's are
 815          * above the starting address given.
 816          */
 817         vma = find_vma(mm, addr);
 818
 819         while (vma) {
 820                 next = vma->vm_next;
 821
 822                 /*
 823                  * Check if the starting address would match, i.e. it's
 824                  * a fragment created by mprotect() and/or munmap(), or it
 825                  * otherwise it starts at this address with no hassles.
 826                  */
 827                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 828                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff) {
 829
 830
 831                         size = vma->vm_file->f_dentry->d_inode->i_size;
 832                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 833                         /*
 834                          * We discovered the size of the shm segment, so
 835                          * break out of here and fall through to the next
 836                          * loop that uses the size information to stop
 837                          * searching for matching vma's.
 838                          */
 839                         retval = 0;
 840                         vma = next;
 841                         break;
 842                 }
 843                 vma = next;
 844         }
 845
 846         /*
 847          * We need look no further than the maximum address a fragment
 848          * could possibly have landed at. Also cast things to loff_t to
 849          * prevent overflows and make comparisions vs. equal-width types.
 850          */
 851         while (vma && (loff_t)(vma->vm_end - addr) <= size) {
 852                 next = vma->vm_next;
 853
 854                 /* finding a matching vma now does not alter retval */
 855                 if ((vma->vm_ops == &shm_vm_ops || is_vm_hugetlb_page(vma)) &&
 856                         (vma->vm_start - addr)/PAGE_SIZE == vma->vm_pgoff)
 857
 858                         do_munmap(mm, vma->vm_start, vma->vm_end - vma->vm_start);
 859                 vma = next;
 860         }
 861
 862         up_write(&mm->mmap_sem);
 863         return retval;
 864 }
 865
 866 #ifdef CONFIG_PROC_FS
 867 static int sysvipc_shm_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
 868 {
 869         off_t pos = 0;
 870         off_t begin = 0;
 871         int i, len = 0;
 872
 873         down(&shm_ids.sem);
 874         len += sprintf(buffer, "       key      shmid perms       size  cpid  lpid nattch   uid   gid  cuid  cgid      atime      dtime      ctime\n");
 875
 876         for(i = 0; i <= shm_ids.max_id; i++) {
 877                 struct shmid_kernel* shp;
 878
 879                 shp = shm_lock(i);
 880                 if (shp) {
 881 #define SMALL_STRING "%10d %10d  %4o %10u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 882 #define BIG_STRING   "%10d %10d  %4o %21u %5u %5u  %5d %5u %5u %5u %5u %10lu %10lu %10lu\n"
 883                         char *format;
 884
 885                         if (!vx_check(shp->shm_perm.xid, VX_IDENT)) {
 886                                 shm_unlock(shp);
 887                                 continue;
 888                         }
 889                         if (sizeof(size_t) <= sizeof(int))
 890                                 format = SMALL_STRING;
 891                         else
 892                                 format = BIG_STRING;
 893                         len += sprintf(buffer + len, format,
 894                                 shp->shm_perm.key,
 895                                 shm_buildid(i, shp->shm_perm.seq),
 896                                 shp->shm_flags,
 897                                 shp->shm_segsz,
 898                                 shp->shm_cprid,
 899                                 shp->shm_lprid,
 900                                 is_file_hugepages(shp->shm_file) ? (file_count(shp->shm_file) - 1) : shp->shm_nattch,
 901                                 shp->shm_perm.uid,
 902                                 shp->shm_perm.gid,
 903                                 shp->shm_perm.cuid,
 904                                 shp->shm_perm.cgid,
 905                                 shp->shm_atim,
 906                                 shp->shm_dtim,
 907                                 shp->shm_ctim);
 908                         shm_unlock(shp);
 909
 910                         pos += len;
 911                         if(pos < offset) {
 912                                 len = 0;
 913                                 begin = pos;
 914                         }
 915                         if(pos > offset + length)
 916                                 goto done;
 917                 }
 918         }
 919         *eof = 1;
 920 done:
 921         up(&shm_ids.sem);
 922         *start = buffer + (offset - begin);
 923         len -= (offset - begin);
 924         if(len > length)
 925                 len = length;
 926         if(len < 0)
 927                 len = 0;
 928         return len;
 929 }
 930 #endif